Water Resources Systems Analysis

WATER RESOURCES
SYSTEMS ANALYSIS
Mohammad Karamouz
Ferenc Szidarovszky
Banafsheh Zahraie
LEWIS PUBLISHERS
A CRC Press Company
Boca Raton London New York Washington, D.C.
Library of Congress Cataloging-in-Publication Data
Karamouz, Mohammad.
Water resources systems analysis / Mohammad Karamouz, Ferenc Szidarovszky,
Banafsheh Zahraie.
p. cm.
Includes bibliographical references and index.
ISBN 1-56670-642-4 (alk. paper)
1. Water resources development. 2. System analysis. I. Szidarovszky, Ferenc. II.
Zahraie, Banafsheh. III. Title.
TC409.K37 2003
333.91—dc21
2002043472
This book contains information obtained from authentic and highly regarded sources. Reprinted material
is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable
efforts have been made to publish reliable data and information, but the author and the publisher cannot
assume responsibility for the validity of all materials or for the consequences of their use.
Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, microfilming, and recording, or by any information storage or
retrieval system, without prior permission in writing from the publisher.
The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for
creating new works, or for resale. Specific permission must be obtained in writing from CRC Press LLC
for such copying.
Direct all inquiries to CRC Press LLC, 2000 N.W. Corporate Blvd., Boca Raton, Florida 33431.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation, without intent to infringe.
Visit the CRC Press Web site at www.crcpress.com
© 2003 by CRC Press LLC

Lewis Publishers is an imprint of CRC Press LLC
No claim to original U.S. Government works

International Standard Book Number 1-56670-642-4
Library of Congress Card Number 2002043472
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
Preface
In the past decades, the management of water resources has become more focused
on mid- and long-term planning for water conservation and demand management,
diversion and water transfer, and all phases of water resources development projects.
It has become clear that structural solutions are not always the most efficient and
economically sound alternatives. Nonstructural means of water resources manage-
ment go beyond utilization of existing structures and the physical limits of water
resources systems to include technical, social, political, and economic aspects for
better resource management.
In this book, applications of a systems approach to water resources planning
and management are presented, and potential conflicts between various sectors and
disciplines affected by the development projects are discussed. Water resources
systems analysis involves modeling of quantity and quality aspects of surface water
and groundwater. Understanding the principles of simulation, optimization tech-
niques, and multiple-criterion decision making as well as engineering economics
and time series analysis are prerequisites for water resources systems analysis (see
Chapters 2 to 5). The book is written for graduate courses in water resources in
civil, environmental, and agricultural engineering studies. The objective of each
chapter has been to incorporate the latest developments in surface and groundwater
systems optimization and modeling. The modeling techniques are discussed with
some field applications.
Chapter 1 presents basic definitions about hydrologic and environmental sys-
tems. This chapter also summarizes major conflict issues and the framework for a
sustainable decision support system in water resources planning and management.
Chapters 2 and 3 summarize the methods and concepts of decision making and
elements of probabilistic and statistical methodology, as well as their use in modeling
risk and uncertainty. Mathematical methods for conflict resolution are also explained
in these chapters.
Chapter 4 presents the principles of engineering economics and their application
to water resources planning and operation, as well as methods for incorporating the
time/money relationship. Chapter 5 discusses basic principles of hydrologic time
series modeling and types of statistical models related to water resources. Chapter
6 summarizes various elements of river basin modeling, including data and infor-
mation processing, as well as objectives, constraints identification, and details of the
modeling process. Management tools such as simulation and optimization models
and a number of computer packages that have been used in water resources planning
and management projects are also presented in this chapter.
Chapter 7 discusses the main characteristics of groundwater systems, ground-
water flow equations, and groundwater modeling and provides applications of opti-
mization models in groundwater management. Chapter 8 summarizes various phases
of river–reservoir systems modeling, including data and information processing, and
deals with such issues as water supply, flood control, power generation, and instream
flow requirements. Formulation of simulation and optimization models for water
resources planning and management in river–reservoir systems are also presented
in this chapter.
Chapter 9 discusses principles of water quality analysis, including major types
of pollutants, water quality criteria, and water quality monitoring. Water quality
management in rivers and reservoirs is discussed in the second part of this chapter,
and groundwater pollutants and principles of groundwater quality management are
also covered. Chapter 10 presents major concerns of water resources engineers
regarding the design and operation of hydroelectric power plants; it also provides a
framework for long- and short-term operation models. Coordination between hydro
and thermal power plants is also discussed in this chapter. Chapter 11 summarizes
various aspects of managing water demands for domestic, industrial, and agricultural
purposes. Wastewater treatment and water reuse issues as they relate to water man-
agement are also discussed in this chapter. Chapter 12 defines climatic, hydrologic,
and agricultural droughts, and the basic steps for analyzing drought for drought
management studies are demonstrated.
It is our hope that this book can serve water resources communities around the
world and add significant value to systems analysis techniques for water resources
planning and management.
Mohammad Karamouz
Ferenc Szidarovszky
Banafsheh Zahraie
Acknowledgments
Many individuals have contributed to the preparation of the book. The initial
impetus for the book was provided by Professors Mark H. Houck and Jacques W.
Delleur through their class notes provided when they taught at Purdue University.
We are indebted to Mr. Reza Kerachian, Ph.D. Candidate of Amirkabir University
(Tehran Polytechnic), for his dedicated and considerable contribution to Chapters
6, 7, 8, and 9. The authors would also like to thank other doctorate students of
Amirkabir University (Tehran Polytechnic); Shahab Araghy-Nejad for his contri-
bution to Chapters 5 and 12; Ali Moridi for his contribution to Chapter 11; Koroosh
Behzadian for his contribution to Chapter 10; and Arman Ganji of Shiraz University
for his contribution to conflict resolution example problems. Special thanks to Neda
Parvini for preparing figures and artwork for the book. The reviews and constructive
comments of Professors Richard McCuen, Mark H. Houck, Lindell Ormsbee,
Nathan Buras, G. V. Loganathan, and Dr. Charles Howard are greatly appreciated.
About the Authors
Mohammad Karamouz is Professor of Civil Engineering at Amirkabir University
(Tehran Polytechnic) and a visiting professor in the Department of Civil Engineer-
ing/Hydrology and Water Resources at the University of Arizona, Tucson. He
received a B.S. degree in civil engineering from Shiraz University, an M.S. in water
resources and environmental engineering from George Washington University, and
a Ph.D. in hydraulic and systems engineering from Purdue University. He is a
Licensed Professional Engineer in the state of New York and a Fellow of American
Society of Civil Engineers. Among many professional achievements, he was the
dean of engineering at Pratt Institute and the founder of Arch Consulting Co. in
New York City. His areas of research and teaching interests include integrated water
resources planning and management, drought analysis and management, decision
support systems, streamflow forecasting, reservoir operation, conjunctive use of
surface and groundwater, and genetic algorithm modeling. He has more than 100
technical publications, books, and book chapters. Professor Karamouz serves nation-
ally and internationally as a consultant to both private and governmental agencies.
Ferenc Szidarovszky is a graduate of the Eotvos University of Sciences, Budapest,

Hungary. He has two Ph.D. degrees, one in mathematics and the other in economics.
The Hungarian Academy of Sciences has awarded him with the degrees Candidate
in Mathematics as well as Doctor of Engineering Sciences. After moving to the U.S.,
he became a professor in the Systems and Industrial Engineering Department at the
University of Arizona, Tucson. His research interests include numerical analysis,
systems theory, decision theory, including multi-criteria decision making, conflict
resolution, and their applications in natural resources management and economics.
He is the author of more than 200 research publications and authored or co-authored
8 books in Hungarian and 9 books in English.
Banafsheh Zahraie is an assistant professor at the Department of Civil Engineering

at the University of Tehran. She received a B.S. in civil engineering, an M.S. in
water resources engineering, and a Ph.D. in water resources management from
Amirkabir University (Tehran Polytechnic). She was a research associate at the
Department of Hydrology and Water Resources at the University of Arizona through
a World Lab fellowship. She is an international expert in application of systems
engineering in hydropower reservoirs and has developed many state of the art
computer-modeling packages and decision-support systems for large-scale water
resource systems. She is a member of American Society of Civil Engineers and
American Geophysical Union. Her research interests include decision support sys-
tems, MCDM, stochastic hydrology, and systems analysis.
Contents
Chapter 1 Introduction ..........................................................................................1
1.1 Introduction ......................................................................................................1
1.2 Availability of Water on Earth and Water Use................................................4
1.3 Water Resources Systems: Basic Concepts.....................................................4
1.4 System Definition and Properties ....................................................................6
1.5 Hydrologic Systems and Modeling Approaches .............................................9
1.5.1 Hydrologic Cycle .................................................................................9
1.5.2 Hydrologic Variables and Parameters................................................11
1.6 Conflict Issues in Water Resources Planning and Management...................12
1.7 Decision Tools in Water Resources Management .........................................13
References................................................................................................................19
Chapter 2 Decision Making: Optimization and Conflict Resolution.................21

2.1 Introduction ....................................................................................................21
2.2 Single-Criterion Optimization........................................................................22
2.3 Multiple-Criteria Optimization ......................................................................30
2.3.1 Sequential Optimization.....................................................................35
2.3.2 The ε-Constraint Method ...................................................................36
2.3.3 The Weighting Method ......................................................................36
2.3.4 Direction-Based Method ....................................................................41
2.3.5 Distance-Based Methods....................................................................44
2.4 Group Decision Making.................................................................................48
2.5 Conflict Resolution ........................................................................................52
2.6 Special Methodology .....................................................................................55
2.6.1 Dynamic Programming ......................................................................55
2.6.2 Genetic Algorithms ............................................................................60
2.7 Problems.........................................................................................................63
References................................................................................................................66
Chapter 3 Decision Making under Uncertainty..................................................67

3.1 Introduction ....................................................................................................67
3.2 Probabilistic Methods ....................................................................................67
3.2.1 Random Variables ..............................................................................67
3.2.2 Central Tendencies and Variances .....................................................75
3.2.3 Joint Distributions ..............................................................................79
3.3 Simulation Models .........................................................................................82
3.3.1 Neural Networks ................................................................................82
3.3.2 Observers............................................................................................84
3.3.3 Stochastic Simulation.........................................................................86
3.4 Measures of Reliability..................................................................................89
3.5 Stochastic Processes.......................................................................................95
3.6 Markov Chains ...............................................................................................96
3.7 Optimization under Uncertainty ..................................................................100
3.7.1 Chance-Constrained Programming ..................................................101
3.7.2 Stochastic Dynamic Programming ..................................................102
3.8 Risk, Uncertainty, and Bayesian Decision Theory......................................103
3.8.1 Risk and Value of Information ........................................................104
3.8.2 Bayesian Decision Making ..............................................................105
3.8.3 Conjugate Distribution Families ......................................................106
3.9 Fuzzy Sets and Parameter Imprecision .......................................................108
3.10 Problems.......................................................................................................112
References..............................................................................................................115
Chapter 4 Water Resources Economics............................................................117

4.1 Introduction ..................................................................................................117
4.2 Benefits and Costs........................................................................................117
4.3 Money–Time Relationship...........................................................................119
4.3.1 Future Value and Present Value of Money ......................................120
4.3.2 Future Value of Money and End-of-Period Cash Flows.................121
4.3.3 Present Value of Money and End-of-Period Cash Flows................121
4.3.4 Future Value of Money and End-of-Period Uniform Gradient
Cash Flows .......................................................................................122
4.3.5 Present Value of Money and End-of-Period Uniform Gradient
Cash Flows .......................................................................................122
4.3.6 End-of-Period Cash Flows and End-of-Period Uniform Gradient
Cash Flows .......................................................................................123
4.4 Economic Analysis of Multiple Alternatives...............................................126
4.4.1 Present Worth Method .....................................................................126
4.4.2 Future Worth Method.......................................................................128
4.4.3 Annual Worth Method .....................................................................129
4.4.4 The Internal Rate of Return (IRR) Method ....................................129
4.4.5 The External Rate of Return Method..............................................130
4.4.6 Comparing Alternatives Using the Rate of Return Method............131
4.5 Economic Evaluation of Projects Using the Benefit–Cost Ratio
Method .........................................................................................................132
4.6 Public vs. Private Projects and Goods.........................................................135
4.7 Economic Models ........................................................................................136
4.7.1 Production Functions .......................................................................136
4.7.2 Resource-Supply Functions .............................................................137
4.7.3 Demand Functions ...........................................................................137
4.8 Efficiency and Equity in Economic Analysis..............................................138
4.9 Problems.......................................................................................................141
References..............................................................................................................143
Chapter 5 Time Series Analysis........................................................................145

5.1 Introduction ..................................................................................................145
5.2 Stochastic Processes.....................................................................................145
5.3 Hydrologic Time Series Modeling: Basic Steps .........................................147
5.4 Data Preparation...........................................................................................147
5.5 Methods of Parameter Estimation ...............................................................152
5.5.1 Method of Moments.........................................................................152
5.5.2 Method of Least Squares .................................................................153
5.5.3 Method of Maximum Likelihood ....................................................154
5.6 Goodness of Fit Tests...................................................................................155
5.6.1 Chi-Square Goodness of Fit Test.....................................................155
5.6.2 Kolmogorov–Smirnov Goodness of Fit Test...................................157
5.6.3 Tests of Normality ...........................................................................158
5.6.4 Tests of Independence in Time........................................................161
5.7 Akaike’s Information Criterion (AIC).........................................................166
5.8 Data Generation and Forecasting ................................................................167
5.9 Autoregressive Modeling .............................................................................167
5.10 Moving Average Process..............................................................................171
5.11 Autoregressive Moving Average Modeling .................................................172
5.11.1 Data Generation and Forecasting Using ARMA Models ...............173
5.12 Autoregressive Integrated Moving Average (ARIMA) Modeling ..............173
5.12.1 Time Series Forecasting Using ARIMA Models ............................175
5.13 Multivariate and Disaggregation Modeling of Time Series........................175
5.13.1 Multivariate Modeling......................................................................175
5.13.2 Disaggregation Modeling.................................................................179
5.14 Problems.......................................................................................................181
References..............................................................................................................183
Appendix A............................................................................................................185
Appendix B............................................................................................................187
Chapter 6 River Basin Modeling ......................................................................191

6.1 Introduction ..................................................................................................191
6.2 Watershed: Definition and Characteristics ..................................................191
6.3 Watershed Geomorphology..........................................................................191
6.3.1 Drainage Area ..................................................................................191
6.3.2 Watershed Length.............................................................................193
6.3.3 Watershed Slope...............................................................................193
6.3.4 Watershed Shape ..............................................................................193
6.4 Soil Characteristics ......................................................................................195
6.4.1 Soil Texture ......................................................................................195
6.4.2 Soil Structure....................................................................................195
6.4.3 Soil Moisture....................................................................................195
6.5 Data Requirements.......................................................................................196
6.5.1 Hydrologic and Climatologic Data..................................................196
6.5.2 Physical Characteristics ...................................................................197
6.5.3 Hydrologic Characteristics...............................................................198
6.5.4 Inflow and Outflow ..........................................................................200
6.5.5 Demand Points .................................................................................200
6.5.6 Control Points ..................................................................................201
6.5.7 Critical Points...................................................................................201
6.5.8 Water Quality Requirements for Various Uses................................201
6.5.9 Return Flows and Drainage .............................................................201
6.5.10 Sampling Network ...........................................................................202
6.6 Objectives .....................................................................................................202
6.6.1 Water Supply....................................................................................202
6.6.2 Flood Control ...................................................................................204
6.6.2.1 Flood Control Alternatives ...............................................205
6.7 Management Tools .......................................................................................207
6.7.1 Streamflow and Excess Water Estimation .......................................207
6.7.1.1 Streamflow Estimation in Homogenous Basins...............207
6.7.1.2 Unit Hydrograph As a Tool.............................................209
6.7.1.3 Potential Evapotranspiration Estimation:
Thornthwaite Method .......................................................210
6.7.1.4 Infiltration Estimation: Green–Ampt Method ..................211
6.7.2 Simulation Models ...........................................................................212
6.7.2.1 Hydrologic Simulation Models ........................................212
6.7.2.2 Object-Oriented Simulation Models.................................224
6.7.3 Optimization Models........................................................................226
6.7.3.1 Water Allocation in Multi-User Systems .........................226
6.7.3.2 River Quality Management ..............................................228
6.7.3.3 Calibration of Hydrologic Simulation Models ................229
6.8 Conflict Issues in River Systems .................................................................229
6.9 Problems.......................................................................................................232
References..............................................................................................................235
Chapter 7 Groundwater Management...............................................................239

7.1 Introduction ..................................................................................................239
7.2 Characteristics of Groundwater Systems.....................................................239
7.2.1 Hydraulic Properties of Soils and Groundwater .............................239
7.3 Groundwater Flow Equations ......................................................................243
7.3.1 The Continuity Equation..................................................................243
7.3.2 Equation of Motion in Groundwater ...............................................246
7.3.3 The Groundwater Flow Equation ....................................................248
7.4 Groundwater Modeling ................................................................................250
7.4.1 The Flow Net ...................................................................................250
7.4.2 Numerical Methods..........................................................................252
7.4.2.1 Finite-Difference Numerical Model Basics .....................253
7.4.2.2 Finite-Element Methods ...................................................263
7.4.2.3 Grid Design.......................................................................269
7.4.2.4 Parameter Estimation and Model Calibration..................269
7.4.2.5 Model Verification ............................................................270
7.4.2.6 Predictions and Post Audits..............................................270
7.5 Management Tools .......................................................................................271
7.5.1 Groundwater Simulation Models.....................................................271
7.5.1.1 Graphic Groundwater .......................................................271
7.5.1.2 MODFLOW ......................................................................271
7.5.2 Safe Yield Estimation.......................................................................272
7.5.3 Optimization Models for Groundwater Management .....................272
7.5.3.1 Optimization Model for Groundwater Operation ............273
7.5.3.2 Optimization Model for Capacity Expansion ..................274
7.5.3.3 Optimization Model for Water Allocation .......................276
7.5.3.4 Optimization Model for Conjunctive Water Use .............277
7.6 Conflict Issues in Groundwater Systems.....................................................284
7.7 Problems.......................................................................................................290
References..............................................................................................................293
Chapter 8 River–Reservoir Systems Modeling ................................................295

8.1 Introduction ..................................................................................................295
8.2 Data Collection and Processing...................................................................295
8.2.1 Inflow................................................................................................295
8.2.2 Water Demands ................................................................................296
8.2.3.1 Types of Dams ..................................................................298
8.2.3.2 Ancillary Facilities............................................................299
8.2.3.3 Hydropower Facilities.......................................................300
8.3 Objectives .....................................................................................................301
8.3.1 Water Supply: Reservoir Storage Zones..........................................302
8.3.2 Power Generation.............................................................................302
8.3.3 Flood Control ...................................................................................303
8.3.4 Navigation Requirements.................................................................303
8.3.5 Environmental Water Rights ............................................................303
8.4 Capacity Expansion......................................................................................304
8.4.1 Interbasin Water Transfer.................................................................304
8.4.2 Optimal Sizing of a Single Reservoir System ................................307
8.4.3 Reliability-Based Design .................................................................309
8.4.5 Sedimentation and Dredging ...........................................................310
8.5 Reservoir Operation Optimization...............................................................311
8.5.1 Water Supply with Monthly Varying Release .................................311
8.5.2 Flood Control ...................................................................................313
8.6 Application of Deterministic Dynamic Programming in Reservoir
Operation: DPR Model ................................................................................316
8.7 Stochastic Reservoir Operation Modeling...................................................319
8.7.1 Stochastic Linear Programming.......................................................320
8.7.1.1 Chance Constraint Reservoir Operation Model ...............320
8.7.2 Stochastic Dynamic Programming Model.......................................322
8.7.2.1 Reservoir Storage Discretization ......................................323
8.7.2.2 Reservoir Inflow Discretization........................................324
8.7.3 BSDP and DDSP Models ................................................................328
8.8 Multiple Reservoir Operation Modeling .....................................................334
8.8.1 Cascade Reservoirs ..........................................................................334
8.8.2 Parallel Reservoirs ...........................................................................335
8.9 Reservoir Operating Rules...........................................................................337
8.10 Cost Functions..............................................................................................340
8.11 Efficiency of Operating Policies..................................................................343
8.11.1 Reliability .........................................................................................344
8.11.2 Resiliency .........................................................................................344
8.11.3 Vulnerability.....................................................................................345
8.12 Conflict Issues in River–Reservoir Systems................................................346
8.13 Problems.......................................................................................................356
References..............................................................................................................360
Chapter 9 Water Quality Management .............................................................363

Introduction............................................................................................................363
9.1 Principles of Water Quality Systems Analysis............................................363
9.1.1 Water Pollutants and Their Principal Sources.................................363
9.1.1.1 Oxygen-Demanding Material ...........................................364
9.1.1.2 Nutrients............................................................................364
9.1.1.3 Pathogens ..........................................................................364
9.1.1.4 Suspended Solids ..............................................................364
9.1.1.5 Salts...................................................................................365
9.1.1.6 Toxic Metals and Organic Compounds............................365
9.1.1.7 Heat ...................................................................................365
9.1.2 Water Quality Criteria and Standards for Water Use......................365
9.1.2.1 Drinking Water Supply .....................................................366
9.1.2.2 Industrial Water Supply ....................................................366
9.1.2.3 Agricultural Water Supply................................................366
9.1.2.4 Aquatic Life ......................................................................367
9.1.3 Pollutant Material Balance in Surface Water ..................................367
9.1.4 Water Quality Monitoring................................................................369
9.1.4.1 Water Quality Variable Selection .....................................370
9.1.4.2 Location of Water Quality Monitoring Stations ..............371
9.2 Surface Water Quality Management............................................................382
9.2.1 River Water Quality Management ...................................................382
9.2.2 Low Flow Determination .................................................................383
9.2.3 Computer Models for Streamwater Quality Simulation .................384
9.2.3.1 QUAL2E Model ...............................................................386
9.2.3.2 WASP5 ..............................................................................386
9.2.4 Simple Streamwater Quality Models...............................................387
9.2.4.1 Oxygen Sag Model ...........................................................387
9.2.4.2 Simple Water Quality Model for Multiple-Discharge
Streams..............................................................................393
9.2.5 Water Quality Management Models for Rivers and Streams .........395
9.2.5.1 Deterministic Approach ....................................................395
9.2.5.2 Stochastic Approach .........................................................408
9.2.6 Lake and Reservoir Water Quality Management ............................412
9.2.6.1 Stratification and Turnover ...............................................413
9.2.6.2 Eutrophication...................................................................414
9.2.7 River–Reservoir Simulation Model Packages .................................414
9.2.7.1 HEC-5Q ............................................................................414
9.2.8 Reservoir Water Quality Management ............................................416
9.2.9 Conflict Resolution in Surface Water Quality Management...........419
9.3 Groundwater Quality Management .............................................................422
9.3.1 Introduction ......................................................................................422
9.3.2 Sources of Groundwater Contamination .........................................422
9.3.3 Contaminant Transport in the Saturated Zone ................................422
9.3.4 Optimization Model for Groundwater Quality Management..........426
9.4 Problems.......................................................................................................428
References..............................................................................................................438
Chapter 10 Hydroelectric Systems Analysis ....................................................441

10.1 Introduction ..................................................................................................441
10.2 Basic Definitions ..........................................................................................441
10.3 Components of Hydropower Plants.............................................................443
10.4 Data Requirements.......................................................................................445
10.4.1 Hydrologic Data...............................................................................445
10.4.3 Power Load and Reliability of Power Systems...............................446
10.4.4 Non-Power Operating Criteria.........................................................449
10.5 Methods for Estimating Power Potential.....................................................449
10.5.1 The Flow–Duration Method ............................................................449
10.5.2 Sequential Streamflow Routing (SSR) Method...............................453
10.6 Hydropower Reservoir Operation................................................................459
10.6.1 Long-Term Planning Models ...........................................................463
10.6.2 Mid-Term Planning Models.............................................................465
10.6.3 Short-Term Planning Models...........................................................465
10.6.4 Real-Time Operation Models ..........................................................466
10.7 Hydrothermal Coordination .........................................................................467
10.8 Conflict Issues in the Operation of Hydropower Systems..........................474
10.9 Problems.......................................................................................................479
References..............................................................................................................481
Chapter 11 Water Demand Analysis and Management ...................................483
11.1 Introduction ..................................................................................................483
11.2 Water Use and Demand: Basic Definitions.................................................484
11.3 Water Demand Forecasting for Regional and National Planning...............486
11.4 Water Supply Issues.....................................................................................488
11.4.1 Rainwater Collection........................................................................488
11.4.2 Desalination......................................................................................488
11.4.3 Artificial Recharge ...........................................................................490
11.4.4 Groundwater Dams ..........................................................................490
11.4.5 Weather Modification.......................................................................491
11.5 Flow Measurement Equipment ....................................................................491
11.5.1 Flow Measurement in Pipes ............................................................491
11.5.2 Flow Measurement in Open Channels and Rivers..........................492
11.6 Municipal Water Demand ............................................................................492
11.7 Municipal Water Demand Estimation and Forecasting...............................492
11.8 Elements of Urban Water Distribution Networks .......................................496
11.9 Municipal Water Demand Management Measures .....................................498
11.9.1 Water Loss Reduction ......................................................................498
11.9.2 Economic Incentives and Water Pricing..........................................500
11.9.3 Institutional Measurements and Effective Legislation ....................500
11.9.4 Water Reuse .....................................................................................501
11.10 Agricultural Water Demand ........................................................................501
11.11 Irrigation Methods.......................................................................................503
11.12 Drainage and Salinity Control ....................................................................508
11.13 Agricultural Water Demand Management..................................................509
11.13.1 Increasing Irrigation Efficiency.....................................................509
11.13.2 Development of the Proper Crop Mix..........................................510
11.13.3 Soil Development and Management.............................................511
11.14 Industrial Water Demand Management ......................................................511
11.15 Environmental Water Demands ..................................................................512
11.16 Water Reuse ................................................................................................513
11.17 Treatment Options for Wastewater Reuse ..................................................515
11.17.1 Preliminary Process.......................................................................515
11.17.2 Primary Treatment.........................................................................515
11.17.3 Secondary Treatment.....................................................................517
11.17.4 Advanced Wastewater Treatment..................................................518
11.18 Gray Water: Definition and Reuse Issues...................................................519
11.19 Conflict Issues in Water Demand Management .........................................520
11.20 Problems......................................................................................................525
References..............................................................................................................526
Chapter 12 Drought Management ....................................................................529

12.1 Introduction ..................................................................................................529
12.2 Drought Definition .......................................................................................529
12.3 Drought Preparedness/Management ............................................................530
12.4 Climatic Drought..........................................................................................532
12.4.1 Point Evaluation ...............................................................................532
12.4.1.1 Fitting Statistical Distribution ..........................................532
12.4.1.2 Analysis of Moving Averages ..........................................534
12.4.1.3 Point Evaluation of Drought Events on a Monthly
Scale..................................................................................537
12.4.2 Regional Analysis of Climatic Drought ..........................................545
12.4.2.1 Weighting Method ............................................................545
12.4.2.2 Median of Ratios Method.................................................547
12.4.2.3 Combined Method ............................................................547
12.5 Prediction of Drought Trends ......................................................................549
12.6 Hydrologic Drought: Definition and Background.......................................550
12.6.1 Run Theory.......................................................................................552
12.6.2 Probability Distribution Function of Drought Characteristics ........552
12.6.3 Reservoir Operation and Demand Allocation .................................553
12.7 Agricultural Drought: Definition and Background .....................................553
12.7.1 Thornthwaite Water Balance Model ................................................553
12.7.2 Agricultural Drought Indicators.......................................................554
12.7.2.1 Potential Evapotranspiration.............................................555
12.7.2.2 Soil Moisture Condition ...................................................555
12.7.2.3 Runoff Calculation............................................................556
12.7.2.4 Potential Climatic Values..................................................557
12.8 Geostatistic Tools in Drought Studies .........................................................565
12.8.1 Kriging .............................................................................................567
12.8.2 Spatial Kriging in Drought Studies .................................................570
12.8.3 Spatiotemporal Kriging....................................................................571
12.9 Problems.......................................................................................................574
References..............................................................................................................577
Index......................................................................................................................579
1 Introduction
1.1 INTRODUCTION
Increasing demand for water, higher standards of living, depletion of resources of
acceptable quality, and excessive water pollution due to agricultural and industrial
expansions have caused intense social and political predicaments. Since the beginning
of the 19th century, the population of the world has tripled, nonrenewable energy
consumption has increased by a factor of 30, and the industrial production has
multiplied by 50 times. Although progress has improved the quality of life, it has
caused significant environmental destruction in such a magnitude that could not be
predicted.
A question that should be addressed is whether in the next decades development
could be done in a way that is economically and ecologically sustainable. We cannot
answer this question unless we have a vision of the future and our planning schemes
are environmentally responsible and sensitive toward the major elements of our
physical environment, namely air, water, and soil. Among these elements, water is
of special importance. Excessive use and misuse of surface and groundwater and
pollution of these vital resources by residential, agricultural, and industrial waste-
water has threatened our well-being. Planning for sustainable development of water
resources means water conservation, waste and leakage prevention, improved effi-
ciency of water systems, improved water quality, water withdrawal and usage within
the limits of the system, a level of water pollution within the carrying capacity of
the streams, and water discharge from groundwater within the safe yield of the
system. In other words, we are seeking a balance among our physical being, our
ability to manage our resources, and the limitations imposed by the environment.
Water is a sustainable resource, and the need for integrated water resources
management is on the agenda of every state. As one of the three vital elements of
air, soil, and water, the role of water in the conservation of the environment is
receiving increased global attention. Figure 1.1 illustrates the vulnerability of various
nations to water scarcity by using a composite index based on available water
resources and current use, population, reliability of water supply, and national
income (World Resources, 1998).
In general, water resources development studies can be classified into planning,
operation, and management. Long-term management of water resources systems
requires considering the competing users of the system along with the future supply
and demands. Then, by taking into account these considerations, modeling operation
of the systems will be helpful to the decision makers in charge of allocating
resources with higher efficiency. Systems need to be expanded because of the
growing demands for resources and services; therefore, capacity expansion is
another part of the long-term water resources development studies when existing
1
2 Water Resources Systems Analysis
WATER RESOURCES
VULNERABILITY
INDEX (1995)
No Vulnerability
Low Vulnerability
Medium Vulnerability
High Vulnerability
No data
FIGURE 1.1 Vulnerability of nations to water scarcity. (From World Resources Institute,
World Resources: A Guide to the Global Environment, a joint publication by The World
Resources Institute, The United Nations Environment Program, The United Nations Devel-
opment Program, and The World Bank, Oxford University Press, New York, 1998.)
facilities for storing, withdrawing, and transferring water to users are not adequate
with an acceptable reliability.
Figure 1.2 shows different steps in the water resources planning process. As
shown in this figure, the process can be classified into the following phases:
1. Problem definition and data collection and processing

2. Modeling
3. Decision making
4. Construction
5. Continuous monitoring of the system
In the first phase, the objectives and constraints are identified. Data collection and
processing are important parts of this process and include:
• Climatic, hydrologic, hydraulic, and environmental characteristics of the

study area affecting the supply and the demands
• Economic, institutional, and legal conditions affecting water allocation
policies
• Structural and physical characteristics of the rivers and reservoirs and
their appurtenant facilities that have impacts on the carrying capacity of
the system
In the modeling phase, simulation and optimization models are used for finding
possible alternatives. In the next phase, conflict resolution and multi-criterion deci-
sion analysis techniques are used for comparing different alternatives. As seen in
Figure 1.2, the selection of final alternatives, in many cases, depends on social and
Introduction 3
FIGURE 1.2 Process of water resources planning.
economic (nontechnical) issues related to institutional framework and distribution

of financial resources in a region. Political issues may also cut off the process if the
policy makers do not effectively participate in the planning process. In the last phase,
monitoring, evaluation, and feedback to decision makers are carried out in order to
modify the planning schemes and the operating policies.
TABLE 1.1
Water Reserves on the Earth
Percentage of
Global Reserves (%)
Volume (1012 m3) Total Water Freshwater
World ocean 1,338,000 96.5 —

Groundwater 23,400 1.7 —
Freshwater 10,530 0.76 30.1
Soil moisture 16.5 0.001 0.05
Glaciers and permanent snow cover 24,064 1.74 68.7
Antarctic 21,600 1.56 61.7
Greenland 2340 0.17 6.68
Arctic islands 83.5 0.006 0.24
Mountainous regions 40.6 0.003 0.12
Ground ice/permafrost 300 0.022 0.86
Water reserves in lakes 176.4 0.013 —
Fresh 91 0.007 0.26
Saline 85.4 0.006 —
Swamp water 11.47 0.0008 0.03
River flows 2.12 0.0002 0.006
Biological water 1.12 0.0001 0.003
Atmospheric water 12.9 0.001 0.04
Total water reserves 1,385,984 100 —
Source: Shiklomanov, I., in Water in Crisis, P. Gleick, Ed., Oxford University Press, New York, 1993,
chap. 2. With permission.
1.2 AVAILABILITY OF WATER ON EARTH AND WATER USE

As can be seen in Table 1.1, oceans, the greatest water reserve on Earth, represent
96.5% of the total reserve. An examination of the other reserves shows that shares
of permanent snow covers and groundwater that are major sources of freshwater add
up to 3.44% of the total water reserve. The spatial distribution of water availability
in different regions can be a detriment to regional and local development plans.
The total world freshwater supply is estimated to be 41,022 billion m3 (World
Resources, 1998). Table 1.2 shows the freshwater supply of different continents and
annual withdrawals for domestic, industrial, and agricultural water uses. As can be
seen in this table, annual withdrawal is about 8% of the total freshwater supply, and
the agriculture sector, with 69% usage, has the highest rate among the water uses.
1.3 WATER RESOURCES SYSTEMS: BASIC CONCEPTS

Each water resources system consists of different elements of two distinct environ-
ments: one is a physical, chemical, and biological environment and the second is a
cultural environment with social, political, economic, and technological dimensions
Introduction
TABLE 1.2
Total Water Supply in Different Continents
Annual Internal Annual Withdrawal Sector Withdrawal

Water Resourcesa Year Total Water Resources Domestic Industrial Agricultural
Continents (1012 m3) of Data (1012 m3) (%) (%) (%) (%)
Africa 3.996 1995 0.15 4 7 5 88

Europe 6.235 1995 0.46 7 14 55 31
North America 5.309 1991 0.51 10 13 47 39
Central 1.057 1987 0.10 9 6 8 86
America
South America 9.526 1995 0.10 1 18 23 59
Asia 13.206 1987 1.63 12 6 9 85
Oceania 1.614 1995 0.02 1 64 2 34
World 41.022 1987 3.240 8 8 23 69
a The average annual flow of rivers and recharge of groundwater generated from endogenous precipitation.
Source: World Resources Institute, World Resources: A Guide to the Global Environment, a joint publication by The World Resources Institute,
The United Nations Environment Program, The United Nations Development Program, and The World Bank, Oxford University Press, New
York, 1998.
5
(White et al., 1992). The physical environment includes interdependent water bodies
and structures, each impacting the state and performance of the others. The cultural
environment encompasses the various social constraints that are mainly focused on
us and our interactions with physical environment. The physical and cultural environ-
ments are inseparable. In other words, natural resources and especially water resources
systems cannot be modeled effectively without considering social constraints. Over
the course of this book, we try to show how different elements of these two environ-
ments and their interactions are in conflict and how they can be modeled.
For modeling, one of the most common approaches is to divide the system into
pieces (disintegration phase) to see what it is composed of and how it works. It is
an example of the reductionist approach, which reduces complex phenomena to their
component parts and examines them in detail. The problem with this approach
becomes apparent when we try to put the system back together again. The process
of turning our focus from analysis to synthesis can become more practical by
considering the following items:
• The interactions among components and their role in the entire system
should be defined before the system is broken into pieces. Knowledge
about these interactions plays an important role in the disintegration phase
and in the analysis of the entire system.
• The approaches that are used for modeling different components should
be consistent.
• The terminologies and methodologies for defining and assessing the per-
formance of each component should be comparable.
The second phase in systems analysis is to integrate different components of the

system in order to assess the performance of the system as a whole. Integrated water
resources planning and management focus not only on the performance of individual
components but also on the performance of the entire system of components (Loucks,
1996).
1.4 SYSTEM DEFINITION AND PROPERTIES

Any structure or device, including different interactive components (real or abstract),
that causes an output reference to a specific input in a given time can be called a
system. The common characteristics of any system in general and water resources
systems specifically can be summarized as follows:
• All systems have some structure and organization.

• Systems are all generalizations, abstractions, or idealizations of the real
world with different levels of complexities.
• Functional and structural relationships exist between components of the
system.
• All systems show some degree of integration.
• Input–output relations and the nature of them are important characteristics
of systems.
Introduction 7
Functional relationships reflect both physical and social functioning of the system.
Physical functioning includes the flow and transfer of some material, while social
functioning involves social, political, and economic preferences. Specifications of
systems can be classified as follows:
• Inputs
• Governing physical laws
• Initial and boundary conditions
• Outputs
Inputs, outputs, and major characteristics of the systems are usually defined by
variables and parameters. A variable is a system characteristic that assumes different
values measured at different times. A parameter is also a system characteristic, but
it normally does not change with time. Different values of variables can be viewed
as possible states of the nature or as alternative futures. Variables or parameters can
be classified as:
• Lumped, which do not change in space

• Distributed, which vary in one or more space dimensions
Memory, which is another important characteristic of a system, is the length of time

in the past for which the input could have an impact on the output. A system may
have different levels of memory:
• Zero memory — The state and output depend only on the input in the
present time.
• Finite memory — The state, output, and behavior depend only on the
history of the system for a specific time span (memory)
• Infinite memory — The state and output depend on the entire history of
the system
Systems are generally classified as deterministic or stochastic. In deterministic

systems, the same input gives the same output, whereas stochastic systems contain
one or more elements for which the relationship between input and output is prob-
abilistic rather than deterministic.
There are also other detailed classifications of systems. They are as follows:
• Continuous systems — The output is produced continuously.

• Discrete systems — The output changes after finite intervals of time.
• Quantized systems — The output values change only at certain discrete
intervals of time and hold a constant value between these intervals.
• Natural systems — The inputs and outputs and other state variables are
measurable and are not controlled.
• Devised systems — The input may be both controllable and measurable.
• Simple systems — No feedback mechanisms exist in these systems.
• Complex systems — Feedback is built into these systems.
• Adaptive systems — These systems learn from their past history to

improve their performance.
• Causal systems — An output cannot occur earlier than the corresponding
input (cause and effect).
• Simulation systems — These are realization systems and are similar to
causal systems.
• Stable systems — If the input is bounded, the output is also bounded and
vice versa.
• Damped systems — The output of the system dies out without ever
crossing the time scale.
Most of the hydrologic systems are stable and causal systems and are heavily
damped. Figure 1.3 shows examples of outputs of continuous, discrete, and quantized
systems.
More details about continuous and discrete variables are presented in the fol-
lowing sections of this chapter. Considering the three basic components of a system
(inputs, system operators, and outputs), as shown in Figure 1.4, systems analysts
could face several different types of problems:
• Design problems — Inputs and system are known and the output must be
quantified.
• System identification problems — Inputs and outputs are known and the
system itself must be identified.
• Detection problems — System and outputs are known and inputs must be
identified.
• Synthetic problems (simulation) — Inputs and outputs are known and the
performance of models must be tested.
Hydrologists primarily deal with design and synthetic problems.
Time Time Time
a) Continuous systems b) Discrete systems c) Quantized systems
FIGURE 1.3 Outputs of continuous, discrete, and quantized systems.
Input operator Output
FIGURE 1.4 Schematic representation of systems operation. (From Chow, V. T., Maidment,
D. R., and Mays, L. W., Applied Hydrology, McGraw-Hill, New York, 1988. With permission.)
Introduction 9
1.5 HYDROLOGIC SYSTEMS AND MODELING

APPROACHES
Complex systems can be decomposed into subsystems, each having an input–output
linkage as a component. Hydrologic systems can be considered as a subsystem of
water resources representing the physical functioning of that system in a region. A
hydrologic system is defined as a structure or volume in space, surrounded by a
boundary, that accepts water and other inputs, operates on them internally, and
produces them as outputs (Chow et al., 1988).
Two approaches can be used for modeling hydrologic systems:
• Theoretical approach, which focuses on modeling the physics of the

system
• Empirical approach, which focuses on using historical observations of
different components of the hydrologic cycle
Utilizing the first approach results in more accurate models and outputs, thus improv-
ing our knowledge about different hydrologic phenomena; however, modelers are
usually faced with a lack of accurate information about inputs of the system and a
wide range of natural complexities.
Models are simplified representations of systems. Loucks et al. (1981) refer to
model development as an art that requires judgment in abstracting from the real
world of the components that are important to the decision-making process. Math-
ematical models, which are the main focus of this book, can be classified into the
following categories:
• Empirical vs. theoretical

• Lumped vs. distributed
• Deterministic vs. stochastic
• Linear vs. nonlinear
Details of these models are explained in different chapters of this book. Before
proceeding with the details of systems of analysis, the holistic view of water
resources planning and management could only be meaningful if it is observed
within the context of the hydrologic cycle, at least in a catchment scale.
1.5.1 HYDROLOGIC CYCLE

The hydrologic cycle is defined as “the pathway of water as it moves in its various
phases through the atmosphere to the Earth, over and through the land, to the ocean,
and back to the atmosphere” (National Research Council, 1991). The hydrologic
cycle can be considered as a closed system for Earth, because the total amount of
water in the cycle is constant.
The hydrologic cycle is generally described in terms of six major components
and some subcomponents, which are shown in Figure 1.5:
FIGURE 1.5 Schematic diagram of the hydrologic cycle.
• Precipitation (P)
• Precipitation on the oceans (Po)
• Precipitation on ice (Pi)
• Precipitation on land (Pl)
• Infiltration (I)
• Evaporation (E)
• Evaporation from oceans (Eo)
• Evaporation from ice (Ei)
• Evapotranspiration from plants (Ts)
• Evaporation from soil moisture (Eg)
• Surface runoff (R)
• Goundwater appearing as surface runoff (Rg)
• Subsurface flow (Rs)
• Groundwater flow (G)
Water budget is an accounting of the inflow, outflow, and storage of water in a specific
hydrologic system. Figure 1.6 is a schematic diagram of a system representing the
hydrologic cycle of a watershed. As can be seen in this figure, Rin, Gin, Rout, and Gout
are surface and groundwater inflows and outflows, respectively. Surface inflow in
this figure represents an inter-basin water transfer. Rg and Eg are groundwater appear-
ing as surface runoff and evaporation from soil moisture, respectively.
The surface water budget (ΔSs) and groundwater budget (ΔSg) can be estimated
as follows:
ΔSs = Pl + Rin − Rout + Rg − Ts − I (1.1)
ΔSg = Gin − Gout − Rg − Eg + I (1.2)

Introduction 11
Ts
P1
Eg
Rin Rg
Rout
Gin
Gout
I
FIGURE 1.6 Schematic diagram of a system representing the hydrologic cycle of a watershed.
1.5.2 HYDROLOGIC VARIABLES AND PARAMETERS

All natural physical processes are subject to variability. For example, rainfall inten-
sity, flood magnitude, or low flows in droughts have wide variations. To study these
variations and incorporate them in the planning and operation of water resources,
engineers gather and investigate samples of data.
The hydrologic cycle is composed of various phenomena, such as precipitation,
runoff, infiltration, evaporation, evapotranspiration, and abstraction. Different char-
acteristic variables, which can simply be called hydrologic variables, have been
defined to describe each of these phenomena. Depth or intensity of rainfall in
different time steps of a rainstorm, monthly inflow discharge to a reservoir, or daily
evaporation are some examples of hydrologic variables (Shahin et al., 1993). A
dataset consists of a number of measurements of a phenomenon, and the quantities
measured are variables.
A continuous variable can have any value on a continuous domain; examples
include volume of water flowing in a river or the amount of daily evaporation
measured in a climatic station. A discrete variable represents an interval or the
number of occurrences within each interval of time and space; the number of rainy
days in a certain period of time (e.g., a year) is an example of discrete hydrologic
variables.
The hydrologic variables can also be classified as qualitative or quantitative. A
qualitative variable can be expressed as a real number in a sensible way; type of
soil is an example of a qualitative hydrologic variable. A quantitative variable can
be measured as a real number; the number of rainy days in a year and rainfall
intensity in a day are examples of discrete and continuous quantitative hydrologic
variables, respectively.
Hydrologic variables usually vary in time and space. A time series is a sequence
of values arranged in order of their occurrence in time. Conflicts can arise when
trying to analyze the impact of various hydrologic variables on each other and model
them during the decision-making process.
1.6 CONFLICT ISSUES IN WATER RESOURCES

PLANNING AND MANAGEMENT
Water resources systems usually have multiple objectives, many of which are usually
in conflict. For example, a major conflict issue in the operation of river–reservoir
systems arises when the reservoir is not capable of supplying all the demands;
supplying instream requirements is in conflict with water conservation and supply
objectives, and monthly variations of power loads (energy demand) and water
demands do not follow the same pattern of decisions. For example, the operation of
groundwater resources can be a multifaceted problem with many economic, hydro-
logic, hydraulic, water quality, or environmental objectives and constraints that must
be satisfied. A common conflict issue in groundwater systems planning and operation
arises when the aquifer supplies water to different demand points for different
purposes and constraints on groundwater table fluctuations and quality aspects must
be maintained. Let’s consider the following example:
• Several demand points exist and groundwater that is supplied to one of

these demand points cannot be used by the others; therefore, a major
conflict issue may occur.
• Excessive discharge of aquifer can cause problems, such as settlement of
ground surfaces and buildings.
• Recharge of aquifer with polluted water such as infiltration of agricultural
wastewater, disposal of sewage or sludge from water treatment plants, and
sanitary landfills can produce environmental problems especially for aqui-
fers intended for water supply purposes and causes some irreversible
conflicts.
The optimal operating policies for operation of groundwater resources can be devel-
oped using hydraulic or water quality response equations, well characteristics, and
hydraulic gradient and water demand requirements, as well as conflict resolution
models that consider the supply, demand, and physical characteristics of the systems.
Besides conflicts in the objectives of water users, other important issues must be
considered from an institutional point of view. Even though water resources planning
for surface and groundwater operation might be done by a specific agency, such as
the department of water supply, the water allocation schemes defined and imposed
by this particular agency could be highly affected by the political and institutional
strength of other agencies, such as the Department of Agriculture, various industries,
and the Department of Environmental Protection. On the other hand, these and many
other entities are affected by water policies because they:
• Use water resources for producing specific goods and products.

• Use water resources for providing some services.
• Are responsible for public health and environmental protection.
Table 1.3 presents an example of how the quantitative aspects of conflict issues in
water supply between different agencies can be quantified. This example is based
Introduction 13
on organizational setups in some developing countries. The qualitative aspects of

conflict issues are shown in Table 1.4. As can be seen in these tables, the various
agencies that directly use water or might be affected indirectly by water supply
policies are in conflict in varying degrees.
The priorities and favorable ranges of water supply for each of these agencies
should be considered in formulating a conflict resolution scheme; therefore, the first
step in conflict resolution studies is to recognize all of the conflict issues and the
responsible agencies. Each of these departments has its own set of priorities for
allocating water to different demands. When formulating the conflict-resolution
algorithm, methods such as Nash bargaining theory (NBT) and multiple-criteria
decision making (MCDM) can be used (see Chapter 2).
These methods can provide the water resources systems analyst with a better
understanding of the mechanisms of cooperation, the roots of and the reasons for
conflicts, and the resolutions required in various political and social settings. They
can also promote a better understanding of the issues and help the stakeholders to
comprehend better with the potential consequences of preferences and choices.
Throughout this book, conflict issues as they relate to the topics of particular chapters
are discussed, and the algorithms for resolving conflicts based on qualitative and
quantitative measures are presented.
1.7 DECISION TOOLS IN WATER RESOURCES

MANAGEMENT
In the past, different tools have been used for modeling water resources systems,
including simulation and optimization models. Yeh (1985) presented a state-of-the-
art review of reservoir management and operation models in general. Yakowitz
(1982) reviewed the application of dynamic programming in water resources plan-
ning and management. These efforts have been mainly focused on development of
efficient tools for decision making in the area of water resources development and
management. The continuing evolution of information technology, including hard-
ware and software developments, has created an environment for better utilization
of these tools.
The application of a systems approach to water resources planning and manage-
ment has been established as a progressive dimension in the field of water resources
engineering. Two approaches have been significantly enhanced in recent years. The
first one is focused on the water-related data availability of variables in time and
space domain affecting the uncertainty of water resources decision making. The
second approach focuses on the complexity of the water resources domain and the
complexity of modeling tools in an environment characterized by rapid technological
development (Simonovic, 2000).
The decision support system (DSS) is a powerful tool for application of a systems
approach in real-world water resources planning and management. The basis of DSS
development was developed in the 1960s and 1970s in the field of management
information system (MIS). While MIS relies on working with databases, DSS com-
bines software and hardware facilities designed by systems engineers to be used to
14
TABLE 1.3
Conflict among Agencies over the Quantitative Aspects of Water Supply
Water Supply Quantitative Aspects (water demands)
Departments and Agencies Domestic Agriculture Health Industry Environment
Department of Energy/Water E (4), B (3), C (4), M (4) B (4), C (4), E (2) E (4), M (3) B (4), M (3) B (3), C (3), E (2),
Supply (A) M (4)
Industrial and mining sector (B) E (4), A (3), C (3), M (3) A (4) E (3) A (4), C (4), E (3), M (4), A (3)
M (3), W (1)
Department of Agriculture (C) A (4), E (4), B (3), A (4), A (4), M (3) M (3), E (3) B (4) A (3), E (2), M (4)
Water Resources Systems Analysis

M (4)
Health Department (E) B (3), C (4), A (4), M (4) A (2) A (4), B (3), C (3), B (3) A (2), C (2)
M (3)
Department of Environmental A (4), B (3), E (4), C (4) C (3) A (3), C (3) A (3), B (3) A (4), B (4), C (4)
Protection (M)
Note: The degree of conflict is indicated in parentheses; i = 5 is the most severe, and i = 1 is the least severe.
Introduction
TABLE 1.4
Conflict among Agencies over the Qualitative Aspects of Water Supply
Water Supply Qualitative Aspects
Fishery and
Domesticated
Departments and Agencies Domestic Agriculture Health Industry Environment Animal
Department of Energy/Water E (4), M (3) C (4), E (2), M (4) E (4), M (3) B (4), M (4), E (2) C (3), E (3), M (4) E (4), M (4)
Supply (A)
Industrial and mining sector (B) E (4), M (3) C (3), E (3), M (4) E (5), M (5) A (4), C (5) E (3) E (4), M (4)
Department of Agriculture (C) M (2), E (4) A (4), B (3), M (2) E (4) B (5), E (3), M (4) A (3), M (4) E (3), M (4)
Health Department (E) A (4), B (4), C (4) A (2), B (3), M (3) A (4), B (5), C (4) C (3), A (2) A (3), B (3) B (4), A (4), C (3)
Department of Environmental A (3), B (3), C (2) C (2), E (3), B (4), A (3), B (5) A (4), C (4) A (4), B (4), C (4) A (4), B (4), C (4)
Protection (M) A (4)
Note: The degree of conflict is indicated in parentheses; i = 5 is the most severe, and i = 1 is the least severe.
15
make better decisions. The concept of DSSs developed as a result of the intersection
of two trends (Stohr and Konsynski, 1992). The first one is the growing belief that
existing information systems, despite their success in automation of operating tasks
in organizational setups, have failed to assist the decision makers with many higher
level tasks. The second trend is continuous improvement in computing hardware
and software that has made it possible to place meaningful computing power into
the development of databanks and complex heuristics. Several attempts have been
made to develop a more flexible framework for these systems. The spatial decision
support system (SDSS), adaptive decision support system (ADSS), and intelligent
decision support system (IDSS) are some new DSS developments.
Figure 1.7 shows an example of a DSS framework for sustainable water resources
planning and management. Access or Oracle DBMS, Visual Basic, HTML, and/or
ASP (Java) can be used to define interfaces and develop a decision support system
with all elements of water resources including: supply, demand, institutions, knowl-
edge base, and conflict assessment. This system consists of six modules as follow:
• Database management
• Streamflow modeling and forecasting
• Demand management (modeling)
• Operating models
• Drought management
• Real-time operation
As can be seen in Figure 1.7, the various types of data and information — including
climatic, hydrologic, hydraulic, and physical characteristics of a water resources
system, as well as assumptions and the results of planning and operation alternatives
— are managed in the database management module. The streamflow modeling and
forecasting module considers statistical and conceptual forecast models, as well as
heuristics for forecast modification, by incorporating long-range climate signals.
Built-in evaluation of hydroclimatic data for improved water supply forecasting,
implementation of the probabilistic forecasts models, and correlation schemes of
weather and streamflow data are the major capabilities of this module. In the demand
management module the emphasis is on a reduction in the quantity of water that is
unaccounted for, optimal crop mixture, irrigation management policies, and optimal
policies for conjunctive use of surface and groundwater resources. Using these two
modules, the decision maker is able to develop forecast models for water supply
and demand and utilize them in real-time operation of water resources systems.
In the operating models module, optimization and simulation models are used
to develop long-term and mid-term planning and operation management policies.
The drought management module is designed in order to develop water resources
management policies during droughts. Because of the occurrence of severe droughts
in many countries in recent years, drought monitoring and management have been
given more attention by investigators. In the real-time operation module, optimal
policies and heuristics developed in different modules of the system will be applied
in the real-time operation of water resources systems. Through the use of these two
modules, the user will be able to define:
planning and management.
FIGURE 1.7 Framework of a decision support system (DSS) for sustainable water resources
Introduction
Conflict Resolution
Multiple Criterion
Decision Making Past Performance ARIMA Models
Deterministic Dynamic of the Models (Midrange)
Database Programming
Bayesian Decision
Management Theory National Weather Streamflow
Stochastic Dynamic Performance Service Modeling and
Programming Evaluation (Short Term) Forecasting
Fuzzy Set Theory Set of Operating
Neural Network or
Hydrologic and Operating Models Other Simulation Policies and NRCS Regression
Rule Curves
Climatic Data Techniques Tradeoffs Forecast
Statistical Model
Water Allocation Object Oriented
Schemes Programming Weekly
Hydraulic/ Genetic Algorithm Aggregation
Physical
Characteristics Decomposition/Linear User Interface
Aggregation Q+A Sessions
Water Demand
Snow Calculation
Data User Interface
Uncertainty in Climatic Conceptual Model
Parameters Forecast
Assumptions Rainfall-Runoff
Model
Uncertainties Due to Uncertainty of
Uncertainty of Power Generation Market Shifts or Water Supply and Develop
Demand Uncertainties Economic Parameters Forecast Forecast Series
User Interface (Q+A Sessions):

User Accepting, Modifying, and Rejecting the Model’s Recommendations
Elasticities of – Unaccounted Water
Develop Tradeoffs at Demand for – Leak Protection Surface Water Domestic and
Different Levels of Water – Public Awareness Industrial
Acceptable Risk Forecast of Supply –Quality Control Ground Water Demands
User Interface
and Demand User Interface
– Crop Mixture Demand
Water Water Optimum – Cropping Sequence Surface Water Agricultural
Conflict Management
Conservation Allocation Cropping –Irrigation Efficiency Demands (Modeling)
Resolution Ground Water
Schemes Schemes Mixture –Low Volume Irrigation
– Crop Net Water Demand
Demand – Market Prices
User Interface
Develop Demand Series Low Flow Augmentation Environmental
Instream Flow Demands
Water Quality Assessment Ground Water
Climatic Recharge
Drought Environmental Accounting
A Drought is
Forecasted
Drought Drought Hydrologic
Management Monitoring Drought Local Contingency Plans
for Floods
Contingency Regional A Decision(s)
Agricultural Plans
Drought Drought Drought Water Allocation Real Time
Analysis Forecasting National Schemes Operation
Water Develop Tradeoffs
Conservation Decisions at Different Levels
of Acceptable Risk User Interface
Schemes and Q+A Sessions
Rationing
User Interface
17
• Drought indicators
• Contingency plans for droughts and floods
• Water allocation schemes accepted by conflicting groups of water users
• Tradeoffs at different levels of acceptable risk
The DSS should incorporate recent advances in optimization under uncertainty and
simulation and be capable of:
• Evaluation of alternatives
• Showing the impact of policies when different interest groups are search-
ing for compromise solutions
• Incorporating uncertainty of supply and demand, uncertainty due to mar-
ket shifts, and uncertainty due to inability of models to perform
• Incorporating parameter uncertainty and uncertainty due to vagueness of
situations and/or boundaries
• Forecasting water supply and demand
• Surveying users’ perceptions
• Supporting a variety of cognitive skills and styles and the level of knowl-
edge of different users
• Assisting users to develop and use their own cognitive skills
• Developing schemes for changing public attitudes for sustainable water
management
• Identifying new and improved scientific information for changing the
market for water as an economic commodity
• Defining yardsticks to measure sustainability of the system with respect
to the decisions
The system should be sufficiently flexible to allow the use and adaptation by users
with various levels of experimental knowledge. Algorithms, heuristics, optimization,
and simulation models in the DSS should be utilized in such a way that they examine
the impact of various policies based on four different decision-making platforms:
• Decisions under certainty

• Decisions under risk
• Decisions under uncertainty
• Decisions under conflict
The DSS should also be empowered with a variety of techniques to identify the
most effective methodologies for continuous feedback and interactive stakeholder
input into the knowledge base. Different mathematical and computational tools
should be considered in the various modules of the system, such as:
• Use of Bayesian decision theory to incorporate new data into the proba-
bilistic decision process
• Use of fuzzy set theory to assign membership functions based on the
opinion of the stakeholder, the analyst, and the decision maker about a
given situation
Introduction 19
• Utilizing genetic algorithms and other optimization techniques to search

through alternatives/operating policies
• The use of neural networks and other simulation techniques to test the
performance of different alternatives or scenarios and to develop forecast
schemes
• The use of scenarios, object-oriented programming, multicriterion deci-
sion making, and simulation to test the ability of a decision-making system
based on how it reacts to a given emergency situation such as a drought
or flood
REFERENCES
Chow, V. T., Maidment, D. R., and Mays, L. W., Applied Hydrology, McGraw-Hill, New York,
1988.
Kottegoda, N. T. and Rosso, R., Statistics, Probability, and Reliability for Civil and Environ-
mental Engineers, McGraw-Hill, New York, 1997.
Loucks, D. P., Surface water resources systems, in Water Resources Handbook, Mays, L. W.,
Ed., McGraw-Hill, New York, 1996, chap. 15.
Loucks, D. P., Stedinger, J. R., and Haith, D. A., Water Resources Systems Planning and
Analysis, Prentice-Hall, Englewood Cliffs, NJ, 1981.
National Research Council, Opportunities in the Hydrologic Sciences, National Academy
Press, Washington, D.C., 1991.
Shahin, M., Van Oorschot, H. J. L., and De Lange, S. J. D., Statistical Analysis in Water
Resources Engineering, Applied Hydrology Monographs, Balkema, Rotterdam, 1993.
Shiklomanov, I., World fresh water resources, in Water in Crisis, Gleick, P., Ed., Oxford
University Press, New York, 1993, chap. 2.
Simonovic, S. P., Tools for water management: one view of the future, Water Int., 25(1),
76–78, 2000.
Stohr, E. A. and Konsynski, B. R., Information Systems and Decision Processes, IEEE
Computer Society Press, Los Alamitos, CA, 1992.
White, I. D., Mottershead, D. N., and Harrison, S. J. (1992), Environmental Systems: An
Introductory Text, Second ed., Chapman & Hall, New York.
World Resources Institute, World Resources: A Guide to the Global Environment, a joint
publication by The World Resources Institute, The United Nations Environment
Program, The United Nations Development Program, and The World Bank, Oxford
University Press, New York, 1998.
Yakowitz, S., Dynamic programming applications in water resources, Water Resources Res.,
18(4), 673–696, 1982.
Yeh, W., Reservoir management and operation models: a state-of-the-art review, Water
Resources Res., 21(12), 1797–1818, 1985.
2 Decision Making:
Optimization and
Conflict Resolution
In our private and professional lives, decisions must often be made, and decision-
making tools are commonly used in the quantitative sciences. This chapter gives an
introduction to the major methods and concepts of this very important scientific field.
2.1 INTRODUCTION
Decision making is the science of choice. Selecting the best technology for a
particular application, developing flood protection alternatives, and optimizing the
operation of a reservoir are all problems of choice. The first element of any decision-
making problem is the decision maker. For a particular problem, we might have a
single person who is responsible for deciding what to do, or several people or
organizations may be involved in the decision-making process. In the first case, we
have only one decision maker; in the second case, multiple decision makers. When
more than one decision maker is present, then they might have different priorities,
objectives, and desires, thus no decision outcome is likely to satisfy every decision
maker. In such cases, a collective decision has to be made, and the outcome depends
on how the different decision makers take the interests of others into account. In
other words, the outcome depends on their willingness to cooperate with each other.
The options from which the selection is made are called alternatives and the set
of all possible alternatives is called the decision space. In many cases, the decision
space has a finite number of elements. For example, selecting a technology from
four possibilities results in a decision space with four elements. In many other cases,
the decision alternatives are characterized by decision variables that represent certain
values about which we are making a decision. For example, reservoir capacity can
be any real value between the smallest and largest possibility. If the decision space
is finite, then the decision-making problem is discrete, and if the decision alternatives
are characterized by continuous variables then the problem is continuous.
All decision-making problems are faced with limitations resulting from financial
considerations and technological, social, and environmental restrictions, among oth-
ers. These limitations are called the constraints of the problem. The alternatives
satisfying all constraints are feasible, and the set of all feasible alternatives is the
feasible decision space. If the decision space is finite, then the construction of the
feasible decision space is very simple. We have to check the feasibility of each
21
alternative by determining whether or not it satisfies all restrictions. If the alternatives

are characterized by decision variables, the constraints are usually presented as
certain equations or inequalities containing the decision variables.
Because we have to make a choice from a given set of feasible alternatives we
need to measure how good those alternatives are. The goodness of any alternative
can be characterized by its attributes. Most of these attributes can be described
verbally, such as how clean the water is or how well it satisfies customers. In
mathematical modeling, we need measurable quantities describing the goodness of
each alternative with respect to the attributes. Such measures are the criteria. For
example, water quality can be measured by pollutant concentrations, consumer
satisfaction by supplies, and so on. In all decision problems, we want to accomplish
or avoid certain things. To what degree we accomplish our goals and avoid unfa-
vorable consequences should be among the criteria. In classical optimization models
we usually have only one criterion to optimize; however, in most decision-making
problems we are faced with several criteria that frequently can conflict with each
other. We can assume that all criteria are maximized, because otherwise the criterion
can be multiplied by –1. For example, cost and quality in any problem are conflicting
criteria, as better quality requires higher cost. When only one criterion is involved,
the problem is a single-criterion optimization problem. In the presence of more than
one criteria, we have a multiple-criteria decision problem. In the case of multiple
decision makers, we might consider the problem as multiple-criteria decision mak-
ing, where the criteria of all the decision makers are considered to be the criteria of
the problem.
2.2 SINGLE-CRITERION OPTIMIZATION

In this section the fundamentals of single-criterion decision-making are briefly
outlined. For mathematical simplicity, let us introduce the following notations. Let
X denote the feasible decision space, and let x ∈ X be any feasible alternative. If X
is a finite set, then the value of X can be any one of the number 1, 2, …, N, when
N is the number of the elements of X. If X is a continuous set characterized by
continuous decision variables, then x is a vector, x = (x1, …, xn), where n is the
number of the decision variables and x1, x2, …, xn are the decision variables. If f
denotes the single criterion, then we assume that f is defined on X and, for all x ∈
X, f(x) is a real number. This condition is usually defined as f : X → R. Because f
is real valued, all possible values of f(x) can be represented on the real line, and the
set:
{
H = f ( x ) x ∈ X}
is called the criterion space. Notice that the elements of X show the decision choices,
and the elements of H give the consequences of the choices in terms of the values
of the evaluation criterion. We can simply say that X shows what we can do and H
shows what we can get.
Decision Making: Optimization and Conflict Resolution 23
If X is finite and the number of its elements is small, then by comparing the f(x)
values the best alternative can be easily identified. If the number of elements of X
is large or even infinity but X is a discrete set (for example, when it is defined by
decision variables with integer values), then the methods of discrete and combina-
torial optimization are used. This methodology can be found in any textbook of
advanced optimization techniques.
The most simple single-criterion optimization problem with continuous variables
is linear programming. In order to give a general formulation, let n be the number
of the decision variables and x1, x2, …, xn be the decision variables. It is assumed
that the criterion and all constraints are linear. The criterion is sometimes also called
the objective function, and it is assumed to have the form:
f ( x ) = c1 x1 + c2 x2 + ... + cn xn = c T x (2.1)
where c1, c2, …, cn are real numbers, c = (c1, …, cn), and
⎡ x1 ⎤
⎢x ⎥
⎢ 2⎥
x=⎢⋅ ⎥ (2.2)
⎢ ⎥
⎢⋅⎥
⎢⎣ xn ⎥⎦
It is also assumed that all constraints are linear and all decision variables are
nonnegative. If any of the decision variables have only negative values, then by
introducing the new variable:
xi− = − xi (2.3)
it becomes nonnegative. If a decision variable can be both positive and negative,

then we can introduce two new variables:
⎧⎪ xi if xi ≥ 0
xi+ = ⎨ (2.4)
⎪⎩0 otherwise
and
⎧⎪− xi if xi < 0
xi− = ⎨ (2.5)
⎪⎩0 otherwise,
and then both of them are nonnegative, and
xi = xi+ − xi− (2.6)
Linear constraints are equalities or inequalities. Equality constraints have the

general form:
ai1 x1 + ai 2 x2 + .... + ain xn = bi (2.7)
and inequality constraints can be written as:
≥
ai1 x1 + ai 2 x2 + .... + ain xn or bi (2.8)
≤
Next we show that all constraints can be rewritten as less than or equal to (≤)
types of conditions. If it is a ⬉-type constraint, then we have nothing to do. Con-
straints with ≥ inequalities can be multiplied by –1, and then the ≥ inequality changes
to ≤. Equality constraint (2.7) is equivalent to two inequality constraints:
ai1 x1 + ai 2 x2 + .... + ain xn ≤ bi
ai1 x1 + ai 2 x2 + .... + ain xn ≥ bi
where the second relation can be multiplied by –1 to obtain:
− ai1 x1 − ai 2 x2 − .... − ain xn ≤ − bi
By repeating the above procedure for all variables and constraints, if necessary,
the resulting problem will have only nonnegative variables and ≤ type of inequalities
in the constraints. Hence, any linear programming problem can be rewritten in this
way to its primal form:
Maximize c1 x1 + c2 x2 + ⋯ + cn xn
Subject to x1, x2 ,⋯, xn ≥ 0

a11 x1 + a12 x2 + .... + a1n xn ≤ b1
a21 x1 + a22 x2 + .... + a2 n xn ≤ b2 (2.9)
⋮
am1 x1 + am 2 x2 + .... + amn xn ≤ bm
Note that when the objective function represents cost or contained pollutant amount,
for example, smaller objective values represent better choices. In such cases, the
objective function is minimized. In the above formulation, we assume maximization,
as minimization problems can be rewritten into maximization problems by multi-
plying the objective function by –1.
In the very simple case of two decision variables, Eq. (2.9) can be solved by
the graphical approach. As an illustration, consider the following example.
Example 2.1
The combination of three technologies is used to remove a certain pollutant from

wastewater. The three technologies remove 1, 2, and 3 g/m3 of the pollutants, respec-
tively. The third technology variant seems to be the best, but it cannot be applied to
more than 50% of the wastewater being treated. The costs of applying the technology
variants are $5, $3, and $2 per cubic meter. If 1000 m3 must be treated in a day, and
at least 1.5 g/m3 of pollutant has to be removed, then a simple linear programming
formulation can be used to model this optimization problem. Let x1 and x2 be the
amount of wastewater (in m3) for which technologies 1 and 2 are used. Then, 1000
– x1 – x2 m3 is the amount to which technology 3 is applied. Clearly,
x1, x2 ≥ 0 (2.10)
x1 + x2 ≤ 1000 (2.11)
The condition that the third technology cannot be used in more than 50% of the
treated wastewater requires that
1000 − x1 − x2 ≤ 500
that is,
x1 + x2 ≥ 500 (2.12)
The total removed pollutant amount (in grams) is:
x1 + 2 x2 + 3(1000 − x1 − x2 )
which has to be at least 1.5 g/m3 of the total 1000 m3 being treated, so:
x1 + 2 x2 + 3(1000 − x1 − x2 ) ≥ 1500
This inequality can be rewritten as:
2 x1 + x2 ≤ 1500. (2.13)
The total cost is given as
5 x1 + 3 x2 + 2(1000 − x1 − x2 ) = 3 x1 + x2 + 2000
Because 2000 is a constant term, it is sufficient to minimize 3x1 + x2 or maximize

–3x1 – x2. Constraint (2.12) is a ≥ type of relation which is multiplied by –1 in order
to reduce it to a ≤ type of constraint. After this modification, we obtain the following
primal-form problem:
Maximize f = −3 x1 − x2
Subject to x1, x2 ≥ 0
x1 + x2 ≤ 1000
(2.14)
− x1 − x2 ≤ −500
2 x1 + x2 ≤ 1500
Each constraint can be represented in the two-dimensional space as a half-plane, the

boundary of which is determined by the straight line representing the constraint as
an equality. For example, the equality x1 + x2 = 1000 is a straight line, with the x1
intercept = 1000 and the x2 intercept = 1000. Because the origin, the (0,0) point,
satisfies the inequality, the half-plane that contains the origin represents the points
that satisfy this inequality constraint. The intercept of the half-planes of all con-
straints gives the feasible decision space, as is shown in Figure 2.1. Notice that this
set has five vertices: (500,0), (750,0), (500,500), (0,1000), and (0,500) and the set
is the convex hull, the smallest convex set containing the vertices.
From the theory of linear programming, we know that if the optimal solution is
unique, then it is a vertex, and in the case of multiple optimal solutions a vertex
always exists among the optimal solutions. Therefore, it is sufficient to find the best
vertex that gives the highest objective function value. Simple calculation shows that:
f(500,0) = –1500
f(750,0) = –2250
f(500,500) = –2000
f(0,1000) = –1000
f(0,500) = –500
and because –500 is the largest value, the optimal decision is x1 = 0, x2 = 500, and
1000 – x1 – x2 = 500. The minimal cost is, therefore, 3x1 + x2 + 2000 = $2500.
X2
(0,1500)
(0,1000)
(0,500) (500,500)
X1
(500,0) (750,0) (1000,0) (1500,0)
X 1+X2=1000
X 1+X2=500
2X 1+X2=1500
FIGURE 2.1 Illustration of the feasible decision space.
In the case of more than two decision variables, the graphical representation of
the feasible decision space is impossible. A computational method known as the
simplex method performs a systematic search on the vertices of the feasible decision
space until the optimal vertex is found.
The concept of the simplex method can be summarized as follows: Any vertex
of the feasible decision space corresponds to a basis of the column space of the
coefficient matrix of the linear constraints. The variables associated with the col-
umns of the basis are called basic variables. It can be shown that moving from one
vertex to any of its neighbors is equivalent to exchanging one of the basic variables
with a non-basic variable. In the simplex method, it is always guaranteed that in
each step the objective function increases, which can be ensured by the following
conditions. Let aij be the (i,j) element of the coefficient matrix; bi , the ith right-hand
side number; and cj, the coefficient of xj in the objective function. Then, xi is replaced
by xj if cj > 0, aij > 0, and
bi ⎧⎪ b ⎫⎪
= min l ⎨ l ⎬
aij ⎩⎪ alj ⎭⎪
An initial basic can be found by introducing artificial variables to all ≥ and = type
of constraints. For ≤ type of constraints, the slack variables play the same roles. In
the first phase of the simplex method, all artificial variables are removed from the
basis by successive exchanges, after which we have an initial feasible basic solution
that is then successively improved in the second phase of the simplex method until
optimum is reached. Many professional software packages implement the simplex
method, so it is often used in practical applications.
In solving large-scale linear programming problems, we often encounter spe-
cially structured coefficient matrices. The most common structure is as follows:
⎛ A0,1 A0,2 ⋯ A0,k −1 A0,k ⎞

⎜ A1,1 ⎟
⎜ ⎟
⎜ A2,2 ⎟
⎜ ⋱ Ak −1,k −1 ⎟
⎜⎜ ⎟
⎝ Ak ,k ⎟⎠
where Ai, j (0 ≤ i ≤ k ,1 ≤ j ≤ k ) are given matrices. When applying the simplex

method in such cases, we can take advantage of the special structure, and in each
simplex step (moving to a neighboring vertex) we are able to preserve this special
structure.
Another special method is the decomposition technique, where the optimal
solution of the large problem can be formulated by using a sequence of the solutions
of much smaller problems, the sizes of which are determined by the sizes of the
blocks Ai,i (i = 1, 2, …, k). The most popular decomposition technique is the
Dantzig–Wolf method, a discussion of which can be found in almost all texts of
optimization. The classical text of Dantzig (1963) is a good source for further
reading.
If the objective function and one or more constraints are nonlinear, then the
problem becomes one of nonlinear programming. In the case of only two variables,
they can be solved by the graphical approach; however, the feasible decision space
may not have vertices, and even if it does the optimal solution might not be a vertex.
In such cases, the curves of the objective function with different values have to be
compared. This procedure is shown next.
Example 2.2
For illustrative purposes, consider the previous example but assume that instead of
minimizing the linear function 3x1 + x2, a nonlinear function:
f ( x ) = x12 + x22
is minimized. In order to find the feasible solution that minimizes this objective
function, we first illustrate the solutions, which give certain special objective function
values. The points with zero objective function values satisfy the equation:
f ( x ) = x12 + x22 = 0
when the only solution is x1 = x2 = 0, or the origin. Points with an objective function
value of 10,000 satisfy the equation:
f ( x ) = x12 + x22 = 10, 000
which is a circle with the origin at its center and a radius of 100. This circle has no
feasible point, so 10,000 is a too small value for the objective function. By increasing
the value of f(x), we always get a similar circle with the same center, but with
increasing radius (see Figure 2.2). By increasing the radius of the circle, we see that
the smallest feasible radius occurs when the circle and the line x1 + x2 = 500 just
touch each other. The tangent point is (250,250), which is therefore the optimal
solution of the problem. In more complicated and higher dimensional cases, com-
puter software must be used, and several professional packages are available to
perform this task.
The most popular nonlinear optimization method is linearization, where all
nonlinear constraints as well as the objective function are linearized if necessary.
The resulting linear programming problem is then solved by the simplex method.
Let g(x1, …, xn) be the left-hand side of a constraint or the objective function.
Furthermore, let ( x10, …, xn0 ) be a feasible solution. Then, the linear approximation
of g is given by the linear Taylor’s polynomial:
∑ ∂x ( x ,..., x )( x − x )
∂g
g( x1,..., xn ) ≈ g( x10 ,..., xn0 ) + 0
1
0
n r
0
r
r =1 r
If all nonlinear constraints and, if necessary, the objective function are linearized
around the point ( x10, …, xn0 ) and the optimal solution of the resulting linear pro-
gramming problem is ( x11, …, xn1 ), then the original problem is linearized again
around this solution and the new linear programming problem is solved. In this
way, an iteration procedure is obtained, and it is known from the theory of
nonlinear programming that the solutions of the successive linear programming
problems converge to the solution of the original nonlinear optimization problem
X2
1500
1000
500
X1
500 1000 1500
FIGURE 2.2 Illustration of nonlinear optimization.
under reasonable conditions. For more details of nonlinear programming, see, for
example, Hadley (1964) or Mangasarian (1962).
2.3 MULTIPLE-CRITERIA OPTIMIZATION

As in the previous session, let X denote the feasible decision space, and let (f1, …, fI )
be the criteria. Set X reflects the decisions that can be made; however, in the presence
of multiple criteria we need to represent the set of all possible outcomes. The set:
H = {( f1 ( x ), ⋯, fI ( x )) x ∈ X} (2.15)
is called the criteria space and shows what we can get by selecting different
alternatives.
Example 2.3
Assume that selection has to be made from four available technologies to perform
a certain task. These technologies are evaluated by their costs and convenience of
TABLE 2.1
Data for Discrete Problem
Technology Cost Convenience
1 –3.0 70
2 –3.5 90
3 –3.8 95
4 –3.2 85
FIGURE 2.3 Criteria space in the discrete case.
usage. The cost data are given in $1000 units, and the convenience of usage is
measured in a subjective scale between 0 and 100, where 100 is the best possible
measure. The data are given in Table 2.1. We multiplied the cost data by –1 in order
to maximize both criteria.
Similar to the single-criterion case, the criteria are sometimes called the objective
functions. The criteria space has four isolated points, in this case as shown in
Figure 2.3. It is clear that none of the points dominates any other point, as lower
cost is always accompanied by lower convenience; that is, increasing any objective
should result in decreasing the other one. Therefore, any one of the four technologies
could be accepted as the solution.
Example 2.4
Consider now a modified version of our earlier first example. Assume that two
pollutants are removed by a combination of three alternative technologies. These
technologies remove 3, 2, and 1 g/m3 of the first pollutant and 2, 1, and 3 g/m3 of
the second pollutant. The total amount of the two pollutants removed is:
3 x1 + 2 x2 + 1(1000 − x1 − x2 ) = 2 x1 + x2 + 1000
and
2 x1 + x2 + 3(1000 − x1 − x2 ) = − x1 − 2 x2 + 3000
If no condition is given on the usage of the third technology, no constraint on the

minimum amount to be removed exists, and no cost is considered, then this problem
can be mathematically formulated as:
Maximize f1 ( x ) = 2 x1 + x2 , f2 ( x ) = − x1 − 2 x2
(2.16)
x1 + x2 ≤ 1000.
The feasible decision space is shown in Figure 2.4.

The criteria space can be determined as follows. Because:
f1 = 2 x1 + x2 and f2 = − x1 − 2 x2 .
we have
2 f1 + f2 − f − 2 f2
x1 = and x2 = 1 .
3 3
The constraints x1, x2 ≥ 0 can be rewritten as:
2 f1 + f2 − f1 − 2 f2
≥ 0 and ≥0
3 3
that is,
2 f1 + f2 ≥ 0 and f1 + 2 f2 ≤ 0.
X2
1000
X
X1
1000
FIGURE 2.4 Feasible decision space for Example 2.4.
The third constraint, x1 + x2 ≤ 1000 now has the form:
2 f1 + f2 − f1 − 2 f2
+ ≤ 1000
3 3
that is, f1 – f2 ≤ 3000. Hence, set H is characterized by the following inequalities:
2 f1 + f2 ≥ 0
f1 + 2 f2 ≤ 0 (2.17)
f1 − f2 ≤ 3000
and is illustrated in Figure 2.5. Notice that H is the triangle with vertices (0,0),
(1000, –2000), and (2000,–1000), and the points on the linear segment connecting
the vertices (0,0) and (2000,–1000) cannot be dominated by any point of H because
none of the objectives can be improved without worsening the other. This, however,
does not hold for the other points of H when both objectives can be improved
simultaneously. Therefore, all points between (0,0) and (2000,–1000) can be
accepted as the best solutions.
f2
3000
(0,0)
f1
H (2000,-1000)
(1000,-2000)
-3000
FIGURE 2.5 Criteria space for Example 2.4.
Any point of f ∈ H is called dominated if there is another point f 1 ∈ H such

that f 1 ≥ f and there is strict inequality in at least one component. Similarly, a
point f ∈ H is called nondominated if there is not another f1 ∈ H. In multiple-
criteria decision making, nondominated points serve as the solutions of the problem.
In the case of single-objective optimization problems, optimal solutions are selected.
Optimal solutions are always better than any nonoptimal solution, and in the case
of several optimal solutions they are equivalent to each other in the sense that they
give the same objective function value. In the case of nondominated solutions, these
properties no longer are valid. Consider Figure 2.5, where the point (0,0) is non-
dominated but is no better than the point (1000, –2000), which is dominated, because
1000 > 0, thus providing higher value in the first objective function. Also, the points
(0,0) and (2000,–1000) are both nondominated and differ in both objective values.
If a single-objective optimization problem has multiple optimal solutions, it does
not matter which one is selected, as they give the same objective function value. In
multiple-criteria problems, the different nondominated solutions give different objec-
tive function values, so it is very important to decide which one of them is chosen,
as they lead to different outcomes. To single out a special point from the set of
nondominated solutions, additional preference information is needed from the deci-

sion makers to decide on trade-offs between the objectives. For example, in Figure
2.5 both (0,0) and (2000,–1000) are nondominated. We can say that (0,0) is better
than (2000,–1000) if a 1000-unit loss in the second objective is not compensated
by a 2000-unit gain in the first objective. This kind of decision cannot be made
based solely on the concepts discussed before. Depending on the additional prefer-
ence and/or trade-off information obtained from the decision-makers, different solu-
tion concepts and methods are available. These topics are the subjects of the fol-
lowing discussions.
2.3.1 SEQUENTIAL OPTIMIZATION

This method is based on the ordinal preference order of the objectives. Assume that
the objectives are numbered so that fI is the most important, f2 is the second most
important, and fI is the least important. In applying the method, first f1 is optimized.
If the optimal solution is unique, then the procedure terminates regardless of the
values of the less important other objectives. Otherwise, f1 is kept on its optimal
level and f2 is optimized. If there is a unique optimal solution, then it is selected as
the solution of the problem; otherwise, both f1 and f2 are kept optimal, and f3 is
optimized, and so on, until there is a unique optimal solution or fI is already
optimized.
Example 2.5
Consider again the problem given in Example 2.3 and represented by Table 2.1. If
cost is more important than convenience, then the least expensive alternative, which
is technology 1, is selected. If convenience is more important, then the third tech-
nology alternative is chosen. Note that the other two nondominated technologies,
alternatives 2 and 4, cannot be chosen by this method.
Example 2.6
In the case of Example 2.4 and Figure 2.5 we have a similar situation. If f1 is more
important than f2, then point (2000,–1000) is the choice with decision variables:
2 f1 + f2 − f − 2 f2
x1 = = 1000 and x2 = 1 =0
3 3
If f2 is the more important objective, then the point (0,0) is the choice, and the
decision:
2 f1 + f2 − f − 2 f2
x1 = = 0 and x2 = 1 =0
3 3
is made, in which case only the third alternative pollutant removal technology is used.
2.3.2 THE ε -CONSTRAINT METHOD

In the case of sequential optimization, the procedure often terminates before all
objectives are optimized. Therefore, less important objectives may have very unfa-
vorable values. In order to avoid such possibilities, the following method can be
utilized. Assume that the most important objective is specified, and minimal accept-
able levels are given for all other objectives. If f1 is the most important objective,
and (ε2, ε3, …, εI) are the minimum acceptable levels, then the solution is obtained
by solving the following single-objective optimization problem:
Maximize f1 ( x )
Subject to x ∈X
f2 ( x ) ≥ ε 2 (2.18)
⋮
fI ( x ) ≥ ε I
Example 2.7
In the case of Example 2.3, assume that cost is more important than convenience,
but we want to have at least 90 in convenience. This means that technologies 1 and
4 are not feasible, but technologies 2 and 3 both satisfy these minimal requirements.
The less expensive among these two remaining alternatives is technology 2, so that
is our choice.
Example 2.8
Assume that in Example 2.4, f1 is the more important objective, but we require that
f2 ≥ –500. With this additional constraint, the criteria space is further reduced to its
subset, which has only such points that satisfy this additional condition. The reduced
set is shown in Figure 2.6. If f1 is maximized on the reduced criteria space, then the
point (1000,–500) is obtained with decision variables:
2 f1 + f2 − f − 2 f2
x1 = = 500 and x2 = 1 =0
3 3
That is, pollutant removal technologies 1 and 3 are used at an equal rate.
2.3.3 THE WEIGHTING METHOD

When applying this method, all objectives are taken into account by importance
weights supplied by the decision makers. Assume that (c1, c2, …, cI) are the relative
importance factors of the objectives which are specified by the decision makers on
FIGURE 2.6 Reduced criteria space for Example 2.8.
some subjective basis or are obtained by some rigorous method such as pair-wise
comparisons. It is assumed that ci > 0 for all i, and
∑c = 1 .
i =1
i
When applying the weighting method, the composite objective is a weighted average
of all objectives, where the weights are the relative importance factors (c1, …, cI ).
Hence, the single objective optimization problem:
Maximize c1 f1 ( x ) + c2 f2 ( x ) + ... + cI fI ( x )
Subject to x ∈X (2.19)
is solved. The method in this form presents a major difficulty. First, the different
objectives have different units. For example, cost is measured in dollars, convenience
ui
f1
0 mi Mi
FIGURE 2.7 Utility function of objective fi.
is given as a subjective measure, pollutant removal is given in g/m3, and so on.

Therefore, the composite objective has no practical meaning, as we are adding apples
and oranges. The second difficulty is based on the observation that, by changing the
unit of any of the objectives, we multiply its weight by the conversion constant so
the optimal solution usually changes. Both difficulties can be overcome by trans-
forming all objectives into a common dimensionless satisfaction scale. This trans-
formation is called the normalizing procedure, and it can be done by a simple linear
transformation or by using special utility functions supplied by the decision makers.
The simple linear normalizing procedure is outlined first. Let fi denote an objec-
tive function. Let mi and Mi denote its smallest and largest possible values, respec-
tively. These values may be computed by simply minimizing and maximizing fi over
the feasible decision space X, or their values can be supplied by the decision makers
based on subjective judgments. The normalized objective function can then be
obtained as:
fi ( x ) − mi
fi ( x ) = (2.20)
Mi − mi
In many practical cases, the decision makers are able to provide a utility function
to each objective that shows their satisfaction levels for the different values of that
objective. Such a function is illustrated in Figure 2.7, where mi gives the least
satisfaction level (0) and Mi gives the highest satisfaction level (1, or 100%). Then,
objective fi is replaced by the corresponding utility value:
f i ( x ) = ui ( fi ( x )) (2.21)
Notice that Eq. (2.20) corresponds to the linear utility function:
ui ( fi ) = ( fi − mi ) / ( Mi − mi )
We always require that the lowest possible utility value be 0 and the highest 1,
so fi ( x ) ∈[0,1] for all x ∈ X. Then, Eq. (2.19) can be modified:
Maximize c1 f 1 ( x ) + c2 f 2 ( x ) + ... + cI f I ( x )
Subject to x ∈X (2.22)
In this new composite objective, only dimensionless values are added, and the linear
combination of the normalized objectives can be interpreted as an average satisfac-
tion level based on all objectives.
Example 2.9
Assume that in Example 2.3 the decision makers consider $4000 as the worst
possible cost, and they believe that a cost of $3000 would give them 100% satis-
faction. Then m1 = –4, and M1 = –3, and, if linear utility function is assumed between
these values, then the linearization is based on the linear transformation:
f1 + 4
f 1=
−3 + 4
so the cost values in Table 2.1 must be modified accordingly. The convenience data
are already given in satisfaction levels (in %), so these numbers are divided by 100
to transform them into the unit interval [0,1]. The normalized table is provided in
Table 2.2.
TABLE 2.2
Normalized Objectives for Discrete Problem
Technology Cost Convenience
1 1.0 0.70
2 0.5 0.90
3 0.2 0.95
4 0.8 0.85
Assume that equal importance is given to the objectives; that is c1 = c2 = 0.5.

Then, the composite objective values for the four technology variants are as follows:
0.5(1) + 0.5(0.7) = 0.85
0.5(0.5) + 0.5(0.9) = 0.70
0.5(0.2) + 0.5(0.95) = 0.575
0.5(0.8) + 0.5(0.85) = 0.825.
Thus, the first alternative is the choice.
Example 2.10
Consider again the problem outlined in Example 2.4. From Figure 2.5, we see that
the smallest values of the objectives are m1 = 0 and m2 = –2000; the largest values
are M1 = 2000 and M2 = 0. If the simple normalization is performed, then the new
objectives are:
f1 ( x ) f ( x ) + 2000
f 1( x ) = and f 2 ( x ) = 2
2000 2000
The modified criteria space can be obtained by transforming the vertices and con-
structing the convex hull generated by the new vertices. Notice that by using the
above transformation we have:
(0, 0) ֏ (0,1)
(1000, −2000) ֏ (0.5, 0)
(2000, −1000) ֏ (1, 0.5)
These new vertices and the transformed criteria space H are illustrated in Figure
2.8. Assume next that equal importance is assigned to the objectives c1 = c2 = 0.5.
Then, we minimize 0.5 f1 + 0.5 f2 over the set H. It is easy to see that the vertex
(1,0.5) gives the highest composite objective value. The values of the original
objectives can be easily obtained by solving the following equations:
f1 f + 2000
f1 = = 1 and f 2 = 2 = 0.5
2000 2000
which yields f1 = 2000 and f2 = –1000. The decision variables are then obtained:
2 f1 + f2 − f − 2 f2
x1 = = 1000 and x2 = 1 =0
3 3
f2
(0,1)
0.5
H (1,0.5)
f1
(0.5,0) 1
FIGURE 2.8 Transformed criteria space for Example 2.4.
These equations for x1 and x2 were defined earlier in Example 2.4.
2.3.4 DIRECTION-BASED METHOD

In many practical cases, the decision makers are able to specify a point in the criteria
space (which might not be feasible in some cases) and a direction in which the
objectives have to be improved. The given point might be the current state of the
system that has to be improved, or it can be some subjective worst possibility. If f *
is the given point and ∨ > 0 is the direction of improvement, then this concept can
be mathematically modeled as the following single-objective optimization problem:
Maximize t
(2.23)
Subject to f * + t∨ ∈ H
The graphical representation of this method is shown in Figure 2.9. Starting from
point f *, we move in direction of ∨ as far as possible. This method has two major
difficulties. First, the obtained solution might be dominated, so the result cannot be
accepted in such cases. Figure 2.10 shows this possibility. Second, this method
f2
Optimal Solution
f1
v
-
-f *
H
FIGURE 2.9 Illustration of the direction-based method.
cannot be applied for discrete problems, as there is usually no other feasible point
along the path we move in the improvement.
Example 2.11
Consider again the criteria space shown in Figure 2.5. Assume f * is selected as the
vector having the smallest values of the objectives in the components; f * =
(0,–2000). If the two objectives have to be improved equally, then ∨ = (1,1). The
graphical solution of problem (2.23) is shown in Figure 2.11. The solution is the
intercept of the two lines with equations:
f2 = f1 − 2000 and f1 + 2 f2 = 0
f2
f1
H -f *
v
- optimal solution that
is dominated
FIGURE 2.10 Illustration of the direction-based method.
It is easy to see that:
4000 2000
f1 = and f2 = −
3 3
and the decision variables are:
2 f1 + f2 − f − 2 f2
x1 = ≈ 666.67 and x2 = 1 =0
3 3
Direction-based methods are sometimes applied when an ideally best point is

known, which is not feasible. In that case, we start relaxing the values of the objective
functions in a given direction until a feasible point is reached. Let f * denote the
f2
1000 2000
f1
Optimal Solution
-1000
-2000
-f *
FIGURE 2.11 Solution by a direction-based method.
given point and ∨ > 0 be the given direction of relaxation. Equation (2.23) is now
modified as:
Minimize t
(2.24)
Subject to f * − t∨ ∈ H
The graphical representation of this method is shown in Figure 2.12.
Example 2.12
In the case of the previous example select f * = (2000,0) and ∨ = (1,1). Then, the
line of relaxation coincides with the line of improvement, so the optimal solution
is the same as in the previous example.
2.3.5 DISTANCE-BASED METHODS

Assume now that an ideally best point f * is computed or obtained from the decision
makers. If f * is feasible, then the corresponding decision variables are obtained by
solving the equations:
f ( x) = f *
f2
-f *
-v
-
Optimal Solution
f1
H
FIGURE 2.12 Illustration of problem (2.24).
Otherwise, we are looking for the feasible point that has minimal distance from the
ideal point. If ρ( f ( x ), f * ) denotes the distance between points f (x) and f *, then
this concept can be mathematically modeled as the single-objective optimization
problem:
Minimize ρ( f ( x ), f ∗ )
(2.25)
Subject to x ∈X
The graphical illustration of this method is very similar to Figure 2.12, where the
optimal solution has the smallest distance from the ideal point. In practical applica-
tions, one of the following distances is usually used:
I
∗
ρ1 ( f ( x ), f = ∑ c f ( x) − f
i =1
i i i
∗
(2.26)
I
∗
ρ2 ( f ( x ), f ) = ∑ c ( f ( x) − f )
i =1
i i i
∗ 2
(2.27)
and
{
ρ∞ ( f ( x ), f ∗ ) = max ci fi ( x ) − fi∗
i
} (2.28)
where c1, c2, …, cI are given positive weights and f1* , f2* ,…, fI * are the components
of f *. The ρ1 distance represents full compensation among the objectives, as a unit
increase in | fi ( x ) − fi* | can be compensated by ci /cj units of | f j ( x ) − f j * | in order
to have the same distance. In the case of distance, ρ2 partial compensation is
assumed. The compensating amounts are proportional to the quantities, and in the
case of ρ∞ no compensation is assumed, as only the largest value of ci | fi ( x ) − fi* |
affects the distance. Smaller values do not count, so they can have any smaller values
without changing the distance. Because weights are included in the distance func-
tions, the objective functions are usually normalized in a way similar to the weighting
method.
Example 2.13
Consider again the normalized discrete problem of Example 2.9. Assume f * = (1,1)
and equal weights are selected. By selecting the ρ1 distance we have:
ρ1 ( f (1), f ∗ ) = 0.51 − 1 + 0.51 − 0.7 = 0.15
ρ1 ( f (2), f ∗ ) = 0.51 − 0.5 + 0.51 − 0.9 = 0.3
ρ1 ( f (3), f ∗ ) = 0.51 − 0.2 + 0.51 − 0.95 = 0.425
ρ1 ( f ( 4), f ∗ ) = 0.51 − 0.8 + 0.51 − 0.85 = 0.175
The smallest distance occurs in the first case, so the first alternative is selected.
Example 2.14
We will consider now the normalized criteria space shown in Figure 2.8. Let the
ideal point be selected again as f * = (1,1), and let us choose the ρ∞ distance with
equal weights. In this case, we have the following nonlinear programming problem:
⎧ 2x + x 2000 − x1 − 2 x2 ⎫
Minimize max ⎨1 − 1 2 , 1 − ⎬
⎩ 2000 2000 ⎭
x1 + x2 ≤ 1000
where we used the normalizing equations derived in Example 2.10. This is a non-
linear optimization problem; however, we are able to rewrite it as a higher dimen-
sional linear programming problem. If we introduce the variable M for the objective
function, then:
2 x1 + x2 2000 − x1 − 2 x2
1− ≤ M and 1 − ≤M
2000 2000
We can eliminate the absolute values by having:
2 x1 + x2 2000 − x1 − 2 x2
−M ≤ 1 − ≤ M and − M ≤ 1 − ≤M
2000 2000
These inequalities can be further simplified as:
2 x1 + x2 − 2000 M ≤ 1
2 x1 + x2 + 2000 M ≥ 1
x1 + 2 x2 + 2000 M ≥ −1
x1 + 2 x2 − 2000 M ≤ −1
Therefore, the above nonlinear optimization problem is equivalent to the three-

dimensional linear programming problem:
Minimize M
x1 + x2 ≤ 1000
2 x1 + x2 − 2000 M ≤ 1
2 x1 + x2 + 2000 M ≥ 1
x1 + 2 x2 + 2000 M ≥ −1
x1 + 2 x2 − 2000 M ≤ −1
The application of the simplex method gives the optimal solution:
6000
x1 = ≈ 666.67 and x2 = 0
9
In some applications, an ideally worst point is given and the feasible point that
is as far as possible from this point in the criteria space is selected. If f * denotes
this point, then this concept can be mathematically modeled as follows:
Maximize ρ( f ( x ), f ∗ )
Subject to x ∈ X (2.29)
where ρ is any distance of I-dimensional vectors.

Finally, we mention that a comprehensive summary of multiple-criteria decision-
making methods can be found, for example, in Szidarovszky et al. (1986).
2.4 GROUP DECISION MAKING

In this subsection, we assume multiple decision makers, with each decision maker
having a single objective function to maximize. One way of solving such problems
is to consider the problem as a multiple-criteria decision problem, where the objec-
tives of the different decision makers form the multiple objective of the problem. If
the decision makers have equal importance, then equal weights are selected; other-
wise, the importance, or powers, of the decision makers are the weights.
Another approach is based on social choice procedures. Assume that I decision
makers have to choose a commonly acceptable alternative from a finite set. Assume
also that each decision maker has his own ranking of the alternatives. The data then
can be presented in a table, where the rows correspond to the decision makers and
the columns show the decision alternatives. Each row of the table represents the
rankings of the decision maker, so each row has a permutation of the integers 1, 2,
…, n, where n is the number of alternatives. Such a table is shown in Table 2.3. The
last column gives the weights of the decision makers. We assume again that:
ci ≥ 0 and ∑c = 1
i =1
i
In the case of plurality voting, we compute for each alternative the total weight
for those decision makers who consider the alternative to be the best, and the alter-
native with the largest such number is selected as the common choice. Mathematically,
this method can be described as follows. For each alternative, we first compute:
I
Ak = ∑ c f (a ) (k = 1,…, n)
i =1
i ik (2.30)
TABLE 2.3
Input for Social Choice
Alternatives
Decision Makers 1 2 … n Weights
1 a11 a12 … a1n c1

2 a21 a22 … a2n c2
⋮ ⋮ ⋮ … ⋮ ⋮
I aI1 aI2 … aIn cI
where
⎧⎪1 if aik = 1
f ( x) = ⎨
⎪⎩0 otherwise,
and then alternative k0 is selected if:
Ak0 = max{ Ak } (2.31)

k
Notice that this method considers only first rankings.

The Borda count takes all rankings into consideration. For each alternative we
first compute the average weighted ranking of each alternative, and the one with the
smallest average value is selected as the social choice. So, in the first step we
compute:
I
Bk = Σ ci aik (2.32)
i =1
for k = 1, 2, …, n. Alternative k0 is the social choice if:
Bk0 = min{Bk } (2.33)

k
The Hare system is based on the successive deletion of less desirable alternatives.
First, we check to see if an alternative is selected as being the best with total weights
of at least 0.5; this alternative is the social choice, and the procedure terminates.
Otherwise, the alternative with the smallest such total weight is deleted, the table is
adjusted, and we go back to the first step and check again. If one among the remaining
alternatives has a total weight of at least 0.5 for its rankings of being best, then this
alternative is the social choice. Otherwise, we delete an alternative again. This
procedure continues until in the first step we have a clear choice. First we compute
A1, …, An by using Eq. (2.30). If, for a k, Ak ≥ 0.5, then alternative k becomes the
social choice. Otherwise, let:
Ak0 = min{Ak} (2.34)

k
Then, we delete alternative k0 from the table, so n becomes n – 1, and we modify

the aik values as follows:
⎧aik if aik < aik0

⎪
aik = ⎨ (2.35)
⎪⎩aik − 1 if aik > aik0
Then, we go back to the first step, computing again the new Ak numbers.
When applying pair-wise comparisons, the decision makers first have to agree
on the order of comparisons. Let k1, k2, …, kn be the order, which is a permutation
of integers 1, 2, …, n. Alternatives k1 and k2 are compared first by computing the
total weight of the decision makers who prefer alternative k1 to k2. If this total weight
is at least 0.5, then k1 is considered better than k2; otherwise, k2 is considered better
than k1. Then, the winner is compared to k3. The winner of this comparison is
compared to k4, and so on. The winner of the last comparison is considered to be
the social choice. When comparing two alternatives k and l we compute the value
N(k,l). Alternative k is considered better than l if N(k,l) ≥ 0.5:
{
N (k, l ) = Σ ci aik < ail
i
} (2.36)
Example 2.15
Assume that five decision makers have to choose the best design plan for a water
resources project. The data are given in Table 2.4. The last column shows that
decision makers 1 and 2 have higher priorities than the third, who is more important
than the fourth and fifth decision makers. In applying plurality voting, we first
compute:
A1 = 0.3, A2 = 0.4, A3 = 0, A4 = 0.3
Because A2 is the largest value, the second plan is considered to be the social choice.
For a Borda count we compute the Bk values:
B1 = 3(0.3) + 1(0.3) + 4(0.2) + 2(0.1) + 4(0.1) = 2.6

B2 = 1(0.3) + 4(0.3) + 3(0.2) + 3(0.1) + 1(0.1) = 2.5
B3 = 2(0.3) + 3(0.3) + 2(0.2) + 4(0.1) + 2(0.11) = 2.5
B4 = 4(0.3) + 2(0.3) + 1(0.2) + 1(0.1) + 3(0.1) = 2.4
The smallest value occurs at the fourth alternative, so it is the social choice again.
TABLE 2.4
Data for Example 2.15
Alternatives
Decision Makers (1 2 3 4) Weights
1 3 1 2 4 0.3
2 1 4 3 2 0.3
3 4 3 2 1 0.2
4 2 3 4 1 0.1
5 4 1 2 3 0.1
TABLE 2.5
Modified Data for Example 2.15
Alternatives
Decision Makers (1 2 4) Weights
1 2 1 3 0.3
2 1 3 2 0.3
3 3 2 1 0.2
4 2 3 1 0.1
5 3 1 2 0.1
When applying the Hare system, we again consider the Ak numbers computed
in plurality voting. Because none of these values is at least 0.5, no choice is made.
The smallest value occurs at the third alternative, so it is deleted. The modified table
is shown in Table 2.5.
We recompute first the Ak values from the modified table:
A1 = 0.3, A2 = 0.4, A4 = 0.3
None of these values is at least 0.5, so no clear choice is made. A1 and A4 are
both the smallest, so we delete one of them (say, alternative A4) from the table. The
new table is given in Table 2.6. Because A1 = 0.4 and A2 = 0.6, the second alternative
is the social choice.
Assume that in applying pair-wise comparisons the decision makers agree in
the order 1, 2, 3, and 4. Alternatives 1 and 2 are compared first:
N(1,2) = 0.3 + 0.1 = 0.4 < 0.5
so alternative 2 is considered better. Then, alternative 2 is compared to alternative 3:
N(2,3) = 0.3 + 0.1 + 0.1 = 0.5

TABLE 2.6
Second Modified Table for Example 2.15
Alternatives
Decision Makers (1 2) Weights
1 2 1 0.3
2 1 2 0.3
3 2 1 0.2
4 1 2 0.1
5 2 1 0.1
which shows that no preference exists between these alternatives. However, in the
comparison order the decision makers wanted to bring alternative 3 into the sequence
of comparisons later than alternative 2, so they gave higher priority to the third
alternative. Therefore, it is compared to the last one:
N(3,4) = 0.3 + 0.1= 0.4 < 0.5
Therefore, the fourth alternative is the social choice. The interested reader may find
more details of voting and social choice procedures in Taylor (1995).
2.5 CONFLICT RESOLUTION

As we mentioned earlier, the presence of multiple decision makers makes the
decision-making process much more difficult, as they usually have different priorities
and goals; therefore, reaching a compromise among the conflicting objectives is
sometimes very difficult. Several approaches are available to solve conflict situations.
One might consider the problem as a multiple-objective optimization problem that
addresses the objectives of the different decision makers. Conflict situations can be
also modeled as social choice problems in which the rankings of the decision makers
are taken into account in the final decision. A third way of resolving conflicts was
offered by Nash, who considered a certain set of conditions the solution has to satisfy
and proved that exactly one solution satisfies his “fairness” requirements. This
section outlines the Nash bargaining solution.
Assume that we have I decision-makers. Let X be the decision space and
fi : X ֏ R be the objective function of decision maker i. The criteria space is defined
as:
{
H = u u ∈ R I , u = (ui ), ui = fi ( x ) with some x ∈ X }
It is also assumed that when the decision makers are unable to reach an agree-
ment, all decision makers will get low objective function values. Let di denote this
value for decision-maker i, and let d = (d1, d2, …, dI ); therefore, the conflict is
completely defined by the pair (H,d), where H is the set of all possible outcomes
and d represents the outcomes if no agreement is reached. Thus, any solution of the
conflict depends on both H and d. Let the solution be denoted as a function of H
and d: ϕ (H,d). It is assumed that the solution function satisfies the following
conditions:
1. The solution has to be feasible: ϕ (H,d) ∈ H.

2. The solution has to provide at least the disagreement outcome to all
decision makers: ϕ (H,d) ≥ d.
3. The solution has to be nondominated; that is, there is no f ∈ H such that
f ≠ ϕ (H,d) and f ≥ ϕ (H,d).
4. The solution must not depend on unfavorable alternatives; that is, if H1
⊂ H is a subset of H such that ϕ (H,d) ∈ H1, then ϕ (H,d) = ϕ (H1,d).
5. Increasing linear transformation should not alter the solution. Let T be a
linear transformation on such that T(f) = (α1 f1 + β1, …, αI fI + βI), with
αi > 0 for all i, then ϕ (T,(H),T(d)) = T( ϕ (H,d)).
6. If two decision makers have equal positions in the definition of the conflict,
then they must get equal objective values at the solution. Decision makers
i and j have equal position if di = dj, and any vector f = ( f1, …, fI ) ∈ H
if and only if ( f 1, …, f I) ∈ H, with f i = fj, f j = fi, f l = fl, for l ≠ i, j.
Then we require that ϕ i (H,d) = ϕ j (H,d).
Before presenting the method for finding the solution that satisfies these prop-
erties, some remarks are in order. The feasibility condition requires that the decision
makers cannot get more than the amount available. No decision maker would agree
to an outcome that is worse than the amount he would get anyway without the
agreement. This property is given in condition 2. The requirement that the solution
must be nondominated shows that no better possibility is available for all. The fourth
requirement says that if certain possibilities become not feasible but the solution
remains feasible, then the solution must not change. If any of the decision makers
changes the unit of his objective, then a linear transformation is performed on H.
The fifth property requires that the solution must remain the same. The last require-
ment shows a certain kind of fairness by stating that if two decision makers have
the same outcome possibilities and the same disagreement outcome, then there is
no reason to distinguish among them in the final solution.
If H is convex, closed, and bounded, and there is at least one f ∈ H such that
f > d, then there is a unique solution f * = ϕ( H , d ) , which can be obtained as the
unique solution of the following optimization problem:
Maximize ( f1 − d1 ) ( f2 − d2 ) ... ( fI − d I )
Subject to fi ≥ di (i = 1, 2,…, I )
(2.37)
f = ( f1,…, fI ) ∈ H
The objective function is called the Nash product. Notice that this method can be
considered as a special distance-based method when the geometric distance is max-
imized from the disagreement point d.
Example 2.16
Consider again the discrete problem given earlier in Table 2.1. Assume that cost and
convenience are the objectives of two different decision makers. Assume that d =
(–4,50), where –4 is considered as the worst possibility of cost and 50 is the worst
possibility for convenience. The values of the Nash products for the four alternatives
are as follows:
(–3 + 4) (70 – 50) = 20
(–3.5 + 4) (90 – 50) = 20
(–3.8 + 4) (95 – 50) = 9
(–3.2 + 4) (85 – 50) = 28
Because the last alternative gives the largest value, the fourth technology variant is
considered to be the solution.
Example 2.17
Consider next the continuous example shown in Figure 2.5, and assume that f1 and
f2 are the objectives of two decision makers, such as might occur when the removal
of two kinds of pollutants is controlled by two different officials. Assume that d =
(0,–2000) is the worst value of the two objectives. In this case, Eq. (2.37) can be
rewritten as:
Maximize f1 ( f2 + 2000)
Subject to ( f1 , f2 ) ∈ H
A simple computer study shows that the optimal solution is:
f1 = 2000 and f2 = −1000,
and the corresponding decision variables are given as:
2 f1 + f2 − f1 − 2 f2
x1 = = 1000, x2 = =0
3 3
In many practical cases, conflict resolution is needed when the decision makers have
different powers. In such cases, problem (2.37) becomes:
( f1 − d1 ) ( f2 − d2 ) ... ( f1 − dI ) I
c1 c2 c
Maximize (2.38)
Subject to fi ≥ di (i = 1, 2, ..., I )
( f1,…, fI ) ∈ H
Here, (c1, c2, …, cI) show the relative powers of the decision makers. The solution
of this problem is usually called the nonsymmetric Nash bargaining solution.
Conflict resolution is a special chapter in game theory, so any text or monograph
on game theory is a potential source for further details. See, for example, Forgo
et al. (1999).
2.6 SPECIAL METHODOLOGY

2.6.1 DYNAMIC PROGRAMMING
In water resources planning, dynamism of the system is a major concern that must
be taken into account in modeling and optimization. Dynamic programming is the
most popular tool for optimizing dynamic systems.
Let t denote time, where t = 0, 1, 2, …, T. Let xt be the state of the system at
time period t, and let ut denote the decision (or control) at the same time period. It
is assumed that the state changes according to the state-transition relation:
xt +1 = f t( xt , ut ) (t = 0,1,…, T − 1) (2.39)
where ft : X × U → X is a given function, and X and U are the state space and control
space, respectively. It is assumed that at each time period an objective function
gt(xt ,ut) is given that is real valued, and we are looking for a sequence of controls
(u0, u1, …, uT) such that an overall objective:
∑ g (x , u )
t =0
t t t (2.40)
is optimal.
Without loss of generality, we may assume that function (2.40) is minimized;
otherwise, we have to multiply it by –1. This problem can be solved as a nonlinear
programming problem:
Minimize ∑ g (x ,u )
t =0
t t t
Subject to xt ∈ X ,⎫⎪
⎬ t = 0,1,..., T (2.41)
ut ∈U ⎭⎪
xt +1 = ft ( xt , ut ), t = 0,1,..., T − 1
In most applications, this problem has many decision variables and constraints, so
its direct solution presents some computational difficulties. Instead of solving one
large-scale problem, the dynamic programming (DP) procedure solves many smaller
sized problems. The most popular DP method computes a sequence of optimal return
functions VT (x), VT–1(x), …, V0(x) using the recursive equation:
Vt ( x ) = min{ ft ( x, u) + Vt +1 ( ft ( x, u))}
(2.42)
u ∈U
with the initial term VT+1(x) = 0 for all x ∈ X. Let u(x) denote the optimizer of the
right-hand side of Eq. (2.42), and let x0 denote the initial state. Then, it can be proved
that the sequence:
x 0* = x 0
ut* = ut ( xt* ) (2.43)
xt*+1 = ft ( xt* , ut* )
gives a global optimum of the objective function, Eq. (2.40).
Example 2.18
Assume x0 = 0; the state transition relation is:
xt +1 = xt + ut , t = 0,1,..., T ,
X = U = R, and gt ( xt , ut ) = ( xt − ut + 1)2 + xt2
In this case:
VT ( x ) = min gT ( x, u) = min( x − u + 1) 2 + x 2
u u
showing that:
uT ( x ) = x + 1, and VT ( x ) = x 2
Then,
{(
VT −1 ( x ) = min ( x − u + 1)2 + x 2 + ( x + u)2
u
) }
{
= min 3 x 2 + 2u 2 + 2 x − 2u + 1 = 3 x 2 + 2 x +
u
} 1
2
with
1
uT −1 ( x ) = ,
2
Similarly,
u ⎩
( )
VT −2 ( x ) = min ⎧⎨ ( x − u + 1) 2 + x 2 + 3( x + u) 2 + 2( x + u) + ⎫⎬
1
2⎭
= min ⎧⎨5 x 2 + 4ux + 4u 2 + 4 x +

3⎫ 3
⎬ = 4x + 4x +
2
u ⎩ 2⎭ 2
with UT–2(x)= –x/2. This recursive reasoning should continue until u0(x) is obtained,
and then the optimal decision sequence is obtained by Eq. (2.43).
In most applications, the optimization problem shown in Eq. (2.42) cannot be
solved analytically, so functions Vt(x) and ut(x) can be determined only numerically.
For such cases, discrete dynamic programming (DDP) is used, where the state and
decision spaces are discretized: X = {x ( i )} and U = {u( j )}. In this discrete case, recur-
sive Eq. (2.42) is replaced by its discrete counterpart:
∧ ∧
V t ( x ( i ) ) = min ⎧⎨ ft ( x ( i ) , u ( j ) ) + V t +1 ( ft ( x ( i ) , u ( j ) ))⎫⎬ (2.44)
j ⎩ ⎭
with the initial term VˆT +1 ( x ( i ) ) = 0 for all i. The difficulty in applying this procedure
is that ft ( x ( i ) , u( j ) ), which represents the new state, does not need to be among the
discrete nodes. This problem can be overcome by using an interpolation technique
between the nodes. The main problem of using this method is known as the curse of
dimensionality. As an illustration, assume that T = 10, and the state space is discretized
into 10 levels in each of its 10 components. Then, the total number of nodes is 1010.
Therefore, DDP is used for only up to four or five decision and state variables.
In order to overcome the problem of dimensionality, successive approximation
algorithms were developed, the most highly regarded techniques being differential
dynamic programming (DIFF DP), discrete differential dynamic programming
(DDDP), and state incremental dynamic programming (IDP). When applying any
of these methods, an initial approximation of the optimal policy must be given by
the user, and this approximate solution is improved in each step until convergence
occurs. However, as is usually the case in nonconvex optimization, the limit can be
only a local optimum or the approximating sequence is divergent.
For the sake of simplicity let {xt} and {ut} denote the sequence of states and
decisions of any iteration step and let {xtnew}, {utnew} be the successor obtained by
applying an iteration step. In the case of DDDP, a discrete dynamic programming
algorithm is used with the additional condition that for all t,
xt − xtnew ∞
<ε (2.45)
where ε > 0 is a specified tolerance level, and … ∞ is the maximum norm of the
vectors. The additional constraints vastly reduce the number of discrete decision
variable values which has to be tested in each step. However, iteration steps are
repeated many times. Constraint (2.45) can also be described as the optimum of
problem (2.44), which is determined only in a corridor of width ε centered about
the predecessor. When applying IDP, we require that for all t,
xtnew = xt + α t ei (2.46)
where ei is the ith basis vector, so xtnew differs from xt only in its ith coordinate.
This additional constraint reduces the solution of optimization problem (2.42) to a
one-dimensional line search. This procedure is very effective if the state transition
relation is invertible in ut; that is, there is a function ht : X × X ֏ U such that
Eq. (2.39) is equivalent to the following equation:
ht ( xt , xt +1 ) = ut (2.47)
Then, Eq. (2.42) has the special form:
⎧⎪ ⎫⎪
( )
∧ ∧
V t (α t ) = min ⎨gt xt + α t ei , ht ( xt + α t ei , xt +1 + α t +1e j ) + V t +1 (α t +1 )⎬ (2.48)
α t +1
⎪⎩ ⎪⎭
where only αt+1 is unknown. This one-dimensional optimization problem can be

solved by any line-search algorithm.
We now take a look at the DIFF DP procedure by introducing the following
notation:
δxt = xtnew − xt
(2.49)
δut = utnew − ut .
Let the quadratic Taylor’s polynomial approximation of fT (xt ,ut) be:
1 1
QT ( xT , uT ) = D δx 2 + ET δxT δuT + FT δuT2 + GT δxT + HT δuT + KT (2.50)
2 T T 2
where for the sake of simplicity we assume that X and U are one-dimensional sets.
QT is minimal if the first-order condition holds:
ET δxT + FT δuT + HT = 0
implying that:
HT ET
δuT = − − δx (2.51)
FT FT T
∧
The approximate optimal value function V T ( x T ) is now obtained as:
∧ 1 E2 E H
V T ( xT ) = QT ( xT , uT ( xT )) = ( DT − T )(δxT )2 + (GT − T T )(δxT )
2 FT FT
1
= A (δx )2 + BT (δxT )
2 T T
For t < T, the general step is similar to the previous construct. Assume that the
approximate optimal return function:
∧ 1
V t +1 ( xt +1 ) = A (δx ) 2 + Bt +1 (δxt +1 ) (2.52)
2 t +1 t +1
is already determined. Let Qt(x,u) be the quadratic Taylor’s polynomial approxima-

tion of gt (x,u) + V̂ t+1(ft (x,u)) about (xt ,ut):
1 1
Qt ( xt , ut ) = Dt δxt2 + Et δxt δut + Ft δut2 + Gt δxt + Ht δut + Kt (2.53)
2 2
The optimum of this function can be obtained as it is presented in Eq. (2.50), with
T being replaced by t. After this step is continued until t = 0, the successor policy
can be obtained as:
x 0new = x 0 , δ x 0 = 0
and then
Ht Et
δut = − − δxt
Ft Ft
utnew = ut + δut
+1 = ft ( x t
xtnew new
, utnew )
for t = 0, 1, 2, …, T. It can be proved that the DIFF DP method has quadratic

convergence:
2
u new − u * < C u − u *
where u* is the optimal policy and C > 0 is a constant.

The survey paper by Yakowitz (1982) gives a summary of the different versions
of dynamic programming with applications to water resources planning and
management.
2.6.2 GENETIC ALGORITHMS

Genetic algorithms are very often used to find optimal solutions to computationally
difficult problems. The basis of these methods is the similarity between biological
and computational processes, especially in regard to how biological reproductive
processes work. Genes and computers are similar to each other in the sense that
they can record, copy, and disperse information. Just as in biological evolution,
genetic algorithms work iteratively; many generations are created successively and
only the fittest solutions survive. Using biological terminology, within the breeding
population individual solutions are called chromosomes, which are encoded as strings
(usually binary string). Individual features found in each chromosome are genes,
and the value of a feature in any given chromosome is the allele.
In creating chromosomes we have to map the decision space into a set of encoded
strings. The encoding mechanism depends on the problem being solved. One possible
way is to use the binary representation of the decision variables. We also have to
decide on the size of the breeding pool; that is, on the number of chromosomes
being involved in the process. A larger number of chromosomes increases diversity,
but the algorithm becomes slower. An initial population is first selected randomly.
The fitness of each string is often evaluated based on the objective functions,
the constraints, and the encoding mechanism. A usual way is to add a penalty term
to the objective function if any chromosome becomes not feasible by violating one
or more constraints.
Pairs of chromosomes are formed randomly, and they are then subjected to
certain genetic manipulations that modify the genes in the parent chromosomes. A
typical procedure is crossover, which swaps a part of the genetic information con-
tained in the two chromosomes. Usually a substring portion is selected randomly in
the chromosomes, and the genes (which are the string elements — for example,
bits) within that substring are exchanged. In this way, new offsprings are created to
replace the parent chromosomes. The particular structure of the crossover mechanism
is usually problem specific and produces meaningful and feasible solutions.
There is no guarantee that the new pairs of chromosomes will be better than the
parent chromosomes. To overcome this difficulty, mutations are allowed to occur
which reverse one or more genes in a chromosome. This simple step may reintroduce
certain genes into the solution that may be essential for the optimal solution and
were lost from the breeding population in previous stages. Frequent use of mutations
makes the search more broadly random, resulting in slower convergence of the
algorithm.
By repeating the above steps iteratively, we sequentially replace either the entire
previous population or only the less fit members of it. The cycle of creation, eval-
uation, selection, and manipulation is repeated iteratively until an optimal or an
acceptable solution is reached.
Example 2.19
Assume function f(x) = x – x2 is maximized in the [0,1] interval. First, the decision
space is discretized and each discretized alternative is encoded. For illustration
purposes, we have four-bit representations. The population is given in Table 2.7.
The initial population is selected next. We have chosen six chromosomes as shown
in Table 2.8, which also presents their actual values and the corresponding objective
function values.
The worst objective function values are obtained at 0001 and 1111. A simple
crossover procedure is performed by interchanging the two middle bits to obtain
0111 and 1001. The new population is given in Table 2.9. The worst objective
function value is obtained at 0011, but two chromosomes have the second worst
objective values: 1100 and 0100. Randomly selecting the first, we have the chro-
mosome pair 0011 and 1100. By interchanging their middle bits we get 0101 and
1010. The resulting new population is shown in Table 2.10.
Selecting the pair 0100 and 0101, we cannot use the same crossover procedure
as before, as they have the same middle bits. Because the two chromosomes differ
in only one bit, interchanging any substring of them will result in the same chro-
TABLE 2.7
Population of Chromosomes
0000 0110
0001 1010
0010 1100
0100 0111
1000 1011
0011 1101
0101 1110
1001 1111
TABLE 2.8
Initial Population
Chromosomes Value Objective Function
0001 0.0625 0.0586

1100 0.75 0.1875
1011 0.6875 0.2148
1111 0.9375 0.0586
0100 0.25 0.1875
0011 0.1875 0.1523
TABLE 2.9
First Modified Population
0111 0.4375 0.2461

1100 0.75 0.1875
1011 0.6875 0.2148
1001 0.5625 0.2461
0100 0.25 0.1875
0011 0.1875 0.1523
TABLE 2.10
Second Modified Population
0111 0.4375 0.2461

0101 0.3125 0.2148
1011 0.6875 0.2148
1001 0.5625 0.2461
0100 0.25 0.1875
1010 0.625 0.2344
mosomes; therefore, mutation will be applied to them. By reversing the first two
bits, we have 1000 and 1001, but the second chromosome is already in the population.
Therefore, mutation is again used to reverse the second bit to obtain 1100. The
resulting modified population is shown in Table 2.11.
It is interesting to see how the best objective function is evolving from population
to population. Starting from the initial table with the value 0.2148, the value
increased to 0.2461 and then remained the same. In Table 2.11 it again increased to
0.25. Because this value is the global optimum, it will not increase in any further
TABLE 2.11
Third Modified Population
0111 0.4375 0.2461

1100 0.75 0.1875
1011 0.6875 0.2148
1001 0.5625 0.2461
1000 0.5 0.25
1010 0.625 0.2344
steps. In practical applications, we stop the procedure if no or very small improve-

ment occurs after a certain (user-specified) number of iterations.
More details on genetic algorithms and possible applications can be found in
Goldberg (1989).
2.7 PROBLEMS
2.1 Graphically solve the following linear programming problem:
Minimize 2 x1 + x 2
Subject to x1 , x 2 ≥ 0
x1 + 3 x 2 ≥ 6
3 x1 + x 2 ≥ 6
2.2 Consider the same feasible decision space as in the previous problem and
minimize the nonlinear objective function:
(a) x12 + x 22
(b) x1 x 2 .
2.3 Find the primal form for the following linear programming problem:
Minimize x1 + 2 x2 − x3
x1 − x2 + 2 x3 ≥ 2
x1 + x2 = 3
2 x1 − x3 ≤ 5
2.4 Consider the following problem with two objective functions:
Maximize f1 = x1 + x 2 , f2 = x1 − x 2
x1 + 3 x 2 ≤ 6
2 x1 − x 2 ≥ 1
(a) Graphically show the decision space.

(b) Graphically show the criteria space.
2.5 Solve the previous problem with sequential optimization if f1 is more

important than f2.
2.6 Select f1 as the more important objective in problem 2.4. Give the range
of ε2 that leads to a feasible solution. Solve the problem for ε2 = 0.
2.7 Solve problem 2.4 by using the weighing method, with c1 = c2 = 0.5.
What are the solutions with and without normalizing the objectives?
2.8 Select equal weights in problem 2.4. Solve it by minimizing distance from
the ideal point. What are your solutions with the ρ1, ρ2, and ρ∞ distances?
2.9 Select v = (1,1) and f * = (2,0), and apply the direction-based method to
problem 2.4.
2.10 Consider the following problem with 2 objectives and 4 alternatives.
Alternatives Objective 1 Objective 2
1 0 3
2 1 2
3 2 1
4 3 0
Assume both objectives are maximized and c1 = c2 = 0.5. Solve the

problem with the application of:
(a) Weighting method
(b) Maximizing distance from the nadir (0,0) and selecting the ρ1
distance.
(c) Repeat part (b) with the ρ2 distance.
(d) Repeat part (b) with the ρ∞ distance.
2.11 Consider the following social choice problem:
Alternatives
Decision Makers 1 2 3 4 Weights
1 1 3 2 4 0.3
2 4 1 2 3 0.2
3 4 3 2 1 0.2
4 1 4 3 2 0.2
5 3 2 1 4 0.1
Find the solution based on plurality voting.

2.12 Solve problem 2.11 using Borda counts.
2.13 Solve problem 2.11 using the Hare system.
2.14 Solve problem 2.11 using pair-wise comparisons, where the order of
comparisons is 1, 2, 3, 4.
2.15 Assume that d = (0,0) and H is characterized by the following set of
inequalities:
x1 , x 2 ≥ 0
2 x1 + x 2 ≤ 4
x1 + 3 x 2 ≤ 6
Find the Nash solution, Eq. (2.37), of this problem.

2.16 Assume c1 = 1/3 and c2 = 2/3. Find the nonsymmetric Nash solution, Eq.
(2.38), of the previous problem.
2.17 The following directed graph shows the possible arcs for a person moving
from one point to another and the time periods required to walk through
each arc. Assume somebody starts at point A and wants to reach point B
in the shortest time period. Find the best path that minimizes the total time.
PROBLEM 2.17
2.18 Assume that for t = 1, 2, 3, 4, gt (xt, ut) = xt2 + ut, xt ∈[0,1], ut ∈[–1,1],
and for all t, xt+1 = xt + ut + 1. Solve Eq. (2.41) by dynamic programming.
2.19 Repeat Example 2.19 for minimizing f(x) =x2 + x+1 in interval [–1,1].
2.20 Solve problem 2.19 for minimizing x3 – 99x + 6 on the integer points of
interval [0,100]. You may use the binary representation of integers.
References
Dantzig, G. B., Linear Programming under Uncertainty, Manage. Sci., 1(3), 157–168, 1955.
Dantzig, G. B., Linear Programming and Extensions, Princeton University Press, Princeton,
NJ, 1963.
Forgo, F., Szep, J., and Szidarovszky, F., Introduction to the Theory of Games, Kluwer
Academic Publishers, Dordrecht, The Netherlands, 1999.
Goldberg, D. E., Genetic Algorithms in Search, Optimization, and Machine Learning, Add-
ison-Wesley, Reading, MA, 1989.
Hadley, A., Nonlinear and Dynamic Programming, Addison-Wesley, Reading, MA, 1964.
Luce, R. D. and Raiffa, H., Games and Decisions, Wiley, New York, 1967.
Mangasarian, O. L., Nonlinear Programming, McGraw-Hill, New York, 1962.
Rubinstein, R. Y., Simulation and Monte Carlo Method, Wiley, New York, 1981.
Szidarovszky, F. M. and Bahill, A. T., Linear Systems Theory, 2nd ed., CRC Press, Boca
Raton, FL, 1992.
Szidarovszky, F., Gershon, M., and Duckstein, L., Techniques of Multi-Objective Decision
Making in Systems Management, Elsevier, Amsterdam, 1986.
Taylor, A. D., Mathematics and Politics, Springer-Verlag, New York, 1995.
Vajda, S., Probabilistic Programming, Wiley, New York, 1972.
Yakowitz, S., Dynamic programming applications in water resources, Water Resources Res.,
18(4), 673–696, 1982.
3 Decision Making under
Uncertainty
This chapter is devoted mainly to the elements of probabilistic and statistical meth-
odology and their use in modeling risk and uncertainty. We have also included a
special section on neural networks, which are the most popular tools when models
are uncertain or if no physical model is available.
3.1 INTRODUCTION
In all fields of natural resources management, we face uncertainty arising from
natural phenomena that cannot be predicted accurately. Any prediction is uncertain,
and in the mathematical modeling of hydrological and water resources systems this
uncertainty has to be taken into account. Of the two primary ways to model uncer-
tainty, one method is based on considering all uncertain elements as random variables
and on the use of probabilistic and statistical models. The other approach is based
on the theory of fuzzy sets, where uncertainty is modeled by membership functions.
This chapter provides an introduction to this very important methodology.
3.2 PROBABILISTIC METHODS

Random events and random outcomes are usually characterized by random variables,
the properties of which are the subject of the next subsection.
3.2.1 RANDOM VARIABLES

In natural resources management, we often face quantities for which the values are
random, thus they are called random variables. A random variable is discrete if its
possible values form a finite or a countable set. A discrete random variable, then, is
completely defined by listing its possible values and giving the occurring probabil-
ities: X = x1, x2, …, and for k = 1,2, …, pk = P(x = xk). The pk probability values
are all taken from the unit interval [0,1], and their sum equals unity:
0 ≤ pk ≤ 1 for k = 1,2,..., and ∑ p =1

k
k (3.1)
67
Example 3.1
The Bernoulli variable has two possible values: 1 and 0, with probabilities p1 =
P(X = 1) = p and p0 = P(X = 0) = q (= 1 – p). For any random event A we can assign
a Bernoulli variable X = 1 if A occurs; otherwise, X = 0, and p1 = P(X = 1) = P(A) = p
and p0 = P(X = 0) = 1 – p.
Example 3.2
Assume an experiment is repeated n times, independent of each other, and in each

case an event A has the same probability, p. Let X denote the number showing how
many times A occurs among the n trials. The possible values of X are 0, 1 , 2, …,
n, with probabilities:
⎛ n⎞
pk = P( X = k ) = ⎜ ⎟ p k q n−k
⎝ k⎠
This random variable is called binomial.

If X1, X2, …, Xn are independent Bernoulli variables with identical parameter p,
then it is easy to see that X = X1 + X2 + … + Xn is binomial with parameters n and p.
Example 3.3
Assume that an experiment is repeated independently, and an event A has the same
p probability in each case. Let X denote the number of repeats until A first occurs.
The possible values of X are 1, 2, 3, … (until infinity), and pk = P(X = k) = q k–1p,
where q = 1 – p as before. This variable is geometric.
Example 3.4
Consider again the model of the previous example, and let r ≥ 1 be a given positive
integer. Now, let X denote the number of trials until A occurs r times. Clearly, the
case of r = 1 coincides with the geometric variable. The possible values of X are r,
r + 1, r + 2, … (until infinity), and
⎛ k − 1⎞ k −r r
pk = P( X = k ) = ⎜ ⎟q p
⎝ r − 1⎠
for k = r, r + 1, r + 2, …. This random variable is negative binomial.
Example 3.5
Assume a box contains N items, S of which are defective (S < N). If we randomly
choose n items from the box without replacement, and X denotes the number of
Decision Making under Uncertainty 69
defective items in the selected sample, then X is hypergeometric. The possible values
of X are 0, 1, 2, …, min{S;n}, and the occurring probabilities are:
⎛ S⎞ ⎛ N − S⎞
⎜ ⎟⎜ ⎟
⎝ k⎠ ⎝ n − k ⎠
pk = P( X = k ) =
⎛ N⎞
⎜ ⎟
⎝n ⎠
If we modify the above example by assuming that sampling is done with replacement,
then each time an item is taken from the box we have the same p = S/N probability
for that item to be defective. Therefore, we have the binomial distribution with
parameters n and p.
Example 3.6
Assume that in a binomial variable n → ∞, p → 0 so that np is kept on a constant

λ level. The limiting distribution is called Poisson. The possible values of X are 0,
1, 2, … (until infinity) with probabilities:
λk − λ
pk = e
k!
Let A be any subset of the possible values of a discrete random variable X. Then:
P( X ∈ A) =
{
∑ }
P( X = x )
k xk ∈A
k (3.2)
where the summation is done for all values of k such that xk ∈ A. For example, if
the possible values of X are integers, then:
k2
P(k1 ≤ X ≤ k2 ) = ∑ P( X = k )
k = k1
(3.3)
where we assume that both k1 and k2 belong to the range of X.

The distribution function of a random variable is defined for all real values x:
F( x ) = P( X ≤ x ) = ∑
{k xk ≤ x }
P( X = x k ) (3.4)
F(x)
1
0 if x <0
F(x) = q if 0 ≤ x < 1
q 1 if x ≥ 1
0 1 x
FIGURE 3.1 Distribution function of Bernoulli variables.
Clearly, if the possible values of X are integers, then
F( x ) = ∑ P( X = k )
{k k ≤ x}
(3.5)
The graph of F(x) for any discrete random variable is a piecewise constant
function, starting with zero, and discontinuity occurs only at the values x1, x2, …,
and they have a jump of height pk = P(X = xk) at each point xk. Figure 3.1 shows
F(x) for a Bernoulli variable.
Continuous random variables are defined directly by their distribution functions
F(x), which are assumed to be continuous at each x. Then, each particular value has
zero occurring probability: P(X = x) = 0 for all x; furthermore,
P( a ≤ x ≤ b) = P( a < x ≤ b) = P( a ≤ x < b) = P( a < x < b) = F(b) − F( a). (3.6)
The distribution functions of mixed distributions have discontinuities and are

not piecewise constant functions. They are not very common in water resources
applications.
The density function of a continuous variable is defined as the derivative of F(x):
f ( x ) = F ′( x ) (3.7)
which exists except as a zero-measured set. By the fundamental theorem of integral

calculus we see that
F(b) − F( a) =
∫ f ( x)dx
a
so probabilities given in Eq. (3.6) can be also obtained by integrating the density
function.
The distribution function and density function of any continuous variable satisfy
the following properties:
lim F( x ) = 0, lim F( x ) = 1, and

x →−∞ x →+∞
f ( x ) = F ( x ) ≥ 0, furthermore
'
∫ f ( x)dx = 1
−∞
Example 3.7
A random variable is uniform in a finite interval [a,b] if its distribution function is:
⎧0 if x≤a
⎪
⎪x − a
F( x ) = ⎨ if a<x≤b (3.8)
⎪b − a
⎪⎩1 if x>b
By differentiation,
⎧ 1 if a < x < b
⎪
f ( x) = ⎨ b − a (3.9)
⎪0 otherwise
⎩
If α,β ∈ [a,b], then:
β−a α−a β−α

P(α ≤ X ≤ β) = F(β) − F(α ) = − = (3.10)
b−a b−a b−a
showing that this probability depends only on the length β – α of the interval [α,β]
and is independent of the location of this interval. This is the reason for calling this
variable uniform.
Example 3.8
A continuous random variable is called exponential if:
⎧⎪0 if x ≤ 0
F( x ) = ⎨ (3.11)
⎪⎩1 − e − λx if x > 0
where λ > 0 is a given parameter. It is easy to see that
⎪⎧λe
− λx
if x > 0
f ( x) = ⎨ (3.12)
⎪⎩0 otherwise.
The exponential distribution has the forgetfulness property, which means the fol-
lowing: If 0 < x < x + y, then
P( X > x + y X > y) = P( X > x ) (3.13)
showing that if X, for example, represents the lifetime of any entity, then at any age
y surviving an additional x time periods does not depend on the age.
Because of the forgetfulness property, exponential variables are only very seldom
used in lifetime modeling. For such purposes, a gamma or a Weibull distribution is
used, which are introduced next.
Example 3.9
Let α, λ > 0 be two given parameters. The gamma variable that has these parameters
is defined by the density function:
⎧ λe − λx (λx )α −1
⎪ if x>0
f ( x) = ⎨ Γ (α ) (3.14)
⎪0 otherwise,
⎩
where
Γ(α ) =
∫x
0
α −1 − x
e dx
is the gamma function. Let α = n be an integer and X1, X2, …, Xn be independent

exponential variables with the same parameter λ. Then, X1 + X2 + … + Xn follows
a gamma distribution with parameters α = n and λ.
The distribution functions of gamma variables cannot be given in closed-form,
simple expressions except in very special cases. They have been tabulated, so
function tables are available. Alternatively, numerical integration can be used.
Example 3.10
Let λ, β > 0 be given parameters. The Weibull distribution with these parameters is
defined by the distribution function:
⎧⎪1 − e − λx β
if x≥0
F( x ) = ⎨ (3.15)
⎩⎪0 otherwise.
By differentiation,
⎧⎪λβx β−1e − λxβ if x>0

f ( x) = ⎨ (3.16)
⎪⎩0 otherwise.
It is easy to see that, if Y is exponential with parameter λ, then X = Y1/β is Weibull

with parameters λ and β.
Example 3.11
The standard normal variable is given by the density function:
x2
1 −2
ϕ( x ) = e ( −∞ < x < ∞) (3.17)
2π
The distribution function cannot be given by an analytic form. It is tabulated

and is usually denoted by φ(x). A general normal variable is obtained as follows.
Let µ and σ > 0 be two given parameters and Z be a standard normal variable. Then,
X = σZ + µ follows a normal distribution with parameters µ and σ. It is easy to see
that, in general,
x−µ
F( x ) = φ( ) (3.18)
σ
and
( x − µ )2
1 −
f ( x) = e 2 σ2
(3.19)
σ 2π
Figure 3.2 shows the density and distribution function of the most frequently used
distributions.
In statistical methods, we often use distributions arising from the normal vari-
able. The three most important such variables are given next.
Example 3.12
If Z1, Z2, …, Zn are independent standard normal variables, then:
Xn = Z12 + Z22 + ⋯ + Zn2 (3.20)

F(x)
1
f(x)
(A)
x
a b
1
F(x)
(B)
f(x)
1 F(x)
0.5
(C) f(x)
FIGURE 3.2 (A) Uniform distribution; (B) exponential distribution; (C) normal distribution.
is the chi-square distribution with n degrees of freedom. The corresponding density

function is the following:
⎧ −x n
−1
⎪ e1 2 ⎛ x ⎞ 2
⎪⎪ 2 ⎝ 2⎠
if x > 0
f ( x) = ⎨
Γ⎛ ⎞
n (3.21)
⎪ ⎝ 2⎠
⎪
⎪⎩0 otherwise.
By comparing this density to the gamma density given in Eq. (3.14) it is clear that
chi-square is a special gamma distribution with α = n/2 and λ = 1/2.
Example 3.13
Let Z be a standard normal variable and X a chi-square variable with n degrees of
freedom. Then,
Z
Tn = (3.22)
X
n
is the t-distribution with n degrees of freedom. It can be shown that the density
function of Tn is:
n + 1⎞
Γ⎛ −
n +1
1 ⎝ 2 ⎠ ⎛ x2 ⎞ 2
f ( x) = ⎜ 1 + ⎟ ( −∞ < 0 < ∞) (3.23)
nπ ⎝ n⎠
Γ⎛ ⎞
n
⎝ 2⎠
The shape of f (x) is very similar to that of the standard normal density function.
Example 3.14
Let Xn and Xm be independent chi-square variables with n and m degrees of freedom,

respectively. Then,
Xn
Fn,m = n (3.24)
Xm
m
is the F-distribution with n and m degrees of freedom. The density function of Fn,m
is as follows:
⎧ ⎛ n + m ⎞ n 2 m 2 ( n−2 ) 2
⎪ Γ⎝ 2 ⎠ n m x
⎪⎪ n+m x>0
f ( x ) = ⎨ Γ ⎛ n ⎞ Γ ⎛ m ⎞ ( m + nx ) 2 (3.25)
⎪ ⎝ 2⎠ ⎝ 2 ⎠
⎪
⎪⎩0 otherwise.
The distribution functions of the above three distributions cannot be presented in

simple equations so they have been tabulated and their particular values can be found
in the tables. For more details and tables, see Milton and Arnold (1995) or Ross
(1987).
3.2.2 CENTRAL TENDENCIES AND VARIANCES

Consider first a discrete random variable X with values x1 < x2 < x3 < … and occurring
probabilities pk = p(X = xk). Central tendencies are characterized by expectation,
median, and mode. The expectation of X is defined as:
E( x ) = ∑x p
k
k k (3.26)
by assuming that
∑x k
k pk < ∞
That is, E (x) is a weighted average of the possible values of X where the weights
are the occurring probabilities.
The median of X is a particular value xl such that:
1
p1 + p2 + ⋯ + pl −1 < ≤ p1 + p2 + ⋯ + pl −1 + pl (3.27)
2
That is, the median is the xl value for which
1
P( X ≤ xl −1 ) < ≤ P( X ≤ xl )
2
or where the cumulative probability first reaches or exceeds 1/2.

The mode of X is the value xm such that
pm = max{pk } (3.28)
k
which is the value of X that has the highest probability of occurring. If more than
one value has the same largest probability, then the mode consists of all such values.
Now let g be any real valued function. Then, more generally,
E( g( x )) = ∑ g( x ) p
k
k k (3.29)
assuming that
∑ g( x ) p
k
k k <∞
As special cases, E (X), E (X2), E (X3), … are called the first, second, third, …
moments of X. The real valued function:
m(t ) = E etX =( ) ∑e k
txk
pk (3.30)
is the moment-generating function of X. It is easy to see that
E( X ) = m' (0)
E( X 2 ) = m'' (0) (3.31)
E( X 3 ) = m''' (0)
and so on; therefore, the moment-generating function is very useful in computing

the moments.
In the case of a continuous random variable X, let f denote the density function.
Then, the expectation of X is given as:
E( X ) =
∫ xf ( x)dx
−∞
(3.32)
and, more generally, for any integrable function g,
E( g( x )) =
∫ g( x) f ( x)dx
−∞
(3.33)
The median of X is the value x such that
1
F( x ) = (3.34)
2
If F has a constant segment at the 1/2 level, then the smallest x value satisfying this
equation is chosen. The mode of X is the maximizer of its density function; that is,
the mode of X is x* if:
f ( x * ) = max{ f ( x )} (3.35)
x
If the maximizer is not unique, then all of them form the mode. The definition of
the moment-generating function is analogous to Eq. (3.30):
( )= ∫e
m( t ) = E e tX tx
f ( x )dx (3.36)
−∞
and because Eq. (3.31) remains true in the continuous case, it can be easily used to
compute the moments of the distribution.
The variability of a random variable is characterized by its variance. If we use

the notation µ = E(X), then the variance of X is defined to be:
[
Var( X ) = E ( X − µ ) 2 ] (3.37)
It can be shown that Var(X) can be easily computed from the first two moments:
( )
Var(X) = E X 2 − µ 2 (3.38)
In Table 3.1 we summarize the expectations and variances of the most popular
distribution types. More details can be found, for example, in Ross (1987).
TABLE 3.1
Expectations and Variances
Name E(x) Var(x)
Bernoulli p pq
Binomial np npq
Geometric 1 q
p p2
Negative binomial r rq
p p2
Hypergeometric s⎛ n −1 ⎤
1 − ⎞ ⎡1 −
s s
n n
N N ⎝ N ⎠ ⎣⎢ N − 1 ⎦⎥
Poisson Λ λ
Uniform a+b (b − a) 2
2 12
Exponential
1 1
λ λ2
Gamma α α
λ λ2
Weibull 1 2
− ⎛ 1⎞ − ⎛ 2⎞
λ β Γ ⎜1 + ⎟ λ β Γ ⎜1 + ⎟ − µ 2
⎝ β⎠ ⎝ β⎠
Normal µ σ2
Chi-square n 2n
t 0 (if n > 1) n
(if n > 2)
n−2
F m m 2 (2 m + 2 n − 4)
(if m > 2) (if m > 4)
m−2 m( m − 2 ) 2 ( m − 4 )
3.2.3 JOINT DISTRIBUTIONS

Let X and Y be two discrete distributions with possible values x1,x2, … and y1,y2,
…, respectively. Their joint distribution is defined by the joint probabilities:
(
pij = P ( X = xi ) ∩ (Y = y j ) ) (3.39)
Similar to the single-variables case, these pij values satisfy the following
relations:
0 ≤ pij ≤ 1 for all i and j,
∑∑ p
i j
ij = 1.
(3.40)
If A ⊆ R2 is any set, then the probability that the pair (X,Y) belongs to A is given
by the sum:
P(( X , Y ) ∈ A) = ∑ pij (3.41)

{ }
( i , j ) ( xi , y j )∈A
The marginal probabilities of X and Y can be easily computed as:
pix = P( X = xi ) = ∑p j
ij (3.42)
and
p Yj = P(Y = y j ) = ∑p i
ij (3.43)
Variables X and Y are independent if, for all i and j,
pij = pix ⋅ p jy (3.44)
If g is any bivariable function, then:
E[g( X , Y )] = ∑ ∑ g( x , y ) p
i j
i j ij (3.45)
where we assume that
∑ ∑ g( x , y ) p
i j
i j ij <∞
similar to the single-variable case.

Continuous joints distributions are characterized by a joint density function f,
which satisfies the following properties:
f ( x, y) ≥ 0 for all x and y,
and
∞ ∞
∫ ∫ f ( x, y)dydx = 1
−∞ −∞
Again, let A ⊆ R2 be any set. Then, similar to Eq. (3.41),
P(( X , Y ) ∈ A) =
∫∫ f ( x, y)dydx
A
(3.46)
The marginal density functions of X and Y are obtained as:
fX ( x) =
∫ f ( x, y)dy
−∞
(3.47)
and
fY ( y) =
∫ f ( x, y)dx
−∞
(3.48)
Variables X and Y are independent if, for all x and y,
f ( x, y) = f x ( x ) f y ( y) (3.49)
Let g be any integrable bivariable function. Then,
∞ ∞
E[g( X , Y )] =
∫ ∫ g( x, y) f ( x, y)dydx
−∞ −∞
(3.50)
Expectations and variances satisfy the following important properties:
1. E(αX + β) = αE(X) + β for all real α and β.

2. E(X + Y) = E(X) + E(Y).
3. If X and Y are independent, then E(XY) = E(X)E(Y).
4. Var(αX + β) = α2 Var(X) for all real α and β.
5. If X and Y are independent, then Var(X + Y) = Var(X) + Var(Y).
Note that, without independence of X and Y, properties (3) and (5) might not be true.
The covariance of X and Y is defined as:
Cov( X, Y ) = E( XY ) − E( X )E(Y ) (3.51)
Clearly, if X and Y are independent, then property (3) implies that Cov(X,Y) = 0.
The correlation between X and Y is determined in the following way:
Cov( X , Y )
Corr( X , Y ) = (3.52)
Var( X )Var(Y )
If X and Y are independent, then Corr(X,Y) = 0, and always –1 ≤ Corr(X,Y) ≤ 1.

It is easy to show that if a deterministic, linear functional relation exists between
X and Y such that with some constants α ≠ 0 and β, Y = αX + β, then:
⎧1 if α > 0
Corr( X , Y ) = ⎨
⎩−1 if α < 0
Based on the above properties, correlation is usually used to measure the strength
of the relation between two random variables. If ⏐Corr(X,Y)⏐ is small, then the
relation is very weak; if Corr(X,Y) is close to +1 or –1, then we conclude that a
strong relation exists between X and Y.
The conditional distribution of X given the value of Y is defined as the conditional
probabilities
(
P X = xi Y = y j = ) pij
pYj
(3.53)
in the discrete case, and as the conditional density function:
f x ( x y) =
f ( x, y)
(3.54)
fY ( y)
in the continuous case. The conditional distribution of Y given the values of X is

defined analogously. With any fixed value of Y, we can obtain the conditional
expectation of X by using the conditional distribution given previously. In the discrete

case,
(
E X Y = yj = ) 1
p Yj ∑x p
i
i ij (3.55)
assuming that
∑x p
i
i ij <∞
and in the continuous case,
∫ xf ( x, y)dx
1
E( X Y = y) = (3.56)
fY ( y)
−∞
The conditional expectation of Y given the value of X can be obtained in a similar

manner. These conditional expectations are usually referred to as regression curves.
We mention, finally, that Milton and Arnold (1995) and almost all introductory
textbooks on probability theory offer extended discussions of joint distributions and
their characteristics.
3.3 SIMULATION MODELS

When analyzing complex systems, the physical model and the mathematical descrip-
tion is so complicated that no solution algorithm can be developed. In many of these
cases, no mathematical model is even available. The only choice we have then is
simulation of the model. This section first introduces the most popular model sim-
ulation method, neural networks. After a model is formulated, its parameters are
determined by observations, measurements, predictions, or computations. Such data
usually are not available directly or are uncertain. In the first case, some kind of
observer is used, and in the second case the usual method is to utilize probabilistic
methods and simulation. Natural phenomena can very seldom be described accu-
rately; for example, future floods, rainfall, and so on can be modeled only by
probabilistic methods. The second part of this section will introduce the fundamen-
tals of observers and stochastic simulation.
3.3.1 NEURAL NETWORKS

Every mathematical model is an input–output relation in a certain sense, where the
input is the set of model parameters and our actions and the output is the consequence
of our actions. In optimization models, the best action (input) is selected based on
FIGURE 3.3 A simple neural network.
the given criteria. Input–output relations can be very conveniently modeled by the
neural network methodology.
A simple neural network is shown in Figure 3.3, where the input variables are
in the nodes of the input layer and the output variables are in the nodes of the output
layer; we assume two input and output variables. In order to provide a high level of
flexibility in model formulation, we also introduce hidden layers in the middle; these
nodes correspond to hidden or artificial variables.
Let z1 and z2 denote the artificial variables corresponding to the hidden layer
nodes. In the case of linear neural networks, we assume that for all hidden variables:
zj = ∑w x
i =1
ij i (3.57)
where m is the number of input variables. In our case, m = 2. The coefficients wij
are constants, which are determined by a training procedure to be explained later.
The relations between the output and hidden variables are also linear:
yk = ∑w
j =1
jk zj (3.58)
where l is the number of hidden variables and we have n output variables. The
coefficients w jk are also estimated by the training process. For the nonlinear neural
network, Eqs. (3.57) and (3.58) have certain nonlinear elements. The parameters of
the nonlinear functions are also determined by the training process. In most appli-
cations, more than one hidden layer is introduced, and relations similar to Eqs. (3.57)
and (3.58) are assumed to exist between the variables of the consecutive hidden
layers. After the parameters of these function relations are estimated, it is very simple
to compute the output y = (y1, …, yn) for any input vector x = (x1, …, xm). We can
simply substitute the components of x into Eq. (3.57) to obtain the hidden variables,
and then these results are substituted into Eq. (3.58) to obtain the output variables.
In the case of more hidden layers, additional similar steps are necessary. Because
the input–output relation is easy to compute, the neural network model can be
considered as a simple subroutine of input–output computations in any descriptive
or optimization procedure.
The theoretical basis of the previously described neural network structure is the
famous Kolmogorov’s theorem (see, for example, Hecht–Nielsen, 1990), which
states that any continuous function f : [0,1] ֏ R n can be implemented exactly by
m
a three-layer neural network with m input nodes, n output nodes, and (2m + 1)
hidden nodes, where Eqs. (3.57) and (3.58) have certain special (usually irrational)
nonlinear elements. If these nonlinear functions are approximated by linear relations,
then we need more terms; therefore, more hidden variables are required which might
be structured in more than one hidden layer (Bose and Liang, 1996; Hecht-Nielsen,
1990).
As has been mentioned earlier, the application of neural network methodology
depends on the estimation of the model parameters wij and w jk . The most frequently
applied methodology is training the network, which is an iteration procedure that
successively improves the parameter values. One possible method is the following.
In the first step, we change the weights between the output layer and the hidden
layer by using the following equation:
w jk (t + 1) = w jk (t ) + β E k (t ) yk (t ) (3.59)
where β is the learning rate, yk (t) is the actual output, and Ek (t) = dk (t) – yk (t) is the
difference between the desired and actual outputs. The same equation cannot be
used to estimate the weights between the input and hidden layers; the hidden
variables are unknown, so we do not have their values. Instead, the following rule
is used:
wij (t + 1) = wij (t ) + β z j (t )∑w k

jk (t ) Ek (t ) (3.60)
This updating process is used iteratively for a large number of input–output pairs
until the error terms Ek (t) become sufficiently small.
3.3.2 OBSERVERS
In analyzing practical systems such as those arising in water resources management,
some state variables cannot be directly measured. However, by constructing a special
feedback structure the state can be made indirectly available. Such feedback struc-
tures are called observers. Let
ẋ = Ax + Bu
(3.61)
y = Cx
FIGURE 3.4 Feedback structure of an observer.
be a time-invariant continuous linear system. Assume that its input u(t) and output
y(t) are available for any t ≥ 0, but we have no way to measure its state variable,
x(t). By building an identical system, we hope that its state will be a good estimate
for the state of the original system; however, in the case of unstable systems this is
not the case. Assume that the copy of the system is governed by the systems equation:
ż = Az + Bu (3.62)
Subtract Eq. (3.62) from (3.61) to see that
x˙ − z˙ = A( x − z )
so the discrepancy between the states satisfies a homogeneous equation. It is well

known from systems theory (see, for example, Szidarovszky and Bahill, 1992), that
x (t ) − z(t ) = e A( t −t0 ) ( x (t0 ) − z(t0 ))
so x(t) – z(t) may become very large for increasing values of t. This difficulty can
be overcome by the feedback structure shown in Figure 3.4.
By subtracting the systems equations of the original and copy systems we obtain:
x˙ − z˙ = ( Ax + Bu) − ( Az + Bu − K y e ) = A( x − z ) + K (Cx − Cz ) = ( A + KC )( x − z ) (3.63)
If the original system is observable, then a constant matrix K exists such that
all eigenvalues of A + KC have negative real parts implying that x(t) – z(t) → 0 as
t → ∞. That is, for large values of t, the online measurements of the state of the
copy system are the same as for the state of the original system, so we are able to
observe x(t) by measuring the value of z(t). This theoretical basis for constructing
observers is known as the eigenvalue placement theorem (see, for example,

Szidarovszky and Bahill, 1992, pp. 352–353).
3.3.3 STOCHASTIC SIMULATION

The consequences of events affected by random elements are usually assessed by
using simulation methods. The basis of such methods is the random number gener-
ators of computers which produce uniformly distributed, independent random num-
bers between 0 and 1. However, most random elements in nature have different
distributions, so these random numbers cannot be used directly. A transformation
method is usually used to obtain random values from other than uniform distributions.
Consider first a discrete random variable with values x1, x2, … and probabilities
of occurring p1, p2, …. Divide the [0,1] intervals into subintervals [0, p1), [p1, p1 +
p2), [p1 + p2, p1 + p2 + p3), and so on, and let I1, I2, … denote these intervals. If U
is a uniform variable in [0,1], then P(U ∈ Ik) = pk.
Notice that I1 ∪ I2 ∪ … = [0,1), and if U ≠ 1, then U belongs to exactly one of
the subintervals. Define the random variable X as:
X = xk if U ∈ Ik (3.64)
This is the random variable we wish to generate, as its values are the xk numbers,
k = 1, 2, …, and P(X = xk) = P(U ∈ Ik) = (p1 + p2 + … + pk) – (p1 + p2 + … + pk–1) = pk.
Example 3.15
Let X be a Bernoulli variable with probability p. That is, the value of X is either 1
or 0, and
P( X = 1) = p and P( X = 0) = q = (1 − p)
In this case I1 = [0,p) and I2 = [p,1). So, from a uniform random number, X can be
obtained as:
⎧1 if U < p
X=⎨ (3.65)
⎩0 otherwise.
Example 3.16
Binomial variables can be generated by using the above general method. However,
we can get the same result by using the following simple idea. Let X1, X2, …, Xn be
independent Bernoulli variables with the same parameter p. Then, X = X1 + X2 + …
+ Xn is a binomial variable with parameters n and p. Therefore, the simple procedure
of the previous example is used n times, and the resulting random numbers are added.
Example 3.17
Poisson variables have only one parameter, λ, and take on values 0, 1, 2, … with
probabilities pk = P(X = k) = (λk/k!)e–λ. One may proceed with the general method
and taking advantage of the recursive relation pk = pk–1λ/k. However, exponential
random variables can be effectively used here as well. The way to simulate expo-
nential variables will be discussed later.
Let X1, X2, … be a sequence of independent exponential random variables with
the same λ as a parameter, so E(Xk) = 1/λ for all k, and add the Xk values until:
k −1 k
∑ X <1≤ ∑ X
i =1
i
i =1
i (3.66)
Then X = k gives a random value of the Poisson variable with the same parameter
value λ.
Continuous random variables are generated by using their cumulative distribu-
tion functions. Let X be a random variable with distribution function F(x), let U be
uniform in [0,1], and let u be a random value of U. Then, the solution of equation
F(x) = u for x gives a random value of X. In order to see this, consider the distribution
function of X at any real value t:
P( X ≤ t ) = P( F( X ) ≤ F(t )) = P(U ≤ F(t )) = F(t )
because U is uniform in [0,1]. Here, we assumed that F is strictly increasing;

therefore, equation F(x) = u has a unique solution for all 0 < u < 1. The events U = 0
and U = 1 occur with zero probability.
Example 3.18
If X is exponential with parameter λ, then F(x) = 1 – e–λx, so we have to solve
equation:
1 − e − λx = u
which implies that:
1
x=− ln(1 − u) (3.67)
λ
Example 3.19
Random values for a Weibull variable can be obtained simply by using exponential
variables and the definition of a Weibull distribution. It is well known that if Y is
exponential with parameter α, then X = Y1/β with some β > 0 is a Weibull variable
with parameters α and β. Then,
1/ β
x = ⎡⎢− ln(1 − u)⎤⎥
1
(3.68)
⎣ λ ⎦
gives a random Weibull value, where u is a uniform value in [0,1].
Example 3.20
It is well known that if Xk (k = 1, 2, …, n) are independent exponential variables

with the same λ parameter, then X1 + X2 + … + Xn is a gamma variable with
parameters α = n and β = 1/λ. Hence, we must generate independent uniform variable
values u1, u2, …, un, then:
n n
X= ∑
k =1
xk = −
1
λ ∑ ln(1 − u ) = − λ1 ln[(1 − u )(1 −u )⋯(1 − u )]
k =1
k 1 2 n (3.69)
follows the given gamma distribution.
Example 3.21
For a normal distribution, the general method cannot be used easily, as the standard
normal distribution function Φ is not given in analytic form; it is only tabulated.
Therefore, the solution of the equation:
x − µ⎞
Φ⎛ =u
⎝ σ ⎠
requires the use of a function table, which can be included in the general software.
An easier approach is offered by the central limit theorem, which implies that if u1,
u2, …, un are independent uniform numbers in [0,1], then for large value of n,
∑u − 2
n
k
Z= k =1
(3.70)
n
12
is a standard normal value. So if µ and σ2 are given, then
X = σZ + µ (3.71)
follows the normal distribution with mean µ and variance σ2.

Consider now any process that depends on random elements. Let X1, X2, …, Xn
denote the random variables of the process. Assume that the outcome is a function
of these variables Y = f(X1, …, Xn), where function f might be a given function or
represents a computational procedure such as given by neural networks. For any
random values x1(i ) , ⋯, xn( i ) of X1, X2, …, Xn, the corresponding values of Y can be
obtained as y (i) = f ( x1(i ), ⋯, xn( i )). Let N be the number of simulations, so i = 1, 2, …,
N. The values y (1), y (2), …, y (N) can be considered as a random sample of Y, so
standard statistical methods can be used to examine the characteristics of the random
outcome. In most cases, we are interested in E(Y), which is estimated by the sample
mean:
E(Y ) ≈
1
N ∑y
i =1
(i )
=y (3.72)
so the sample mean is accepted as the expected value. The accuracy of this estimation
can be characterized by its variance:
()
Var y =
1
N
(
Var f ( X1, ⋯, Xn ) ) (3.73)
which converges to zero as N → ∞. If N is sufficiently large, then y can be considered

a normal variable as the result of the central limit theorem. Therefore, we have for
any ε > 0,
⎛ ε N y − E(Y ) ε N ⎞
( ) (
P y − E(Y ) < ε = P − ε < y − E(Y ) < ε = P⎜ −
⎝ σ
)<
σ/ N
<
σ ⎠
⎟
where σ2 = Var(f(X1, …, Xn)). If Φ again denotes the standard normal distribution

function, then this probability can be further simplified as:
⎛ε N⎞ ⎛ ε N⎞ ⎛ε N⎞
Φ⎜ ⎟ − Φ⎜ − ⎟ = 2 Φ⎜ ⎟ −1 (3.74)
⎝ σ ⎠ ⎝ σ ⎠ ⎝ σ ⎠
This is the probability that y approximates the true expectation E (Y) within
the error bound ε. Notice that as N → ∞, this probability value converges to
2Φ (∞) – 1 = 1.
Rubinstein (1981) gives a comprehensive summary of random number gener-
ators and simulation methods.
3.4 MEASURES OF RELIABILITY

Reliability theory is concerned with the statistical properties of the lifetimes of
products, machinery, hydrological objects, systems, and so on. Because the lifetime
depends on the particular object or system we consider, it must be considered to be

a random variable. Let X be the time until the first failure occurs. It is well known
that the distribution function of X is defined as:
F(t ) = P( X ≤ t ), (3.75)
which represents the probability that the object or system breaks down before time
period t. The reliability function gives the probability that it will work without
breaking down at least until time period t:
R(t ) = P( X > t ) = 1 − P( X ≤ t ) = 1 − F(t ) (3.76)
In reliability theory, the Weibull distribution is assumed to model lifetime most

appropriately. In this case, (see Example 3.10):
β
F(t ) = 1 − e − αt , (t > 0) (3.77)
where α, β > 0 are given parameters. The density function is obtained by simple
differentiation:
β
f (t ) = α β t β−1e − αt (t > 0 ) (3.78)
Notice that in the special case of β = 1, f(t) = αe–αt, which is the density function
of the exponential variable. The exponential variable is very seldom used in reliability
studies, as it has the so-called forgetfulness property, which was already discussed
earlier and is repeated here for convenience. If X is an exponential variable, then
for all t, τ >0:
P( X > t + τ X > τ) = P( X > t ) (3.79)
This relation shows that the probability of the working condition in any time length
t is independent of how long the system was working before. Equation (3.79) can
be shown as:
P( X > t + τ) R(t + τ) e − α ( t + τ )
P( X > t + τ X > τ) = = = − ατ = e − αt = R(t ) = P( X > t )
P( X > τ) R( τ) e
In most practical cases the breakdown probabilities increase in time as the system
becomes older, so the exponential variable is inappropriate.
The hazard-rate is given as:
P(t ≤ X ≤ t + Δt t ≤ X )
ρ(t ) = lim (3.80)
Δt →0 Δt
and shows how often failures will occur after time period t. It is clear that:
P(t ≤ X ≤ t + Δt ) F(t + Δt ) − F(t ) 1 f (t )

ρ(t ) = lim = lim ⋅ = (3.81)
Δt →0 P(t ≤ X )Δt Δt →0 Δt R(t ) R(t )
Therefore, the hazard rate can be computed as the ratio of the density function and
the reliability function.
Example 3.22
Assume X follows the Weibull distribution. Then:
β
R(t ) = 1 − F(t ) = e − αt
Therefore, from Eq. (3.81) we have:
β
αβt β−1e − αt
ρ(t ) = β = αβt β−1 (3.82)
e − αt
which is an increasing polynomial of t showing that failure will occur more fre-
quently for larger values of t. In the special case of an exponential distribution, β = 1
and ρ(t) = α is a constant.
An important problem in reliability engineering is to reconstruct F(t) or the
reliability function if the hazard rate is given, and we now offer a solution to this
problem. Notice first that:
ρ(t ) =
f (t )
=−
(1 − F(t ))′
1 − F (t ) 1 − F (t )
Integrate both sides in the interval [0,t] to obtain:
∫ ρ(τ) dτ = [− ln(1 − F(τ))]

t
= − ln(1 − F(t )) + ln(1 − F(0))
t
0 τ=0
Because F(0) = 0, the second term equals zero, so:
∫ ρ(τ) dτ
t
ln(1 − F(t )) = −
0
implying that
⎛ ⎞
∫ ρ(τ) dτ⎟⎠
t
1 − F(t ) = exp⎜ −
⎝ 0
and, finally,
⎛ ⎞
∫ ρ(τ) dτ⎟⎠
t
F(t ) = 1 − exp⎜ − (3.83)
⎝ 0
Example 3.23
Assume first that the hazard rate is constant, with ρ(t) = α. Then,
⎛ ⎞
∫ αdτ⎟⎠ = 1 − e
t
F(t ) = 1 − exp⎜ − − αt
⎝ 0
showing that the distribution is exponential. Similarly, if ρ(t) is a monomial, ρ(t) =

αβtβ–1, then:
⎛ ⎞
∫ αβτ [ ]
t
dτ⎟ = 1 − exp⎛⎝ − ατβ ⎞⎠ = 1 − e − αt
β
F(t ) = 1 − exp⎜ − β −1 t
⎝ 0 ⎠ 0
demonstrating that the distribution is Weibull.

We know that F(t) is increasing, F(0) = 0, and lim t→∞
F (t) = 1. Similarly, R(t) is
decreasing, R(0) = 1, and limt→∞
R (t) = 0. The basic analytical properties of the hazard
rate are now examined. If β > 2 and X is a Weibull variable, then ρ(0) = 0, lim t→∞
ρ (t)
= ∞, and ρ(t) is a strictly increasing and strictly convex function; if β = 2, then ρ(t)
is linear; if 1 < β < 2, then ρ(t) is strictly increasing and strictly concave; if β = 1,
then ρ(t) is a constant; and if β < 1, then the hazard rate is decreasing in t, in which
case defective systems tend to fail early. So, we have shown that the hazard rate
decreases for a well-made system. Notice that:
f ′(t )(1 − F(t )) + f (t )

2
p ′( t ) =
(1 − F(t ))2
which is positive if and only if the numerator is positive:
f (t )
2
f ′( t ) > − = − f (t ) ρ(t ) (3.84)
1 − F (t )
In this case, ρ(t) is locally increasing; otherwise, ρ(t) decreases.

The reliability of a component can be considered in the way shown here;
however, the reliability of a system with several components depends on the con-
figuration of the components in the entire system. Many systems are arranged in a
series, some have a parallel configuration, and others are a combination of the two.
R1(t) R2(t) Rn(t)
FIGURE 3.5 Series combination of components.
R1(t)
R2(t)
Rn(t)
FIGURE 3.6 Parallel combination of components.
In a series system, the system fails if any of its components fails; in a parallel system,
failure occurs when all components fail simultaneously.
Consider first the series combination shown in Figure 3.5. If Ri (t) denotes the
reliability function of component i (i = 1, 2, …, n), then the reliability function of
the system is given as:
(
R(t ) = P( X > t ) = P ( X1 > t ) ∩ ( X2 > t ) ∩ ⋯ ∩ ( Xn > t ) )
where X is the time when the system fails, and (X1, …, Xn) are the same for the
components. If we assume that the failures of the different components occur inde-
pendently of each other, then:
R(t ) = P( X1 > t ) P( X2 > t ) ⋯ P( Xn > t ) = Π Ri (t )

n
(3.85)
i =1
Notice that the inclusion of a new component in the system makes the reliability
function smaller, as it is multiplied by the new factor, Rn+1 (t), which is less than one.
Consider next the parallel system of Figure 3.6. The system fails, if all compo-
nents fail, so
(
P( X ≤ t ) = P ( X1 ≤ t ) ∩ ( X2 ≤ t ) ∩ ⋯ ∩ ( Xn ≤ t ))
If the components fail independently of each other, then
R(t ) = 1 − F(t ) = 1 − P( X ≤ t ) = 1 − Π P( Xi ≤ t ) = 1 − Π Fi (t ) = 1 − Π(1 − Ri (t )) (3.86)

n n n
i =1 i =1 i =1
where F and Fi are the cumulative distributions of time until the first failure of the
system and component i, respectively. Notice that by including a new component,
the reliability function increases, as the second term is multiplied by 1 – Rn+1 (t),
which is less than one.
In most cases, however, the connections between systems components are a mix
of series and parallel. Then, Eqs. (3.85) and (3.86) are combined appropriately as
is shown in the following example.
Example 3.24
Figure 3.7 shows a system assembled with five components, and components 2 and
3 are parallel. Using Eq. (3.86) we obtain:
( )
3
R2 (t ) = 1 − Π 1 − R2i (t )
i =1
and
( )(
R3 (t ) = 1 − 1 − R31 (t ) 1 − R32 (t ) )
Therefore, Eq. (3.85) implies that the reliability function of the system is as follows:
( ) ( )
R(t ) = R1 (t )⎡1 − Π 1 − R2i (t ) ⎤ ⋅ ⎡1 − Π 1 − R3i (t ) ⎤ ⋅ R4 (t ) ⋅ R5 (t )
3 2
⎣⎢ i =1 ⎦⎥ ⎢⎣ i =1 ⎦⎥
As a particular example, assume that:
R1 (t ) = R4 (t ) = R5 (t ) = e − t
and
R2i (t ) = R3j (t ) = e −2 t (1 ≤ i ≤ 3,1 ≤ j ≤ 2)
R 21(t)
R 31(t)
R 1(t) R 22(t) R 4(t) R 5(t)
R 32(t)
R 23(t)
FIGURE 3.7 Combined connections in a system.

and the reliability of the system at t = 0.1 has to be determined. At t = 0.1,
R1 (0.1) = R4 (0.1) = R5 (0.1) = e −0.1 = 0.9048
R2i (0.1) = R3j (0.1) = e −0.2 = 0.8187
and we have:
R2 (0.1) = 1 − 0.18133 = 0.9940
and
R3 (0.1) = 1 − 0.18132 = 0.9671
Hence,
R(0.1) = 0.90483 (0.9940)(0.9671) = 0.7121 = 71.21%
We mention finally that Milton and Arnold (1995) provide much more details
of the fundamentals of reliability theory.
3.5 STOCHASTIC PROCESSES

Suppose that a certain event occurs at random time points 0 ≤ t1 < t2 < …. These
events constitute a stochastic process. For example, the times when earthquakes,
floods, rainfalls, and so on occur represent stochastic processes. The process math-
ematically can be completely defined if we know the distribution functions of t1,
t2 – t1, t3 – t2, …. A very important characteristic of stochastic processes is the
number of events, N(t), that occur in the time interval [0,t]. The most frequently
used stochastic process is the Poisson process, which is defined by the following
properties:
1. N(0) = 0.
2. The number of events occurring in mutually exclusive time intervals are
independent.
3. The distribution of the number of events occurring in any given time
interval depends on only the length of the interval and not on its location.
P( N ( Δt ) = 1)
4. lim =λ (3.87)
Δt →0 Δt
where λ > 0 is a constant.
P( N ( Δt ) ≥ 2)
5. lim =0 (3.88)
Δt →0 Δt
Condition (1) shows that the process starts at t = 0. If 0 < t1 < t2 < t3 < t4, then
condition (2) states that the number of events in interval [t1,t2], which is N(t2) – N(t1),
is independent of the number of events in interval [t3,t4], which is N(t4) – N(t3). This
condition is called the independent increment assumption. Condition (3) shows that
the distribution of N(t + s) – N(t) depends on only s and is independent of t. This
condition is known as the stationary increment assumption. Conditions (4) and (5)
state that in a small interval of length Δt, the probability that one event occurs is
approximately λΔt, but the probability that at least two events occur is approximately
zero.
Under conditions (1) to (5), for k = 0, 1, 2, …,
(λt )k − λt
P( N (t ) = k ) = e (3.89)
k!
That is, N(t) has a Poisson distribution with parameter λt. The random variables
X1 = t1, X2 = t2 − t1, X3 = t3 − t2 ,⋯
represent the sequence of inter-arrival times. Let their distribution functions be

denoted by F1, F2, F3, …. Then, clearly,
F1 (t ) = P(t1 < t ) = 1 − P(t1 ≥ t ) = 1 − P( N (t ) = 0) = 1 − e − λt (3.90)
from Eq. (3.89); therefore, X1 is exponentially distributed with expectation 1/λ. For
all k ≥ 2 and s, t ≥ 0,
P( Xk > t Xk −1 = s) = P(0 event occurs in( s, s + t ) Xk −1 = s)
= P(0 event occurs in(s, s + t )) (3.91)

= P(0 event in(0, t )) = P( N (t ) = 0) = e − λt
so the distribution function of Xk is the same as that of X1 and is given in Eq. (3.90).
Hence, all variables X1, X2, X3, … are exponential with the same parameter λ, and
they are independent of each other. More details on the Poisson process are offered,
for example, in Ross (1987).
3.6 MARKOV CHAINS

Consider a dynamic system with discrete time scales, t = 0, 1, 2, …. Assume that
at each time period the state of the system can be any one from the finite set {S1,
S2, S3, …, Sn}, and at each time period the state changes randomly. Let X(t) denote
the state at time period t and assume that the transition probabilities:
{
Pij = P X (t + 1) = S j X (t ) = Si } (3.92)
are independent of t. Notice that Pij is the probability that the state at time period
t + 1 is Si given that the state in the previous time period t is Si. It is convenient to
form the transition matrix:
⎛ P11 P12 ⋯ P1n ⎞

⎜ ⎟
⎜ P21 P22 ⋯ P2 n ⎟
P=⎜ ⎟ (3.93)
⎜⋮ ⋮ ⋮ ⎟
⎜P Pn 2 ⋯ Pnn ⎟⎠
⎝ n1
which has the following fundamental properties:
1. 0 ≤ Pij ≤ 1, for all i and j.

n
2. ∑ P = 1 for all i.
j =1
ij
Any matrix satisfying these conditions is called a stochastic matrix.

At any time period t, let pi (t) denote the probability that the system is in state Si:
pi (t ) = P( X (t ) = Si ) (3.94)
Clearly,
3. 0 ≤ pi (t) ≤ 1 for all i and t;

n
4. ∑ p (t) = 1 for all t.

i =1
i
Next, define vector p(t )T = ( p1 (t ), p2 (t ),..., pn (t )) as a row vector. Any n-element

vector satisfying these two conditions is called a probability vector. Notice that pi (t)
gives the probability that the system is in state Si at time period t. By using the
transition matrix P, we are able to establish a direct relation between the consecutive
state-probability vectors p (t) and p (t + 1). By using the well-known idea of
conditioning, it is easy to see that:
(
p j (t + 1) = P X (t + 1) = S j ) ∑ P( X(t + 1) = S
= j )
X (t ) = Si P( X (t ) = Si )
i =1
(3.95)
n
= ∑ P p (t )
i =1
ij i
showing that
p(t + 1)T = p(t ) T P (3.96)
or by taking the transpose of both sides:
p(t + 1) = P T p(t ) (3.97)
Notice that this is a homogeneous, discrete, linear system with state-vector p(t )
at time period t, so all methods known from systems theory can be applied here.
Let p (0)T be the probability vector of the initial time period. Then, by the
repeated application of Eq. (3.96), we obtain:
p(1)T = p(0)T P,
p(2)T = p(1)T P = p(0)T P2
and so on, and in general:
p(t )T = p(0)T P t (3.98)
for all t ≥ 1.
Example 3.25
Assume that the weather in a city can be characterized on any day as sunny (S1),
cloudy (S2), or rainy (S3). If it is sunny one day, then the probability that it will
remain sunny or become cloudy is 50% for either. If it is cloudy one day, then the
probability that the next day will be sunny is 40%; cloudy, 30%; and rainy, 30%. If
it is rainy one day, then the probability that it will be rainy again the next day is
50%; cloudy, 40%; and sunny, 10%. Based on these probability values we can
construct the transition matrix:
⎛ 0.5 0.5 0 ⎞
P = ⎜⎜ 0.4 0.3 0.3⎟⎟
⎜ ⎟
⎝ 0.1 0.4 0.5⎠
Assume that one day the weather is sunny; that is,
p(0)T = (1, 0, 0)
Then, from Eq. (3.96), with t = 0 we obtain:
⎛ 0.5 0.5 0 ⎞
p(1)T = (1, 0, 0)⎜⎜ 0.4 0.3 0.3⎟⎟ = (0.5, 0.5, 0)
⎜ ⎟
⎝ 0.1 0.4 0.5⎠
⎛ 0.5 0.5 0 ⎞
p(2) = (0.5, 0.5, 0)⎜⎜ 0.4 0.3 0.3⎟⎟ = (0.45, 0.40, 0.15)
T
⎜ ⎟
⎝ 0.1 0.4 0.5⎠
and so on. For example, the components of p(2)T show that two days later the
probability that the weather will be sunny is 45%; cloudy, 40%; and rainy, 15%.
A Markov chain is regular, if P t > 0 for some possible integer t. In the previous
example,
⎛ 0.45 0.4 0.15⎞

p = ⎜⎜ 0.35 0.41 0.24⎟⎟ > 0
2
⎜ ⎟
⎝ 0.26 0.37 0.37⎠
Thus, our weather example is described by a regular Markov chain. Regular Markov
chains satisfy the following condition:
5. There is a unique probability vector p T such that:
pT = pT P (3.99)
which is called the long-term probability vector of the system.

6. For arbitrary initial state probability vector p (0)T,
lim p(0)T P t = p T (3.100)

t →∞
That is, the left-hand side limit always exists and equals p T .
7. The limit:
lim P t = P (3.101)
t →∞
is an n × n matrix for which each of the rows is equal to p T .

Equation (3.99) and the additional equation:
∑ p =1
i =1
i
provide a system of linear equations to determine the long-term probability vector.

Notice that Eq. (3.99) consists of n linear equations, which are linearly dependent,
as by adding the components of both sides we have:
pT 1 = 1
and
p T P1 = p T ( P1) = p T 1 = 1
The additional equation serves as the nth required independent equation in order to
guarantee a unique solution.
Example 3.26
In the previous example, Eq. (3.99) and the additional equation give the linear system:
p1 = 0.5 p1 + 0.4 p2 + 0.1 p3

p2 = 0.5 p1 + 0.3 p2 + 0.4 p3
p3 = 0.3 p2 + 0.5 p3
p1 + p2 + p3 = 1.
It is easy to see that the unique solution is the following:
23 25 15
p1 = ≈ 0.365, p2 = ≈ 0.397, p3 = ≈ 0.238
63 63 63
The fundamentals of Markov chains and special cases are discussed, for example,
in Luenberger (1979).
3.7 OPTIMIZATION UNDER UNCERTAINTY

In hydrology, the consequences of our decisions as well as our options depend on
random elements not known at the time when the decisions are made. In this section,
two popular methodologies will be discussed that work well to solve optimization
problems with random and uncertain factors.
3.7.1 CHANCE-CONSTRAINED PROGRAMMING

Consider an optimization problem for which both the objective function and the
constraints depend on a random variable:
Maximize ( )
f x, ξ
(3.102)
Subject to g( x, ξ) ≥ 0
where x ∈ Rn is the decision vector, ξ is a random variable, f is a real-valued function,

and g ∈ Rm. It is usual to replace the stochastic objective function by its expectation
with respect to ξ ; however, this expectation-optimizing concept itself does not
require the feasibility of the optimal solution. For any decision x, there is always a
chance that it is not feasible for certain values of ξ . Therefore, the constraints are
substituted by the requirements that the probability of infeasibility has to be below
a certain tolerance level. This condition can be posed on each individual constraint
or on the entire constraint set. In the case of infeasibility, losses might occur, and
their expectations are usually added to the expected value of the original objective
function. Therefore, we now consider the following problem:
Maximize [ ( )] [ ( ( ))]
E f x, ξ − E L g x, ξ
(3.103)
Subject to P( g( x, ξ) ≥ 0) ≥ 1 − ε
where L is a loss function with zero value if the constraints are all satisfied. This
problem is now a deterministic nonlinear optimization problem that can be solved
by standard methodology. Let gi denote the ith component of g. Then, alternatively,
the constraint of Eq. (3.103) might be replaced by the set of constraints:
(( ) )
P gi x, ξ ≥ 0 ≥ 1 − ε i (for all i) (3.104)
We now introduce the notation:
{ ( )
F( x ) = ξ g x, ξ ≥ 0 } (3.105)
and let p denote the density function of ξ . Then, the constraint of Eq. (3.103) can
be rewritten as:
∫ p(ξ)d ξ ≥ 1 − ε
F( x )
(3.106)
and, similarly, the objective function is also an integral:
∫ [ f ( x, ξ) − L(g( x, ξ))] p(ξ)d ξ (3.107)
In practical applications, numerical integration is used in computing the objective

function values (see, for example, Szidarovszky and Yakowitz, 1978). A good survey
of stochastic optimization methodology is offered in Wets (1983).
3.7.2 STOCHASTIC DYNAMIC PROGRAMMING

In the case of DP formulation, we assumed that the state transition relation:
xt +1 = ft ( xt , ut ) (3.108)
uniquely determines the next state given the current state and decision. In the
stochastic version (SDP), this deterministic function is replaced by a probabilistic
transition function pt (x,xt,ut), which gives the probability density function of xt+1 for
fixed t, xt, and ut . Because of the presence of stochastic elements, a decision strategy
is a sequence of function ut = Dt (x), which selects the decision depending on the
particular state xt = x occurring at time period t. We also replace the deterministic
objective function by its expectation:
⎡ T ⎤
E⎢ ∑
⎢⎣ t =0
gt ( xt , xt +1 , Dt ( xt ))⎥
⎥⎦
(3.109)
where we assume that gt also depends on the next state. In the deterministic case,
xt+1 is uniquely determined by xt and ut, so xt+1 would be a superfluous variable. The
stochastic version of the DP recursive equation is the following:
Vt ( x ) = max ⎡⎢
u ⎣ ∫ (g ( x, x' , u) + V
t t +1 ( x' ))⎤⎥ pt ( x' , x, u)dx ′
⎦
(3.110)
with the choice of VT+1 (x) = 0 for all x. Let u = St (x) denote an optimal choice.
Then, Dt (x) = St (x) is optimal for objective (3.109). The corresponding discrete
stochastic dynamic programming (DSDP) algorithm is based on the recursion:
⎧⎪ ⎫⎪
( )
Vˆt x ( i ) = min ⎨
j ∑ g (x
⎪⎩ k
t
(i )
) ( ) ( )
, x ( k ) , u( j ) + Vˆt +1 x ( k ) Pt x ( k ) , x ( i ) , u( j ) ⎬
⎪⎭
(3.111)
where Pt is the probability mass function obtained from the density function pt and
the discretization scheme being used. The curse of dimensionality can be made less
burdensome by using some version of DDDP or IDP. The survey article of Yakowitz
(1982) has more details of the methodology discussed here and applications in to
water resources management.
3.8 RISK, UNCERTAINTY, AND BAYESIAN DECISION

THEORY
Decision problems under uncertainty are usually formulated as:
Maximize f ( x, ξ)
(3.112)
Subject to x ∈ X (ξ)
where ξ is the notation for the uncertain phenomenon. Notice that both the objective
function and the feasible set depend on ξ. First, we show that the dependence of
X(ξ) on ξ can be formally eliminated. For this purpose, we define:
X = U X (ξ) (3.113)
ξ∈∑
where ∑ is the set of all possible values of ξ, and we let the modified objective
function be:
⎪⎧ f ( x, ξ) if x ∈ X (ξ)
g( x, ξ) = ⎨ (3.114)
⎩⎪−∞ otherwise.
Thus, the problem can be reformulated as:
Maximize g( x, ξ)
(3.115)
Subject to x ∈ X.
In the classical approach, the maxmin rule is most frequently applied. For each
decision alternative x we compute the worst possible outcome:
gw ( x ) = inf g( x, ξ) (3.116)
ξ
Then, the alternative with the highest worst outcome, x*, is selected as the decision:
gw ( x * ) = sup gw ( x ) = sup inf g( x, ξ) (3.117)

x x ξ
This solution concept is based on worst-case analysis in which decision makers

try to protect themselves as much as possible against the worst possibility that might
occur. In many cases, worst possibilities occur only very seldom, so protection
against such events seems to be overprotection in most cases. Another approach to
choose a single decision from X is to assign a weight function w(ξ) over all possible
values of ξ and choose the decision that maximizes:
∫ g( x, ξ)w(ξ)dξ (3.118)
∑
and if ∑ contains only finite or countable many values, then the integral is replaced
by the summation:
∑ g( x, ξ)w(ξ)
ξ
(3.119)
It seems reasonable to assign a particular weight function w(ξ) that reflects the
relative likelihood of the different values in ∑ being the true value of ξ. With this
reasoning, we have revealed the basic concept of statistical decision making.
3.8.1 RISK AND VALUE OF INFORMATION
In statistical decision making, we always assume that the uncertainty in ξ is based

on its randomness. Let h denote the density function of ξ (or the joint density
function if ξ has more than one component). According to Eqs. (3.118) or (3.119),
the expected value of the objective function is maximized:
Maximize E[g( x, ξ)]

(3.120)
Subject to x ∈ X.
Let x* be the optimal solution. For each particular value of ξ let x(ξ) be the
maximizer of g(x,ξ). Because the true value of ξ is unknown, we cannot select the
corresponding optimal decision x(ξ); instead, x* is the choice. This lack of informa-
tion results in the loss (or we face with risk) of:
(
L(ξ) = g( x (ξ), ξ) − g x * , ξ ) (3.121)
for each particular value of ξ. Because the true value of ξ is not known, we can
compute only its expectation:
[ ] [(
E[ L(ξ)] = E g( x (ξ), ξ) − E g x * , ξ )] (3.122)
which gives the average loss occurring by not knowing the value of ξ. In other
words, this is the value of the knowledge of the true value of ξ, or it is the value of
the perfect information on ξ.
Most decisions in water resources management are made repeatedly over time.
As time progresses, new information about the uncertain factors is gained and their
distributions can be updated so more accurate decisions can be made. The method-
ology used in such situations is known as Bayesian decision making, the elements
of which are the subject of the next subsection.
3.8.2 BAYESIAN DECISION MAKING

Let again ξ denote the uncertain model parameter. Before additional information is
available its density function is believed to be h(ξ). This density function is called
the prior distribution, because it is the distribution of ξ before additional observations
are made. Let θ be a random variable or vector that is related to the parameter ξ,
and assume that a value of θ is being observed. Then, the posterior density of ξ can
be obtained by using Bayes’ theorem:
h(ξ θ) =
( )
l θ ξ h(ξ)
(3.123)
∫ l(θ ξ' )h(ξ' )dξ'
∑
where l(θ⏐ξ) is the conditional density function of θ given the value of ξ. If ξ has
a discrete distribution, then the denominator of the right-hand side is replaced by a
sum. So, after observing θ, the density function of ξ is updated and a more accurate
decision can be made. The benefit of the additional information θ is given by the
difference:
( ) (
g x ** , ξ − g x * , ξ ) (3.124)
where x** is the optimal decision based on the updated density function h(ξ⏐θ).
Because ξ is unknown and an updated density is available, the expected benefit is
computed as:
∫ ∫ [g( x , ξ) − g( x , ξ)]h(ξ θ)dξl (θ)dθ

** *
(3.125)
Θ ∑
where l (θ) is the denominator of the right-hand side of Eq. (3.123), and Θ is the
set of the values of θ. Based on Eq. (3.125), we can decide if it is worthwhile to
wait for an additional observation by comparing the expected benefit to the cost of
collecting the additional observation and any other losses occurring by delaying the
decision.
3.8.3 CONJUGATE DISTRIBUTION FAMILIES

Assume that an observation is made on a random variable or vector θ as was assumed
in the previous section. Usually, it is a random sample from one of the standard
distributions such as Bernoulli, binomial, or Poisson. All of these distributions
depend on certain parameters. For the Bernoulli variable, the parameter is p; for
binomial, n and p; for Poisson, λ; and so on. Let π denote the parameter or parameter
vector of the distribution in general. A convenient property of a family of prior
distributions is that it is closed under sampling, which means that, if the prior
distribution belongs to the family, then for any sample size and observation values
the posterior distribution also belongs to the same family. A distribution family with
this property is a conjugate family of prior distributions.
In selecting a conjugate family in a particular application we have to consider
the following. The family has to be small enough so it can be described by only a
few parameters, so computationally it is tractable. The family also has to be rich
enough to allow the decision makers to find within the prior family a distribution
that fits adequately the prior distribution.
Assume first that ξ has a Bernoulli distribution with possible values 1 and 0 and
occurring probabilities P(ξ = 1) = π and P(ξ = 0) = 1 – π. The most commonly
used conjugate family for π is the family of beta distributions with density function:
Γ (α + β) α −1
π (1 − π)β−1 (0 ≤ π ≤ 1) (3.126)
Γ (α )Γ (β)
where α and β are constants. Let θ1, θ2, …, θn be a random sample of the Bernoulli
distribution and assume that its parameter π follows a beta distribution with constants
α and β. Then, the posterior distribution of π given the observed values θ1, θ2, …,
θn is also a beta distribution with constants:
n n
α+ ∑ i =1
θi , β + n − ∑θ
i =1
i
Assume next that ξ is normally distributed with mean m and precision p (which
is 1/σ2), so its density function is:
p ( ξ − m )2
p −
e 2 ( −∞ < ξ < ∞) (3.127)
2π
Assume that precision p is known and the prior distribution of m is normal with
mean µ and precision π. If θ1, θ2, …, θn is a random sample of ξ, then the posterior
distribution of m is also normal with mean:
πµ + p ∑θ
i =1
i
µ' =
π + np
and precision π′ = π + np.

The conjugate family becomes more complicated when both m and p are
unknown. In this case, we have the following result. Assume that ξ is normally
distributed with mean m and precision p, both of which are unknown, and their joint
distribution is as follows: The conditional distribution of m given the value of p is
normal with mean µ and precision τp, where τ is a positive constant and the marginal
distribution of p is gamma with parameters α and β; that is, the marginal density
function of p is given as:
β α α −1 −βp
p e ( p > 0) (3.128)
Γ(α )
Let θ1, θ2, …, θn be a random sample of ξ. Then, the posterior joint distributions
of m and p are as follows. The conditional distribution of m given the value of p is
again a normal distribution with mean:
τµ + ∑θ
i =1
i
τ+n
and precision (τ + n)p, and the marginal distribution of p is again gamma with
parameters
( )
2
n
τn θ − µ
∑ (θ − θ )
n 1 2
α' = α + and β' = β + +
2 2 i =1
i
2( τ + n )
where
∑θ
1
θ= i
n i =1
The most commonly used conjugate families other than the ones presented above
are given, for example, in De Groot (1970).
3.9 FUZZY SETS AND PARAMETER IMPRECISION

Parameter uncertainty in water resources and hydrologic modeling is handled by
randomization or by fuzzification. The previous section discussed randomization
and the corresponding probabilistic methods; in this section, the fundamentals of
fuzzy sets and fuzzy decisions are introduced.
Let X be the set of certain objects. A fuzzy set A in X is a set of ordered pairs:
A= {( x, µ A }
( x )) , x ∈ X (3.129)
where µA : X ֏ [0,1] is called the membership function and µA (x) is the grade of
membership of x in A. In classical set theory, x either belongs to A or does not belong
to A, so µA (x) equals 1 or 0, respectively.
The basic relations and operations of fuzzy sets can be defined as follows:
1. A fuzzy set A is empty, if µA (x) = 0 for all x ∈ X.

2. A fuzzy set A is called normal, if:
sup µ A ( x ) = 1 (3.130)
x
Notice that any nonempty fuzzy set can be normalized by introducing the
normalized membership function:
µ A ( x)
µ A ( x) = (3.131)
sup µ A ( x )
x
3. The support of a fuzzy set A is defined as:
{
S( A) = x x ∈ X , µ A ( x ) > 0 } (3.132)
4. We say that fuzzy sets A and B are equal, if:
µ A ( x ) = µ B ( x ) (for all x ∈ X ) (3.133)
5. A fuzzy set A is a subset of B, if for all x ∈ X,
µ A ( x) ≤ µ B ( x) (3.134)
It is clear that A=B if A is a subset of B and B is also a subset of A. The

relation that A is a subset of B is denoted as A ⊆ B .
6. A′ is the complement of A, if for all x ∈ X:
µ A' ( x ) = 1 − µ A ( x ) (3.135)
Notice that this is the straightforward generalization of the concept of

complement of classical set theory, as x ∈ A if and only if µA (x) = 1 and
x ∉ A′, so µA′ (x) = 0. Similarly, x ∉ A if and only if µA′ (x) = 1 and µA (x) = 0.
7. The intersection of fuzzy sets A and B is defined as:
µ A∩ B ( x ) = min{µ A ( x ); µ B ( x )} ( for all x ∈ X ) (3.136)
Notice that conditions (5) and (7) imply that A ⊆ B if and only if A ∩ B
= A.
8. The union of fuzzy sets A and B is given as:
µ A∪ B ( x ) = max{µ A ( x ); µ B ( x )} (for all x ∈ X ) (3.137)
From conditions (5) and (8), we see that A ⊆ B if and only if A ∪ B = B.

9. The algebraic product of fuzzy sets A and B is denoted by AB and is
defined by the relation:
µ AB ( x ) = µ A ( x )µ B ( x ) (for all x ∈ X ) (3.138)
10. The algebraic sum of fuzzy sets A and B is denoted by A + B and has
the membership function:
µ A+ B ( x ) = µ A ( x ) + µ B ( x ) − µ A ( x )µ B ( x ) (for all x ∈ X ) (3.139)
11. A fuzzy set A is convex, if for all x, y ∈ X and λ ∈ [0,1]:
µ A (λx + (1 − λ ) y) ≥ min{µ A ( x ); µ A ( y)} (3.140)
If A and B are convex, then it is easy to see that A ∩ B is also convex.

12. A fuzzy set A is concave, if A′ is convex. It is easy to see that if A and B
are concave, then A ∪ B is also concave.
13. Let X and Y be two sets and f : X ֏ Y be a mapping from X to Y. Assume
that A is a fuzzy set in X. The fuzzy set B induced by mapping f is defined
in Y with membership function:
µ B ( y) = sup µ A ( x ) (3.141)
x ∈f −1 ( y )
A(x) B(x)
FIGURE 3.8 Membership functions of A and B.
A+B (x)
A∪B (x)
A∩B (x)
AB (x)
FIGURE 3.9 Operations of fuzzy sets A ∩ B, A ∪ B, AB, A + B.
where
f −1 ( y) = {x x ∈ X , f ( x ) = y}
Example 3.27
Let X = [–∞,∞] and fuzzy sets A and B be defined by the membership functions
shown in Figure 3.8. The membership functions of A ∩ B, A ∪ B, AB, and A + B
are presented in Figure 3.9.
Example 3.28
Let X = [–1,1] and assume that
⎧1 + x if − 1 ≤ x ≤ 0
⎪
µ A ( x ) = ⎨1 − x if 0 < x ≤1
⎪
⎩0 otherwise.
Let f(x) = x2. We will now determine the membership function of the fuzzy set
induced by function f. Notice first that f([–1,1]) = [0,1], so µB (y) = 0 if y ∉ [0,1].
If y ∈ [0,1], then
f −1 ( y) = { y, − y }
FIGURE 3.10 Illustration of µA (x) and µB (y).
and because
µ A ( y ) = 1 − y = µ A ( − y ),
µ B ( y) = sup µ A ( x ) = 1 − y .
x ∈f −1 ( y )
Both µA (x) and µB (y) are illustrated in Figure 3.10.

Uncertain model parameters can be considered as fuzzy sets, and the result of
any series of operations is a fuzzy set. Uncertainty in constraints and in the objective
functions also can be expressed by fuzzy sets.
Let X be the set of decision alternatives. A fuzzy constraint is defined by a fuzzy
set C in X, and the membership function value µC (x) shows to what degree an
alternative satisfies the constraint C. Similarly, any objective function can be iden-
tified by a fuzzy set G, and µG (x) shows to what degree an alternative represents an
attainment of the goal represented by this objective function. Any multiple-criteria
decision-making problem can be defined in a fuzzy environment as the set of fuzzy
goals (G1, …, GM) and fuzzy constraints (C1, …, CN). The solution of the problem
is the decision alternative that satisfies the constraints Cj to as high a level as possible
and gives the best possible attainment of the goals. In other words, the intersection
of the fuzzy sets (G1, …, GM) and (C1, …, CN) must have the highest membership
value. Therefore, we consider:
{ }
µ D ( x ) = min µ G1 ( x ), ⋯, µ GM ( x ); µ C1 ( x ), ⋯, µ CN ( x ) (3.142)
and find alternative x that maximizes this membership function.
Example 3.29
Assume we have six alternatives, two constraints, and two objectives. The member-
ship values for all alternatives with respect to the constraints and objectives are given
TABLE 3.2
Membership Values in Fuzzy Decision
Alternatives
1 2 3 4 5 6
µG 1 0.9 0.85 0.95 0.8 0.65 0.9

µG 2 0.95 0.70 0.92 0.9 0.75 0.95
µC 1 0.75 0.8 0.9 0.95 0.9 0.75
µC 2 0.8 0.9 0.91 0.85 0.8 0.8
µD 0.75 0.7 0.9 0.8 0.65 0.75
in Table 3.2. The last row gives the values of µD by selecting the smallest numbers
of all columns. The largest value of µD appears in the third column, so the third
alternative is the best.
Interested readers are referred to Bellman and Zadeh (1970) or to the section
on fuzzy decisions in Szidarovszky et al. (1986).
3.10 PROBLEMS
3.1 Assume that the number of occurrences in time period [0,1] has Poisson
distribution with expectation λ = 3. Find the probability that the first
occurrence will happen before t = 1/2.
3.2 A density function of a random variable X is given as:
⎧⎪a( x + 1) if 0 < x < 2

f ( x) = ⎨
⎩⎪0 otherwise.
(a) Find the value of a.

(b) Find F(X).
(c) Compute E(X) and Var(X).
3.3 Find the mode and the median of the random variable of the previous
problem.
3.4 A discrete random variable takes on the values 0, 1, 2, 3 with probability
P(X = k) = a/k2 (k ≥ 1) and P(X = 0) = 2a, where a > 0 is a parameter.
(a) Find the value of a.
(b) Find E(X) and Var(X).
(c) Find the mode and median.
3.5 Find the moment generating function of the random variable of
problem 3.2.
3.6 A joint probability table is given below:
Y
X 0 1
0 0.1 0.3
1 0.1 0.2
2 0.2 A
(a) Find the value of A.

(b) Are X and Y independent?
(c) Compute E(X), E(Y), Var(X), and Var(Y).
(d) Compute Cov(X,Y) and Corr(X,Y).
3.7 A joint density function is given as:
⎧⎪ A( x + y) if 0 < x < y < 1

f ( x, y) = ⎨
⎪⎩0 otherwise.
Solve the previous problem for this continuous case.

3.8 Consider a neural network with two input nodes, two output nodes, and
three hidden nodes. Generate data by selecting the input values x1 and x2
from the unit interval [0,1] and using the output functions y1 = x12 + x22
and y2 = x12 − x 22 . Use back propagation to train the network.
3.9 Construct an observer for system:
⎛1 1⎞ ⎛ 1⎞
x=⎜ ⎟ x+⎜ ⎟ u
⎝0 1⎠ ⎝ 2⎠
y = (1, 1) x
Select matrix K such that the eigenvalues of the feedback system become
–1 and –2.
3.10 Consider the distribution function of a random variable X:
⎧0 if x≤0
⎪⎪
F( x ) = ⎨ x 2 if 0 < x ≤1
⎪
⎪⎩1 if x ≤1
Derive an algorithm to generate random values of X from a uniform

variable.
3.11 Solve the previous problem if:
⎧1 − 1 if x > 1
⎪
F( x ) = ⎨ x
⎪0 if x ≤ 1
⎩
3.12 Find the reliability functions and hazard rates for the random variables of
problems 3.10 and 3.11.
3.13 A parallel system contains 10 elements with common hazard rate ρ(t) =
0.1. Find the hazard rate of the system.
3.14 Solve the previous problem if the system is a series combination of 10
same elements.
3.15 Consider a Poisson process with λ = 2. Find the probability that there are
exactly three events before t = 2.
3.16 Find the long-term probability vector if the probability for the Markov
chain with transition matrix is:
⎛ 0.2 0.6 0.2⎞

P = ⎜ 0.4 0.4 0.2⎟
⎜ ⎟
⎝ 0 0.6 0.4⎠
3.17 Assume that the initial probabilities are:
P1 (0) = 0.1, P2 (0) = 0.6, and P3 (0) = 0.3
Find the probabilities Pi (3), i = 1, 2, 3.

3.18 Consider the following linear programming problem:
Maximize x1 + x 2
2 x1 + x 2 ≤ ξ
x1 + 2 x 2 ≤ 3
where ξ is normally distributed with µ = 3 and σ2 = 1. Give the chance

constraint formulation of the problem, and find the optimal solution.
3.19 Consider the following optimization problem:
Maximize x1 + ξ x 2
x1 + 2 x 2 ≤ 4
2 x1 + x 2 ≤ 4
where ξ is a uniform random variable in [0,1]. Formulate Eq. (3.120) and

find the optimal solution.
3.20 Repeat Example 3.29 for the following fuzzy decision problem:
Alternatives
1 2 3 4 5 6
µG 1 0.8 0.65 0.7 0.9 0.9 0.85

µG 2 0.75 0.72 0.69 0.8 0.8 0.9
µG 3 0.7 0.68 0.65 0.75 0.8 0.8
µG 4 0.81 0.8 0.6 0.85 0.6 0.95
where we have six alternatives to select from and four fuzzy goals or
constraints.
References
Bellman, R. and Zadeh, L. A., Decision making in a fuzzy environment, Manage. Sci., 17(4), 141–164,
1970.
Bose, N. K. and Liang, P., Neural Network Fundamentals with Graphs, Algorithms, and Applications,
McGraw-Hill, New York, 1996.
DeGroot, M. H., Optimal Stochastic Decisions, McGraw-Hill, New York, 1970.
Hecht-Nielsen, R., Neurocomputing, Addison-Wesley, Reading, MA, 1990.
Luenberger, D., Introduction to Dynamic Systems, John Wiley & Sons, New York, 1979.
Milton, J. S. and Arnold, J. C., Introduction to Probability and Statistics, 3rd ed., McGraw-Hill, New York,
1995.
Ross, Sh. M., Introduction to Probability and Statistics for Engineers and Scientists, John Wiley & Sons,
New York, 1987.
Rubinstein, R. Y., Simulation and the Monte Carlo Method, John Wiley & Sons, New York, 1981.
Szidarovszky, F. and Bahill, A. T., Linear Systems Theory, 2nd ed., CRC Press, Boca Raton, FL, 1992.
Szidarovszky, F. and Yakowitz, S., Principles and Procedures of Numerical Analysis, Plenum, New York,
1978.
Szidarovszky, F., Geshon, M., and Duckstein, L., Techniques for Multi-Objective Decision Making in
Systems Management, Elsevier, Amsterdam, The Netherlands, 1986.
Wets, R. J.-B., Stochastic programming: solution techniques and approximation schemes, in Mathematical
Programming: The State of the Art, Bachem, A., Groetschel M., and Korte, B., Eds., Springer-
Verlag, Berlin, 1983, pp. 506–603.
Yakowitz, S., Dynamic programming applications in water resources, Water Resources Res., 18(4),
673–696, 1982.
4 Water Resources
Economics
4.1 INTRODUCTION
Economics consists of a set of analytical tools that can be used for balancing
competing objectives by allocating the scarce resources. Engineering economy is
concerned with the economic aspects of engineering; it involves the systematic
evaluation of costs and benefits of technical projects. Environmental economics is
the application of the principles of economics to the study of how environmental
resources are developed and managed.
The entire process of planning, design, construction, operation, and maintenance
of water resources systems entails many important and complex decisions. Besides
technological and environmental considerations, economic principles play a signif-
icant role in making these decisions. The principles of engineering economics guide
the decision makers in selecting the best planning and/or operational decisions.
This chapter discusses the principles of engineering economics and their appli-
cation to water resources planning and operation. Major objectives of economic
analysis for water resources development projects are also discussed, and various
methods for incorporating the money–time relationship are presented.
4.2 BENEFITS AND COSTS

Water resources development projects have two sides: on the one side, they create
value and on the other side they encounter costs. The value side of the analysis is
based on the fact that individuals have preferences for goods and services. The value
of a good to a person is what that person is willing to pay for that good. Thus, the
value of a good or service is tied to willingness to pay. The marginal willingness to
pay is what each person is willing to pay for an additional unit of a good or service.
The costs associated with different economic activities can be classified as fixed
and variable. Fixed costs are not affected by the range of operation or activity level.
General management and administrative salaries and taxes on facilities are some
examples of fixed costs. Variable costs are those associated with the quantity of
output or other measures of activity level. An incremental cost or incremental revenue
is the additional cost or revenue that results from increasing the output of a system
by one unit.
117
Example 4.1
A contractor has the choice of two sites from which to deliver gravel to a construction
site. The cost of transferring gravel to the site of an earthfill dam is estimated to be
$1.15/m3/km. The costs associated with the two sites can be summarized as follows:
Cost Site A Site B
Distance from construction site 6 km 4.3 km

Monthly rental of site $1000 $5000
Cost to set up and remove equipment $15,000 $25,000
Transfer cost $1.15/m3/km $1.15/m3/km
The contract requires 50,000 km3 of gravel to be delivered by the contractor, and it
is estimated that 4 months will be required for the job. Compare the two sites in
terms of their fixed, variable, and total costs. For the selected site, how many cubic
meters of material must the contractor deliver before starting to make a profit if he
is paid $8.05/m3 delivered to the construction site?
Solution: As shown in the following table, the rent and plant setup are fixed
costs, whereas the transfer cost is a variable cost.
Costs Fixed Variable Site A Site B
Rent (4 months) √ $4000 $20,000

Plant setup √ $15,000 $25,000
Material transfer cost √ 6(50,000)($1.15) = 4.3(50,000)($1.15) =
$345,000 $247,250
_________________________ ___________________________
Total $364,000 $292,250
As can be seen in the table, site B, which has higher fixed costs, has the smaller
total cost. The contractor begins to make a profit at the point where total revenue
equals the total cost of the material delivered. For site B, it can be written that:
4.3($1.15) = $4.95 in variable cost per cubic meter delivered.

Total cost = Total revenue.
$45,000 + $4.95X = $8.05X.
X = 14,516 m3 delivered.
Therefore, by selecting site B, the contractor will begin to make a profit on the job
after delivering 14,516 m3 of material.
Another classification of costs is private vs. social costs. The private costs are
the costs experienced by the decision makers directly, but the social costs include
all costs of an action, no matter who experiences them (Field, 1997). The difference
between social and private costs is the external cost or environmental cost. These
costs are called “external” because, although they are real costs to some members
of society, the firms or agencies that are making decisions and taking actions do not
Water Resources Economics 119
normally consider them when they make decisions. Water resources, which are open
access resources, are important sources of external or environmental costs.
Example 4.2
Suppose three similar industries are situated on a lake. The industries use the water
of the lake and discharge sewage back into the lake. Before using the water, the
industries treat the water, which costs about $40,000 per year for each industrial
unit. Suppose a new industrial unit is planning to start operating on the lake. Because
of the untreated sewage being added to the lake by this unit, each of the existing
industries will have to spend an additional $20,000 each year for water treatment.
How much is the external cost of the new industrial unit operation?
Solution: Because each industrial unit will spend $20,000 per year for the
additional treatment, the external cost is $60,000 ($20,000 × 3).
A similar definition is used by economists for external benefits. An external
benefit is a benefit enjoyed by somebody who is outside of, or external to, decisions
being made about consuming resources or using services. For example, consider a
private power company that has constructed a dam for producing hydropower. Even
though the dam produces downstream benefits such as mitigating both floods and
low flows, the power company is not reimbursed for them; therefore, these benefits
are classified as external benefits.
Although an economist seeks to evaluate all consequences of an action in
commensurate monetary units, many values cannot be measured within such a
framework. For example, many actions have direct physical effects on humans such
as loss of health or life, emotional effects such as loss of prestige or personal integrity,
and psychological effects such as environmental changes. These effects cannot be
measured in monetary units; however, various economists have devoted a great deal
of energy to quantifying these positive and negative effects within a monetary
framework. A value that cannot be expressed in monetary terms is called intangible
or irreducible (James and Lee, 1971).
4.3 MONEY–TIME RELATIONSHIP

Most engineering activities involve the commitment of capital for extended periods;
therefore, the effects of time should be considered. The same amount of money
spent or received at different times has different values because opportunities are
available to invest the money in various enterprises to produce a return over a period.
Generally, the future value of a particular amount of money will be larger than
the existing amount today. Financial institutions are willing to pay interest on
deposits because they can lend the money to investors. If an amount of money is
deposited in a bank, interest accrues at regular time intervals. These time intervals
are called interest periods. The interest rate can be defined as the rate at which
money increases in value from the present to the future. The discount rate refers to
the rate by which the value of money is discounted from the future to the present.
The interest accrued in a single interest period is called simple interest. If the earned
1 2 N-1 N
P(Given)
FIGURE 4.1 General cash flow diagram relating present equivalent and future equivalent of
single payments.
interest is not withdrawn at the end of an interest period and is redeposited with the
original investment in the next interest period, the interest thus accrued is referred
to as compound interest.
The following notation is used in this section:
i: Interest rate per interest period

N: Number of compounding periods
P: Present sum of money; the equivalent value of one or more cash flows at
a reference point in time called the present
F: Future sum of money; the equivalent value of one or more cash flows at
a reference point in time called the future
A: End-of-period cash flows (or equivalent end-of-period values) in a uni-
form series continuing for a specific number of periods, starting at the
end of the first period and continuing through the last period
G: End-of-period uniform gradient cash flows
4.3.1 FUTURE VALUE AND PRESENT VALUE OF MONEY

Figure 4.1 is a general cash flow diagram relating present, P, and future, F, single
sums separated by N periods with an interest rate of i percent per period. As can be
seen in this figure, P dollars is invested at the beginning of period one. This amount
will grow to P + Pi = P(1 + i) by the end of period one. Therefore, the future amount
after N periods is equal to:
F = P(1 + i ) N (4.1)
The quantity (1 + i)N is the single-payment compound amount factor. The (F/P,
i%, N) functional symbol is also used in this book for (1 + i)N. For finding P when
F is given, the following relation can be used:
( )
N
P = F⎛
1 ⎞
= F(1 + i ) − N = F P F , i%, N (4.2)
⎝1+ i⎠
A A A A
N
1 2 N-1
P F
FIGURE 4.2 General cash flow diagram relating series of cash flows to present and future
equivalent values.
The quantity (1 + i)–N is the single-payment present worth factor.
4.3.2 FUTURE VALUE OF MONEY AND END-OF-PERIOD

CASH FLOWS
Consider the cash flow diagram shown in Figure 4.2. We can treat each of cash flows
separately and find their future values as follows:
[
F = A (1 + i )
N −1
+ (1 + i )
N −2
+ ⋯ + (1 + i ) + (1 + i )
1 0
]
which can be reduced to:
⎡ (1 + i ) N − 1 ⎤
F = A⎢ ⎥ = A F A , i%, N ( ) (4.3)
⎣ i ⎦
The quantity (F/A, i%, N) is the uniform series compound amount factor. For
finding A, when F is given, the following relation can be used:
⎡ ⎤
A = F⎢
(1 +
i
) N
− 1
⎥ = F A F , i%, N( ) (4.4)
⎣ i ⎦
(A/F, i%, N) is called sinking fund factor or capital recovery factor. This factor
indicates the number of dollars one can withdraw in equal amounts at the end of
each of N periods if $1 is initially deposited at i% interest.
4.3.3 PRESENT VALUE OF MONEY AND END-OF-PERIOD

CASH FLOWS
Using Eqs. (4.1) and (4.3) we obtain (see Figure 4.2):
⎡ (1 + i ) N − 1 ⎤
P = A⎢ N ⎥ = A (
P , i%, N ) (4.5)
⎣ i(1 + i ) ⎦
A
(P/A, i%, N) is the uniform series present worth factor. To find A when P is
given, the following relation can be used:
⎡ i(1 + i ) N ⎤
A = P⎢
(1 + ) N
− 1
(
⎥ = P A P , i%, N ) (4.6)
⎣ i ⎦
The quantity (A/P, i%, N) is the capital recovery factor.
4.3.4 FUTURE VALUE OF MONEY AND END-OF-PERIOD UNIFORM

GRADIENT CASH FLOWS
Figure 4.3 shows a cash flow diagram of a sequence of end-of-period cash flows
increasing by a constant amount, G, in each period. The future equivalent of sequence
of cash flows shown in Figure 4.3 can be computed as follows:
( ) ( ) ( ) (
F = G F A , i%, N − 1 + G F A , i%, N − 2 + ⋯ + G F A , i%, 2 + G F A , i%, 1 )
⎡ (1 + i ) N −1 − 1 (1 + i ) N −2 − 1 (1 + i)2 − 1 (1 + i)1 − 1 ⎤
F = G⎢ + +⋯+ + ⎥
⎣ i i i i ⎦
which reduces to:
G⎡ ⎤
N −1
∑
NG
F= ⎢ (1 + i) j ⎥ −
i ⎢ ⎥⎦ i
⎣ j =0 (4.7)
F=
G F
i
(
A , i%, N −
NG
i
)
4.3.5 PRESENT VALUE OF MONEY AND END-OF-PERIOD UNIFORM
GRADIENT CASH FLOWS
The following relation can be used for estimating the present equivalent value of
cash flows shown in Figure 4.3:
G 2G ( N − 2)G ( N − 1)G
P= + +⋯+ +
(1 + i)2 (1 + i)3 (1 + i) N −1 (1 + i) N
FIGURE 4.3 General cash flow diagram relating the uniform gradient of increasing cash
flow to future and present equivalent values.
which reduces to:
⎡ (1 + i ) N − 1 − Ni ⎤
P = G⎢
⎣ i (1 + i )
2 N
⎦
(
⎥ = G P G , i%, N ) (4.8)
The factor (P/G, i%, N) is the gradient to present equivalent conversion factor
or the discount gradient factor.
4.3.6 END-OF-PERIOD CASH FLOWS AND END-OF-PERIOD

UNIFORM GRADIENT CASH FLOWS
From Eq. (4.6), it can be written that:
⎡ i(1 + i ) N ⎤
A = P⎢ ⎥
⎣ (1 + i ) − 1 ⎦
N
⎡ (1 + i ) N − 1 − Ni ⎤
= G⎢
⎣ i(1 + i ) − i ⎦
N (
⎥ = G G , i%, N
A ) (4.9)
The factor (A/G, i%, N) is the uniform series gradient factor. Table 4.1 provides a
summary of previous sections about discrete compounding interest factors.
TABLE 4.1
Relations between General Cash Flow Elements Using Discrete
Compounding Interest Factors
Relations Interest Factor Factor Name
F = (1 + i)N (F/P,i%,N) Single payment compound amount
1
P= (P/F,i%,N) Single payment present worth
(1 + i )
N
(1 + i ) − 1
N
F = A⋅ (F/A,i%,N) Uniform series compound amount

i
(1 + i ) − 1
N
P = A⋅ (P/A,i%,N) Uniform series present worth

i (1 + i )
N
i
A = F⋅ (A/F,i%,N) Sinking fund
(1 + i ) − 1
N
i (1 + i )
N
A = P⋅ (A/P,i%,N) Capital recovery

(1 + i ) − 1
N
(1 + i ) − 1 − Ni
N
P = G⋅ (P/G,i%,N) Discount gradient

i (1 + i )
2 N
(1 + i ) − 1 − Ni
N
A = G⋅ (A/G,i%,N) Uniform series gradient

i (1 + i ) − i
N
Example 4.3
Suppose you deposit $6000 today, $3000 after 2 years, and $4000 after 5 years into
a bank account paying 5% interest per year. How much money will you have in
your account after 10 years?
Solution: The general cash flow diagram is shown in Figure 4.4. The money in
the account after 10 years can be estimated as summation of the future values of
single payments as follows:
( ) (
F = 6000 F P , 5%, 10 + 3000 F P , 5%, 8 + 4000 F P , 5%, 5 ) ( )
= $6000(1.6289) + $3000(1.4775) + $4000(1.2763) = $19311.1
Example 4.4
A water resources project produces benefits that amount to $10,000 in year 1 and
increase on a uniform gradient to $100,000 in year 10, after which the benefits
increase at a uniform gradient of $5000 per year to $175,000 in year 25. The value
of the benefits then remains constant at $175,000 each year until the end of the
F=?
1 2 3 4 5 6 7 8 9
10
$3000 $4000
$6000
FIGURE 4.4 Cash flow diagram for Example 4.3.
project life, which is 40 years. What is the present worth of these benefits at a 4%
interest rate? Figure 4.5 is a cash flow diagram for this example.
Solution:
1. Present equivalent value of benefits for years 1 through 10:
10, 000⎛ , 4%, 10⎞ + 10, 000⎛ , 4%, 10⎞ = 10, 000 × 41.9919 = $419, 919
P P
⎝G ⎠ ⎝A ⎠
2. Present equivalent value of benefits for years 11 through 25 is:
105, 000 ⎛ , 4%, 15⎞ ⎛ , 4%, 10⎞ + 5, 000 ⎛ , 4%, 15⎞ ⎛ , 4%, 10⎞
P P P P
⎝A ⎠⎝ F ⎠ ⎝G ⎠⎝ F ⎠
= 105, 000 × 11.11839 × 0.67556 + 5, 000 × 69.735 × 0.67556 = 1, 024, 220
3. Present equivalent value of benefits for years 26 through 40 is:
175, 000 ⎛ , 4%, 15⎞ ⎛ , 4%, 25⎞ = 175, 000 × 11.11838 × 0.37512 = 729, 877
P P
⎝A ⎠⎝ F ⎠
The total present worth is the sum of these three values, which is equal to $2,174,016.
More details about interest rates that vary with time and interest formulas for
continuous compounding can be found in DeGarmo et al. (1997) and Au and Au
(1992).
4.4 ECONOMIC ANALYSIS OF MULTIPLE

ALTERNATIVES
The main goal of economic analysis is to find out whether a capital investment and
the costs associated with the project within the project lifetime can be recovered by
revenues. In addition, it is important to find out whether the investment is sufficiently
attractive considering the risks involved and potential alternative uses. The five
methods explained briefly in this section are present worth, future worth, annual
worth, internal rate of return, and external rate of return. When comparing alterna-
tives, the selection of one alternative might exclude the others; these alternatives are
mutually exclusive. Different projects usually have different useful lives, which
should be reflected in the economic analysis.
4.4.1 PRESENT WORTH METHOD

In this method, the alternative that has the largest present worth (PW) of the dis-
counted sum of benefits minus costs over its life is selected:
∑ ( B − C )⎛⎝ F , i%, t⎞⎠

P
PW = t t (4.10)
t =1
where:
Ct is the cost of each alternative in year t.

Bt is the benefit of each alternative in year t.
N is the study period in years or the planning horizon.
i is the interest rate.
By using this method to evaluate investment alternatives, the one with the greatest
positive equivalent worth or least negative equivalent worth is selected. (Cost alter-
natives are those that will generate negative cash flows, except for a possible positive
cash flow element derived from disposal of assets at the end of the useful life of the
project; see DeGarmo et al. 1997).
When using this method, the following items should be considered:
• The same discount rate should be used for all alternatives.

• The same time base should be used for estimating the present worth of
alternatives.
• The same study period (planning horizon) should be used.
Selection of the study period is an important step in comparing projects with different
life spans. The study period is the selected time over which mutually exclusive
alternatives are compared. The repeatability assumption can be used when the project
lives are different.
Example 4.5
Consider two water resources development projects that are being proposed to supply
the water demand of an industrial complex. Table 4.2 shows the characteristics of
these projects. Find the best alternative when the minimum attractive rate of return
(MARR) = 5%.
Solution: In this example, by using the repeatability assumption the study period
is set at 20 years, which is the least common multiple of lives, and we consider two
cycles of alternative B. By applying the present worth method, it can be written that:
A = −$40, 000 − $160⎛ , 5%, 20⎞ + $4, 000 × ⎛ , 5%, 20⎞

P P
⎝A ⎠ ⎝A ⎠
The present worth of alternative A is:
A = −$40, 000 − $1, 994 + $49, 845 = $7, 855
TABLE 4.2
Data for Example 4.5
Costs and Benefits Project A Project B
Construction costs $40,000 $25,000

Operation and maintenance costs $160 $100
Economic life 20 years 10 years
Annual benefits $4000 $4000
The present worth of alternative B is:
B = −$25, 000 − $100⎛ , 5%, 10⎞ + $4, 000⎛ , 5%, 10⎞

P P
⎝A ⎠ ⎝A ⎠
−$25, 000⎛ , 5%, 10⎞ − $100⎛ , 5%, 10⎞ ⎛ , 5%, 10⎞

P P P
⎝F ⎠ ⎝A ⎠⎝ F ⎠
+$4, 000⎛ , 5%, 10⎞ ⎛ , 5%, 10⎞ = $8, 257

P P
⎝A ⎠⎝ F ⎠
Therefore, project B should be chosen.
4.4.2 FUTURE WORTH METHOD

In this method, all benefits and costs of different alternatives are converted into their
future worth figures. Then, the alternative with the greatest future worth or the least
negative future worth is selected. When using this method, the following items should
be considered:
• The same discount rate should be considered for all alternatives.

• The same time base should be used for estimating the future worth of
alternatives.
• The same study period should be used for all alternatives.
When selecting the study period, two options may be considered. One option is to
use the repeatability assumption, explained in the previous section. The second
option is to use the co-terminated assumption. Based on this assumption, if an
alternative has a useful life shorter than the study period, the estimated annual cost
of the activities might be used during the remaining years. Similarly, if the useful
life of an alternative is longer than the study period, a reestimated market value is
normally used as the terminal cash flow at the end of the co-terminated life of the
project (DeGarmo et al., 1997).
Example 4.6
Referring back to Example 4.5, assume that you want to invest your money in one
of two alternatives. Consider that the study period is 20 years. Use the repeatability
assumption and select the best alternative.
Solution: The future worth of alternative A can be estimated as follows:
FW of A = −$40, 000⎛ , 5%, 20⎞ − $160⎛ , 5%, 20⎞ + $4, 000⎛ , 5%, 20⎞
F F F
⎝P ⎠ ⎝A ⎠ ⎝A ⎠
= −$40, 000 × 2.6533 − $160 × 33.0660 + $4, 000 × 33.0660 = $20, 841
For estimating the future worth of alternative B, it can be assumed that all cash
flows will be reinvested until the end of the study period, so the future worth of the
project can be estimated as follows:
⎡ ⎛F ⎞ ⎛F ⎞⎤⎛ F ⎞
FW of B = ⎢−$25, 000⎜ , 5%, 10⎟ + $4 , 000⎜ , 5%, 10⎟ ⎥ ⎜ , 5%, 10⎟
⎣ ⎝P ⎠ ⎝A ⎠⎦⎝ P ⎠
⎛F ⎞ ⎛F ⎞ ⎛F ⎞
−$100⎜ , 5%, 20⎟ − $25, 000⎜ , 5%, 10⎟ + $4 , 000⎜ , 5%, 10⎟
⎝A ⎠ ⎝P ⎠ ⎝A ⎠
= [−$25, 000 × 1.6289 + $4 , 000 × 12.5779] × 1.6289
−$100 × 33.066 − $25, 000 × 1.6289 + $4 , 000 × 12.5779 = $21902
Therefore, alternative B should be selected.
4.4.3 ANNUAL WORTH METHOD

In this method, the costs and benefits of all alternatives are converted to uniform
annual figures, then the alternative with the greatest annual worth or the least negative
worth, in the case of cost alternatives, is selected. When using this method, projects
with different economic lives can be compared without considering the previously
explained assumptions about the study period. It should be noted that when applying
this method, the same discount rate should be used for all alternatives, just as for
the present worth and future worth methods.
Example 4.7
Solve Example 4.5 by using the annual worth method.
Solution:
AW of A = −$40, 000⎛ , 5%, 20⎞ − $160 + $4, 000 = $632

A
⎝P ⎠
Aw of B = −$25, 000⎛ , 5%, 10⎞ − $100 + $4, 000 = $663

A
⎝P ⎠
Alternative B is selected because of its higher annual worth.
4.4.4 THE INTERNAL RATE OF RETURN (IRR) METHOD

The internal rate of return (IRR) method is one of the most widely used methods
for economic analysis. In this method, the IRR is defined as the discount rate that
will set the net present value or the net future value of the cash flow profile equal
to zero. In this method, the projects that have IRR less than the minimum attractive
rate of return (MARR) will be rejected. The present worth and annual worth methods
are usually used to find the IRR.
Example 4.8
Suppose you have invested $1000 in a project. Based on the contract, you will receive
benefits of $500 and $1500, 3 and 5 years later, respectively. How much is the IRR
of the investment?
Solution: Using the present worth method, it can be written that:
–$1, 000 + $500 ( P / F, i%, 3) + $1, 500 ( P / F, i%, 5) = 0

⇒ i = 16.90%
So, if the MARR is less than 16.90%, the investment will not be rejected.
In this method, it is assumed that the recovered funds are reinvested at IRR. In
many real situations, it is not possible to reinvest the money at IRR rather than at
MARR. In such cases, the external rate of return (ERR) method should be used.
4.4.5 THE EXTERNAL RATE OF RETURN METHOD

As mentioned in the previous section, it is sometimes impossible to reinvest the
recovered money at the internal rate of return. The external rate of return method
takes into account the interest rate, e, that is external to a project and at which net
cash flows generated by the project over its life can be reinvested. This method can
be summarized in the following steps:
1. The present value of all net cash flows is estimated using the e percentage
discount rate.
2. The future value of all net cash flows is estimated using the e percentage
interest rate.
3. The external rate of return is estimated by setting these two quantities
equal.
Example 4.9
Consider the cash flow diagram shown in Figure 4.6. If the external rate of return
is considered 15% and the MARR is equal to 20%, is the project rejected?
Solution: Based on the ERR method, it can be written that:
[ ]
ABS ( −$1, 000 − $500 ( P / F, 15%, 1)) × ( F / P, i ′%, 6)
= ($600 − $100) × ( F / A, 15%, 5) ⇒ i ′

i ′ = 15.3%
Because i′ is less than MARR, the project is rejected.

$600 $600 $600 $600 $600
$100 $100 $100 $100 $100

$500
$1,000
4.4.6 COMPARING ALTERNATIVES USING THE RATE OF RETURN

METHOD
When comparing different alternatives using the rate of return method, the same study
period should be considered. Projects that have a rate of return lower than the MARR
would be rejected. The more costly alternative would be selected only if the incremental
rate of return exceeds the MARR; otherwise, the least costly alternative is selected.
Example 4.10
Two different types of pumps (A, B) can be used in a pumping station. Table 4.3
shows the costs and benefits of each of the pumps. Compare the two pumps using
a MARR = 10%.
Solution: The study period is 12 years. By using the repeatability assumption,
it can be written that:
Present worth of costs of A = $200 + ($200 – $50)(P/F, 10%, 6) – $50

(P/F, 10%, 12) = $269
Present worth of benefits of A = $95(P/A, 10%, 12) = $647
Present worth of costs of B = $700 – $150(P/F, 10%, 12) = $652
Present worth of benefits of B = $120(P/A, 10%, 12) = $818
TABLE 4.3
Data Used in Example 4.10
Economic Characteristics Pump A Pump B
Initial investment $200 $700

Annual benefits $95 $120
Salvage value $50 $150
Useful life (year) 6 12
The graphical method can be used to select one of the pumps. In this method, a line
is drawn between the two points representing the present costs vs. present benefits
PW of costs
(102)
8 B(652,818)
7
6 A(269,647)
5
4
3
2
1
45
1 2 3 4 5 6 7 (102) PW of benefits
FIGURE 4.7 Comparing the two projects in Example 4.10 by the rate of return method.
for A and B. The slope of this line is compared with a 45° line drawn on the graph.
As can be seen in Figure 4.7, the present worth of costs and the present worth of
benefits are equal for points on the line having a 45° slope. When the slope of the
line connecting two projects is less (more) than 45°, the project with the lower
(higher) initial cost would be selected. As shown in Figure 4.7, pump A, which has
a lower initial cost, should be selected.
4.5 ECONOMIC EVALUATION OF PROJECTS USING

THE BENEFIT–COST RATIO METHOD
The benefit-cost ratio method has been widely used in the economic analysis of
water resources projects. This method is based on the calculation of the ratio of
benefits to costs. The time value of money should be considered; therefore, this ratio
should be estimated based on the equivalent worth of discounted benefits and costs.
For this purpose, the annual worth, present worth, or future worth of benefits and
costs can be used. One of the basic formulations for the benefit–cost ratio method
is as follows:
B
B / C factor = (4.11)
I +C
where:
B is the net equivalent benefits.

C is the net equivalent annual costs, usually classified as operation and
maintenance costs.
I is the initial investment.
For any project to remain under consideration, its benefit–cost ratio should exceed
one. Therefore,
B
>1
I+C
Then,
B − ( I + C) > 0
Therefore, the benefit–cost criterion will eliminate all those projects for which
equivalent amounts are less than zero. Another formulation for this method, which
reflects the net gain expected per dollar invested, is as follows:
B−C
B / C factor = (4.12)
I
In order to consider the salvage value associated with an investment, the fol-
lowing relation can also be used in applying the benefit–cost method:
B−C
B / C factor = (4.13)
I−S
where S is the salvage value of the investment.
Example 4.11
Referring back to Example 4.10, assume that the annual costs of operation and
maintenance for pumps A and B are equal to $50 and $60, respectively. Compare
the two pumps using the benefit–cost ratio method.
Solution: Considering the salvage value of the pumps, Eq. (4.13) can be used.
Because the useful lives of the projects are different, it is easier to use the annual
worth method, as follows:
95 − 50
Pump A: B / C =
200( A / P, 10%, 6) − 50( A / F, 10%, 6)
45
= = 1.14
200 × 0.2296 − 50 × 0.1296
120 − 60
Pump B: B / C =
700( A / P, 10%, 12) − 150( A / F, 10%, 12)
60
= = 0.63
700 × 0.1468 − 150 × 0.0468
It is apparent that the benefit–cost ratio for pump B is less than one; therefore, pump
A should be selected.
In order to compare multiple mutually exclusive alternatives by the benefit–cost
ratio method, the principle of incremental analysis should be applied. For this
purpose, the following steps should be taken:
1. Arrange the alternatives by increasing order of their equivalent worth of

costs (denominators, or I + C, I, and I – S in Eqs. (4.11), (4.12), and
(4.13), respectively).
2. Consider the alternative with the denominator of least value as the baseline.
3. Estimate an incremental B/C ratio (ΔB/ΔC) based on the difference
between the equivalent worth of benefits and costs of the second least
cost alternative and the baseline.
4. If the ratio is equal to or greater than one, the higher equivalent cost
alternative becomes the new baseline; otherwise, the last baseline alter-
native is maintained.
5. Repeat steps (3) and (4) until the last alternative has been compared.
Example 4.12
Three flood control projects have been proposed for a river. Each of the projects has
a useful life of 25 years and the interest rate is 5% per year. The benefits and costs
of the projects are listed in Table 4.4. Compare the alternatives using the benefit–cost
ratio method.
Solution: Present worth of costs and benefits for the projects can be estimated
as follows:
Present worth of costs of A = $8,000,000 + $700,000 (P/A, 5%, 25) =

$17,865,730
Present worth of costs of B = $9,500,000 + $650,000 (P/A, 5%, 25) =
$18,661,035
Present worth of costs of C = $11,000,000 + $630,000 (P/A, 5%, 25) =
$19,879,157
Present worth of benefits of A = $1,500,000 (P/A, 5%, 25) = $21,140,850
Present worth of benefits of B = $1,600,000 (P/A, 5%, 25) = $22,550,240
Present worth of benefits of C = $1,750,000 (P/A, 5%, 25) = $24,664,325
TABLE 4.4
Benefits and Costs of Three Flood Control Alternatives in Example 4.12
(in $millions)
Benefits and Costs Project A Project B Project C
Capital investment 8 9.5 11

Annual operation and maintenance costs 0.7 0.65 0.63
Annual expected reduction in flood damage (benefits) 1.5 1.6 1.75
For all the projects the benefit–cost ratio is greater than one; therefore, all the projects
are acceptable. For comparing projects A and B, it can be written that:
$22, 550, 240 − $21, 140, 850

ΔB / ΔC( B − A) = = 1.77
$18, 661, 035 − $17, 865, 730
Because the estimated benefit–cost ratio is greater than one for each project, project
B should be considered as the baseline because it has a higher present worth of costs
than project A. For comparing projects B and C, it can be written that:
$24, 664, 325 − $22, 550, 240

ΔB / ΔC(C − B) = = 1.73
$19, 879, 157 − $18, 661, 035
The estimated ratio is greater than one; therefore, project C is recommended.
4.6 PUBLIC VS. PRIVATE PROJECTS AND GOODS

Water resources projects are usually classified as public projects, which means they
are financed and operated by federal, state, or local governmental organizations.
These projects may be of any size but are usually very large. These projects are
subject to the principals of engineering economics, but a number of special charac-
teristics should be included in the economic evaluation of these projects. Some of
the basic differences between public water resources projects and privately owned
projects could be summarized as follows:
• Purposes of the project and conflict issues. Public water resources

projects are usually multipurpose. For example, a reservoir might supply
water and power demands and instream flows, in addition to controlling
floods to prevent damage that may occur downstream of the reservoir in
high-flow seasons. Some of these objectives are in conflict, which is a
common issue. For example, holding enough water in the reservoir to
control flooding is in conflict with water conservation schemes for sup-
plying demands in low-flow seasons. Conflict issues in regard to water
resources development projects are explained in the next several chapters
of this book. The main purpose of water resources in public sector is to
minimize social costs. Private projects usually provide specific goods and
services that are not in conflict.
• Stakeholder community. While private projects are often related to a
limited community of stakeholders, water resources projects usually affect
a large number of people. Besides the direct impact of supplying water
demand for users of the water resources system, other issues such as
environmental impacts or controlling power network fluctuations also
have broad effects on regional or national scales.
• Project life. The life of private projects is relatively short, while the life
of public projects can be as long as 20 to 100 years.
• Measurement of the project efficiency. For private projects, the rate of

return on capital is usually considered as a measure of efficiency, while
for water resources development projects the efficiency is almost always
very difficult to measure. This is mainly because of different short- and
long-term effects of water resources projects that cannot be easily quan-
tified in monetary units. In contrast, the benefits of private projects are
usually monetary or relatively easy to equate to monetary terms.
• Political issues. Funding for public projects is often subject to a strong
political influence, while this is not an important issue for most private
projects. In addition to public and private projects, goods also can be
classified as public and private. Two main characteristics of public goods
are that they are nonrival and nonexclusive (Mansfield, 1994). For nonrival
goods, the marginal cost of providing goods to an additional consumer is
zero. This means that one person’s use does not preclude enjoyment by
the others. Agricultural and industrial uses of water are rivals in consump-
tion, while the aesthetic and recreational values of a beautiful stream are
nonrivals. Also, we consider these goods to be nonexclusive because
people cannot be excluded from consuming the goods. Costs required to
keep those not entitled from using the pubic good are exclusion costs.
Exclusion costs for water resources are usually very high; therefore, we
usually consider them to be nonexclusive.
4.7 ECONOMIC MODELS

Economic models relate the needs of consumers to the resources that can be used
to satisfy these needs. These models include demand functions for goods and ser-
vices, supply functions for resources, and production functions for goods and ser-
vices. Each of these elements is briefly explained in the following sections.
4.7.1 PRODUCTION FUNCTIONS

Production functions refer to production of useful products by utilizing resources
and certain technologies. For example, supplying water for irrigation might require
specific structures for diversion and transferring water to agricultural lands. The
production technologies are subject to change and advancements, which cannot be
predicted accurately. For economic activities that significantly affect environment
and natural resources, the tradeoff between producing goods and services and envi-
ronmental quality is an important issue. This tradeoff is illustrated in Figure 4.8 by
a production possibility curve. Considering the technological capabilities over a
specific time period, each society may choose to locate itself at a specific point on
this curve. Therefore, the issue becomes one primarily of social choice and depends
on the value that people in a particular society place on the environment and
economic outputs (Field, 1997). As can be seen in the figure, the current level of
economic output (C1) can be increased to C2 only at the cost of a decrease in
environmental quality from E1 to E2.
c2 Goods
c1
E2 E1 Environmental quality
FIGURE 4.8 Production possibilities curve. (Adapted from Field, B. C., Environmental Eco-
nomics: An Introduction, McGraw-Hill, New York, 1997.)
Price per unit resources
Units of resources
FIGURE 4.9 Resource-supply function.
4.7.2 RESOURCE-SUPPLY FUNCTIONS

The resource-supply function may be defined as the quantity of resources made
available per unit time as a function of the price per unit of resources. As can be
seen in Figure 4.9, if the price for a given resource is high, the resource supplier
will be willing to deliver a large quantity. In order to develop the water supply
resources, development of well fields, interbasin water transfer projects, etc. could
be considered, if the price of water is estimated accurately. The distribution of water
resources development costs should be considered when making decisions about
development alternatives and water allocation schemes. When a scarcity occurs,
more water might be allocated to users who are capable of paying high prices for
water, such as industries; however, political and legal issues may significantly affect
this process (Viessman and Welty, 1985).
4.7.3 DEMAND FUNCTIONS

Demand functions indicate the quantity of goods and services that the consumer
would demand at any particular price. This is another way of looking at the marginal
willingness to pay relationship. Figure 4.10 shows an example of demand functions
for water resources. The price–quantity relation for domestic use is very steep. It
means that even if the price of water significantly increases, people would still satisfy
their needs for cooking, washing, and maintaining health standards. At the same
FIGURE 4.10 Examples of demand function. (Adapted from Viessman, W. and Welty, C.,
Water Management: Technology and Institutions, Harper & Row Publishers, New York, 1985.)
time, people are not going to use much more water than they really need no matter
how inexpensive the water is. The demands for products derived from water are
more sensitive to price changes.
The demand functions that engineers and planners encounter are more complex
because of a variety of factors that affect the demand functions. The demand func-
tions can be formulated as follows (Mays, 1996):
QW = QW ( PW , Pa , P, Y , Z ) (4.14)
Where QW refers to the consumers’ water demands in a specific period of time,

PW is the price of water, Pa is the price of an alternative water resource, P is an
average price index representing all other goods and services, Y is the consumer
income, and Z is a vector representing other factors such as climate and consumer
preferences.
The responsiveness of consumer demands to variations in each of the above-
mentioned variables is an important concept in economic analysis and is referred to
as demand elasticity. One of the most important aspects of demand elasticity is price
elasticity, which can be defined as the percentage change in quantity consumed if
price is changed 1%. Almost all estimates of long-run price elasticity of domestic
water demand in the United States seem to fall between –0.3 and –0.7, meaning that
other factors being held constant, a 10% increase in price would lead to a 3 to 7%
decrease in the amount purchased (Mays, 1996).
4.8 EFFICIENCY AND EQUITY IN ECONOMIC

ANALYSIS
A goal of economic analysis is to understand human behavior when facing scarcity
to determine how to manage scarcities of resources and products. In the case of
water resources management problems, allocation of water resources among com-
peting users is the main issue. The two main effects of allocating scarce sources are
classified as efficiency and distribution (equity). Economic efficiency is an index for
measuring whether the allocation of resources among competing users has maxi-
mized profits or welfare level. For economic activities that use natural resources,
TABLE 4.5
Distribution of Costs and Benefits in Example 4.13
Group 1 Group 2 Group 3
Benefits and Costs ($/person/year)
Decrease in costs of water supply (benefits) 60 80 120

Costs of the project 40 60 80
Difference 20 20 40
such as water, or affect these resources by polluting wastes, efficiency should be

measured by incorporating social costs.
Equity effects refer to allocation of costs and benefits to individuals and different
groups in the society. Distribution of costs and benefits based on equity can follow
a vertical or horizontal distribution. If the benefits of a water resources development
project allocated to different groups of people with different levels of income are
equal, then the distribution is horizontal. In vertical distribution, the costs and
benefits are allocated to different groups of people proportional to the level of their
income. In the vertical distribution of costs and benefits, if more profits are allocated
to groups with higher levels of income, the program is regressive. In contrast,
progressive programs allocate more profit to groups with lower levels of income.
Example 4.13
A water supply project is implemented in a region. Three groups of people with the
same levels of income will use this project. Table 4.5 shows the distribution of costs
and benefits among these groups. Is the distribution of costs and benefits vertical or
horizontal?
Solution: As shown in Table 4.5, for groups 1 and 2 the difference between
benefits and costs is equal; therefore, a horizontal distribution has been applied to
groups 1 and 2 but the difference between costs and benefits for the third group is
twice that of the other two groups. In the next example, the effects of vertical
distribution are explained.
Example 4.14
Three water supply projects are implemented in a region. The water demand of
people in a big city, a small city, and a village will be supplied by these projects.
The levels of income and distributions of costs and benefits for these projects are
presented in Table 4.6. Discuss the distributions of costs and benefits and determine
which projects are progressive and regressive.
Solution: As shown in Table 4.6 for the first program, the difference between
costs and benefits is 1% of income for the people in both cities and in the village;
therefore, the distribution of costs and benefits is vertical in this program. The second
program is a regressive program, because the profits allocated to the groups with
higher levels of income are higher. The third program is progressive, because it has
TABLE 4.6
Distribution of Costs and Benefits in Three Water Supply Projects
in Example 4.14
Village Small City Big City
(Pop. 5000) (Pop. 20,000) (Pop. 50,000)
Water supply program 1

Decrease in water supply from other 150 (3) 300 (1.5) 600 (1.2)
sources (benefits)
Costs of project implementation ($) 100 (2) 100 (0.5) 100 (0.2)
Difference 50 (1) 200 (1) 500 (1)
Decrease in water supply from other 150 (3) 1400 (7) 5500 (11)
sources (benefits)
Costs of project implementation ($) 100 (2) 800 (4) 3000 (6)
Difference 50 (1) 600 (3) 2500 (5)
Decrease in water supply from other 700 (14) 2200 (11) 3000 (6)
sources (benefits)
Costs of project implementation ($) 200 (4) 1000 (5) 1500 (3)
Difference 500 (10) 1200 (6) 1500 (3)
Note: Numbers in the parentheses are percents of income
allocated less profit to the people with higher income. It should be noted, however,
that a project might be regressive or progressive overall, but the distributions of costs
and benefits might be different. For example, the second program is regressive, but
the costs have been distributed progressively.
In quantifying economic efficiency, the differences between public and private
goods should also be considered. For example, consider two groups of consumers for
a good or service. The demand curves for these groups are represented by DA and
DB. The total demand for these groups when the good is a private good can be
estimated by horizontally summing the demand curves for each group. The total
demand curve (D) and the demand curves for two consumers are shown in Figure 4.11.
price
DA DB D
supply
c d e
f g h output
FIGURE 4.11 Efficient output of a private good.

price D
DB
DA supply
j n
k c
m p
o q
output
FIGURE 4.12 Efficient output of a public good.
The efficient output for this case is h because at this output the marginal benefit
to each consumer of an extra unit of the good equals its marginal cost. Assuming
that the marginal benefit can be measured as the maximum amount that each group
will pay for the extra unit, then the marginal benefit for groups A and B would be
cf and dg, respectively. The marginal cost of the extra unit is eh for the output rate
equal to h. Because cf = dg = eh, the marginal benefit to each group equals the
marginal cost (Mansfield, 1994).
Figure 4.12 shows a similar case for a public good. The total demand curve can
be obtained by summing the demand curves for the two groups vertically. Because
both groups of consumers consume the total amount of the good, the combined price
paid by the two groups of consumers is the sum of the price paid by each group.
The efficient output of a public good occurs when the marginal social benefits
equal the marginal social costs. For the output rate of q, the marginal social benefit
from an extra unit of output is the vertical sum of om and ok, which equals oj.
Because the marginal social cost of an extra unit of output is qn and because oj is
equal to qn, oq is the optimal rate of supply for the public good.
4.9 PROBLEMS
4.1 What is the future amount that will be available in 4 years if $1000 is
invested at 11% per year simple interest?
4.2 A company lends $10,000 at 10% interest rate for 10 years. At the end
of year 10, the entire principal plus interest is invested at a 12% rate of
interest for 10 years. How much will accumulate at the end of a 20-year
period?
4.3 Draw a cash flow diagram for problem 4.2 from the viewpoint of the
lending company.
4.4 A certain type of equipment is intended to be used in a project for which
the economy study is based on a 25-year period; it is estimated that it
will require replacement every 10 years. What salvage value should be
used if the initial cost is $10,000? Use a 10% interest rate.
4.5 Following are three alternatives for supplying water to a community for
25 years, after which the economic lives and period of analysis terminate.
Using the costs presented in the table and a 5% discount rate, compare
the three projects using the present worth method.
Costs Timing (Year) Project A ($) Project B ($) Project C ($)
Initial investment 0 2,000,000 1,000,000 1,500,000

10 0 1,000,000 1,200,000
20 0 1,000,000 0
Operation and 1–10 7000 4000 6000
maintenance costs 10–20 8000 7000 8000
20–25 9000 9000 9000
4.6 Compare the projects in problem 4.5 using the benefit–cost ratio method.
4.7 Compare the projects in problem 4.5 using the rate of return method.
4.8 Consider the following two mutually exclusive alternatives for a water
diversion project:
Project A Project B
Capital investment $200,000 $300,000

Annual revenues $100,000 $140,000
Annual expenses $44,000 $86,000
Useful life (years) 10 15
Assume the study period is 15 years and repeatability is applicable. Using

the annual worth method, which project do you recommend?
4.9 Solve problem 4.8 using the present worth method.
4.10 The net cash flows are shown below for three alternatives for water
resources development projects. The MARR is 12% per year and the study
period is 10 years. Which alternative is economically preferred based on
the external rate of return method? (External rate of return = MARR =
12% per year.)
End of Year Project A ($) Project B ($) Project C ($)
0 171,200 126,400 143,600

1 14,800 24,200 20,000
? ? ? ?
? ? ? ?
? ? ? ?
10 14,800 24,200 20,000
4.11 Three water supply projects are implemented in a region. Water demand
of people in a big city, a small city, and a village will be supplied through
these projects. Level of income and distribution of costs and benefits for
these projects are presented in the following table. Discuss the distribution
of costs and benefits and determine which projects are progressive and
regressive.
Village Small City Big City

(Pop. 1000) (Pop. 20,000) (Pop. 70,000)

Decrease in water supply from 30 1400 7700
other sources (benefits)
Costs of project implementation 20 800 4200
Difference 10 600 3500
REFERENCES
Al-Futaisi, A. and Stedinger, J. R., Hydrologic and economic uncertainties and flood-risk
project design, ASCE J. Water Resources Plan. Manage., 125(6), 314–324, 1999.
Arnold, F. S., Economic Analysis of Environmental Policy and Regulation, John Wiley and
Sons, New York, 1995.
Au, T. and Au, T. P., Engineering Economics for Capital Investment Analysis, Boston: Allyn
and Bacon, 1983.
Creel, M. and Loomis, J., Recreation value of water to wetlands in the San Joaquin Valley:
linked multinomial logic and count data trip frequency model, Water Resources Res.,
28(10), 1992.
DeGarmo, E. P., Sullivan, W. G., Bontadelli, J. A., and Wicks, E. M., Engineering Economy,
Prentice-Hall, Upper Saddle River, NJ, 1997.
Feder, G. and Le Moigne, G., Managing water in sustainable manner, Finance Dev., 31(2),
June 1994.
Field, B. C., Environmental Economics: An Introduction, McGraw-Hill, New York, 1997.
Hartwick, J. H. and Olewiler, N. D., The Economics of Natural Resources Use, Addison-
Wesley, Reading, MA, 1998.
Hoehn, J. P. and Krieger, D. J., Economic analysis of water services investments and tariffs
in Cairo, Egypt, ASCE J. Water Resources Plan. Manage., 126(6), 345–350, 2000.
James, L. D. and Lee, R. R., Economics of Water Resources Planning, McGraw-Hill, New
York.
Kahn, J. R., The Economic Approach to Environmental and Natural Resources, 2nd ed.,
Dryden Press, Troy, MO, 1998.
Kizner, I. M., Entrepreneurial discovery and the competitive market process: an Austrian
approach, J. Econ. Lit., 35(1), 60–85, 1997.
Loomis, J. B., Integrated Public Lands Management: Principles and Applications on National
Forests, Parks, Wildlife Refuges, and BLM Lands, Columbia University Press, New
York, 1993.
Loomis, J. B. and Walsh, R. G., Recreation Economic Decisions: Comparing Benefits and
Costs, Department of Agriculture and Resources Economics, Colorado State Univer-
sity, Fort Collins, CO, 1997.
Loomis, J. B., Environmental valuation techniques in water resources decision making, ASCE
J. Water Resources Plan. Manage., 126(6), 339–344, 2000.
Maksimovic, C. and Tejada-Guibert J. A., Eds., Frontiers in Urban Water Management:
Deadlock or Hope, IWA, London, 2001.
Mansfield, E., Managerial Economics: Theory, Application, and Cases, W.W. Norton, 1996.
Mansfield, E., Microeconomics: Theory and Applications, W. W. Norton, New York, 1994.
Mays, L. W., Water Resources Handbook, McGraw-Hill, New York, 1996.
Pudney, S., Modeling Individual Choice: The Econometrics of Corners, Kinks, and Holes,
Basel-Blackwell, Oxford, 1989.
Recio, B., Rubio, F., Lomban, J., and Ibanez, J., An econometric irrigated crop allocation
model for analyzing the impact of water restriction policies, J. Agric. Water Manage.,
42, 47–63, 1999.
Sagoff, M., The Economy of the Earth, Cambridge University Press, Landon, 1988.
Shabman, L. and Stephenson, K., Environmental valuation and its economic critics, ASCE J.
Water Resources Plan. Manage., 126(6), 382–388, 2000.
Tietenberg, T., Environmental and Natural Resources Economics, Harper Collins, New York,
1992.
United Nations, Proceedings of the Expert Group Meeting on Water Pricing, Water Resources
Series, No. 55, United Nations, New York, 1981.
USACE, Guidelines for Risk and Uncertainty Analysis in Water Resources Planning, Water
Resources Support Center, Institute of Water Resources, U.S. Army Corps of Engi-
neers, Fort Belvoir, VA, 1992a.
USACE, Guidelines and Procedures for Risk and Uncertainty Analysis in Corps Civil Works
Planning, Water Resources Support Center, Institute of Water Resources, U.S. Army
Corps of Engineers, Fort Belvoir, VA, 1992b.
USWRC, Economics and Environmental Principles for Water and Related Land Resources
Implementation Studies, U.S. Water Resources Council, U.S. Government Printing
Office, Washington, D.C., 1983.
Viessman, W. and Welty, C., Water Management: Technology and Institutions, Harper & Row
Publishers, New York, 1985.
5 Time Series Analysis
5.1 INTRODUCTION
The concept of random variables has been used in the field of hydrology and water
resources since the beginning of the 20th century, but in recent years a great deal
of effort has been devoted to improve the following:
1. Understanding the stochastic nature of hydrologic variables such as

streamflows
2. Modeling procedures
3. Developing new statistical models
4. Parameter estimation techniques
5. Model evaluation and fitness tests
Operation policies for water resources systems could not be applied in real-time
decision making without forecasting the future state of the resources. For example,
consider a reservoir that supplies water for various purposes; the amount of each
scheduled release depends on the probable range of inflow to the reservoir. Because
of a lack of adequate knowledge about physical processes in the hydrologic cycle,
many investigators have expanded the application of statistical models to forecasting
and generation of synthetic data. Synthetic data also help in incorporating uncer-
tainties and probable extreme events.
New advancements in this area have been mainly focused on coupling physical
characteristics of the water resources systems and the effects of large-scale climate
signals on water resources and early prediction of rainfall and streamflow. This
chapter discusses the basic principles of hydrologic time-series modeling and several
types of statistical models. It should be noted that the details of statistical modeling
are presented in many references, such as Salas et al. (1988) and Brockwell and
Davis (1987). An introduction to these models and their application in water
resources systems analysis are presented here.
5.2 STOCHASTIC PROCESSES

All natural physical processes are subject to variability. For example, rainfall inten-
sity, flood magnitude, or low flows during severe droughts can vary widely. To study
these variations and incorporate them in the planning and operation of water
resources, samples of data are collected. A dataset consists of a series of measure-
ments of a phenomenon, and the quantities measured are termed variables. To allow
for the possible unpredictable nature of future observations, it is assumed that each
observation is a realized value of a certain random variable.
145
The hydrologic cycle is composed of phenomena such as precipitation, runoff,

infiltration, evaporation, evapotranspiration, and abstraction. Characteristic vari-
ables, which can also be referred to as hydrologic variables, have been defined to
describe each of these phenomena. Depth or intensity of rainfall at different time
steps of a rainstorm, monthly inflow discharge to a reservoir, or daily evaporation
are some examples of hydrologic variables, which usually vary in time and space
(Shahin et al., 1993). A time series is a sequence of values arranged in their order
of occurrence in time.
A process is a mathematical description of the behavior of a phenomenon in
one or more dimensions in space and/or time. Because all hydrologic phenomena
change in space or time, they are hydrologic processes. If a process contains a
random component, it is a stochastic process, which is a family of random variables,
{xt, t ∈ T}, defined on a probability space. Stochastic processes are subdivided into
stationary and nonstationary. A stochastic process is stationary if the expected values
of statistical descriptors do not change over time. If a time series is stationary, the
series should be divided into a number of nonoverlapping subseries and the expected
values of statistical descriptors of each series should be the same for each of the
subseries.
Hydrologic variables are mostly nonstationary due to variations that are the result
of natural and human activities. In hydrologic modeling, these variations can be
classified as:
• Trend is a unidirectional gradual change (increasing or decreasing) in the
average value of the variable. Changes in nature caused by human activities
are the main reason for the over-several-years trends. The trends are usually
smooth, and we should be able to represent it by a continuous and differ-
entiable function of time. Trend, Tt, is usually considered to be determin-
istic, and it can be modeled by linear or polynomial functions, such as:
Linear function: Tt = a + bt
Polynomial function: Tt = a + bt + ct 2 + ⋯ + dt m
Power functions: Tt = a − br t (0 < r < 1) and Tt = 1 / ( a + br t ) (0 < r < 1)
• Jump is a sudden positive or negative change in the observed values.

Human activities and natural disruptions are the main reasons for jumps
in hydrologic series.
• Periodicity represents cyclic variations in a hydrologic time series. These
variations are repetitive over fixed intervals of time.
• Randomness represents variations due to the uncertain nature of the sto-
chastic process. The random component of the hydrologic time series can
be classified as autoregressive or purely random.
In statistical modeling of time series, it is assumed that the time series is purely
random; therefore, steps should be taken to remove trends, jumps, and periodicity
from the data.
Time Series Analysis 147
5.3 HYDROLOGIC TIME SERIES MODELING: BASIC

STEPS
A systematic approach to hydrologic time series modeling could consist of the
following main steps (Salas et al., 1988):
• Data preparation: The first step is to remove trends, periodicity, and

outlying observations and fit the data to a specific distribution by applying
the proper transformations.
• Identification of model composition: The next step is to decide upon the
use of a univariate or multivariate model, or a combination, with disag-
gregation models. This decision can be made based on the characteristics
of the water resources system and existing information.
• Identification of model type: Various types of models, such as AR (autore-
gressive), ARMA (autoregressive moving average), and ARIMA (autore-
gressive integrated moving average), can be selected in this step; details
of these models are presented in this chapter. Statistical characteristics of
the time series and the modeler input and knowledge about the types of
models are the key factors in identification of model type.
• Identification of model form: The form of the selected model should be
defined based on the statistical characteristics of the time series such as
seasonal and non-seasonal characteristics. The periodicity of the data and
how it can be considered in the structure of the selected model is the main
issue in this step.
• Estimation of model parameters: Several methods, such as method of
moments and method of maximum likelihood, can be used for estimating
the model parameters.
• Testing the goodness of fit of the model: In this step, assumptions such as
test of independence and normality of residuals should be checked. The
statistics used for verifying these assumptions are briefly explained in this
chapter.
• Evaluation of uncertainties: When evaluating uncertainties, the model,
parameter, and natural uncertainties in data should be analyzed separately.
Model uncertainty may be evaluated by testing whether there are signif-
icant differences between the statistics of actual data and forecasted and
generated data by alternative models. Parameter uncertainty may be deter-
mined from the distribution of estimated parameters.
5.4 DATA PREPARATION

The main tasks in data preparation phase can be summarized as:
1. Trends removal
2. Removal of outlying observations
3. Removal of periodicity
4. Fit a well-known distribution to the data by applying proper transforma-
tions if needed.
The principal assumption in using statistical models for hydrologic time series is
that the variable is considered to be absolutely random. Therefore, if the series
contains unnatural components, such as regulated release from reservoirs or trends,
they should be removed first. Trend and seasonality are usually detected by the time
sequence plot of the series. They can also be characterized by sample autocorrelation
functions, which are slowly decaying and nearly periodic.
For example, consider a nonseasonal series, xt, which contains trend components
as follows:
xt = Tt + zt (5.1)
where Tt and zt are trend and random components, respectively. The trend component
can usually be defined as a function of time as:
(
Tt = f t, ⋯, t m , β1 , ⋯, β m ) (5.2)
where (β1, ⋯, βm) are the parameters to be estimated. These parameters can be
estimated based on the least-square method by minimizing the following sum:
∑ (x − T )
2
t t (5.3)
t
For this purpose, the partial derivatives of the above sum with respect to each of the
parameters, (β1, ⋯, βm), is set to zero. The random series can then be obtained by
subtracting the trend from the observed data using Eq. (5.1).
Example 5.1
A series of annual water consumption in a city for a period of 10 years is shown in
Table 5.1. Estimate a linear trend function and generate the random series of water
consumption by removing the trend.
Solution: The linear trend function can be defined as:
Tt = α t + β (5.4)
In the least-square method, the following sum should be minimized:
10 10
∑ ( xt − Tt ) = ∑ ( x − α t − β)
2 2
t
t =1 t =1
TABLE 5.1
Annual Water Consumption
in the City for Example 5.1
Water Consumption
Year (million m3/year)
1 53
2 62
3 63
4 73
5 79
6 91
7 95
8 105
9 122
10 131
α and β can be calculated as:
10 10 10
10 ∑ tx − ∑ t∑ x
t =1
t
t =1 t =1
t
α= 2 = 8.55
10
⎛ 10
⎞
10 ∑
t =1
t −⎜
⎝
2
∑ t =1
t⎟
⎠
β = x − αt = 40.4
The random component after removing trend is shown in Table 5.2.
TABLE 5.2
Random Components of Annual Water Consumption for Example 5.1
Water Consumption
Year (million m3/year) Trend Component Random Component
1 53 48.95 53 – 48.95=4.05
2 62 57.5 4.5
3 63 66.05 –3.05
4 73 74.60 –1.6
5 79 83.15 –4.15
6 91 91.70 –0.7
7 95 100.25 –5.25
8 105 108.80 –3.8
9 122 117.35 4.65
10 131 125.90 5.1
The method used in Example 5.1 is known as classical decomposition of the

series into a trend component, a seasonal component, and a random component.
Another method for removing trend and seasonality is differencing, which is widely
used in modeling of hydrologic time series. For this purpose, the first differencing
operator, ∇, is defined as:
∇( xt ) = xt − xt −1 = (1 − B) xt (5.5)
where B is the backward shift operator,
B( xt ) = xt −1 (5.6)
Higher orders of the operators B and ∇ are defined as:
B j ( xt ) = xt − j and ( )
∇ j ( xt ) = ∇ ∇ j −1 ( xt ) , j ≥1 (5.7)
If the first-order difference operator is applied to the linear trend function shown
in Eq. (5.4), then we obtain the constant function ∇(Tt) = α. In a similar way, any
polynomial trend of degree k can be reduced to a constant by application of the
operator ∇k. Therefore, if a polynomial trend is detected in a hydrologic time series
as:
Tt = ∑a t
i=0
i
i
(5.8)
Then, by successive differencing of Equation 5.8, the following expression can be

obtained:
∇ m ( xt ) = m! am + ∇ m zt (5.9)
where zt is the random component.

In most of the parameter estimation methods, it is assumed that the time series
probability distribution is normal, but in many cases the hydrologic time series do
not follow normal distribution, are asymmetrically distributed, or are bounded by
zero. Therefore, it is often necessary to transform those variables to normal before
performing statistical modeling. Different transformations have been used in hydro-
logic time series analysis. For example, it has been found that the log-normal
distribution fits the annual series of precipitation and runoff in the United States
(Salas et al., 1988). If we represent these series by X, the following transformation
yields the normal series Y:
Y = log( X ) (5.10)
If gamma probability distribution function is used for the annual series X, then
the following transformation yields a relatively normal distribution of Y:
Y= X (5.11)
The following two-parameter transformation (first developed by Box and Cox,

1964) has been used by many investigators:
( xt + λ 2 )
λ1
−1
zt = ( λ1 −1)
λ1 > 0 (5.12)
λ1g
and
zt = g ⋅ ln( xt + λ 2 ) λ1 = 0 (5.13)
where λ1 and λ2 are the transformation parameters, g is the geometric mean of xt +

λ2, and xt and zt are the original and transformed series, respectively. The first
parameter, λ1, governs the strength of the transformation. λ1 = 1 corresponds to the
original data and λ1 = 0 to a logarithm. Values are scaled by the geometric mean to
keep the variance constant, thus allowing for a first comparison of mean-squared
errors between two different transformations.
Hashino and Delleur (1981) developed the following transformation similar to
that of Box and Cox:
( x + c)
λ
(λ ) −1
zt = t λ≠0 (5.14)
λ
and
zt( ) = ln( xt + c)
λ
λ=0 (5.15)
where λ and c correspond to λ1 and λ2 in the Box–Cox transformation. The values

in the transformation are not scaled by the geometric mean and therefore do not
permit direct comparison of mean-squared errors between two different transforma-
tions. Delleur and Karamouz (1982) applied the following transformation on annual
data for the Gunpowder and Patasco Rivers in Maryland:
zt = ( x t + a )
b
b≠0 (5.16)
and
zt = log( xt + a) b=0 (5.17)
where a and b are the transformation parameters.
5.5 METHODS OF PARAMETER ESTIMATION

The methods of moments, least squares, and maximum likelihood are the basic
methods used for estimating model parameters and are explained in the following
sections.
5.5.1 METHOD OF MOMENTS

Consider a sample (x1, ⋯, xN). The nth sample moment is defined as follows:
Mn = ⎛ ⎞ ∑x
1 n
(5.18)
⎝ N⎠ i
i =1
If a model is fitted to a sample, then the population moments of the series generated
by this model are functions of model parameters; therefore, the moment parameter
estimates can be obtained by equating sample moments and population moments.
EXAMPLE 5.2
Suppose the following model is used for the sample (x1, ⋯, xN):
x t = α ⋅ zt + β
where zt is an independent random variable with mean zero and variance one; α and
β are the model parameters. Estimate the model parameters using the method of
moments.
Solution: The first and second sample moments can be estimated as:
M1 = ⎛ ⎞ ∑x
1
⎝ N⎠ i
i =1
and
M2 = ⎛ ⎞ ∑x
1 2
⎝ N⎠ i
i =1
The first two population moments of the model are:
E[ X ] = β
and
[ ]
E X 2 = α 2 + β2
By equating the first population and sample moments, it can be written that:
β̂ = ⎛ ⎞ ∑x
1
⎝ N⎠ i
i =1
which is an estimate of the parameter β. By equating the second moments, an

estimate of the second parameter can be obtained:
αˆ =
1
N ∑x
i =1
2
i − βˆ 2
5.5.2 METHOD OF LEAST SQUARES

Consider a sample (x1, ⋯, xN). A model is fitted to the sample as follows:
xˆ t = f ( xt −1 , xt −2 , ⋯, α1 , α 2 , ⋯, α m ) + ε t (5.19)
where εt is the model residual and (α1, α2, ⋯, αm) are the model parameters. The
objective of the least-squares method is to minimize the following sum:
N N
∑ ( xt − xˆt ) ∑ (x − f (x , xt −2 , ⋯, αˆ 1 , αˆ 2 , ⋯, αˆ m ) )
2 2
= t t −1 (5.20)
t =1 t =1
To minimize the above sum, all partial derivatives with respect to estimated values
of the parameters (αˆ 1 , αˆ 2 , ⋯, αˆ m ) should be equal to zero. Therefore, they can be
written as:
N N
∑ ( xt − xˆt ) ∑ ( x − xˆ )
2 2
∂ ∂ t t
t =1 t =1
= 0, ⋯, =0 (5.21)
∂αˆ 1 ∂αˆ m
By solving these equations simultaneously, the model parameters ( αˆ 1 , αˆ 2 , ⋯, αˆ m )

can be estimated.
Example 5.3
Consider that the following model is used for the sample (x1, ⋯, xN):
xt = αxt −1 + ε t
Find the model parameter, α, using the least-squares method.

Solution: The sum of square of errors is estimated as:
∑ ε = ∑ ( x − αx )
2
2
t t t −1
The partial derivatives of the sum with respect to α can be written as:
∑ ( x − αx )
2
∂
∑(x (αxt −1 − xt )) = 0
t t −1
= t −1
∂α
Then the parameter can be estimated as:
α=
∑x x t t −1
∑x 2
t −1
5.5.3 METHOD OF MAXIMUM LIKELIHOOD

Consider the model shown in Equation 5.2. The joint probability of (ε1, ⋯, εN) is
the likelihood function and it can be shown as:
L( ε ) = ∏ f (ε , α , ⋯, α
t =1
t 1 m ) (5.22)
The maximum likelihood of the parameters (α1, ⋯, αm) is obtained by maximizing

the function L(ε). For this purpose, the logarithm of likelihood function can also be
maximized:
N N
Log( L(ε )) = log ∏

t =1
f (ε t , α1 , ⋯, α m ) = ∑ log[ f (ε , α , ⋯, α )]
t =1
t 1 m (5.23)
For maximizing the above sum, partial derivatives with respect to the parameters
(α1, ⋯, αm) should be equated to zero:
∂Log( L(ε )) ∂Log( L(ε ))

= 0, ⋯, =0 (5.24)
∂α1 ∂α m
The maximum likelihood estimate of the parameters can be obtained by simulta-

neously solving the above equations.
5.6 GOODNESS OF FIT TESTS

The goodness of fit tests presented in this section are used to identify the nature of
the unknown frequency distribution of the population. This procedure determines if
a particular shape for the population frequency curve is consistent with the sample
results obtained. Two of the most common test statistics for the goodness of fit are:
1. Chi-square test, for testing the hypothesis of goodness of fit of a given

distribution
2. Kolmogorov–Smirnov statistic, for estimating the maximum absolute dif-
ference between the cumulative frequency curve of sample data and the
fitted distribution function
The following steps should be taken for hypothesis testing (Lapin, 1990):
1. Formulate the null hypothesis (H0). The data represent a specific distri-
bution.
2. Select the test procedure and test statistic (such as chi-square or Kolmo-
grov–Smirnov statistics).
3. Select the significance level and the acceptance and rejection regions for
the decision rule.
4. Compute the value of the test statistic for the sample data.
5. Make the decision.
5.6.1 CHI-SQUARE GOODNESS OF FIT TEST

In this test, two sets of frequencies are compared. First, the data should be divided
into a numbers of categories and then the actual frequencies (fi) in each category
(class) should be compared with the expected frequencies ( fî ), which are estimated
based on the fitted distribution. The test statistic can be estimated as follows:
NC
( f − fˆ )
∑
i i
χ =
2
(5.25)
i =1
fî
TABLE 5.3
Precipitation Data for Example 5.4
Interval Precipitation Depth Classes Actual Frequency
(i) (mm) (fi )
1 0 ≤ P < 5 38
2 5 ≤ P < 10 26
3 10 ≤ P < 15 12
4 15 ≤ P < 20 10
5 20 ≤ P < 25 8
6 25 ≤ P < 30 3
7 30 ≤ P < 35 2
8 35 ≤ P < 40 1
9 40 ≤ P 0
where NC is the total number of categories. This estimation is close enough for
testing the goodness of fit whenever the expected frequency in any category is equal
or greater than 5 (Lapin, 1990). χ2 is chi-square distributed with NC – NP – 1 degree
of freedom, where NP is the number of parameters to be estimated.
Example 5.4
The classified monthly precipitation data measured in a station are divided into nine
categories (see Table 5.3). Previous studies have shown that in spite of assumptions
about distribution of precipitation data, it is exponentially distributed in this station.
Therefore, the probability that the monthly rainfall depth, P, is equal or below x can
be estimated as:
Prob[ P ≤ x ] = 1 − e − λx
Assume λ to be 0.105. Use the chi-square test with a 5% significance level and
comment on the selected distribution for the sample data.
Solution: The expected frequencies are estimated using the selected exponential
distribution as shown in the third column of Table 5.2. For example, for the first
class of the data, the probability of having rain with depth less than 5 mm is:
Prob[ P < 5] = 1 − e −.105×5 = 0.41
The expected frequencies are then estimated based on the difference between cumu-
lative probabilities (see Table 5.4). For example, for the second interval, it can be
written that:
Expected frequency = (0.65 – 0.41) × 100 = 24

TABLE 5.4
Estimated Expected Frequencies and Chi-Square Statistic for the Data
Presented in Example 5.4
Exponential Expected
Actual Cumulative Frequencies
( f̂ )
Interval Frequency Probability at
(i) (fi ) Upper Limit i fi − fî χ2
1 38 0.41 41 –3 0.22
2 26 0.65 24 2 0.17
3 12 0.79 14 –2 0.29
4 10 0.88 9 1 0.11
5 8 0.93 5 3 1.8
6 3 0.96 3
7 2 0.97 1
–1 0.14
8 1 0.99 2
9 0 1 1
Sum 100 — 100 — 2.73
As shown in Table 5.4, the actual frequency for the intervals 6 to 9 is less than 5;
therefore, these classes are grouped together. The chi-square statistic is then esti-
mated as 2.73. The degree of freedom for the test is estimated as:
Degree of freedom = 6 – 1 – 1 = 4
From Appendix A and considering a 5% significance level and 4 degrees of freedom,

the critical value is χ 20.05( 4 ) = 9.488. Because the estimated value of the test statistic
is smaller than the critical value, the null hypothesis that the monthly precipitation
is exponentially distributed is accepted.
5.6.2 KOLMOGOROV–SMIRNOV GOODNESS OF FIT TEST

The chi-square test performs better when the sample size is large enough to include
at least five observations in each class interval. For smaller sample sizes, the
Kolmogorov–Smirnov test performs better. This test is based on the maximum
difference between actual and expected frequencies. The test statistic can be formu-
lated as follows:
D = max F( x ) − Fˆ ( x ) (5.26)
where F is the actual cumulative relative frequency and F̂ is the expected cumulative
relative frequency.
The same steps should be taken for hypothesis testing as discussed in Section
5.5. The estimated value for the test statistic should then be compared with the
critical value selected based on the significance level and number of the data points.
TABLE 5.5
Estimated Expected Frequencies and Kolmogorov–Smirnov Statistic
for the Data Presented in Example 5.5
Expected Cumulative
Actual Actual Cumulative Relative Frequency
( F̂( x))
Interval Frequency Relative Frequency
(i) (fi) (F(x)) D
1 38 0.38 0.41 0.03

2 26 0.64 0.65 0.01
3 12 0.76 0.79 0.03
4 10 0.86 0.88 0.02
5 8 0.94 0.93 0.01
6 3 0.97 0.96 0.01
7 2 0.99 0.97 0.02
8 1 1 0.99 0.01
9 0 1 1 0
Example 5.5
Solve Example 5.4 using the Kolmogorov–Smirnov goodness of fit test.
Solution: Table 5.5 shows the actual and expected cumulative relative
frequencies.
The last column in this table shows the absolute values of the difference
between the actual and the expected cumulative frequencies. As it can be seen, the
maximum value for the test statistic, D, is estimated as 0.03. From Appendix B
and considering 5% significance level and n =100, the critical value, D, is equal
to 0.12. Because the estimated value of the test statistic is smaller than the critical
value, the null hypothesis that the monthly precipitation is exponentially distributed
is accepted.
5.6.3 TESTS OF NORMALITY

The graphical test is the most commonly used test for testing the hypothesis that a
given time series is normal. In this test, the empirical distribution of the series is
plotted on normal probability paper to determine whether the plotted points can be
approximated as a straight line. Figure 5.1 shows the normal probability plot for
streamflows of Salt River in northern Arizona. The streamflow data do not follow
normal distribution.
The chi-square test has also been widely used for testing the normality of data.
The test statistic and the procedure for testing the goodness of fit hypothesis were
presented in Section 5.5.1. Another normality test that is widely used in practice
is the skewness test. The skewness coefficient of a time series xt is estimated as
follows:
99.9
99
95
Cumulative percent
80
50
20
0.1
0 300 600 900 1200 1500
Volume (MCM)
FIGURE 5.1 The normal probability plot for monthly streamflow at Salt River Station on
Salt River in Arizona (1914–1998).
∑ (x − x)
1 3
t
N t =1
γˆ = 3/ 2 (5.27)
⎡1 N
⎤
∑ ( xt − x ) ⎥
2
⎢
⎢⎣ N t =1 ⎥⎦
where N is the total number of sample data and x is the sample mean. The skewness
test is based on the fact that the skewness coefficient of a normal variable is zero.
If the series is normally distributed, γ̂ is asymptotically normally distributed with
mean zero and variance 6/N (Snedecor and Cochran, 1967). Then (1 – α) probability
limits on γ could be defined as:
⎡−u 6 , u1−α / 2 6 N ⎤ (5.28)

⎣⎢ 1−α / 2 N ⎦⎥
where u1−α/2 is the (1 – α/2) quantile of the standard normal distribution. Therefore,
if γ̂ falls within the limits of Eq. (5.4), the hypothesis of normality is accepted;
otherwise, it is rejected. Snedecor and Cochran (1967) recommended the above
TABLE 5.6
Table of Skewness Test for Normality
for Sample Sizes, Less than 150
N α = 0.02 α = 0.01
25 1.061 0.711
30 0.986 0.662
35 0.923 0.621
40 0.870 0.587
45 0.825 0.558
50 0.787 0.534
60 0.723 0.492
70 0.673 0.459
80 0.631 0.432
90 0.596 0.409
100 0.567 0.389
125 0.508 0.350
150 0.464 0.321
175 0.430 0.298
Source: Adapted from Snedecor and Cochran (1967).
procedure when the sample size is greater than 150. For smaller sample sizes, they
suggest comparing the computed coefficient of skewness with the tabulated values,
γα(N), presented in Table 5.6. If γ̂ < γ α ( N ) , the hypothesis of normality is accepted.
Example 5.6
In order to bring the probability distribution of the monthly Salt River streamflows
(see Figure 5.1) closer to a normal distribution, the logarithmic transformation is
applied as follows:
yt = log10 ( xt )
where xt and yt are the original and transformed streamflow series in month t,
respectively. Figure 5.2 shows the normal probability plot for the transformed series
for the period 1914 to 1998. The skewness coefficient of the transformed series is
estimated to be 0.8. Use the skewness test and comment on the normality of the
transformed data.
Solution: As shown in Figure 5.2, the transformed series displays a better fit to
the normal distribution compared with the original series; however, it still cannot
be approximated as a straight line. If the 5% significance level is selected, the 95%
probability limits on the skewness coefficient can be estimated using Eq. (5.4):
99.9
99
95
Cumulative percent
80
50
20
0.1
3.1 3.6 4.1 4.6 5.1 5.6
Monthly streamflows
FIGURE 5.2 The normal probability plot for transformed montly streamflow at Salt River
Station at Salt River in Arizona (1914–1998).
⎡−u 6 , u0.975 6 N ⎤ = ⎡−1.96 6 1020 , 1.96 × 6 1020 ⎤ = [−0.15 , 0.15]

⎢⎣ 0.975 N ⎥⎦ ⎢⎣ ⎥⎦
Because 0.8 does not fall within the above limits, the series is not normally distributed.
5.6.4 TESTS OF INDEPENDENCE IN TIME

The modeling of a hydrologic time series usually assumes that the stochastic com-
ponent, after removing the periodic components and time-dependence structure are
removed, is an independent and normally distributed series (Salas et al., 1988). The
basic statistical representation of the time-dependence structure is the correlogram,
which shows the fluctuations of the autocorrelation coefficient of the series for
different lags. The autocorrelation coefficient is a dimensionless measure of linear
dependence that can be estimated as follows:
N −k
∑ ( x − x)( x
t =1
t t +k −x )
rk = N (5.29)
∑ ( x − x)
2
t
t =1
where rk is the lag-k autocorrelation coefficient and x is the sample mean. Figure 5.3
shows the correlogram of the Salt River monthly stream flows for the period of 1914
0.5
0.25
Autocorrelation
-0.25
-0.5
0 5 10 15 20 25
Lag
FIGURE 5.3 Correlogram of monthly streamflow of Salt River in Arizona (1914–1998).
to 1998. For an independent series, the population correlogram is equal to zero for
k ≠ 0. Even though correlograms of samples of independent series show fluctuations
around zero, they are not necessarily equal to zero due to sampling variability.
Anderson (1942) determined probability limits for the correlogram of independent
series as follows:
• Anderson test of independence in time. Anderson (1942) gave the fol-

lowing limits with 95% and 99% probability levels:
−1 ± 1.645 N − k − 1
rk (95%) = (5.30)
N−k
−1 ± 2.326 N − k − 1
rk (99%) = (5.31)
N−k
where N is the sample size. Based on this test, autocorrelations falling

outside of the probability limit indicate significant time dependence; oth-
erwise, the series can be considered as an independent series.
• Porte Manteau test. The Porte Manteau lack-of-fit test was first devel-
oped by Box and Pierce (1970) as an approximate test of model adequacy.
Hipel and McLeod (1977) and others applied this test for verifying linear
models of hydrologic time series. In this test, instead of checking to see

if autocorrelations fall within the confidence bound, a single statistic that
depends on the autocorrelation function is used. Suppose that the autore-
gressive integrated moving average (ARIMA) (p, d, q) model (discussed
later in this chapter) is selected for representing a time series, where p is
the number of autoregressive terms, q is the number of moving average
terms, and d is the number of differences. As mentioned, this test evaluates
the adequacy of a model by determining whether the residual series of
the model is independent. The following statistic is used in this test:
Q = (N − d) ⋅ ∑ r (ε )
k =1
k
2
(5.32)
where L is maximum lag considered. The statistic Q is approximately chi-

square distributed with L – p – q degrees of freedom. Therefore, if the
estimated test statistic is smaller than the chi-square value, χ α2 ( L − p − q) ,
of a given probability limit (1 – α), εt is an independent series and the
model is adequate. Ljung and Box (1978) argued that under the hypothesis
of model adequacy the cutoff value given by χ α2 (L – p – q) is closer to
the (1 – α) quantile of the distribution of the following statistic:
Q ′ = n(n + 2) ∑ r (ε ) (n − j )
k =1
k
2
(5.33)
Tao and Delluer (1976) used the following statistic for testing the ade-
quacy of a seasonal ARMA model:
L ω
Q1 = N ⋅ ∑∑r
k =1 τ =1
2
k ,τ (ε ) (5.34)
where ω is the number of time intervals per year (seasons). Granger and
Anderson (1978) found examples where the residuals were uncorrelated
while the squared residuals were correlated. The sample autocorrelation
function of squared residuals ε̂ t2 can be estimated as (Brockwell and
Davis, 1987):
N−L
∑ (εˆ t
2
)(
− ε 2 εˆ t2+ L − ε 2 )
ρˆ εε ( L) = t =1
L ≥1 (5.35)
∑( )
N
εˆ t2 − ε 2
t =1
where
N
∑ εˆ 2
t
ε2 = t =1
N
McLeod and Li (1983) showed that:
Qεε = N ( N + 2) ∑ ρˆk =1
2
εε (k ) ( N − k ) (5.36)
has an approximate χ2(L) distribution under the assumption of model

adequacy; therefore, the adequacy of model is rejected at significance
level α if:
Qεε > χ12−α ( L) (5.37)
• Cumulative periodogram test. This test can be used for determining

whether the periodicity is adequately removed from series. The peri-
odogram of sample residual, ε̂ t , can be estimated as follows (Salas et al.,
1988) and MSD stands for mean squared division:
⎡⎛ N
⎞ ⎛
2 N
⎞
2
⎤
( ) 2
MSD h j = 2
N
⎢⎜
⎢⎝ ∑ (
εˆ t ⋅ Cos 2 πh j t ⎟ + ⎜
⎠ ⎝
) ∑ ( )
εˆ t Sin 2 πh j t ⎟
⎠
⎥
⎥
(5.38)
⎣ t =1 t =1 ⎦
where hj = j/N is the harmonic (frequency). The cumulative periodogram

of ε̂ t can then be defined as:
∑ MSD(h )
j =1
j
Pi = i = 1, ⋯, N / 2 (5.39)
σ 2ε
where σ 2ε is the variance of ε̂ t . For an independent series, the plot of Pi

versus i will be scattered around a line joining (0,0) and (0.5N,1). A
periodicity with harmonic hj in the residuals would produce a large value
of MSD(hj) and in turn would appear as a deviation from the straight line,
which is valid for an independent series (Salas et al., 1988). Probability
limit lines using the Kolmogorov–Smirnov statistic can be drawn to test
the significance of such deviations (Box and Jenkins, 1970). Considering
the significance level α, the limit lines for a real independent series are
drawn approximately at distances ± Kα N ′ around the theoretical
TABLE 5.7
Approximate Critical Values of the Kolmogorov–Smirnov
Statistic Kα (Yevjevich 1972a)
α 0.01 0.05 0.1 0.2 0.25

Kα 1.63 1.36 1.22 1.07 1.02
straight line where N′ = (N – 2)/2 or N′ = (N – 1)/2 for even or odd N

values, respectively. If more than αN of the plotted points fall outside the
probability lines, the residual series still has some periodicity; otherwise,
it can be concluded that the residuals are independent. Approximate values
of Kα are given in Table 5.7.
• Turning point test of randomness. Suppose the time series xt are obser-
vations of a random sequence. The turning point at time t is detected when
xt–1 < xt and xt > xt+1 or xt < xt–1 and xt < xt+1. If Tp is the number of turning
points, then the probability of a turning point at time t is 2/3. The expected
number of turning points as stated by Brockwell and Davis (1987) is:
2( n − 2 )
µ Tp = (5.40)
3
The variance of the number of turning points can be determined as follows:
(16n − 29)
σ T2P = (5.41)
90
It can be shown for a random sequence, the number of turning points (Tp)
has an N (µ Tp , σ T2p ) distribution. So the assumption that xt is a random
sequence is rejected if:
Tp − µ Tp
> u1−α 2 (5.42)
σ Tp
where u1−α/2 is the (1 – α/2) probability level of a standard normal

distribution.
Example 5.7
A series of residuals of a model fitted to precipitation data in a station for 20 time
intervals is given in Table 5.8. Use the turning point test and comment on the
randomness of the residuals.
TABLE 5.8
The Series of Residuals for Example 5.7
Month Residual Month Residual
1 0.87 11 0.30
2 0.45 12 0.51
3 0.33 13 0.41
4 0.36 14 0.24
5 0.31 15 0.56
6 0.12 16 0.48
7 0.33 17 0.06
8 0.17 18 0.77
9 0.78 19 0.36
10 0.90 20 0.22
Solution: The turning points occurred in months 3, 4, 6, 7, 8, 10, 11, 12, 14,
17, 18. The expected value and variance of the number of turning points can be
estimated using Eqs. (5.16) and (5.17):
2(20 − 2) (16 × 20 − 29)

µ Tp = = 12 and σ TP = = 1.80
3 90
The statistic for testing the randomness can be defined using Eq. (5.18):
11 − 12
= 0.556
1.8
which is less than the critical value of the N(0,1) distribution with a 5% significance
level (u1–0.05/2 = 1.96), so the assumption of randomness is accepted.
5.7 AKAIKE’S INFORMATION CRITERION (AIC)

Akaike’s information criterion (AIC) (Akaike, 1974), which considers parsimony in
model building, is the primary criterion used for model selection. Parsimony is the
tendency to select minimum number of parameters in order to reduce the chance of
error accumulation. Akaike defined the AIC among competing autoregressive mov-
ing average (ARMA (p,q)) models as follows:
( )
AIC( p, q) = N ⋅ ln σˆ 2 (ε ) + 2( p + q) (5.43)
where N is the sample size and σ̂ 2 (ε ) is the maximum likelihood estimate of the
residual variance. Based on this criterion, the model with minimum AIC is selected.
The goodness of fit tests should be applied to the model with minimum AIC to make
sure the residuals are consistent with their expected behavior. If the model residuals
do not pass the test, the models with higher values of AIC should be checked.
5.8 DATA GENERATION AND FORECASTING

Two major uses of time-series modeling are generation of synthetic values and
forecasting future events. The generation of synthetic time series often consists of
the generation of independent normal variables in a series defined as a random
process. The generated random component of a series will be added to its trend or
deterministic component to reproduce the original time series. The second important
application of time series models is forecasting of future events. The forecast values
are often different from the observed values; therefore, a confidence interval is
assigned for the forecast values.
5.9 AUTOREGRESSIVE MODELING

Autoregressive (AR) models have been used widely in hydrologic time-series mod-
eling. These models incorporate the correlation between time sequences of variables.
These models are the simplest models and their development goes back to the
application of Thomas Fiering and Markov lag-1 models. They can be classified into
the following subsets:
• AR models with constant parameters, which are typically used for mod-
eling of annual series
• AR models with timely variable parameters, which are typically used for
modeling of seasonal (periodic) series
The basic form of the AR model of order p, (AR p), with constant parameters is:
zt = ∑φ z
i =1
i t −i + εt (5.44)
where:
zt is the time-dependent normal and standardized series {N(0,1)}. φi are the

autoregressive coefficients. εt is the time-independent variable (white noise).
p is the order of the autoregressive model.
The above formulation is based on the standardized series, which can be obtained
as follows:
xt − µ
zt = (5.45)
σ
where µ and σ are the mean and standard deviation of the series xt. The parameter
set of the model is:
{µ, σ, φ ,⋯, φ , σ (ε)}

1 p
2
(5.46)
where σ2(ε) is the variance of the time-independent series. The model parameters
can be estimated by solving the following linear equations (Yule–Walker equations)
simultaneously:
ri = φˆ 1ri −1 + φˆ 2 ri −2 + ⋯ + φˆ p ri − p (i ≤ p ) (5.47)
where ri are the sample correlation coefficients of lag i. The parameter σ2(ε) can
also be estimated using the following relation:
Nσˆ 2 ⎛ ⎞
p
σˆ 2 (ε ) =
( N − p) ⎜⎝
1− ∑ φˆ r ⎟⎠
i =1
i i (5.48)
where N is the number of the data and σ̂ 2 is the sample variance.

The stationary condition must be met by the model parameters. For this purpose,
the roots of the following equation should lie inside the unit circle (Yevjevich, 1972):
u p − φˆ 1u p−1 − φˆ 2 u p−2 − ⋯ − φˆ p = 0 (5.49)
In other words, we must have ⏐ui⏐ < 1. In order to forecast or generate annual AR
models, the following relation can be used:
zˆt = φˆ 1 zˆt −1 + ⋯ + φˆ P zˆt − P + σˆ (ε )ζ t (5.50)
where ζ t is the standardized normal variable.
Example 5.8
For an AR(2) model, the parameters have been estimated as φ1 = 0.5 and φ2 = 0.3.
Check the stationary condition of the parameters.
Solution: Using Eq. (5.49), the following expressions can be written:
⎧⎪u1 = 0.85
u 2 − 0.5u − 0.3 = 0 ⇒ ⎨
⎪⎩u2 = −0.35
The roots lie within the unit circle; therefore, the parameters pass the stationary
condition.
Example 5.9
For a sample of 100-year normal and standardized annual inflows to a reservoir, an
AR(2) model is selected. The first and second correlation coefficients are estimated
as r1 = 0.6 and r2 = 0.35. Estimate the model parameters if the variance of the normal
and standardized inflow series is estimated as 1.5.
Solution: Using Eq. (5.47), we write:
⎧⎪r1 = φ1 + φ 2 r1
⎨
⎩⎪r2 = φ 2 + φ1r1
By solving these equations,
r1 (1 − r2 ) r2 − r12
φ1 = and φ2 =
1− r 1
2
1 − r12
Then,
φ1 = 0.61 and φ 2 = −0.02
The variance of model residuals can also be estimated using Eq. (5.48):
100 × 1.5
σˆ 2 (ε ) = (1 − 0.6 × 0.61 + 0.35 × 0.02) = 0.96
(100 − 2)
In order to identify whether an AR(p) is an appropriate model for a specific time
series, it is necessary to estimate and investigate the behavior of the partial auto-
correlation function (PACF) of the series. The PACF represents the time-dependence
structure of the series in a different way. The partial autocorrelation lag k may be
regarded as the correlation between z1 and zk+1, adjusted for the intervening obser-
vations (z2, …, zk). For an AR(p) process, the partial autocorrelation function is the
last autoregressive coefficient of the model φk(k), which can be estimated by rewriting
Eq. (5.33) as follows:
ri = φˆ 1 (k ) ri −1 + φˆ 2 (k ) ri −2 + ⋯ + φˆ k (k ) ri −k (i = 1,…, k )
which represents a set of linear equations,
φˆ 1 (k ) r0 + φˆ 2 (k ) r1 + ⋯ + φˆ k (k ) rk −1 = r1
φˆ 1 (k ) r1 + φˆ 2 (k ) r0 + ⋯ + φˆ k (k ) rk −1 = r2
⋮
φˆ 1 (k ) rk −1 + φˆ 2 (k ) rk −2 + ⋯ + φˆ k (k ) r0 = rk
which can be written as:
⎡1 r1 r2 ⋯ rk −1 ⎤ ⎡ ϕ1 (k ) ⎤ ⎡ r1 ⎤
⎢r 1 r1 ⋯ rk −2 ⎥ ⎢⎢ϕ 2 (k )⎥⎥ ⎢r2 ⎥
⎢ 1 ⎥ ⎢ ⎥
⎢ r2 r1 1 ⋯ rk −3 ⎥ ⎢ϕ 3 (k )⎥ = ⎢r3 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ ⋮ ⋮ ⋮ ⋮ ⎥ ⎢ ⋮ ⎥ ⎢⋮⎥
⎢⎣rk −1 rk −2 rk −3 ⋯ 1 ⎥⎦ ⎢⎣ϕ k (k )⎥⎦ ⎢⎣rk ⎥⎦
Thus, the partial autocorrelation function φk(k) can be determined by successively

applying the above relation. It can be shown that the partial correlogram of an
autoregressive process of order p has peaks at lags 1 through p and then cuts off.
Hence, the partial autocorrelation function can be used to identify the order p of an
AR(p) model.
The basic form of the AR model with periodic parameters is
zν, τ = ∑φ
i =1
z
i , τ ν, τ − i + σ τ (ε )ζ ν,τ (5.51)
where zν,τ is normal and the standardized value in year ν and season τ, φi,τ are periodic
autoregressive coefficients, and στ(ε) is the standard deviation of residuals in
season τ. zν,τ is estimated using the seasonal mean and variances as follows:
x ν, τ − µ τ
z ν, τ = (5.52)
στ
where µτ and στ are the mean and standard deviation of x in season τ, respectively.
The parameter set of the model can be summarized as:
{µ , σ , φ
τ τ 1,τ , ⋯, φ P,τ , σ 2τ (ε ), τ = 1, ⋯, η }
where η is the total number of seasons. Equation (5.47) can be similarly used for
each season:
rk = ∑ φˆ
i =1
r
i ,τ k −i ,τ − min ( k ,i ) ( k > 0) (5.53)
The residual variance can be estimated using the following relation:
σˆ 2τ (ε ) = 1 − ∑ φˆ
j =1
j ,τ ⋅ rˆj ,τ (5.54)
5.10 MOVING AVERAGE PROCESS

The autoregressive models can be used as an effective tool for modeling hydrologic
time series such as low-flow-season streamflow that is mainly supplied from ground-
water and has low variations. But the previous studies have shown that the stream-
flows in high-flow season can be better formulated by adding a moving average
component to the autoregressive component (Salas et al., 1988).
If the series zt is dependent only on a finite number of previous values of a
random variable, εt, then the process can be called a moving average process. The
moving average model of order q (MA(q)) can be formulated as:
zt = ε t − θ1ε t −1 − θ 2 ε t −2 − θ 3 ε t −3 − ⋯ − θ q ε t −q (5.55)
It can also be written as:
zt = − ∑θ ε
j =0
j t− j (θ 0 = −1) (5.56)
where θ1, ⋯, θq are q orders of MA(q) model parameters. The parameter set of the
model can be summarized as:
{µ, θ ,⋯, θ , σ (ε)}

1 q
2
The following relation can be used for estimating the moving average parameters:
⎧ q−k
⎪ ∑
⎪ j =0
θ jθ j +k
⎪ q (k ≤ q)
rk = ⎨
⎪
1+ ∑ θ2j
(5.57)
⎪ j =1
⎪
⎩0 (k > q)
The parameters of the model should satisfy the invertibility condition. This condition,
for example, for a MA(1) process, shows that the first order moving average process
is inverted into an autoregressive process of infinite order, provided that |θ1 < 1|. In
other words, the root of u – θ1 = 0 should be inside the unit circle. To generalize
the invertibility condition to an MA(q), the roots of the following polynomial should
lie inside the unit circle:
u q − θˆ 1u q −1 − θˆ 2 u q −2 − ⋯ − θˆ q = 0 (5.58)
Example 5.10
For an MA(2) model, the parameters have been estimated as θ1 = 0.65 and θ2 = 0.3.
Determine whether the parameters pass the invertibility condition.
Solution: Using Eq. (5.35), we can write:
⎪⎧u1 = 0.96
u 2 − 0.65u − 0.3 = 0 ⇒ ⎨ (5.59)
⎪⎩u2 = −0.31
The roots lie within the unit circle; therefore, the parameters pass the invertibility
condition.
5.11 AUTOREGRESSIVE MOVING AVERAGE

MODELING
The ARMA model of order (p,q) can be defined by combining an autoregressive
model of order p and a moving average model of order q as follows:
zt − φ1 zt −1 − ⋯ − φ p zt − p = ε t + θ1ε t −1 + ⋯ + θ q ε t −q (5.60)
The ARMA(p,q) model can also be shown in the following compact form:
φ( B)zt = θ( B)ε t (5.61)
where φ(z) and θ(z) are the pth and qth degree polynomials:
φ( z ) = 1 − φ1 z − ⋯ − φ p z p (5.62)
θ( z ) = 1 + θ1 z + ⋯ + θ q z q (5.63)
The parameter set of the ARMA(p,q) model can be summarized as:
{µ, θ ,⋯, θ , φ ,⋯, φ , σ (ε)}

1 q 1 p
2
(5.64)
The parameters of the ARMA(p,q) model should satisfy both the conditions of
invertibility and stationarity.
Table 5.9 shows the characteristic behavior of the autocorrelation and of the
partial autocorrelation for the AR, MA, and ARMA processes, which can be effec-
tively incorporated in selecting the type and form of the statistical model.
TABLE 5.9
Characteristic Behavior of AR, MA, and ARMA Processes
Process Autocorrelation Partial Autocorrelation
AR(p) Damped and infinite in extent Peaks at lags 1 through p and then cuts off
exponentials and/or waves
MA(q) Peak at lags 1 through q and then cuts off Damped and infinite in extent
exponentials and/or waves
ARMA(p,q) Irregular in first q-p lags and then damped Irregular in first p-q lags and then damped
and infinite in extent exponentials and/or and infinite in extent exponentials and/or
waves waves
5.11.1 DATA GENERATION AND FORECASTING USING ARMA

MODELS
Once the ARMA model is fitted to a time series, the following procedures can be
used for generating or forecasting the values of that time series. For generation using
the ARMA(p,q) model (Eq. (5.60)), it is necessary to give p initial Z values to the
model. Equation (5.60) may be used recursively to generate the desired numbers of
Zt. By generating a sufficiently long series and deleting 50 or 100 initial terms, the
transient effect of the initial values is negligible (Salas et al., 1988). The synthetic
generated values conserve the statistical properties of the historical data.
Also, ARMA(p,q) can be used to forecast Z values for lead time L. If Zt(L)
denotes the value of Zt at lead time L, the following equations could be used for
forecasting:
Zt ( L) = φ1 Zt + L−1 + φ 2 Zt + L−2 + .... + φ p Zt + L− p − θ1ε t + L−1 − ... − θ q ε t + L−q for L ≤ q (5.65)
Zt ( L) = φ1 Zt + L−1 + φ 2 Zt + L−2 + .... + φ p Zt + L− p for L > q (5.66)
5.12 AUTOREGRESSIVE INTEGRATED MOVING

AVERAGE (ARIMA) MODELING
The ARMA models are suitable for the data that have the following two basic
characteristics:
1. No apparent deviation from stationarity

2. Rapidly decreasing autocorrelation function
If these conditions are not met by a time series, a proper transformation should be
performed to generate time series satisfying the above two conditions. This has usually
been achieved by differencing, satisfying the essence of ARIMA models. This class
of models is very powerful for describing stationary and nonstationary time series.
The nonseasonal form of ARIMA models of order (p,d,q) can be formulated as:
φ( B)(1 − B) zt = θ( B)ε t
d
(5.67)
where φ(B) and θ(B) are polynomials of degree p and q, respectively:
φ( B) = 1 − φ1 B − φ 2 B 2 − .... − φ p B p (5.68)
θ( B) = 1 − θ1 B − θ 2 B 2 − .... − θ q B q (5.69)
The model has p + q + 1 parameters to be estimated:
{φ ,⋯, φ , θ ,⋯, θ , σ (ε)}

1 p 1 q
2
In order to model the seasonal hydrologic time series, the seasonal form of ARIMA
models of nonseasonal order (p,d,q) and of seasonal order (P,D,Q)w with seasonality
w has also been developed. The general multiplicative ARIMA(p,d,q)(P,D,Q)w can
be formulated as:
(1 − Φ1 B w − Φ 2 B 2 w − .... − Φ p B Pw )(1 − φ1 B − φ 2 B 2 − .... − φ p B p )(1 − B w ) D (1 − B) d zt =

(5.70)
(1 − Θ1 B w − Θ 2 B 2 w − ... − Θ Q BQW )(1 − θ1 B − θ 2 B 2 − ... − θ q B q )ε t
where:
εt is an independently distributed random variable.

B is the backward operator; B(zt) = zt–1.
(1 – Bw)D is the Dth seasonal difference of season w.
(1 – B)d is the dth nonseasonal difference.
p is the order of the nonseasonal autoregressive component.
q is the order of the nonseasonal moving average component.
P is the order of the seasonal autoregressive component.
Q is the order of the seasonal moving average component.
Φ is the seasonal autoregressive parameter.
Θ is the seasonal moving average parameter.
φ is the nonseasonal autoregressive parameter.
θ is the nonseasonal moving average parameter.
Equation (5.70) can also be shown in the following condensed form:
( ) (
Φ B w φ( B) 1 − B w ) (1 − B)d zt = Θ( B w )θ( B)ε t
D
(5.71)
Estimating the parameters are done by computer packages. For details of parameter
estimation refer to Salas et al. (1989).
Example 5.11
Formulate the ARIMA(1,0,1)(1,0,1)12.

Solution: Using Eq. (5.67), it can be written that:
(1 − Φ1 B w )(1 − φ1 B)zt = (1 − Θ1 B w )(1 − θ1 B)ε t
or
zt − φ1 zt −1 − Φ1 zt −12 + φ1Φ1 zt −13 = ε t − θ1ε t −1 − Θ1ε t −12 − θ1Θ1ε t −13
The model has nonlinear parameters because of the terms θΘ and φΦ.
5.12.1 TIME SERIES FORECASTING USING ARIMA MODELS

The ARIMA models are nonstationary and cannot be used for synthetic generation
of stationary time series but they are useful for forecasting. The previous ARIMA
model equations can be used for data forecasting. As an example, the forecasting
equations for ARIMA(1,0,1)(1,0,1)12 could be represented as follows:
zt ( L) = φ1 zt −1+ L + Φ1 zt −12 + L − φ1Φ1 zt −13+ L − θ1ε t −1+ L

(5.72)
− Θ1ε t −12 + L − θ1Θ1ε t −13+ L for L ≤ 12
zt ( L) = φ1 zt −1+ L + Φ1 zt −12 + L − φ1Φ1 zt −13+ L for L > 12 (5.73)
5.13 MULTIVARIATE AND DISAGGREGATION

MODELING OF TIME SERIES
5.13.1 MULTIVARIATE MODELING

The planning, design, and operation of water resources systems often involve several
hydrologic time series. In this case, multivariate stochastic analysis and multivariate
modeling are necessary to consider cross-correlations between different time series.
Many investigators, such as Fiering (1964), Matalas (1967), Mejia (1971), Valencia
and Schaake (1973), and O’Connell (1974), have proposed multivariate models.
Disaggregation modeling is often done within a multivariate model framework.
Disaggregation models are used to decompose time series to several subseries that
are temporal or spatial fractions of the key time series. Most applications of disag-
gregation have been in the temporal domain, although some investigators have
applied the same principle in the spatial domain. The first well-known disaggregation
model was presented by Valencia and Schaake (1973). Multivariate and disaggrega-
tion models may be used either for the generation of synthetic time series or for
forecasting. For disaggregation models, the key series must be available as an input
to the model.
An AR(1) multivariate model may be represented as:
Zt = A1 Zt −1 + Bε t (5.74)
where Z is an (n × 1) vector of elements z, which represent the values of n time

series; A1 and B are an (n × n) matrix of parameters obtained by calculating cross
correlation presented in the next section; and ε is an (n × 1) vector of independent,
normally distributed random variables with zero mean and variance one.
Equation (5.74) may be written in the matrix form as:
⎡ Zt(1) ⎤ ⎡ a11 a12 ⋯ a n1 ⎤ ⎡ Zt −1 ⎤ ⎡ b11

(1)
b12 ⋯ b n1 ⎤ ⎡ ε t ⎤
(1)
⎢ ( 2 ) ⎥ ⎢ 21 ⎥ (2) ⎥ ⎢
⎢ ⎥ (2) ⎥
⎢
⎢ Zt ⎥ ⎢ a a 22 ⋯ a 2 n ⎥ ⎢ Zt −1 ⎥ ⎢b 21 b 22 ⋯ b 2 n ⎥ ⎢ε t ⎥
⎢ ⋮ ⎥=⎢ ⋮ ⋮ ⋮ ⋮ ⎥⎢ ⋮ ⎥ ⎢ ⋮
+
⋮ ⋮ ⋮ ⎥⎢ ⋮ ⎥
(5.75)
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ Zt( n ) ⎥ ⎢⎣a n1 a n2 ⋯ a nn ⎥⎦ ⎢⎣ Zt(−n1) ⎥⎦ ⎢⎣b n1 b n2 ⋯ b nn ⎥⎦ ⎢⎣ε (t n ) ⎥⎦
⎣ ⎦
The correlation structure of Zt of the above equation indicates a lag-0 and lag-1
cross-correlation in time. In the case of two time series (n = 2), the formulations for
AR(1) multivariate model are:
Zt(1) = a11 Zt(−11) + a12 Zt(−21) + b11ε (t1) + b12 ε (t 2 ) ⎫⎪

⎬ (5.76)
Zt( 2 ) = a21 Zt(−11) + a22 Zt(−21) + b21ε (t1) + b22 ε (t 2 ) ⎭⎪
The extended AR(2) multivariate model is:
Zt = A1 Zt −1 + A2 Zt −2 + Bε t (5.77)
The parameters of Eq. (5.72) are matrices A1, A2, and B; Salas et al. (1988) provide
details of the mathematical estimation of the model parameters.
The main parameter for estimating the other parameters of multivariate models
is the covariance structure represented by a correlation matrix. The correlation matrix
of lag k for n correlated time series can be obtained as:
⎡ρ11
k , ρ12
k , ..., ρ1kn ⎤
⎢ 21 ⎥
ρ , ρk ,
22
..., ρ2k n ⎥
Mk = ⎢ k (5.78)
⎢ . . . . ⎥
⎢ n1 ⎥
⎢⎣ρk , ρnk 2 , ..., ρnn
k ⎥ ⎦
Where ρijk is the lag-k cross-correlation coefficient between time series Zti and Ztj .
For the multivariate AR(1) model, the multivariate correlation function can be written
by one of the following two expressions:
Mk = A1 Mk −1, k > 0 ⎫
⎪
⎪
⎪⎪
or ⎬ (5.79)
⎪
⎪
⎪
Mk = A1 M0, k ≥ 0 ⎪⎭
k
And, for the multivariate AR(2) model, the multivariate correlation function is:
Mk = A1 Mk −1 + A2 Mk −2 , k > 0 (5.80)
The moment estimates of the parameters A1 and B for the multivariate AR(1) may
be obtained from Eqs. (5.81) and (5.82):
Aˆ1 = Mˆ 1 Mˆ 0−1 (5.81)
and
Bˆ BT = Mˆ 0 − Aˆ1 Mˆ 1T (5.82)
Similarly, for the multivariate AR(2) model, the moment estimate of the parameters
A1, A2, and B may be obtained from Eqs. (5.78) and (5.79) (Salas and Pegram, 1978):
[ ] [ ]
−1
Aˆ1 = Mˆ 1 − Mˆ 2 Mˆ 0 −1 Mˆ 1T × Mˆ 0 − Mˆ 1 Mˆ 0−1 Mˆ 1T (5.83)
[ ] [ ]
−1
Aˆ 2 = Mˆ 2 − Mˆ 1 Mˆ 0 −1 Mˆ 1T × Mˆ 0 − Mˆ 1T Mˆ 0−1 Mˆ 1 (5.84)
[
BBT = Mˆ 2 − Aˆ1 Mˆ 1T + Aˆ 2 Mˆ 2T ] (5.85)
Example 5.12
Table 5.10 provides 18 years of annual flows at two stations of a river basin.
Formulate the multivariate model AR(1) for generating synthetic data for these two
stations. Assume that the data follow normal distribution.
TABLE 5.10
Annual Streamflows of Two Stations (million m3)
Streamflows of Streamflows of
Year Station 1 Station 2
1 150 250
2 200 450
3 330 502
4 401 720
5 550 550
6 662 600
7 445 465
8 551 650
9 302 390
10 203 300
11 405 520
12 470 740
13 502 780
14 608 705
15 402 600
16 501 590
17 460 480
18 230 320
Solution: The correlation coefficients matrices of lag-0 and lag-1 are computed
using Eq. (5.78):
⎡ 1 0.76⎤
M0 = ⎢
⎣0.76 1 ⎥⎦
⎡ 0.41 0.12 ⎤
M1 = ⎢
⎣0.52 0.37⎥⎦
The parameters A1 and BBT are estimated using Eqs. (5.81) and (5.82), respectively:
⎡0.78 −0.47⎤
A1 = ⎢
⎣0.56 −0.05⎥⎦
⎡0.74 0.54⎤
BBT = ⎢
⎣0.54 0.73⎥⎦
Matrix B could then be estimated as:
⎡0.82 0 ⎤
B=⎢
⎣ 0.6 −0.57⎥⎦
After estimating the multivariate AR(1) parameters, Eq. (5.74) could be used to
generate synthetic data for the stations.
5.13.2 DISAGGREGATION MODELING

Disaggregation modeling is a process by which time series are generated depending
on a time series already available. Disaggregation models allow us to preserve
statistics of time series at more than one level. The levels may be represented in
both time and space. Disaggregation models have two major benefits. First, these
techniques reduce the number of parameters in comparison with the other data-
generation models. Second, disaggregation allows for increased flexibility in the
methods used for generation (Salas et al., 1988).
The two basic forms of disaggregation are temporal and spatial models. A
temporal example is the disaggregation of an annual time series into a seasonal time
series. An example of spatial disaggregation is the disaggregation of total natural
flow of a river basin into individual tributary flows.
The general disaggregation is a linear dependence model as follows:
Y = AX + Bε (5.86)
where Y is the subseries or dependent series generated based on the key (independent)
time series X; ε represents the current value from a completely random series; and
A and B are the model parameters. In general, the above variables and parameters
are represented in the form of vectors and matrices.
Two other disaggregation models are the extended and condensed models. The
extended model developed by Mejia and Rousselle (1976) is an extension of the
basic temporal model. An additional term is added to the model to consider seasonal
covariance between seasons of the current year and seasons of the past year. The
model takes the form:
Y = AX + Bε + CZ (5.87)
where Z is a column matrix containing seasonal values from the previous years, and
C is an additional parameter matrix.
Lane (1979) has developed a condensed model that can be written as:
Yτ = Aτ X + Bτ ε + Cτ Yτ−1 (5.88)
The subscript τ denotes the current season being generated. Parameters for the linear,
extended, and condensed disaggregation models can be estimated as follows (Salas
et al., 1988):
For the linear model:
Â = SYX SXX −1 (5.89)
ˆ ˆ T = S − Aˆ ⋅ S
BB (5.90)
YY XY
where Â and B̂ are the estimated parameters. In a temporal disaggregation model,

SYX is the matrix of covariances between the seasonal and annual series, SXX is the
matrix of covariances among the annual series, and SYY is the matrix of covariances
among the seasonal series.
For the extended model, the parameters of the model are estimated as follows:
Â = ( SYX − SYZ SZZ −1SZX )( SXX − SXZ SZZ −1SZX )

−1
(5.91)
(
Cˆ = SYZ − AS
ˆ
XZ SZZ )
−1
(5.92)
(ˆ
ˆ ˆ T = S − AS
BB YY
ˆT ˆ ˆT ˆ ˆT ˆ ˆT
XX A − ASXZ C − CSZX A − CSZZ C ) (5.93)
Or, equivalently,
ˆ
ˆ ˆ T = S − AS ˆ
BB YY XY − CSZY (5.94)
The notation is the same as for the basic linear model. Z consists of the lagged
partial seasonal series, which are used in addition to those used in the linear basic
model. SZX and SZY are the matrix of covariances between the lagged partial seasonal
series and the annual and seasonal series, respectively.
The parameters of the condensed model are estimated as:
[
Aˆ τ = SYX ( τ, τ) − SYY ( τ − 1, τ)SYY −1 ( τ − 1, τ − 1)SYX ( τ − 1, τ) × ]
(5.95)
[ ]
−1
SXX ( τ, τ) − SXY ( τ, τ − 1)SYY −1 ( τ − 1, τ − 1)SYX ( τ − 1, τ)
Cˆ τ = SYY ( τ, τ − 1) − Aˆ τ SXY ( τ, τ − 1)SYY −1 ( τ, τ − 1) (5.96)

and
Bˆ τ Bˆ τ T = SYY ( τ, τ) − Aˆ τ SXX ( τ, τ) − Cˆ τ SYY ( τ − 1, τ) (5.97)
The condensed model has one equation for each season. The current season (τ)
is denoted by τ. As an example, for covariance matrices, SXY(τ,τ – 1) represents the
covariance matrix between the annual series associated with the current season (τ)
and seasonal values associated with the previous season (τ-1). See Salas et al. (1988)
for more details.
5.14 PROBLEMS
5.1 The following table shows the annual rainfall at a rain gauge. Test the
stationarity of the time series in its mean. Define a trend model and the
residuals of the series.
Precipitation Data Precipitation Data

Time Sequences (mm) Time Sequences (mm)
1 200 11 172
2 212 12 180
3 302 13 185
4 212 14 190
5 192 15 302
6 186 16 215
7 148 17 218
8 312 18 158
9 220 19 142
10 195 20 175
5.2 A series of residuals of a model fitted to precipitation data in a station in

20 time intervals is given in the following table. Use the turning point
test and comment on the randomness of the residuals.
Month Residual Month Residual
1 0.87 11 0.78
2 0.28 12 0.82
3 0.32 13 0.14
4 0.34 14 0.22
5 0.33 15 0.65
6 0.15 16 0.48
7 0.18 17 0.32
8 0.17 18 0.07
9 0.21 19 0.24
10 0.22 20 0.25
5.3 The classified monthly precipitation data measured in a station are divided
into eight categories as shown in the following table. Previous studies
have shown that the precipitation data in this station are normally distrib-
uted. Use the chi-square test with a 5% significance level and comment
on the selected distribution for the sample data.
Interval Precipitation Depth Classes Actual Frequency

(i ) (mm) (fi )
1 0≤P<5 27
2 5 ≤ P < 10 24
3 10 ≤ P < 15 9
4 15 ≤ P < 20 8
5 20 ≤ P < 25 5
6 25 ≤ P <30 4
7 30 ≤ P < 35 2
8 35≤ P < 0 1
5.4 A streamflow time series of a river could be shown in the form of the
following equation. Find the extended formulation for this model.
(1 − B12 ) Zt = (1 + 0.2 B)(1 − 0.9 B12 )ε t
5.5 Summarize the main properties of the multivariate and disaggregation

models? What is the benefit of these models in comparison to the univari-
ate model?
5.6 Write the mathematical expression of a first-order and second-order non-
seasonal ARIMA. Why is the seasonal differencing done? Write the math-
ematical expression for the second-order seasonal differencing for a
monthly time series (assume 12 for the number of seasons).
5.7 An ARMA(1,1) model is fitted to the time series of streamflow at a station
with parameters ϕ = –0.6 and θ = –0.4. Plot the autocorrelation functions
of the model from lag-1 to lag-5.
5.8 Estimate the autocorrelation and partial autocorrelation coefficients for
lag-1 to lag-5 for the streamflow data shown in the following table.
Assume that the data is normal. Estimate the parameters of AR(1) and
AR(2) models and standard deviation of the residuals for both models.
5.9 For the data of Problem 5.8, estimate the parameters of the MA(1), MA(2),
and ARMA(1,1). Use AR(1) and ARMA(1,1) models and generate 5
synthetic data.
Discharge Discharge
(m3/sec) Year (m3/sec) Year
14.6 18 59.8 1
44.2 19 63.3 2
131.0 20 57.7 3
73.3 21 64.0 4
46.6 22 63.5 5
39.4 23 38.0 6
41.2 24 74.6 7
41.3 25 13.1 8
62.0 26 37.6 9
90.0 27 67.5 10
50.6 28 21.7 11
41.9 29 37.0 12
35.0 30 93.5 13
16.5 31 58.5 14
35.3 32 63.3 15
30.0 33 74.4 16
52.6 34 34.2 17
References
Akaike, H., A New Look at the Statistical Model Identification, IEEE Trans. on Automatic
Control, AC-19, 6, 716–723, 1974.
Anderson, T. W., Distribution of the Serial Correlation Coefficients, Annals of Math. Stat.,
13(1), 1–13, 1942.
Box, G. E. P. and Cox, D. R., An analysis of transformation, J. Roy. Stat. Soc., B26, 211–252,
1964.
Box, G. E. P. and Jenkins, G. M., Time Series Analysis: Forecasting and Control, Holden-
Day, San Francisco, 1970.
Box, G. E. P. and Pierce, D. A. Distribution of the residual autocorrelations in autoregressive-
integrated moving average time series models, J. Am. Stat. Assoc., 65, 1509–1526,
1970.
Brockwell, P. J. and Davis, R. A., Time Series Theory and Methods, Springer-Verlag, New
York, 1987.
Delleur, J. W., and Karamouz, M., Uncertainty in reservoir operation, Optimal Allocation of
Water Resources, International Association of Hydrological Sciences, Publications
No. 135, Exeter, U.K., 1982.
Fiering, M. B., Multivariate techniques for synthetic hydrology, J. Hydraulic Division, ASCE,
90, HY5, 43–60, 1964.
Granger, C. W. J., and Anderson, A. P., Non-linear time series modeling, Applied Time Series
Analysis, D. F. Findley (ed.), Academic Press, New York, 1978.
Hipel, K. W. and McLeod, A. I., Advances in Box-Jenkins modeling, 1. Model construction,
Water Resources Res., 13(3), 567–575.
Lane, W. L., Applied Stochastic Techniques (LAST Computer Package) User Manual, Division
of Planning Technical Services, Bureau of Reclamation, Denver, CO, 1979.
Lapin L. L., Probability and Statistics for Modern Engineering, PWS-Kent Publishing, Bos-
ton, MA, 1990.
Ljung, G., and Box G. E. P., On a measure of lack of fit in time series models, Biometrika,
(65), 297–303, 1978.
McLeod, A. I., and Li, W. K., Diagnostic Checking of ARMA Time Series Models Using
Squared-Residual Autocorrelations, J. Time Series Anal., 4, 269–273, 1983.
Matalas, N. C., Mathematical assessment of synthetic hydrology, J. Water Resources Res.,
3(4), 937–945, 1967.
Mejia, J. M., On the Generation of Multivariate Sequences Exhibiting the Hurst Phenomenon,
Fort Collins, CO, 1971.
Mejia, J. M. and Rousselle, J., Disaggregation models in hydrology revisited, J. Water
Resources Res., 12(2), 185–186, 1976.
O’Connell, P. E., Stochastic Modeling of Long-Term Persistence in Streamflow Sequences,
Ph.D. thesis, Imperial College, University of London, 1974.
Salas, J. D. and Pegram, G. G. S., A seasonal multivariate multilag autoregressive model in
hydrology, in Modeling Hydrologic Processes, Morel-Seytoux, H., Salas, J. D., Sand-
ers, T. G., and Smith, R. E., Eds., Water Resources Publications, Fort Collins, CO,
1978.
Salas, J. D., Delleur, J. W., Yevjevich, V., and Lane, W. L., Applied Modeling of Hydrological
Time Series, Water Resources Publication, Fort Collins, CO, 1988.
Shahin, M., Van Oorschot, H. J. K., and De Lang, S. J., Statistical Analysis in Water Resources
Engineering, A. A. Balkema, Rotterdam, the Netherlands, 1993.
Snedecor, G.W. and W.G. Cochran, Statistical Methods, 6th edition, Iowa State University
Press, Ames., Iowa, 1967.
Tao, P. C. and Delleur, J. W., Seasonal and nonseasonal ARMA models in hydrology, ASCE
J. Hydraulic Division, 102, HY10, 1541–1559, 1976.
Valencia, D. and Schaake, J. C. Jr., Disaggregation process in stochastic hydrology, J. Water
Resources Res., 9(3), 580–585, 1973.
Yevjevich, V., Stochastic Process in Hydrology, Water Resources Publications, Fort Collins,
CO, 1972.
APPENDIX A
Chi-Square Distribution (The following table provides the values of χ2α that corre-
spond to a given upper-tail area α and a specified number of degrees of freedom.)
TABLE A.1
Degrees
of Upper-Tail Area α
Freedom 0.99 0.98 0.95 0.90 0.80 0.70 0.50
1 0.000157 0.000628 0.00393 0.0158 0.0642 0.148 0.455

2 0.0201 0.0404 0.103 0.211 0.446 0.713 1.386
3 0.115 0.185 0.352 0.584 1.005 1.424 2.366
4 0.297 0.429 0.711 1.064 1.649 2.195 3.357
5 0.554 0.752 1.145 1.610 2.343 3.000 4.351
6 0.872 1.134 1.635 2.204 3.070 3.828 5.348
7 1.239 1.564 2.167 2.833 3.822 4.671 6.346
8 1.646 2.032 2.733 3.490 4.594 5.527 7.344
9 2.088 2.532 3.325 4.168 5.380 6.393 8.343
10 2.558 3.059 3.940 4.865 6.179 7.267 9.342
11 3.053 3.609 4.575 5.578 6.989 8.148 10.341
12 3.571 4.178 5.226 6.304 7.807 9.034 11.340
13 4.107 4.765 5.892 7.042 8.634 9.926 12.340
14 4.660 5.368 6.571 7.790 9.467 10.821 13.339
15 5.229 5.985 7.261 8.547 10.307 11.721 14.339
16 5.812 6.614 7.962 9.312 11.152 12.624 15.338
17 6.408 7.255 8.672 10.085 12.002 13.531 16.338
18 7.015 7.906 9.390 10.865 12.857 14.440 17.338
19 7.633 8.567 10.117 11.651 13.716 15.352 18.338
20 8.260 9.237 10.851 12.443 14.578 16.266 19.337
21 8.897 9.915 11.591 13.240 15.445 17.182 20.337
22 9.542 10.600 12.338 14.041 16.314 18.101 21.337
23 10.196 11.293 13.091 14.848 17.187 19.021 22.337
24 10.856 11.992 13.848 15.659 18.062 19.943 23.337
25 11.524 12.697 14.611 16.473 18.940 20.867 24.337
26 12.189 13.409 15.379 17.292 19.820 21.792 25.336
27 12.879 14.125 16.151 18.114 20.703 22.719 26.336
28 13.565 14.847 16.928 18.939 21.588 23.647 27.336
29 14.256 15.574 17.708 19.768 22.475 24.577 28.336
30 14.953 16.306 18.493 20.599 23.364 25.508 26.336
(continued)
TABLE A.1 (CONTINUED)
Degrees
of Upper-Tail Area α
Freedom 0.30 0.20 0.10 0.05 0.02 0.01 0.001
1 1.074 1.642 2.706 3.841 5.412 6.635 10.827

2 2.408 3.219 4.605 5.991 7.824 9.210 13.815
3 3.665 4.642 6.251 7.815 9.837 11.345 16.268
4 4.878 5.989 7.779 9.488 11.668 13.277 18.465
5 6.064 7.289 9.236 11.070 13.388 15.086 20.517
6 7.231 8.558 10.645 12.592 15.033 16.812 22.457
7 8.383 9.803 12.017 14.067 16.622 18.475 24.322
8 9.524 11.030 13.362 15.507 18.168 20.090 26.125
9 10.656 12.242 14.684 16.919 19.679 21.666 27.877
10 11.781 13.442 15.987 18.307 21.161 23.209 29.588
11 12.899 14.631 17.275 19.675 22.618 24.725 31.264
12 14.011 15.812 18.549 21.026 24.054 26.217 32.909
13 15.119 16.985 19.812 22.362 25.472 27.688 34.528
14 16.222 18.151 21.064 23.685 26.873 29.141 36.123
15 17.322 19.311 22.307 24.996 28.259 30.578 37.697
16 18.418 20.465 23.542 26.296 29.633 32.000 39.252
17 19.511 21.615 24.769 27.587 30.995 33.409 40.790
18 20.601 22.760 25.989 28.869 32.346 34.805 42.312
19 21.689 23.900 27.204 30.144 33.687 36.191 43.820
20 22.775 25.038 28.412 31.410 35.020 37.566 45.315
21 23.858 26.171 29.615 32.671 36.343 38.932 46.797
22 24.939 27.301 30.813 33.924 37.659 40.289 48.268
23 26.018 28.429 32.007 35.172 38.968 41.638 49.728
24 27.096 29.553 33.196 36.415 40.270 42.980 51.179
25 28.172 30.675 34.382 37.652 41.566 44.314 52.620
26 29.246 31.795 35.563 38.885 42.856 45.642 54.052
27 30.319 32.912 36.741 40.113 44.140 46.963 55.476
28 31.391 34.027 37.916 41.337 45.419 48.278 56.893
29 32.461 35.139 39.087 42.557 46.693 49.588 58.302
30 33.530 36.250 40.256 43.773 47.962 50.892 59.703
APPENDIX B
Critical Values of D for Kolmogorov Smirnov Maximum Deviation Test for Good-
ness of Fit (The following table provides the critical values, Dα , corresponding to
an upper-tail probability, α of the test statistic D.)
TABLE B.1
n α = 0.005 α = 0.01 α = 0.025 α = 0.05 α = 0.1
1 0.99500 0.99000 0.97500 0.95000 0.90000

2 0.92929 0.90000 0.84189 0.77639 0.68377
3 0.82900 0.78456 0.70760 0.63604 0.56481
4 0.73424 0.68887 0.62394 0.56522 0.49265
5 0.66853 0.62718 0.56328 0.50945 0.44698
6 0.61661 0.57741 0.51926 0.46799 0.41037
7 0.57581 0.53844 0.48342 0.43607 0.38148
8 0.54179 0.50654 0.45427 0.40962 0.35831
9 0.51332 0.47960 0.43001 0.38746 0.33910
10 0.48893 0.45662 0.40925 0.36866 0.32260
11 0.46770 0.43670 0.39122 0.35242 0.30829
12 0.44905 0.41918 0.37543 0.33815 0.29577
13 0.43247 0.40362 0.36143 0.32549 0.28470
14 0.41762 0.38970 0.34890 0.31417 0.27481
15 0.40420 0.37713 0.33760 0.30397 0.26588
16 0.39201 0.36571 0.32733 0.29472 0.25778
17 0.38086 0.35528 0.31796 0.28627 0.25039
18 0.37062 0.34569 0.30936 0.27851 0.24360
19 0.36117 0.33685 0.30143 0.27136 0.23735
20 0.35241 0.32866 0.29408 0.26473 0.23156
21 0.34427 0.32104 0.28724 0.25858 0.22617
22 0.33666 0.31394 0.28087 0.25283 0.22115
23 0.32954 0.30728 0.27490 0.24746 0.21645
24 0.32286 0.30104 0.26931 0.24242 0.21205
25 0.31657 0.29516 0.26404 0.23768 0.20790
26 0.31064 0.28962 0.25907 0.23320 0.20399
27 0.30502 0.28438 0.25438 0.22898 0.20030
28 0.29971 0.27942 0.24993 0.22497 0.19680
29 0.29466 0.27471 0.24571 0.22117 0.19348
30 0.28987 0.27023 0.24170 0.21756 0.19032
31 0.28530 0.26596 0.23788 0.21412 0.18732
32 0.28094 0.26189 0.23424 0.21085 0.18445
33 0.27677 0.25801 0.23076 0.20771 0.18171
34 0.27279 0.25429 0.22743 0.20472 0.17909
35 0.26897 0.25073 0.22425 0.20185 0.17659
36 0.26532 0.24732 0.22119 0.19910 0.17418
37 0.26180 0.24404 0.21826 0.19646 0.17188
38 0.25843 0.24089 0.21544 0.19392 0.16966
(continued)
TABLE B.1 (CONTINUED)

n α = 0.005 α = 0.01 α = 0.025 α = 0.05 α = 0.1
39 0.25518 0.23786 0.21273 0.19148 0.16753

40 0.25205 0.23494 0.21012 0.18913 0.16547
41 0.24904 0.23213 0.20760 0.18687 0.16349
42 0.24613 0.22941 0.20517 0.18468 0.16158
43 0.24332 0.22679 0.20283 0.18257 0.15974
44 0.24060 0.22426 0.20056 0.18053 0.15796
45 0.23798 0.22181 0.19837 0.17856 0.15623
46 0.23544 0.21944 0.19625 0.17665 0.15457
47 0.23298 0.21715 0.19420 0.17481 0.15295
48 0.23059 0.21493 0.19221 0.17302 0.15139
49 0.22828 0.21277 0.19028 0.17128 0.14987
50 0.22604 0.21068 0.18841 0.16959 0.14840
51 0.22386 0.20864 0.18659 0.16796 0.14697
52 0.22174 0.20667 0.18482 0.16637 0.14558
53 0.21968 0.20475 0.18311 0.16483 0.14423
54 0.21768 0.20289 0.18144 0.16332 0.14292
55 0.21574 0.20107 0.17981 0.16186 0.14164
56 0.21384 0.19930 0.17823 0.16044 0.14040
57 0.21199 0.19758 0.17669 0.15906 0.13919
58 0.21019 0.19590 0.17519 0.15771 0.13801
59 0.20844 0.19427 0.17373 0.15639 0.13686
60 0.20673 0.19267 0.17231 0.15511 0.13573
61 0.20506 0.19112 0.17091 0.15385 0.13464
62 0.20343 0.18960 0.16956 0.15263 0.13357
63 0.20184 0.18812 0.16823 0.15144 0.13253
64 0.20029 0.18667 0.16693 0.15027 0.13151
65 0.19877 0.18525 0.16567 0.14913 0.13052
66 0.19729 0.18387 0.16443 0.14802 0.12954
67 0.19584 0.18252 0.16322 0.14693 0.12859
68 0.19442 0.18119 0.16204 0.14587 0.12766
69 0.19303 0.17990 0.16088 0.14483 0.12675
70 0.19167 0.17863 0.15975 0.14381 0.12586
71 0.19034 0.17739 0.15864 0.14281 0.12499
72 0.18903 0.17618 0.15755 0.14183 0.12413
73 0.18776 0.17498 0.15649 0.14087 0.12329
74 0.18650 0.17382 0.15544 0.13993 0.12247
75 0.18528 0.17268 0.15442 0.13901 0.12167
76 0.18408 0.17155 0.15342 0.13811 0.12088
77 0.18290 0.17045 0.15244 0.13723 0.12011
78 0.18174 0.16938 0.15147 0.13636 0.11935
79 0.18060 0.16832 0.15052 0.13551 0.11860
80 0.17949 0.16728 0.14960 0.13467 0.11787
81 0.17840 0.16626 0.14868 0.13385 0.11716
82 0.17732 0.16526 0.14779 0.13305 0.11645
(continued)
TABLE B.1 (CONTINUED)

n α = 0.005 α = 0.01 α = 0.025 α = 0.05 α = 0.1
83 0.17627 0.16428 0.14691 0.13226 0.11576

84 0.17523 0.16331 0.14605 0.13148 0.11508
85 0.17421 0.16236 0.14520 0.13072 0.11442
86 0.17321 0.16143 0.14437 0.12997 0.11376
87 0.17223 0.16051 0.14355 0.12923 0.11311
88 0.17126 0.15961 0.14274 0.12850 0.11248
89 0.17031 0.15873 0.14195 0.12779 0.11186
90 0.16938 0.15786 0.14117 0.12709 0.11125
91 0.16846 0.15700 0.14040 0.12640 0.11064
92 0.16755 0.15616 0.13965 0.12572 0.11005
93 0.16666 0.15533 0.13891 0.12506 0.10947
94 0.16579 0.15451 0.13818 0.12440 0.10889
95 0.16493 0.15371 0.13746 0.12375 0.10833
96 0.16408 0.15291 0.13675 0.12312 0.10777
97 0.16324 0.15214 0.13606 0.12249 0.10722
98 0.16242 0.15137 0.13537 0.12187 0.10668
99 0.16161 0.15061 0.13469 0.12126 0.10615
100 0.16081 0.14987 0.13403 0.12067 0.10563
6 River Basin Modeling
6.1 INTRODUCTION
This chapter discusses elements of river basin modeling, including data and infor-
mation processing, objectives, and constraints identification, and details of modeling
process. The objectives of river basin modeling that are discussed include water
quality and quantity management and flood control. Also presented in this chapter
are managerial tools, such as simulation and optimization models, and a number of
computer packages that have been widely used in various water resources develop-
ment and management projects.
6.2 WATERSHED: DEFINITION AND

CHARACTERISTICS
The concept of watershed is the foundation of all hydrologic planning and design.
An understanding of watersheds — formed by rivers and the land area they drain
— is basic for understanding the major sources of freshwater all over the world.
Sound water resources management must be undertaken at a whole-watershed level,
rather than just at a local level. As shown in Figure 6.1, many major watersheds
cross national boundaries (World Resources Institute, 1998).
A watershed is also referred to as a catchment basin, which can be regarded as
a system as it has a well-defined boundary and its elements show clear relationships
both structurally (in terms of morphology) and functionally (by virtue of the flow
of matter and energy). The inputs and outputs across the system boundary can also
be clearly distinguished (Figure 6.2). The areal limits of the river system are set by
a major watershed, while minor watersheds define subcatchments within it
(Figure 6.3). The boundary of a watershed is defined by all points that will shed
water to the outlet; it is only necessary to determine which points in a region
contribute water to the outlet, and the most extreme points represent the watershed
boundary. In the next sections some characteristics of a watershed are defined.
6.3 WATERSHED GEOMORPHOLOGY

6.3.1 DRAINAGE AREA
The drainage area is an important watershed characteristic for hydrologic planning
and design because it contains the volume of water generated from rainfall. Deter-
mining the drainage area requires delineation of the watershed boundary. The Geo-
graphic Information System (GIS) is a powerful tool for this purpose.
191
192
Europe
12 - Danube
13 - Dnieper
14 - Dniester
15 - Don
16 - Ebro
41 17 - Elbe 19 21 57
36 18 - Garonne 70 59
21 30 64
35 19 - Glama
38 26 17 29 13
20 - Guadalquivir 22 15 28 50
North and Central America 34 40 21 - Kemijoki 18 12
32 - Brazos 14
37 22 - Loire 27 16 29 23
33 - Colorado 23 - Po 49
33 20 67 54
34 - Columbia 32 24 - Rhine-Maas
39 55
35 - Fraser 25 - Rhone 69
36 - Mackenzie 52 51
26 - Seine 56 68
37 - Mississippi 27 - Tagus 63 60
38 - Nelson 53
66 58
28 - Ural 9 5
39 - Rio Grande
29 - Vistula 65 61
10 6
40 - St. Lawrence 1
43 44 30 - Volga
41 - Yukon 3
31 - Weser
Africa 2
42 1 - Chari Asia and Oceania
47 46 2 - Congo 49 - Amu Darya 60 - Mahanadi

3 - Juba-Shibeli 11 50 - Amur 61 - Mekong
South America 4 - Limpopo 51 - Brahmaputra 62 - Murray
42 - Amazon 45 7 4
5 - Niger 52 - Ganges 63 - Narmada
43 - Magdalena 48 6 - Nile 8 53 - Godavan 64 - Ob
44 - Orinoco 7 - Okavango 54 - Hwang Ho 65 - Salween
45 - Paraná 8 - Orange 55 - Indus 66 - Tapti 62
46 - Sáo Francisco 9 - Senegal 56 - Irrawaddy 67 - Tigris & Euphrates
47 - Tocantins 10 - Volta 57 - Kolyma 68 - Xun Jiang
48 - Uruguay 11 - Zambezi 58 - Krishna 69 - Yantze
59 - Lena 70 - Yenisey
FIGURE 6.1 Major watersheds across national boundaries. (From World Resources Institute, World Resources: A Guide to the Global Environment,
a joint publication by The World Resources Institute, The United Nations Environment Programme, The United Nations Development Programme, and
The World Bank, Oxford University Press, London, 1998.)
River Basin Modeling 193
Evaporation Evapotranspiration
Precipitation
Ov
er
flowland
CHANNE Ov
L er
Infiltration flowland
Perc ble
olation Water ta
flow
ater
GROUNDWATER ZONE r o u ndw
G Channel runoff
FIGURE 6.2 Catchment basin system.
6.3.2 WATERSHED LENGTH

Watershed length is defined as the distance measured from the watershed outlet to
the farthest point on the basin divide. The length is usually a measure of the travel
time of water through the watershed.
6.3.3 WATERSHED SLOPE

Watershed slope is an important factor in the momentum of runoff and flood mag-
nitude. It is the rate of change of elevation with respect to distance along the principal
flow path. According to the simple classification scheme of White et al. (1992),
based on the surface gradient and slope plan form, any catchment basin consists of
no more than five fundamental types of slopes. White et al. (1992) classified these
slopes into two groups: flat (or nearly so) and valley (Figure 6.4). Where it is required
to subdivide the watershed, it is necessary to compute the slopes of each subarea.
For this purpose, the principal flow path for that subarea must be delineated. The
path is divided into reaches, and the river slope in different reaches may be required
for computing the reach travel time.
6.3.4 WATERSHED SHAPE

Watersheds have an infinite variety of shapes, and parameters that reflect the basin
shape are used in hydrologic planning and design. Some typical parameters that
have been developed to reflect watershed shape are as follows:
FIGURE 6.3 Delimitation of catchment watershed.
FIGURE 6.4 Morphological components of the catchment basin.
• Length to the center of area (Lca) is the distance (measured in kilometers)

along the main channel between the basin outlet and the point on the main
channel opposite the center of the area.
• Shape factor (Ll) is estimated by the following equation:
Ll = ( L ⋅ Lca / 2.58)
0.3
(6.1)
where L is the watershed length in kilometers. Ll shows the elongation of

a watershed.
• Circularity ratio (Rc) is estimated by the following equation:
Rc = A A (6.2)
0
where A is area of watershed, and A0 is the area of a circle with perimeter

equal to the perimeter of the basin.
6.4 SOIL CHARACTERISTICS

The three soil phase systems are solid, liquid, and gas. The liquid and gas phases
are represented by the soil water and air, respectively, that occupy the pores and
voids. However solid, liquid, and gas phases do not form a random mosaic in these
three dimensions but are more or less organized to impart a definite vertical and
lateral structure to the system. The properties of the soil solids and of the voids and
their associated soil water and soil air vary vertically and laterally. The vertical
component of the soil assembly is referred to as the soil profile and the soil layers
as its horizon. Soil horizons are referred to as follows (McCuen, 1998):
O horizon — Surface litter consisting primarily of organic matter

A horizon — Top soil consisting of humus and inorganic minerals
E horizon — Leaching zone where percolating water dissolves water-soluble
matter
B horizon — Subsoil below the A or E horizon that contains minerals and
humic compounds
C horizon — A zone consisting primarily of under-composed mineral parti-
cles and rock fragments
R horizon — An impermeable layer of bedrock
6.4.1 SOIL TEXTURE

Soil texture refers to the size of mineral particles in different size categories. On the
base of mean diameter (d), particles of soil can be separated into four classes:
1. Gravel, d ≤ 2 mm
2. Sand, 0.02 ≤ d ≤ 2 mm
3. Silt, 0.002 ≤ d ≤ 0.02 mm
4. Clay, d ≤ 0.002 mm
Water-holding characteristics and the infiltration capacity of the soil can be deter-
mined by the soil texture. The capacity of soil to pass and store infiltrating water
increases with the size of the soil particles as a result of the increase in pore space.
6.4.2 SOIL STRUCTURE

The kind, size, and distribution of soil aggregates and soil voids and pores determine
the structure or fabric of the soil. Soil structure refers to the tendency of the soil
particles to aggregate into lump and clods. The hydrologic response of the watershed,
soil moisture, and water movement through the column of soil can be affected by
the structure of the soil.
6.4.3 SOIL MOISTURE

The water in the unsaturated zone of soil is referred to as soil moisture and is acted
upon by gravitation, surface tension, and molecular forces. Soil moisture is held as
capillary water in the smaller pore spaces of the soil or as hygroscopic water
absorbed on the surface of soil particles. Capillary water can be removed from soil
by applying forces sufficient to overcome the capillary forces. Field capacity is the
moisture content of the soil after drainage. Drainage of water will continue for some
time after a soil is saturated, so field capacity must be defined in terms of a specific
drainage period. The permanent wilting point is a moisture content at which plants
can no longer extract sufficient water from soil for growth. Water below this level
of soil moisture is referred to as unavailable water, and the only moisture presented
in the soil is the thin film or layer of moisture attached to the soil particles, which
is referred to as hygroscopic moisture. Available water is the moisture content of
the soil that exists between field capacity and the permanent wilting point. An
efficient irrigation procedure is to apply water when the moisture content of the soil
approaches the wilting point in an amount sufficient to raise the soil moisture to the
field capacity within the root zone.
6.5 DATA REQUIREMENTS

6.5.1 HYDROLOGIC AND CLIMATOLOGIC DATA
Water resources planning and management requires basin climatic and hydrologic
data. The movement of water through air, sea, lakes, rivers, land, soil, glaciers, and
living organisms forms the hydrologic cycle. The atmospheric part of the cycle
begins when evaporation occurs off of open water surfaces, as well as off of ice,
plants, soil, and all other surfaces recently wetted by precipitation.
A portion of rainfall is intercepted by the leaves and stems of vegetation.
Interception, along with both depression storage and soil moisture, constitutes basin
recharge — that portion of precipitation that does not contribute to surface runoff
or groundwater. Depression storage includes the water retained as puddles in surface
depressions. Some of the rainwater or melted snow overflows as surface runoff and
overland flow. Other water may infiltrate into the soil and flow laterally in the soil
surface to a stream channel as interflow. Overland flow and interflow are usually
grouped together as direct runoff.
Data gathered from gauging stations can be used to determine stream flows and
estimate interflows. Infiltration is the movement of water through the soil surface
and into the soil. The infiltration capacity of a soil at any time is the maximum rate
at which water will enter the soil. The infiltration rate is the rate at which water
actually enters the soil during a storm. Soil texture and structure, moisture content,
land cover and use, and rainfall intensity are the types of data that can be used to
determine water losses due to infiltration.
Climate interacts with soil, rock, plants, animals, surface water, and ice. Within
the hydrologic cycle, water is stored at a number of locations. The major part of it
is stored in the atmosphere, land, oceans, and polar ice caps. Transfers between these
locations are carried out by the processes of evaporation, condensation, precipitation,
runoff, freezing, and melting. The greatest exchanges are those between oceans and
atmosphere. Evaporation is the process by which liquid water is changed into its
gaseous phase (water vapor). The most important climatic data necessary to estimate
the rate of evaporation are the heat energy supply, humidity of the air, characteristics
of the airflow across the surface, and nature of the evaporating surface.
Vegetation provides an extra pathway for water to travel from the ground surface
into the atmosphere. Water vapor is diffused through pores on the leaf surface and is
transferred into the atmosphere. This transfer of water is referred as transpiration
and accounts for a large proportion of moisture losses from vegetated surfaces. The
rate of transpiration loss is governed by two types of factors: extrinsic factors affecting
the rate of evaporation from the water surface and the factors intrinsic to the plant.
For example, transpiration loss from most plants will be large when the atmosphere
is relatively dry. In most plants, leaf stomata open under light conditions and close
at night, thereby introducing a clear diurnal variation in transpiration losses. There-
fore, a significant amount of climatic and hydrologic data are required for estimating
of transpiration losses, the most important of which are temperature, available water,
land cover, wind condition, and geographic latitude. In modeling river basins, a correct
estimate of the water budget and the spatiotemporal distribution of data is a key factor.
Data accuracy and considering the physical and hydrological characteristics of the
watershed is essential to integrated water resources planning of that system.
6.5.2 PHYSICAL CHARACTERISTICS

Watershed and river geomorphology is a basic piece of data necessary for river basin
planning and management. These characteristics are important for determination of
the volume of water that can be generated from precipitation. As discussed previ-
ously, drainage area, watershed length, watershed slope, watershed shape, surface
roughness, soil characteristics, and land cover of the river basin are the most impor-
tant physical characteristics of watershed that should be estimated before undertaking
the river basin planning process.
In addition to watershed characteristics, river geomorphology is frequently used
in hydrologic computations. The major physical characteristics of rivers or watershed
channels are:
1. Length
2. Slope
3. Cross-section
4. Roughness
5. Form
River length and slope can be described by several computational schemes; the
most common are described here.
The river (channel) length is the distance between the river basin outlet and the
end of the river along the main channel (Lc). The river (channel) slope is the
difference in elevation between the upper and lower elevation of river (ΔEc ) over
the length (Lc ).
ΔEc
Sc = (6.3)
Lc
(a) (b) meander scar

erosion
direction
of
migration
point bar main axis
deposition of flow
in channel
FIGURE 6.5 (a) Braided channel, and (b) geomorphological features of a meandering
channel.
In most cases, the river slope is not uniform and a weighted slope can be used that
provides better results:
2
Si = ⎛ ⎞
n
(6.4)
⎝ k⎠
where n is the number of river or channel segments and k is given by the following
equation:
∑ (ΔE
1
k= (6.5)
Li )
0.5
i =1 i
where ΔEi is the difference in elevation between the points defining the upper and
lower segment i of the river, and Li is the length of segment i. The river slope over
each segment is considered relatively constant.
Cross-section information, including cross-section area, wetted perimeters, slope
average velocity, and roughness, is usually necessary for hydrologic analysis, plan-
ning, and management.
Channel roughness affects the hydraulic characteristics of river flow and is
required for a number of hydraulic and hydrologic computations. Manning’s rough-
ness coefficient is used in a number of methods for estimation of water elevation,
velocity of stream flow, and travel time of runoff, thus channel roughness should be
estimated as an important input.
The channel form is another important physical characteristic of rivers. The river
discharge can increase in the downstream direction as the watershed area increases,
thus the channel form changes, becoming deeper and wider. A modeled channel
form can demonstrate how the river adjusts to input of sediment and discharge.
Channel forms can be classified as straight, sinuous, meandering, and braided chan-
nels (see Figure 6.5) (White et al., 1992).
6.5.3 HYDROLOGIC CHARACTERISTICS

Hydrologic characteristics of watersheds reflect the volume of discharge hydrograph
produced by a specific rainfall hyetograph. The following parameters, which are
mainly used in rainfall-runoff simulation models (such as HEC-1), show the timing
of runoff and the shape and the volume of the streamflow hydrograph:
• Time of concentration (tc): Time for a wave to propagate from the most
distant point in the watershed to the outlet
• Lag time (tp): Time from the center of mass of rainfall excess to the peak
of the hydrograph that can be estimated using following methods: (1)
Snyder’s method (Snyder, 1938):
t p = 0.74 Ct ( L Lc )
0.3
(6.6)
where tp is the lag time (hr); L is the length of the main stream from the
outlet to the divide (km); Lc is a watershed shape parameter, which is the
length measured along the main channel from the watershed outlet to a
point on the main channel that is perpendicular to the center of the area
of the watershed (km); and Ct is a watershed storage coefficient, usually
ranging from 1.8 to 2.2. (2) Linsley’s method (Linsley et al., 1982):
0.33
⎛ L Lc ⎞
t p = 0.72 Cl ⎜ ⎟ (6.7)
⎝ S ⎠
where tp is the lag time (hr); L is the length of main stream from the outlet
to the divide (km); Lc is a watershed shape parameter, which is the length
measured along the main channel from the watershed outlet to a point
on the main channel that is perpendicular to the center of the area of the
watershed (km); S is the basin slope; and Cl is a watershed storage
coefficient, usually ranging from 1.2 for mountainous basins to 0.35 for
valleys.
• Storage coefficient (R): Shows the storage capacity of the drainage system
of the watershed. Linsley et al. (1982) developed the following equation
for estimating storage coefficient (R):
⎛ L A⎞
R = 0.49 Cr ⎜ ⎟ (6.8)
⎝ S ⎠
where A is the watershed area, and Cr is an empirical coefficient that

depends on climatic and physical characteristics of the watershed, usually
varying from 6 to 8%.
A correct estimate of these parameters will insure the effectiveness of flood warning
systems as a major component of an overall watershed management system.
Example 6.1
The area of a basin is estimated to be 500 km2. If L = 25 km and Lc = 10 km.
Estimate the lag-time (tp ) using the Snyder and Linsley methods. Assume that
Ct = 1.8, S = 0.02, Cl = 0.8, and Cr = 0.07.
Solution: Using Eqs. (6.6) and (6.7), it can be written that:
tP = 0.74 × 1.8(25 × 10)0.3 = 6.98 hr

0.33
⎛ 25 × 10 ⎞
t p = 0.72 × 0.8⎜ ⎟ = 6.79 hr
⎝ 0.02 ⎠
The lag time values that are estimated by the two methods are close.
Application of the above example could be used to determine the amount of
lead time needed for any flood control measures.
6.5.4 INFLOW AND OUTFLOW

If precipitation exceeds the infiltration and evaporation rates, water will begin to
collect on the surface. If rainfall continues, overland flow will occur on the surface,
and overland flows from different places begin to run together. Water soon collects
into small channels that lead to streams. Other parts of the streamflow may come
from direct precipitation and interactions between ground and surface water.
The river inputs have spatial and temporal variations. The river flow variation
in time is recorded by hydrographs at various locations, and the shape and magnitude
of the hydrographs are related to many factors associated with the rainfall and river
basin characteristics. The storm hydrograph can be separated into direct runoff and
base flow. The seepage from groundwater contribution to the base flow can also be
determined. The river discharge and elevation vary over time at any section and also
in the downstream direction. The flow hydrograph at any point of the river can be
determined using a flow routing procedure from one known river hydrograph at one
upstream point. Hydrologic routing relates input I(t), output O(t), and river storage
S(t) by the continuity equation:
dS(t )
= I (t ) − O(t ) (6.9)
dt
When storage S(t) is related to input and output by the storage function, the river
length should be divided to some extent based on spatial distribution of tributaries,
withdrawals, groundwater–river interactions, and physiologic characteristics of the
river. The outflow of each reach can be determined from the known input hydrograph,
groundwater seepage, and drainage using a flow routing method.
6.5.5 DEMAND POINTS

Determination of demand points and the time series of demand at each point is an
important step in river basin planning. Water allocation models require demand
quantity, along with demand point characteristics such as distance from resources,
elevation, and topographic condition. Demand point characteristics can affect allo-
cated water costs and the required equipment and means for water transfer. Other
information, such as water quality and the physiologic condition of the demand
point, is also needed.
6.5.6 CONTROL POINTS

From a hydraulic standpoint, a control section or control point is a point in the river
at which the flow changes from subcritical to supercritical, passing through the
critical depth. In water resources planning, however, a control point is a section that
experiences a considerable change in discharge or a point of the river that has a
withdrawal tributary and a monitoring or gauging station.
6.5.7 CRITICAL POINTS

Critical points require more attention in river basin planning and management.
Important structures, protected areas, and demand points that must have water of a
known quantity and/or quality at an acceptable reliability level are some examples
of critical points. Critical points must be determined before the planning phase
because they could have considerable effects on planning alternatives such as for
flood control or upstream reservoir operation.
6.5.8 WATER QUALITY REQUIREMENTS FOR VARIOUS USES

Qualitative characteristics of water are important for water resources planning and
management. Water quality is evaluated in terms of its physical, chemical, and
microbiological properties. Quantitative measurement of these characteristics is nec-
essary for determination of water quality. Water quality requirements are determined
based on the intended use of the water; for example, water contaminated by chem-
icals might be harmful for agricultural use but could be suitable for some other uses.
Therefore, the water quality requirements for each type of water use should be
determined, along with assessment of raw water quality and selection of suitable
water treatment processes, if necessary. Water quality is usually evaluated by the
degree to which physical, chemical, and biological standards are met for each water
use. For example, most standards control the turbidity, odor, color, taste, and micro-
biological characteristics of drinking water. More information about water quality
standards and qualitative management of water resources is presented in Chapter 9.
6.5.9 RETURN FLOWS AND DRAINAGE

During or after the precipitation, water entering the ground may be stored temporarily
in the upper layer of soil; it may flow laterally above the groundwater and reach a
river or other surface waters, or it may infiltrate and reach the groundwater. Excess
agricultural water use and water that infiltrates from municipal drainage wells can
reach surface water as return flows. Sometimes groundwater partially or completely
recharges the surface waters resources. For example, between rainstorms, ground-
water discharges to streams. During storms, the stream level rises, and water drains
from the stream to the groundwater, raising the groundwater table. Therefore,
groundwater level and river stages have a dynamic relationship and are generally
related. Installation of piezometers or monitoring wells and the design of a network

for groundwater table monitoring can provide information that is necessary to deter-
mine groundwater and river interaction.
6.5.10 SAMPLING NETWORK

Water quantity and quality monitoring is essential to obtain information on the
quantity and physical, chemical, and biological characteristics of water resources.
The type of information to be gathered is determined based on the monitoring
objectives and on the statistical water sampling procedures and guidelines. Figure
6.6 demonstrates the flow of information and functional summary of water quality
monitoring systems.
One of the most critical factors in a monitoring network is the location of the
monitoring and sampling stations. Data on water quantity, water quality, and ground-
water depth are collected. The factors that influence the selection of monitoring
and sampling stations are different and are determined based on the objectives and
constraints of the monitoring system. For example, in a groundwater monitoring
network, cost factors may lead to the decision to incorporate preexisting wells into
the network. In addition to site or variable selection, sampling frequency is an
important aspect of network design. A high percentage of monitoring network
operation costs are related to sampling frequency. In statistical assessment of qual-
itative and quantitative variables, professional judgment and cost constraints are
often considered as the basis for sampling frequencies. After network design,
operating procedures need to be specified before sampling or data collection. The
procedure includes sample collection, sample and sampling preservation methods,
chain of custody, laboratory analysis methods, quality control, and data verification
certification guidelines.
6.6 OBJECTIVES
6.6.1 WATER SUPPLY
Supplying water of the required quantity and quality is one of the most important
objectives of water resources planning. Determination of current water use and
prediction of future water demands are necessary for water supply planning and
management. Domestic or municipal, industrial, and agricultural uses represent the
major demands for water resources. Hydropower generation, recreation, and fishing
are other types of water use that do not require direct withdrawal from the resources
and are called instream uses. Each of these uses has different water quality require-
ments. Surface and groundwater are two main sources of water supply; in this
chapter, the focus is on surface water.
Rivers and lakes are important resources for public water supplies but surface
waters are being polluted by industrial and municipal wastes, agricultural return
flow, and polluted runoffs. Thus, an expensive treatment process is usually necessary
before this water could be consumed.
WATER QUALITY MONITORING OBJECTIVES
NETWORK DESIGN
• Station location
• Variable Selection
• Sampling frequency
DATA ANALYSIS
SAMPLE COLLECTION • Basic Summary Statistics
• Sampling Technique • Regression Analysis
• Field Measurement • Water Quality Indices
• Sample Preservation • Quality Control Interpretation
• Sampling Point • Time series Analysis
• Sample Transport • Water Quality Methods
LABORATORY ANALYSIS
• Analysis Technique/Chain of INFORMATION UTILIZATION
Custody • Information Needs
• Operational Procedures • Reporting format
• Quality Control • Water Quality Indices
• Data Recording • Operational Procedures
• Utilization Evaluation
DATA HANDLING
• Data Reception
➢ Laboratory
➢ Outside Sources
• Screening and Verification
• Storage and Retrieval
• Reporting
• Dissemination
FIGURE 6.6 Flow of information and functional summary of monitoring system activities.
(From Sanders, T. G. et al., Design of Networks for Monitoring Water Quality, Water Resources
Publications, Littleton, CO, 1983. With permission.)
Wastewater reuse is an important alternative for water supply. Wastewater reuse

may be more attractive in arid areas where water is scarce and water losses are high.
Municipal wastewater is the best resource for water reuse in terms of its quantity
and low variability. Advanced wastewater treatment is required for direct municipal
reuse. In indirect wastewater reuse, wastewater usually is discharged to the ground-
water aquifer and then is withdrawn for potable or other municipal uses. Wastewater
recycling and reuse in industries provides savings in energy and costs. Return flows
from irrigation are a reusable source for agricultural or industrial uses. When reused
water is allocated for agricultural uses, suitable treatment is required because it is

highly contaminated with salt and chemical matters from the use of fertilizers,
pesticides, and herbicides. More details about water reuse are presented in
Chapter 11.
6.6.2 FLOOD CONTROL

A flood or flooding has been defined by the Water Resources Council as a general
or temporary condition of partial or complete inundation of normally dry areas by
the overflow of inland or tidal waters and/or the unusually rapid accumulation of
surface waters (Viessman and Hammer, 1985). During a flood, the overflows from
river or channel banks cover the floodplain with lower velocity than in the main
river, thus the floodplain can store a high quantity of water and can act as a wider
waterway. Floods cause loss of life and property within the river floodplain as well
as added costs due to cleanup of the flooded lands.
Flood control and flood damage reduction are important objectives of river basin
planning. Flood control and floodplain management require hydrologic and hydraulic
analysis of floods. This analysis determines inundated areas, flood elevation, and
characteristics of required hydraulic structures for flood control or flood damage
reduction. The typical requirements for floodplain analysis and planning include
(Hoggan, 1997; Mays et al., 1996):
1. Floodplain information studies collect and analyze information regarding

specific flood events for various return periods. Floodplain information
studies require some hydraulic, morphological, and hydrological data.
Satellite information from floodplain feathers can be used for this purpose.
As an example, data requirements for flood monitoring in basins larger
than 1000 km2 are summarized in Table 6.1.
2. Evaluation of future land use alternatives identifies the potential for flood
damage and the environmental impacts of flood events at different
frequencies.
3. Evaluation of flood reduction measures examines floods with different
frequencies to determine expected flood damage reduction associated with
specific design flows.
4. Design studies determine the scope of structural measures for specific
flood events.
5. Operation studies develop optimal operating policies for flood control
facilities.
Simple steady- or unsteady-state models can be used to determine water surface

elevation and streamflow hydrographs at various locations along the river. Water
surface elevation and streamflow discharge can be used for potential flood damage
estimation. Identification of an alternative or a combination of alternatives for flood
damage reduction is the main objective of flood control planning. The best plan can
be selected based on its net benefit. The increase in the net income of additional
floodplain development and existing floodplain activities is referred to as location
TABLE 6.1
Data Requirements for Flood Studies in Basins Larger than 1000 km2
Hydrological Element Accuracy Resolution (m) Frequency
Floodplain area/boundary ±5% area 10 5 yearsa

Flood extent ±5% 100 <= 4 daysb
Precipitation ±2 mm if <40 mm 5000 6 hours
±5% if >40 mm
Saturated soil area ±5% 100 <= 4 daysb
Soil moisture profile ±10% field capacity 1000 1 day
Snow-covered area/snow line ±5% area 1000 1 day
Snowpack water equivalent ±2 mm if <2 cm 1000 1 day
±10% if >2 cm
Snowpack free water content ±2 mm if <2 cm 1000 1 day
±10% if >2 cm
Snow surface temperature ±1ºC 1000 6 hours
a And after each major flood event.
b Depends on the degree of flooding; with major floods, daily monitoring would be desirable.
Source: Barrett, E. C. and Curtis, L. F., Introduction to Environmental Remote Sensing, Chapman
& Hall, New York, 1992.
benefit and intensification benefit, respectively. This is computed as the sum of the
location benefits and intensification benefits and the flood-inundation-reduction ben-
efit, less the total cost of implementing, operating, maintaining, repairing, replacing,
and rehabilitating the system (Mays, 1996).
6.6.2.1 Flood Control Alternatives
Reduction of flood damage is the main objective of flood control planning. Flood
control alternatives can be classified into two groups: structural and nonstructural.
Structural alternatives represent traditional flood damage reduction by physical
means. In other words, the construction of flood control facilities can be referred as
structural measures. Important measures for structural reduction of flood damage
include:
• Dams and reservoirs. Flood control dams may be constructed across the
river to store floodwaters and to reduce the magnitude of the flood and
the downstream stage of the flood. The stored floodwater can be allocated
to different water demands. Flood control reservoirs also can change the
hydraulic characteristics and flow regime in downstream of the reservoir.
• Levees and floodwalls. Levees and floodwalls are the oldest and commonly
used methods of protection against floods. Levees or dikes are constructed
parallel to rivers to prevent overflow of floodwater to the floodplain.
Floodwalls are usually constructed from concrete and perform the same
function as levees. Levees and floodwalls can increase the peak discharge
of floodwater downstream because natural storage in the floodplain is

decreased.
• Flood diversion. A flood diversion structure or bypass directs excess
floodwater from areas with a high potential for flood damage into an area
or channel that can carry the flood discharge.
• Channel modification. Improvement of the hydraulic capacity of a channel
can lower the water stage and increase the carrying capacity. Channel
improvement lowers the stage of water for different discharges but reduces
natural floodplain storage and results in a higher discharge of flood peak.
Straightening of bends or meanders, lining, widening, and deepening are
the most important methods of channel modification.
• Detention measures. Detention basins are small impoundments with
uncontrolled release that can be created by excavations, small dams, or
walls. Detention measures store floodwater in the river basin before reach-
ing the main channel. Recharging the groundwater and improving the
water quality by reducing sedimentation are among other advantages of
onsite detention measures.
Although structural measures have been effective in flood damage reduction, in

a comprehensive floodplain plan, both structural and nonstructural alternatives
should be evaluated. In a nonstructural approach, flood damage is reduced by
developing operational policies for river basin management. Some nonstructural
flood damage reduction measures are as follows:
• Reservoir operation. In multipurpose reservoirs, temporary storage may

be allocated for flood flow or reduction of outflow discharge. Development
of suitable operating policies for dam gates and spillways that consider
the actual inflow hydrograph and carrying capacity of the downstream
river is important for nonstructural flood damage reduction. More details
about reservoir operation are provided later.
• Watershed improvement. Improving infiltration and groundwater recharge
and reducing erosion are the main objectives of watershed modification
for flood damage reduction. Improvement of vegetation cover, land use,
and terracing management can increase the time of concentration for
watersheds and reduce surface soil erosion by reduction of overland flow
velocity. Although watershed improvement is a long-term and expensive
plan, it can be one of the most effective techniques for protecting of water
and soil resources.
• Floodplain regulation. The development of land use regulations within
main channels and floodplain areas is referred to as floodplain regulation
or management. It may consist of many administrative actions that regulate
floodplain land use based on potential flood damage. In most regions,
government or local agencies restrict floodplain development by floodplain
zoning. For example, in many countries, all property owners located within
a 100-year floodplain must purchase flood insurance, and the lowest floor
or basement of the buildings must be built above the 100-year flood level.
• Flood forecast and flood warning systems. Real-time flood forecasting is

one of the most effective methods for flood damage reduction. Determi-
nation of climatic, hydrologic, and hydraulic and morphological charac-
teristics of the river basin is the first step in flood forecasting. Satellite
remote sensing is beginning to contribute useful information with respect
to soil moisture, runoff from snowmelt, land cover/use, and rainfall that
can be used in flood forecasting and flood warning systems. A flood
warning system can provide an opportunity to act before flood occurs.
Thus, it can reduce flood damage costs and health hazards. These systems
are useful for drainage areas having a concentration time sufficient for
controlling floodwater or reducing flood damage. Flood warning prepared-
ness systems send rainfall and runoff information to processing centers
when critical thresholds have been reached. In the processing center,
downstream discharge and water elevation are calculated, and the time of
occurrence and magnitude of downstream flow are estimated and reported.
Flood warning preparedness plans, then, can be considered to be a critical
and important component of other flood control measures.
• Flood proofing. Flood proofing includes all actions by individuals or small
groups to modify their sites, structures, and facilities to reduce flood
damage. Raising structures, using materials or methods that are less sus-
ceptible to flood damage, preventing water from entering structures, and
building local levees to restrict and direct floodwater are examples of flood
proofing. This approach is an inexpensive method that can be used together
with other flood control measures to reduce the risk of flood damage.
6.7 MANAGEMENT TOOLS

6.7.1 STREAMFLOW AND EXCESS WATER ESTIMATION
Surface water flow is an important element of water resources planning and design
studies for water supply, streamflow forecasting, flood control, irrigation, and res-
ervoir design. Most design problems are caused by designs based on streamflow
hydrographs that consider only peak discharge rates. Streamflow and rating curves
that correlate discharge with flow rate can be determined in the field using stream
velocity measurements and channel geometry. The channel may contain a certain
amount of base flow even in the absence of rainfall. Base flow occurs due to water
stored in watersheds from past storm events and snowpacks and groundwater
contribution and appears as streamflow. Direct runoff from rainfall excess and base
flow make up the total hydrograph. The unit hydrograph transforms the rainfall
excess into direct runoff.
6.7.1.1 Streamflow Estimation in Homogenous Basins
The selection of appropriate methods for streamflow estimation is based on the

characteristics of the river basin. In a drainage area nearly homogenous, the stream-
flow at each point of the watershed main channel is related to the watershed area
above that point. Therefore, the streamflow at site s can be estimated based on
measured flow at gagged point s′ (Loucks et al., 1981):
⎛ As ⎞
Qts = Qts′ ⋅ ⎜ s′ ⎟ (6.10)
⎝A ⎠
where As and As′ are the watershed area above the s and s′ sites. Qts and Qts′ are the
streamflow at sites s and s′ in period t, respectively. Equation (6.10) is not applicable
when the change in spatial distribution of rainfall and runoff is significant. This
method provides an estimate of the annual average flow at ungauged site s, but
seasonal or monthly streamflow can be derived using the ratio of the flow of each
period to the annual flow.
When the records of two gauging stations are available, preferably one upstream
and one downstream of the ungauged site, streamflow at the ungauged site can be
estimated on the basis of the drainage area as follows (Gupta, 1989):
Qts Qta ⎛ Q at Q bt ⎞ ⎛ A s − A a ⎞
= + − ⎜ ⎟ (6.11)
A s A a ⎜⎝ A a A b ⎟⎠ ⎝ A b − A a ⎠
where Qta , Qtb is the streamflow at gauged sites a and b in period t; Qts is the
streamflow at ungauged site s; and Q at , Q bt is the average of the entire record at
gauged sites a and b. The streamflow at an ungauged site can be estimated on the
basis of the distance between the sites as:
Qts Qta ⎛ Qat Qbt ⎞ Ls

= +⎜ − ⎟⋅ (6.12)
As Aa ⎝ Aa Ab ⎠ Lb
where Ls is the distance between stations s and a, and Lb is the distance between
stations b and a. When more than one appropriate gauged site is available, a weighted
average at the gauged site can provide an acceptable estimation of flows of ungauged
sites. The weights are selected based on relative physical and hydrologic similarities
between site s and other gauged sites, such as the relative physical and hydrologic
similarities between site s and other gauged sites; for example, the relative distance
between the ungauged point and gauged sites can be used in estimation of relative
weights.
Example 6.2
The average monthly streamflow at two gauged stations A and B on a river are
measured as 4 and 8.4 m3/s, respectively. The drainage areas at the gauging stations
A and B are 48 and 130 km2, respectively, and the distance between them is 35 km.
Consider an ungauged point that is located 10 km downstream of station A and has
a drainage area of 75 km2. The streamflows in September at stations A and B are 3
and 7 m3/s, respectively. Estimate the streamflow in September at ungauged point.
Solution: For estimating the streamflow in September at the ungauged station,

Eq. (6.10) is used and the streamflow at station A is determined as follows:
3
⎛ 75 ⎞ = 4.68 m
. = 3×⎝
s
QSep
48 ⎠ s
Using Eq. (6.11), we obtain:
s
QSep 4 ⎛ 4 8.4 ⎞ 75 − 48
= + − ×
.
75 48 ⎝ 48 130 ⎠ 130 − 48
m3
. = 6.71
s
QSep
s
and from Eq. (6.12) we have:
s
QSep 4 ⎛ 4 8.4 ⎞ 10
= + − ×
.
75 48 ⎝ 48 130 ⎠ 35
m3
. = 6.65
s
QSep
s
6.7.1.2 Unit Hydrograph As a Tool
Where the drainage basin is not homogenous, the unit hydrograph is recognized as
one of the most effective tools in estimating surface runoff. The unit hydrograph is
the hydrograph that results from one centimeter of uniform excess rainfall over the
entire watershed at a uniform rate and during a certain time period of rainfall. The
theory of unit hydrographs is based on four major assumptions:
1. Rainfall excess of equal duration is assumed to produce hydrographs with

equivalent time basis regardless of the intensity of the rain.
2. The direct runoff characteristic is dependent on previous precipita-
tion/rainfall and is assumed to be a single event.
3. Direct runoff ordinates are directly proportional to the depth of excess
rainfall (principle of linearity).
4. Spatial and temporal distributions of all rainfalls are assumed to be the
same.
When the duration of excess rainfall (T) is very small, the unit hydrograph is
referred to as an instantaneous unit hydrograph (IUH). When the ordinates of a
unit hydrograph represent the ratio of time to peak and peak discharge, this kind
of unit hydrograph is a dimensionless hydrograph. Some theoretical and empirical
formulas that relate peak flow and time to peak to watershed characteristics have
been developed and are referred to as synthetic unit hydrographs. A regionalized

unit hydrograph for ungauged drainage areas can be derived from a gauged water-
shed with similar characteristics.
Although the unit hydrograph has been used in most rainfall-runoff analyses,
care should be taken in applying the unit hydrograph when the principle of linearity
is violated. For example, for long-duration rainfalls that vary in intensity, the prin-
ciple of linearity can be justified. Spatial distribution of rainfall can also change the
shape of a hydrograph.
The unit hydrograph can be applied to calculate direct runoff and the streamflow
hydrograph. After estimation of abstractions, the excess rainfall hyetograph is cal-
culated from the rainfall hyetograph. The direct runoff, Qn, can be computed from
a given unit hydrograph, Un–m+1, and excess rainfall, Pm, using the following discrete
convolution equation:
Qn = ∑ P ⋅U
m =1
m n − m +1 (6.13)
where Un− m +1 is the unit hydrograph ordinate. n indicates the time step in which the
direct runoff is estimated.
6.7.1.3 Potential Evapotranspiration Estimation: Thornthwaite

Method
Different methods have been developed by various researchers in order to estimate

the potential evapotraspiration (ET p), which is a major source of water loss. Thorn-
thwaite and Mather (1955) developed a method in which the input data are precip-
itation (mm), monthly average temperature (°C), and geographic latitude. In this
method, a temporal index (im) is estimated for each month (m):
1.514
θm ⎞
im = ⎛ if θm > 0
⎝ 5⎠ (6.14)
im = 0 otherwise
where θm is average air temperature (ºC) in month m. The monthly potential evapo-
transpiration rate ( ETmp) is then estimated using the following equation:
12
I= ∑i
m =1
m (6.15)
a = 6.75 * 10 −7 I 3 − 7.71 * 10 −5 I 2 + 1.79 * 10 −2 I + 0.49239 (6.16)
ETmp = 16(10θ m / I )a fλ (6.17)

TABLE 6.2
Average Monthly Temperature (θ ) and Temporal Index of Thornthwaite Method
Month Jan. Feb. Mar. April May June July Aug. Sept. Oct. Nov. Dec.
Temperature –4 –2 5.5 12 18 25 28 23 18 10 7 3
(°C)
i 0 0 1.15 3.75 6.91 11.36 13.48 10.01 6.91 2.85 1.66 0.46
where:
fλ is a correction factor based on variation of daylight hours in different

geographic latitudes.
ETmp is the monthly potential evapotranspiration (mm).
I is the annual heat index, excluding subzero temperatures.
Example 6.3
Table 6.2 provides average monthly temperatures in a basin and values of i calculated
using Eq. (6.14). Estimate the monthly evapotranspiration in August.
Solution: The annual heat index I is determined as:
12
I= ∑i
m =1
m, = 58.5
The monthly potential evaporation in August is then estimated assuming fλ = 1.03:
a = (6.75 × 10 −7 )58.53 − (7.71 × 10 −5 )58.52 + (1.792 × 10 −2 )58.5 + 0.492 = 1.41
⎛ 10 × 23 ⎞
1.41
. = 16(1.03)⎝ = 113.6 mm
P
EPAug
58.5 ⎠
6.7.1.4 Infiltration Estimation: Green–Ampt Method
Green and Ampt (1911) developed a method for estimation of the infiltration rate
that is a conceptual representation of the infiltration process. It is assumed that the
wetting front is a sharp boundary dividing the soil with moisture content θi below
from saturated soil with moisture content η above. Using Darcy’s law, the following
equation for estimating cumulative infiltration, F(t), may be solved by the successive
substitution method:
⎛ F (t ) ⎞
F(t ) = Kt + Ψ( η − θ i ) ln⎜1 + (6.18)
⎝ Ψ( η − θ i ) ⎟⎠
where K is the hydraulic conductivity (cm/hr), Ψ is the wetting front soil suction
head (cm), and η is porosity.
Example 6.4
Consider a silt loam soil with porosity 0.5 and initial moisture content 0.16. Assume
that the ponding depth on the surface is negligible, K = 0.65 cm/hr, and Ψ = 16.7
cm. In order to estimate the cumulative infiltration (F) after 2 hr of rainfall, Eq.
(6.18) can be used as follows:
⎛ F (2) ⎞
F(2) = 0.65 × 2 + 16.7(0.5 − 0.16) × ln⎜1 + ⎟
⎝ 16.7(0.5 − 0.16) ⎠
F(2) = 4.75 cm
F(2) is obtained by the method of successive substitution.
6.7.2 SIMULATION MODELS

Simulation is the process of mimicking the dynamic behavior of systems over time.
The results of simulation models can be used to describe future possible states of a
system. Simulation models have been widely used in the area of water resources
planning and management to analyze the impacts of development policies on the
performance of systems. These models can be classified into two categories:
1. Hydrologic simulation models

2. Object-oriented programming models
A brief explanation of each category is presented in the following sections, and some
examples of simulation models that have been widely used by different investigators
are provided.
6.7.2.1 Hydrologic Simulation Models

Hydrologic simulation models have been developed based on the empirical and
analytical equations used to model the interrelations and water exchange between
surface and subsurface water bodies in the hydrologic cycle; some of these equations
were discussed briefly in the previous sections. Simulation models may also incor-
porate statistical correlations, empirical formulas, and probabilistic estimating meth-
ods. In the following sections, some of the most widely used simulation models,
including HEC-1, HEC-2, HEC-5, HEC-6, and MIKE-11, are briefly explained.
6.7.2.1.1 HEC-1 Package
The HEC-1, which is a flood hydrograph estimation package, calculates discharge
hydrographs from single-storm events for basins of different degrees of complexity.
The first version of this model was developed by the Hydrologic Engineering Center
of the U.S. Army Corps of Engineers in 1967. The program has several optional
capabilities, including parameter estimation and model calibration, and uses the
optimization method, multiplan–multiflood analyses, precipitation depth/area

computation, dam-break analyses, and flood damage analyses (Hoggan, 1997). The
main capabilities of this model can be summarized as follows:
• Optimal estimation of unit hydrograph, loss rate, and streamflow routing

parameters from measured data
• Simulation of watershed runoff and streamflow from historical or design
rainfall
• Computation of damage frequency curve and expected annual damages
for various locations and multiple flood control plans
• Simulation of reservoir outflow for dam safety analysis
Simulation of the watershed for rainfall-runoff modeling, an important compo-

nent of river-basin modeling, includes runoff estimation, river and reservoir routing,
and combining hydrographs. Dividing the river basin into smaller sub-basins with
physical and hydrologic parameters that have lower spatial variability is a primary
and basic step for rainfall-runoff modeling. A basin with numerous tributaries and
spatial variability in land use and land cover should be broken down into smaller
sub-basins to fit the constraints and assumptions of the model. Also, it may be
necessary to obtain the information required in terms of level of detail and location
as dictated by the objectives of the study. For this purpose, the following items
should be considered:
• Historical records of gauging stations at the outlets of sub-basins can

significantly improve the accuracy of the calibration process for each sub-
basin and the entire watershed. Results of comparisons between observed
and computed streamflows can be used in estimating model parameters.
• The model parameters are lumped; therefore, the uniformity of physical
and hydrologic characteristics in each sub-basin is an important constraint
in defining sub-basins.
• Spatial distributions of rain gauges in different sub-basins should be
identified.
• Special precipitation patterns that may affect parts of basin should be
determined.
Figure 6.7 shows a schematic of sub-basins and the rainfall-runoff modeling

process using the HEC-1 model. The following steps should be considered when
using this package for rainfall-runoff modeling:
• Analysis of watershed physiography, including delineation of basin and

sub-basin boundaries
• Estimation of basic physical parameters of the sub-basins including:
• Area
• Perimeter
• Slope of the main channel
• Shape factor
214
Sub-basin No. 1
Combing point
Routing Reach No. 1 Gage B
Sub-basin No. 3
C
Routing Reach
No. 2
Sub-basin No. 2
Routing Reach No. 3
C
Combing point Sub-basin No. 4
Gage A
Reservoir Routing
Reservoir outflow
Hydrograph
Routing Reach No. 4

Combing point
Gage C C Sub-basin No. 5
Routing Reach No. 5
C Sub-basin No. 6
Combing point
Gage D
FIGURE 6.7 Schematic of
rainfall-runoff model using
watershed simulation model.
• Estimation of basic hydrologic parameters of the sub-basins including:

• Time of concentration (Snyder method or Linsley method)
• Storage coefficient (Linsley method)
• Snowmelt estimation:
• Degree-day method
• Energy balance method
• Interception and infiltration estimation:
• Initial and constant method (initial loss volume is satisfied, then con-
stant loss rate begins)
• HEC exponential method (loss function is related to antecedent soil
moisture condition and is a continuous function of soil wetness)
• SCS curve number method
• Horton’s equation method
• Green–Ampt method
• River and reservoir routing:
• Maskingum method
• Maskingum–Cunge method
• Kinematic wave method
• Progressive average-lag (Straddle–Stagger) method
• Modified-plus method
HEC-1 is capable of simulating snowfall and snowmelt. Up to ten elevation

zones of equal increments can be specified in each sub-basin. Temperature data at
the lowest elevation zone are reduced by the lapse rate (°C or °F per unit change in
elevation). The model determines whether precipitation falls as snow or as rain by
using the melt temperature (usually ±2°C), which is another input to the model
(Bedient and Huber, 1988). A number of methods are available in this package to
be used for surface runoff computation:
• Direct input of the unit hydrograph

• Snyder unit hydrograph method
• Clark hydrograph method
• SCS method
• Kinematic wave for overland hydrograph
An automatic optimization procedure is also provided in this package that is a

useful tool for trial-and-error parameter selection. Figure 6.8 shows a schematic
diagram of parameter estimation in HEC-1. The objective function is to minimize
the error index subject to certain constraints (ranges) on the parameters:
∑ (QR − QC ) ⎛⎝ N ⎞⎠
2 WT
STDER = i i (6.19)
i =1
Select Initial Estimate of all

Unknown Parameters
Determine Next Estimate of

Parameters
Simulate Rainfall-Runoff
Process to Compute Runoff
No
Compute Value of Objective Yes

Function Satisfied Stop
FIGURE 6.8 Schematic of parameter estimation in HEC-1. (From Hoggan, D. H., Floodplain
Hydrology and Hydraulics, McGraw-Hill, New York, 1997. With permission.)
where:
STDER is the error index.

QRi is the observed hydrograph ordinate for time i.
QCi is the computed ordinate for time i from HEC-1.
WT is a weighting function that emphasizes accurate reproduction of peak
flows.
N is the total number of hydrograph ordinates.
Frequency analysis and regionalization of hydrologic parameters are the bases

of streamflow estimation in ungauged watersheds. The experience and knowledge
of the analyst in regard to the model parameters are important when using this model
for ungauged watersheds. A more detailed description of this model is presented in
the user’s manual for this package (Hydrologic Engineering Center, 1981).
6.7.2.1.2 HEC-2
The water surface profile can be computed by the HEC-2 package, which was
initially developed in the 1970s by the Hydrologic Engineering Center (1982a) of
the U.S. Army Corps of Engineers. The computational procedure is based on solution
of one-dimensional energy equations. The following assumptions are considered in
development of this model:
• Steady-state flow
• Gradually varied flow
• One-dimensional flow with correction for horizontal velocity distribution
• Small channel slope
• Constant friction slope (averaged) between two adjacent cross sections
• Rigid boundary conditions
Data requirements include flow regime, starting elevation, discharge, loss coef-
ficients, cross-sectional geometry, and reach lengths. An overview of input data is
shown in Figure 6.9. In HEC-2, the following two equations are used:
FIGURE 6.9 Overview of HEC-2 input data.
α 2 V22 α V2
WS2 + = WS1 + 1 1 + he (6.20)
2g 2g
⎛ α V2 α V2 ⎞
he = LS f + C⎜ 2 2 − 1 1 ⎟ (6.21)
⎝ 2g 2g ⎠
where:
WS1 and WS2 are the water surface elevations at the ends of the reach.
V1 and V2 are the mean velocities (total discharge/total flow area) at the ends
of the reach.
α1 and α2 are the velocity coefficients for flow at the ends of the reach.
he is the energy head loss.
L is the discharge-weighted reach length.
S f is the representative friction slope for the reach.
C is the expansion or contraction loss coefficient.
Four equations in HEC-2 approximate the friction loss between two cross-sections:
2
⎛ Q + Q2 ⎞
Sf = ⎜ 1 ⎟ average conveyance (6.22)
⎝ K1 + K2 ⎠
⎛ Sf + Sf 2 ⎞
Sf = ⎜ 1 ⎟ average friction slope (6.23)
⎝ 2 ⎠
Sf = Sf1 ⋅ Sf 2 geometric mean friction slope (6.24)
⎛ 2S f ⋅ S f 2 ⎞
Sf = ⎜ 1
⎟ (6.25)
⎝ f1 + Sf 2 ⎠
S
where K1 and K2 represent the conveyance at the beginning and end of the reach.
Conveyance is defined from Manning’s equation as:
AR2 / 3
K= (6.26)
n
The total conveyance for a cross-section is obtained by summing the conveyance

from the left and right overbanks and the channel. The velocity coefficient, α, is
estimated using the following relation:
⎛ A2 ⎞ ⎛ K 3 K3 3
K ROB ⎞
α = ⎜ T3 ⎟ ⎜ 2LOB + CH
2
+ 2 ⎟ (6.27)
⎝ KT ⎠ ⎝ ALOB ACH AROB ⎠
where the subscript T is the cross-sectional total, LOB is the left overbank, CH is
the channel, and ROB is the right overbank. HEC-2 can compute up to 14 profiles
in a single run using the same cross-sectional data. Critical depth is also calculated
automatically for all cross-sections. For more details about this model, see the HEC-2
user’s manual (Hydrologic Engineering Center, 1982).
6.7.2.1.3 HEC-5
HEC-5 was developed in the 1970s at the Hydrologic Engineering Center (1982b)
of the U.S. Army Corps of Engineers and is a simulation model for flood control
and conservation systems. This program was developed to assist in planning studies
for evaluating proposed reservoirs in a system and to assist in estimating the flood
control and conservation storage requirements for each project recommended for
the system. It is also useful for selecting the proper reservoir releases throughout
the system during flood emergencies. The program may be used to determine:
• Flood control and conservation storage requirements for each reservoir in

the system
• Influence of a system of reservoirs on the spatial and temporal distribution
of runoff in a basin
• Evaluation of operational criteria for both flood control and conservation
(including hydropower) for a system of reservoirs
• Expected annual flood damages, system costs, and system net benefits for
flood damage reduction
• Design of a system of existing and proposed reservoirs or other alterna-

tives (including nonstructural alternatives) that results in the maximum
net flood control benefits, as determined by simulation runs for selected
alternative systems
The following methods are available in this package for river and reservoir
routing:
• Stream routing:
• Straddle–Stragger
• Tatum
• Muskingum
• Modified plus
• Working R&D
• Reservoir routing:
• Accounting method
• Surcharge routing
• Emergency release
Expected annual damages can be computed for different control points using
one or more ratios for each of several historical or synthetic floods. Expected annual
damages, costs, and net benefits for non-reservoir alternatives such as levees, channel
improvements, and nonstructural alternatives, including floodproofing and floodplain
zoning, can also be evaluated.
In addition to flood control operation, conservation operation may also be eval-
uated by this model to provide minimum flows at downstream locations based on
required reservoir storage. Hydropower reservoirs can also be evaluated by HEC-5
in order to meet on-site monthly, weekly, daily, and hourly energy requirements or
allocated system energy. For this purpose, the tailwater rating curve and power
generation efficiency are inputs to the model. Hydropower reservoirs can also be
operated based on a rule curve relating plant factors to a percent of conservation
storage. Both constant and varied diversions are included as a function of inflows
or reservoir storage. Benefits for hydropower reservoirs can be calculated based on
input rates for energy, secondary energy, and power shortages (negative benefits).
Table 6.3 shows the reservoir operation priorities in the HEC-5 model (Hydrologic
Engineering Center, 1982), and Figure 6.10 provides various reservoir storage levels
in the model.
6.7.2.1.4 HEC-6
HEC-6, which was first developed in the 1970s at the Hydrologic Engineering Center
of the U.S. Army Corps of Engineers, is a one-dimensional, movable-boundary,
open-channel-flow, numerical model. This model is designed to simulate and predict
changes in river profiles resulting from scour and/or deposition over moderate time
periods. A continuous flow record is partitioned into a series of steady flows of
variable discharge and duration. For each flow, a water surface profile is calculated,
TABLE 6.3
Reservoir Operation Priority in HEC-5 Model
Condition Normal Priority Optional Priority
Flooding at downstream location No release for power requirements Release for primary power
When primary power releases can Release down to top of buffer pool Release down to top of
be made without increasing inactive pool
flooding downstream
Flooding at downstream location No release for minimum flow Release minimum desired
flow
If minimum desired flows can be Release minimum flow between top Same as normal
made without increasing flooding of conservation and top of buffer
downstream pool
If minimum required flows can be Release minimum flow between top Same as normal
made without increasing flooding of conservation and top of inactive
downstream pool
Diversions from reservoirs (expect Divert down to top of buffer pool Divert down to top of
when diversion is a function of inactive pool
storage)
Source: Hydrologic Engineering Center, HEC-5 Simulation of Flood Control and Conservation Systems,
U.S. Army Corps of Engineers, Davis, CA, 1982.
FIGURE 6.10 Reservoir storage levels in HEC-5 model.
thereby providing energy slope, velocity, depth, etc. at each cross-section. Potential
sediment transport rates are then computed at each section. These rates, combined
with the duration of the flow, permit a volumetric accounting of sediment within each
reach. The amount of scour or deposition at each section is then computed and the
cross-sections are adjusted accordingly. The computations then proceed to the next
flow in the sequence, and the cycle is repeated beginning with the updated geometry.
Separation of sediment deposition from the hydraulics of flow is valid in some
circumstances; for example, deposition in deep reservoirs can usually be characterized
as a progressive reduction in storage capacity if the material is rarely entrained once
Left overback Main channel Right overbank

Downstream
X-section
Upstream
X-section
FIGURE 6.11 Schematic illustrating the main channel and right and left overbank reaches
between consecutive cross-section in HEC-6. (From Hydrologic Engineering Center, HEC-6
Scour and Deposition in Rivers and Reservoirs, U.S. Army Corps of Engineers, Davis, CA,
1993.)
it is deposited. HEC-6 is capable of predicting sediment behavior in shallow reservoirs

and most rivers, however, which requires that the interactions between the flow
hydraulics, sediment transport, channel roughness, and related changes in boundary
geometry be considered.
The procedure for clay and silt deposition is based on settling velocity. Different
methods are available for sediment transport functioning of bed material load.
A river system that is modeled by HEC-6 consists of main stem, tributaries, and
local inflow/outflow. A one-dimensional energy equation is used for water surface
profile computations. Manning’s equation and n values for overbank and channel
areas may be specified by discharge or elevation (Figure 6.11). Expansion and
contraction losses are included in the determination of energy losses. The energy
loss coefficients may be changed at any cross-section.
For each discharge in a hydrograph, the downstream water surface elevation can
be determined by either a user-specified rating curve or a time-dependent water
surface elevation. Internal boundary conditions can be imposed on the solution. The
downstream rating curve can be changed at any time, as can internal boundary
conditions.
Flow conveyance limits, containment of flow by levees, ineffective flow areas,
and overtopping of levees are simulated in a manner similar to HEC-2. Supercritical
flow is approximated by normal depth; therefore, sediment transport phenomena
occurring in supercritical reaches are simplified in HEC-6. For more details about
this package, see the user’s manual (Hydrologic Engineering Center, 1993).
6.7.2.1.5 MIKE-11
MIKE-11 is a software package developed at the Danish Hydraulic Institute (1992)
for the simulation of flows, sediment transport, and water quality in estuaries, rivers,
irrigation systems, and similar water bodies. It consists of four major parts:
1. Rainfall-runoff process modeling

2. Modeling of river and estuary hydraulics
3. Noncohesive sediment processes
4. Quality processes and advection-dispersion modeling
Figure 6.12 shows a typical MIKE-11 simulation. The rainfall-runoff process is

modeled using the NAM model, which was developed by the Institute of the Hydro-
dynamics and Hydraulic Engineering at the Technical University of Denmark. The
NAM model considers evapotranspiration, groundwater recharge, base flow, and
groundwater storage and interflows. It also can be used to describe the effects of
irrigation within a watershed, including:
• The overall water balance of the watershed, which is mainly affected by

increased evapotranspiration and possible external water resources for
irrigation
• Local infiltration and groundwater recharge in irrigated areas
• Distribution of watershed runoff on different components (overland flow,
interflow, baseflow), which may be influenced by increased infiltration in
irrigated areas
Water transfer in the root zone is also considered when modeling evapotranspiration.
The meteorological data used in the NAM model are rainfall, potential evapotrans-
piration, and temperature.
In the hydraulic modeling module, MIKE-11 uses an implicit, finite-difference
model for computation of unsteady flows in rivers and estuaries. The model is also
capable of describing subcritical and supercritical flow conditions by means of a
numerical scheme that adapts to local flow conditions. Computational modules are
included for description of flow over hydraulic structures. The formulations can be
applied to loop networks and quasi two-dimensional flow simulation on floodplains.
The computational scheme is applicable for vertically homogeneous flow conditions
ranging from steep river flows to tide-influenced estuaries.
The noncohesive sediment transport module can be run in two modes: explicit
and morphological. In the explicit mode, output is required from the hydrodynamic
module in terms of discharge water levels both in time and space. Results are provided
in the form of volume transport rates and accumulated volumes of deposition or
erosion. The explicit mode is useful where significant morphological changes are
unlikely to occur. In the morphological mode, sediment transport is calculated in
tandem with the hydrodynamic module. The feedback between these modules is
achieved through solution of the sediment continuity equation and updating the bed
resistance and subsequent sediment transport. Results of this mode are in the form of
sediment transport rates, bed-level changes, resistance number, and dune dimensions.
The advection–dispersion modeling module is based on the one-dimensional equa-
tion of conservation of mass of a dissolved or suspended material. The advection–dis-
persion equation is solved numerically using an implicit finite difference scheme
which, in principle, is unconditionally stable and has negligible numerical dispersion.
A correction term has been introduced in order to reduce the third-order truncation
error. This correction term makes it possible to simulate dispersion/convection of
FIGURE 6.12 Typical MIKE-11 simulation. (From Danish Hydraulic Institute, MIKE-11: A
Microcomputer Based Modeling System for Rivers and Channels — Reference Manual,
Denmark, 1992.)
concentration profiles with very steep fronts. The water quality module, which is
coupled with the advection–dispersion module, deals with basic aspects of river water
quality as influenced by human activities, such as oxygen depletion and ammonia
levels resulting from organic matter loadings.
For more details about the models used in the various MIKE-11 modules and
their limitations, see the reference manual (Danish Hydraulic Institute, 1992).
6.7.2.1.6 IRAS
IRAS (Interactive River-Aquifer Simulation Program) is a generic model developed
by Cornell University in 1995. This program can simulate water flows, water storage
volumes, water qualities, and hydroelectric power and energy produced and con-
sumed throughout a surface water or an independent surface and groundwater system
over time. IRAS has been developed to assist those interested in evaluating the
performance or impact of alternative design and operation policies of regional water
resources systems.
The system to be simulated by this model should be represented by a network
of connected nodes and links. The nodes of the network can represent aquifers,
gauge sites, natural lakes, reservoirs, wetlands, confluences, and diversions. A single
node may be a combination of any of these nodes. The links of the network can be:
• Unidirectional (flows only in one direction)

• Bidirectional (flows in either direction or flows dependent on changing
surface elevation or pressure head differences)
IRAS can simulate up to 10 separate independent water quality constituents, and the
system to be simulated can include up to 60 stream and river reaches and up to 60
sites that can represent any feasible combination of aquifers, gauge and monitoring
sites, consumption sites, natural lakes, wetlands, reservoirs, confluences, diversion
and withdrawal sites, hydropower and pumping sites, and wastewater effluent dis-
charge sites. The time resolution can be defined by the user and can include within-
year periods and their durations, as well as the number of simulation time steps
contained in each within-year period. If the demand for water at various sites depends
on local rainfall, streamflow, or storage at those sites, demand-driven targets can be
defined and water allocations to satisfy these demands can vary from year to year.
The reservoir release policy based on storage volume zones (supply-driven
policies) should be defined by the user. For this purpose, storage volume boundaries
and releases associated with the storage volumes in each of the storage volume zones
should be defined by the user. Figure 6.13 shows a reservoir release rule based on
storage volumes in the program. The program is also capable of simulating reservoir
operation based on demand deficits. Releases based on demand deficits, prorated
over the remainder of the within-year period and then multiplied by a source mul-
tiplier for the reservoir, are added to the release based on supply-driven policies.
6.7.2.2 Object-Oriented Simulation Models
Applying object-oriented modeling when utilizing a system dynamic (SD) approach

for modeling water resources systems is a new way of thinking about modeling real-
FIGURE 6.13 Reservoir release rule based on storage volumes in IRAS. (From Loucks, D. P.
et al., IRAS: Interactive River-Aquifer Simulation — Program Description and Operation,
Resources Planning Associate, Inc., Ithaca, NY, 1995. With permission.)
world concepts. SD models provide insight into feedback processes and therefore
give system users a better understanding of the dynamic behavior of systems. The
SD approach is based on the theory of feedback processes. In a feedback system,
future behavior is influenced by its past. Feedback systems can be classified into
two different classes (Ahmad and Simonovic, 2000):
• Negative feedback, which seeks a goal and responds as a consequence of

failing to achieve the goal
• Positive feedback, which generates growth processes where action builds
a result that generates still greater action
The feedback loop indicates how a system might behave because of its internal
feedback loops and the effects that positive and negative feedback loops have on a
system. By separating policy questions from data, object-oriented modeling and a
FIGURE 6.14 Basic building blocks of an object-oriented simulation model. (From Simonovic,
S. P., Water Int., 25(1), 76–88, 2000. With permission.)
system dynamic approach make the model results functionally transparent to all
parties involved in the water management program. The proposed approach is flex-
ible and transparent and allows for easy involvement of stakeholders in the process
of decision analysis about water resources (Simonovic, 2000).
Figure 6.14 shows the basic elements of an object-oriented simulation model.
Stock represents accumulations and can be used as a source — for example, water
stored in the reservoir. The accumulation needs flow (inflow to the reservoir) is
modeled by Flow. Flow and Stock are inseparable and comprise a minimum set of
elements necessary to describe dynamics. Converters convert input to outputs; they
can represent information or material quantities. Connectors link stocks to convert-
ers, stocks to flow regulators, and converters to other converters.
Computer software tools such as STELLA, DYNAMO, VENSIM, and POW-
ERSIM can be used for the development of object-oriented simulation models for
complex water resources systems. For more details see High Performance Systems
(1992), Lyneis et al. (1994), Ventana Systems (1996), and Powersim Corp. (1996).
6.7.3 OPTIMIZATION MODELS

Optimization models are used to find the best way to meet different objectives of
river system management. Such models might be used, for example, to determine
the volume of water that should be allocated to various demand points along the
river or to select the treatment efficiency that should be applied to effluents discharg-
ing to the river in order to meet environmental constraints. Some examples of the
application of optimization models in river system management are discussed in the
following sections.
6.7.3.1 Water Allocation in Multi-User Systems
The costs of excluding potential (excluding cost) beneficiaries from exploiting water
resources along the river and saving it for supplying water rights are an important
component of total river management costs. Exclusion costs are highly correlated
with the level of relative scarcity; where water is scarce, the costs of monitoring and
enforcing regulations for water allocation among users are higher than where water
is plentiful relative to demand. Optimization models are useful tools for development
of water allocation heuristics when water is scarce. The objective function of water
allocation problems can be the minimization of total costs of operation of systems
as follows:
n ND
Min ∑ ∑ (C
t =1 i =1
t ,i − Bt ,i ) (6.28)
where ND is the number of demand points and Ct,i is the total cost of water allocation
to user i in time t, which can be estimated as the costs of water diversion ( Cidiv
,t ) plus
exclusion costs ( Ciexc
,t ) associated with water allocation and controlling the water
consumption by user i:
Ct ,i = Ctdiv
,i + Ct ,i
exc
(6.29)
where Bt,i is the total benefit of water allocation to user i in time t, which can be
determined based on the price of water allocated to various users. As mentioned in
the previous sections, deterministic and stochastic models can be used for optimi-
zation of river–reservoir systems.
As an example of stochastic modeling, consider a water management authority
that should deliver water from a river to municipal and industrial water users. Let
D1 and D2 be the water demands of the municipal and the industrial water users,
respectively. If the demands are supplied, the resulting benefits to the water man-
agement authority can be estimated considering the price of water for these uses
(P1, P2). However, if the water rights are not supplied, water must be delivered from
an alternative source, which would incur greater costs, or the costs associated with
rationing must be incorporated (C1, C2). Let Q be the random variable representing
the available water. The objective function of stochastic optimization model can be
formulated as follows:
⎛ 2
⎞ ⎛ 2
⎞
Maximize ⎜
⎝
∑ i =1
PD
i
i
⎟ − E⎜
⎠ ⎝
∑C F
i =1
i i ,Q ⎟
⎠
(6.30)
where Fi,Q is the amount by which demand Di is not met when the river flow is Q.
To solve this problem with dynamic programming or linear programming techniques,
the probability distribution of Q should be approximated by a discrete distribution.
If Q take values qj with probability pj (j = 1, …, n), the objective function can be
reformulated as follows (Loucks et al., 1981):
⎛ 2
⎞ ⎛ 2 n ⎞
Maximize ⎜
⎝
∑i =1
PD
i
i
⎟ − ⎜⎜
⎠ ⎝
∑ ∑ p C F ⎟⎟⎠
i =1 j =1
j i i, j (6.31)
Subject to q j ≥ D1 + D2 − F1, j − F2, j (6.32)
Di ≥ Fi, j (6.33)
The above constraints should be repeated for all values of j.

FIGURE 6.15 Schematic of multireach river.
6.7.3.2 River Quality Management
Optimization models have been widely used for development of cost-effective water
pollution control strategies. These models usually focus on different measures of
quality of water in receiving wastewaters. Such models are particularly useful for
management efforts that include the entire watershed because they are capable of
accounting for many interactions that occur among the various sources of pollution
in a watershed, the effects of pollution on water quality, and the costs of options for
reducing pollution discharges (ReVelle and McGarity, 1997).
Consider a multireach, multidischarge river as shown in Figure 6.15, where qi
and pi are quantity and quality of water discharged to the river at the beginning of
reach i, respectively. qi′ and pi′ are the same indicators for water diverted at the
beginning of reach i. The optimization model for finding the best treatment efficiency
for each of the discharges (ei) can be formulated as follows:
Min ∑ C (e )
i =1
i i (6.34)
Subject to: Qi = Qi −1 + qi − qi′ (6.35)
i i = Qi −1 ( Pi −1 Ai −1 ) + ei qi pi − qi′ pi′
PQ (6.36)
Pi ≤ Pmax (6.37)
where:
Ci is the cost of applying treatment with a specific efficiency on discharge i.

Qi is the river discharge in reach i.
Pi is the river water quality in reach i.
Ai is the percent of decrease in the quality indicator (Pi ) due to self-purification
of the river in reach i.
Pmax is the maximum or minimum allowed value of the quality indicator.
Details of water quality management models are presented in Chapter 9.

FIGURE 6.16 Comparison between computed and observed hydrographs for calibration of
rainfall-runoff models.
6.7.3.3 Calibration of Hydrologic Simulation Models
Model calibration refers to adjusting the model and parameters to bring the model
outputs as close to the observed values as possible; however, the parameters of
hydrologic models almost always depend on climatic, geographic, and hydrologic
characteristics. Because of the large number of factors affecting hydrologic pro-
cesses and a lack of knowledge about the details of particular hydrologic processes,
parameter estimation is a major source of error in modeling. Optimization algo-
rithms can play an important role in improving the calibration process. Application
of the optimization algorithm in the HEC-1 package for modeling the rainfall-
runoff process was discussed previously; this algorithm is based on minimizing
differences between the computed and observed runoff hydrographs, as shown in
Figure 6.16.
6.8 CONFLICT ISSUES IN RIVER SYSTEMS

River system planning and management is usually a multiobjective problem, with
many objectives in conflict. For example:
• Several demand points are usually located along the river, and water that
is supplied to one of these demand points cannot be used by others;
therefore, the major conflict issue in river system planning arises when
the river flow is less than instream and offstream water requirements.
• Water supply to some demand points, such as agricultural zones, will
increase low-quality return flows and endanger aquatic life.
• The treatment efficiency of pollutant dischargers from various sectors
(e.g., domestic, industrial, agricultural) can result in conflicts among these
sectors stemming from a desire to reduce violations of water quality
standards.
River system planning and management conflicts arise when the water demands
of different sectors are supplied from one river system. In order to examine this
problem, consider a river system that supplies the following demands:
• Domestic water demand

• Agricultural water demand
• Industrial water demand
• In-stream water demand (environmental and recreation water demand)
Water allocation schemes are typically defined and imposed by:
• Department of Water Supply

• Department of Agriculture
• Department of Environmental Protection
• Department of Industries
The priorities and favorable ranges of water supplies for each of these agencies
should be considered when formulating conflict resolution problems; therefore, the
first step in conflict resolution studies is to recognize all of the conflict issues and
the responsible agencies. The next step is to define the acceptable range of allocated
water to each sector and its corresponding utility function. The utility function
assigns a number between 0 and 1 to each value of allocated water; 0 indicates that
this value of allocated water is not acceptable, and 1 indicates that the allocated
water completely satisfies the water demands of the sector. The utility functions are
developed by decision-makers from the various sectors. The final step is to formulate
the conflict resolution problem. For this purpose, different methods can be used, as
briefly explained in Chapter 2. A typical conflict resolution problem in river systems
is explained in the following example.
Example 6.5
Determine the monthly water allocation for domestic, industrial, agricultural, and
recreation demands in the river system shown in Figure 6.17. The average monthly
river discharges upstream of the system, within a 2-year time horizon, are presented
in Table 6.4. The return flow of domestic and industrial sectors is assumed to be
20% of the allocated water, and the initial volume of the lake is 30 million m3. The
utility functions of the sectors are presented in Table 6.4 and Figure 6.18. The utility
values have been normalized between 0 and 1, and higher utility values reflect higher
priorities of the decision-makers or sectors. The shape of the graphed utilities is
considered to be trapezoidal, and the array (a, b, c, d) in Table 6.4 shows the values
of water allocated to the agricultural sector, corresponding to utilities of 0, 1, 1, and
0, respectively.
Solution: The Nash solution of this problem is the unique optimal solution of
the following problem:
4 12
Maximize ∏ ∏( f
i =1 t =1
i ,t − di,t ) wi (6.38)
Subject to di,t ≤ fi,t ≤ fi,∗t ∀i, t (6.39)

FIGURE 6.17 Components of the river system (for Example 6.5).
TABLE 6.4
Monthly River Discharge and Utility Functions of the Agricultural
Sector
River Discharge
Agricultural Demand Upstream of the System
Month (million m3/month) (million m3)
1 (0,10,15,25) 55
2 (0,8,12,20) 53
3 (0,0.01,25,25.01) 55
4 (0,0.01,25,25.01) 60
5 (0,0.01,25,25.01) 65
6 (0,5,10,15) 80
7 (5,10,25,50) 80
8 (15,30,60,90) 85
9 (20,35,60,90) 80
10 (25,35,60,90) 75
11 (30,40,60,90) 73
12 (25,35,60,10) 70
where Wi is the relative weight of sector i, which shows the relative authority of this
sector in water allocation; fi,t, fi*,t , and di,t are the utility function, the ideal point,
and the disagreement point of sector i in period t, respectively. The relative weights
of agencies that are related to agricultural, domestic, industrial, and recreational
demands are assumed to be 1. The water allocated to these sectors based on the
conflict resolution model is presented in Table 6.5.
1.2
Industrial demand
Domestic demand
1 Volume of lake
0.8
Utility
0.6
0.4
0.2
0
0 20 40 60 80
Volume (MCM)
FIGURE 6.18 Utility function of different sectors for water allocated to industry (for Exam-
ple 6.5).
6.9 PROBLEMS
6.1 The length of the main stream of a watershed from the outlet to the divide
is 20 km. The watershed shape parameter (Lc) is 12 km, and the watershed
storage coefficient (Ct) is 2.1. Find the lag time of a watershed by Snyder’s
method.
6.2 Solve the previous problem using Linsley’s method. Assume that the slope
of the watershed is 10%, and the watershed storage coefficient (Cl) is 1.2.
6.3 Find the storage coefficient for the basin in problem 6.2 with respect to
the empirical coefficient, Cr. Assume that the basin area is 50 km2.
6.4 Results of frequency analysis of historical data for the annual peak flows
at the outlet of three basins are as follows: ˙
A B C
Return Period Discharge Damage Discharge Damage Discharge Damage
(yr) (m3/s) (1000$) (m3/s) (1000$) (m3/s) (1000$)
10 240 350 560 250 110 650
25 350 400 650 320 190 680
100 420 620 800 450 250 930
Carrying capacity 100 400 90
(m3/s)
Find the annual expected damage for all basins.

6.5 In problem 6.4, reducing the flood damage by channel modification has
been studied for three basins. Results of this study have shown that the
capital cost of channel improvement in each basin can be estimated as a
function of flood peak discharge and the river carrying capacity:
Ci = α i × (QiP − QiC )3
River Basin Modeling
TABLE 6.5
Water Allocation to Different Sectors Based on the Conflict Resolution Model
Month Industrial Agriculture Domestic Lake Month Industrial Agriculture Domestic Lake
(year 1) (million m3) (million m3) (million m3) (million m3) (year 2) (million m3) (million m3) (million m3) (million m3)
1 17.56 24.78 34.91 13.70 1 26.65 38.15 38.75 8.91

2 17.56 28.13 38.75 10.96 2 18.72 39.82 45.48 9.05
3 23.34 24.78 36.83 8.11 3 19.87 39.82 38.75 11.31
4 18.72 26.45 38.75 19.30 4 33.73 33.14 34.91 6.51
5 19.87 39.82 35.87 14.50 5 18.72 21.44 37.79 12.11
6 24.49 36.48 34.91 5.36 6 19.87 31.47 39.71 9.17
7 18.72 31.47 34.91 22.44 7 17.56 28.13 39.71 19.55
8 17.56 31.47 36.83 39.23 8 19.87 23.11 39.71 44.05
9 23.34 23.11 46.44 35.86 9 23.34 23.11 36.83 31.36
10 17.56 23.11 36.83 35.55 10 19.87 23.11 35.87 31.37
11 21.03 24.78 36.83 35.89 11 16.41 21.44 34.91 30.10
12 18.72 24.78 39.71 31.37 12 16.41 21.44 37.79 29.28
233
where:
Ci is the cost of channel improvement in basin i ($).
QiP is the peak discharge in basin i (m3/sec).
QiC is the carrying capacity of the main river in basin i (m3/sec).
αi is the estimated coefficient for basin i.
α is estimated to be 0.001, 0.0005, and 0.01 for basins A, B, and C,
respectively. In which basin is the channel improvement more cost effec-
tive? The annual maintenance costs are estimated to be equal to 5% of
the capital investment required for each of these basins.
6.6 Consider a basin covered by silt loam soil with porosity 0.5 and initial
moisture content 0.16. Assume that the hydraulic conductivity is
0.65 cm/hr, and the wetting front soil suction head is 16.7 cm. The area
of the basin is 220 km2. The frequency analysis of rainfall data in this
basin is shown in the following table:
Return Period Duration Intensity

(yr) (hr) (cm/hr)
5 5 1
10 4 2
100 5 5
The flood damages at the outlet of the basin are as follows:
Flood Volume Flood Damage

(million m3) (106 $)
10 25
20 40
30 55
40 100
Construction of a dam may reduce the flood damages. The annual costs
of the dam considering the capital investment and annual costs of opera-
tion and maintenance are shown in the following table:
Reservoir volume Annual Cost

(million m3) (106 $)
5 2.5
10 3
15 4
20 7
40 12
If the evaporation rate is estimated to be 10 mm, find the optimal volume

of the reservoir.
6.7 A river supplies water to an industrial complex and agricultural lands
located downstream of the complex. The average monthly industrial and
irrigation demands and the monthly river flows in a dry year are as follows
(numbers are in million m3):
Jan. Feb. March April May June July Aug. Sept. Oct. Nov. Dec.
Industrial 8 8.5 8.5 9 9 9.5 9.5 10 9.5 9 8.5 8.5

Irrigation 0 0 20 50 55 70 70 65 20 10 0 0
Monthly 25 27 40 48 60 42 25 17 14 18 20 25
flows
The price of water for industrial and irrigation uses is $90,000 and
$10,000/million m3, respectively.
(a) Formulate the problem for optimizing the water allocation for this
river.
(b) Solve the problem using linear programming.
6.8 In problem 6.7, the shortages should be supplied from groundwater
resources. The cost of supplying demand from groundwater is a function
of volume of water, which should be extracted. The cost of water extrac-
tion for industrial demands is estimated as Cind = 1000 ⋅ xind
2
, where xind is
the volume of water extraction for industrial complex. The cost of water
extraction for irrigation purposes is estimated as Cirr = 250 ⋅ xirr2
, where
xirr is the volume of water extraction for irrigation. Formulate the problem
for finding the optimal monthly volumes of water allocation.
6.9 In problem 6.7, the industrial complex discharges its wastewater to the
river upstream of the agricultural lands. The monthly wastewater discharge
rate is about 20% of the water use of the complex in each month. In order
to keep the quality of the river flow in an acceptable range for irrigation,
the pollution load of the industrial wastewater should be reduced by
primary treatment, which has additional cost equal to $12,000/million m3.
(a) Formulate the problem for optimizing the water allocation from this
river.
(b) Solve the problem using linear programming.
References
Ahmad, S. and Simonovic, S. P., System dynamics modeling of reservoir operations for
flood management, J. Comput. Civil Eng., 14(3), 190–198, 2000.
Barrett, E. C. and Curtis, L. F., Introduction to Environmental Remote Sensing, Chapman
& Hall, London, 1992.
Bedient, Ph. B. and Huber, W. C., Hydrology and Floodplain Analysis, Addison-Wesley,
Reading, MA, 1988.
Danish Hydraulic Institute, MIKE-11: A Microcomputer Based Modeling System for Rivers
and Channels — Reference Manual, Denmark, 1992.
Green, W. H. and Ampt, G. A., Studies of soil physics, 1: The flow of air and water through
soils, J. Ag. Sci., 4(1), 1–24, 1911.
Gupta, R. S., Hydrology and Hydraulic Systems, Prentice-Hall, Englewood Cliffs, NJ, 1989.
High Performance Systems, Stella II: An Introduction to Systems Thinking, High Perfor-
mance Systems, Inc., Hanover, NH, 1992.
Hoggan, D. H., Floodplain Hydrology and Hydraulics, McGraw-Hill, New York, 1997.
Hydrologic Engineering Center, HEC-1 Flood Hydrograph Package: User’s Manual and
Programmer’s Manual (updated 1985), U.S. Army Corps of Engineers, Davis, CA.
1981.
Hydrologic Engineering Center, HEC-2 Water Surface Profiles Flood Hydrograph Package:
User’s Manual, U.S. Army Corps of Engineers, Davis, CA, 1982a.
Hydrologic Engineering Center, HEC-5 Simulation of Flood Control and Conservation
Systems, U.S. Army Corps of Engineers, Davis, CA, 1982b.
Hydrologic Engineering Center, HEC-6 Scour and Deposition in Rivers and Reservoirs, U.S.
Army Corps of Engineers, Davis, CA, 1993.
Linsley, R. K., Kohler, M. A., and Paulhus, J. L. H., Hydrology for Engineers, 3rd ed.,
McGraw-Hill, New York, 1982.
Loucks, D. P., French, P. N., and Taylor, M. R., IRAS: Interactive River-Aquifer Simulation
— Program Description and Operation, Resources Planning Associate, Inc., Ithaca,
NY, 1995.
Lyneis, J., Kimberly, R., and Todd, S., Professional dynamo: simulation software to facilitate
management learning and decision making, in Modeling for Learning Organizations,
Morecroft, J. and Sterman, J., Eds., Pegasus Computations, Waltham, MA, 1994.
McCuen, R. H., Hydrologic analysis and design, Prentice-Hall, Englewood Cliffs, NJ, 1998.
Powersim Corp. (1996), Powersim 2.5 Reference Manual, Powersim Corporation, Inc.,
Herndon, VA, 1996.
ReVelle, C. and McGarity, A. E., Design and Operation of Civil and Environmental Engi-
neering Systems, John Wiley & Sons, New York, 1997.
Sanders, T. G., Ward, R. C., Loftis, J. C., Steele, T. D., Ardin, D. D., and Yevjevich, V.,
Design of Networks for Monitoring Water Quality, Water Resources Publications,
Littleton, CO, 1983.
Simonovic, S. P., Tools for water management: one view of the future, Water Int., 25(1),
76–88, 2000.
Snyder, F. F., Synthetic Unit Hydrograph, Trans. Am. Geophys. Union, 19, 447–454, 1938.
Thornthwaite, C. W. and Mather, J. R., The water balance, in Publications in Climatology:
Laboratory Climatology, Vol. 8, No. 1, Drexel Institute of Technology, Centerton,
NJ, 1955.
Vetana Systems, Vensim User’s Guide, Ventana Systems, Inc., Belmont, MA, 1995.
Viessman, J. R. and Hammer, M., Water Supply and Pollution Control, Harper & Row,
New York, 1985.
White, I. D., Mottershead, D. N., and Harrison, S. J., Environmental Systems: An Intro-
ductory Text, Chapman & Hall, London, 1992.
Programme, The United Nations Development Programme, and The World Bank,
Oxford University Press, London, 1998.
7 Groundwater
Management
7.1 INTRODUCTION
The term groundwater usually refers to subsurface water located below the water
table in saturated soils and geologic formations. In groundwater studies, an under-
standing of surface water and subsurface, unsaturated soil moisture, and their inter-
actions is important. Groundwater is an important feature of the environment and a
part of the hydrologic cycle; therefore, an understanding of its role in this cycle is
necessary if integrated analyses are to be used in the study of watershed resources
and regional assessment of contamination (Freeze and Cherry, 1979). The main
objective of groundwater management is the optimal allocation of groundwater
resources to water demands, taking into consideration the complex economic, envi-
ronmental, and hydrogeologic constraints and conflicts in objectives. This chapter
discusses the main characteristics of groundwater systems, groundwater flow equa-
tions, and groundwater modeling, and some optimization models for groundwater
management are also presented.
7.2 CHARACTERISTICS OF GROUNDWATER SYSTEMS

7.2.1 HYDRAULIC PROPERTIES OF SOILS AND GROUNDWATER
Groundwater occurs whenever the soil pores are occupied and saturated by water.
An aquifer is a permeable geologic formation that can transmit significant quantities
of water under the usual hydraulic gradient. The saturated geologic unit is referred
to as an aquiclude when it is not capable of transmitting significant volumes of water
under ordinary hydraulic conductivity. An aquitard is permeable enough to transmit
water in significant quantities within regional groundwater modeling but not in
sufficient quantities for installing production wells within them. A confined aquifer
is an aquifer that has been confined between two aquitards; however, when the water
table forms the upper boundary, the aquifer is unconfined.
Hydraulic head at a given point (A in Figure 7.1) in groundwater is defined as
follows:
P
h=Z+ = Z+ψ (7.1)
ρg
239
FIGURE 7.1 Illustration of hydraulic head.
As shown in Figure 7.1, h is the hydraulic head, Z is the distance between the point
(A) and a datum level, P is the fluid pressure from the water column above the point,
ρg (or γ) is the specific weight of water, and ψ is the pressure head.
The loss of head per unit length of flow along a streamline is the hydraulic
gradient and is presented as:
dh
i= (7.2)
dl
The hydraulic conductivity is defined as the volume of water flowing perpendicular

to a unit area of porous media per unit time and under unit hydraulic conductivity.
The hydraulic conductivity (K) combines fluid and porous medium properties as
follows:
kρg
K= (7.3)
µ
where k is the intrinsic permeability and is a function of pore size, and µ is the
dynamic viscosity of water, which is a function of temperature (generally decreasing
when the temperature increases). The hydraulic conductivities of several lithologies
are presented in Table 7.1.
Hydraulic conductivity values usually show spatial variation within a geologic
porous media or with the direction of measurement, termed heterogeneity and iso-
tropy, respectively. In other words, if the hydraulic conductivity is independent of
the position in porous media, the geologic formation is homogenous; if it is inde-
pendent of the direction of measurement in a point, the formation is isotropic. Water
in unsaturated areas is subject to hydraulic gradient, surface tension, and molecular
attraction. The unsaturated hydraulic conductivity is equal to Kψ, where ψ is the
pressure head, which is usually negative in unsaturated media.
In the mid-19th century, Henri Darcy discovered one of the most important
relationships of porous media hydrodynamics. Darcy’s law defines the specific
Groundwater Management 241
TABLE 7.1
Hydraulic Conductivities of Several Lithologies
Lithology Porosity (%) K (cm/sec)
Unconsolidated
Gravel 25–40 10–2 – 102
Sand 25–50 10–4 – 1
Silt 35–50 10–7 – 10–3
Clay 40–70 10–10 – 10–7
Glacial till 10–20 10–10 – 10–4
Indurated
Fractured basalt 5–50 10–5 – 1
Karst limestone 5–50 10–4 – 10
Sandstone 5–30 10–8 – 10–4
Limestone, dolomite 0–20 10–7 – 10–4
discharge (volumetric flow rate per unit area of porous media, perpendicular to the
flow direction) as follows:
dh
q = −K( ) (7.4)
dl
where q is the specific discharge, which has the dimension of velocity; K is the
hydraulic conductivity, and dh/dl is the hydraulic gradient. The negative sign indi-
cates the flow direction, which is equal to the direction of decreasing gradient. The
average pore velocity or seepage velocity that can be used to estimate true travel
time or solute transport is:
q
v= (7.5)
ne
where ne is the effective porosity, which is the fraction of pores in porous medium
that allows the passage of water.
Example 7.1
As shown in Figure 7.2, a semi-impervious aquitard separates an overlying water
table aquifer from an underlying confined aquifer. Determine the rate of flow that
takes place between the aquifers.
Solution: Because the water table is above the piezometric surface and a semi-
impervious (leaky) layer exists, flow will take place from the water table aquifer to
the confined aquifer. Assume that the head at point b in Figure 7.2 is hb, and consider
FIGURE 7.2 Vertical downward flow through soil layers for Example 7.1.
the unit horizontal area through which flow takes place. Between points a and b,
from Eq. (7.4):
12(27-hb )
q=
27
Between points b and c, from Eq. (7.4):
(hb + 4.5) - (24 + 4.5)

q = (0.3)
4.5
From these equations we have:
hb = 26.6m
m3
q = 0.17 per square meter
day
The transmissivity or transmissibility (T) in horizontal or layered aquifers is defined

as:
T = Kb (7.6)
where b is the saturated thickness of the aquifer. Therefore, transmissivity reflects

the flow rate of water transmitted through a unit width of aquifer under a unit
hydraulic gradient.
The storativity (or storage coefficient) of an aquifer is defined as the volume of
water released or taken into a unit surface area per unit time and per unit decrease
or increase in hydraulic head. The storativity can be defined as:
S = Ss b (7.7)
where Ss is specific storage, and b is the saturated thickness of the aquifer. In

confined aquifers, the specific storage is defined as follows:
Ss = ρg(α + nβ) (7.8)
where α and β are aquifer and water compressibility, respectively and n is soil porosity.
7.3 GROUNDWATER FLOW EQUATIONS

The mathematical description of groundwater flow is based on the conservation of
mass, energy, and momentum. In this section, groundwater flow equations are devel-
oped using these principles. The groundwater equations are expressed in terms of
partial differential equations, where the spatial coordinates and time are independent
variables.
7.3.1 THE CONTINUITY EQUATION

The continuity equation as a fundamental law of groundwater flow expresses the
principle of mass conservation as follows:
mass inflow rate – mass outflow rate

= rate of change of mass storage with time (7.9)
Considering an elementary control volume of soil (Figure 7.3) that has the volume
of (Δx, Δy, Δz), the mass of the groundwater (M) in this control volume is:
M = ρ.θ.Δx.Δy.Δz (7.10)
FIGURE 7.3 Mass conservation in elementary control volume (actual volume is three dimen-
sional but it has been shown in only two dimensions for more clarity). (From Delleur, J. W.,
Ed., The Handbook of Groundwater Engineering, CRC Press/Springer-Verlag, Boca Raton,
FL, 1999. With permission.)
where θ is the moisture content of the porous medium. Equation (7.9) can be
rewritten as follows:
∂M
= inflow − outflow (7.11)
∂t
The inflow and outflow can be calculated for each side of the element. For example,
considering the flux vector q = (qx , qy , qz ), the mass of groundwater inflow to the
left side is:
⎛ Δx ∂ρqx ⎞
ρqx ( x − Δx / 2, y, z, t )ΔyΔz ≈ ⎜ ρqx − ⎟ ΔyΔz (7.12)
⎝ 2 ∂x ⎠
A similar equation can be derived for the other side. Considering these equations,
the total inflow minus outflow can be derived as follows:
∂M ⎛ ∂ρqx ∂ρqy ∂ρqz ⎞

= −⎜ + + ⎟ ΔxΔyΔz (7.13)
∂t ⎝ ∂x ∂y ∂z ⎠
Based on Eq. (7.11), the change in storage is given by:
∂M ∂ ⎛ θ ∂ρ ∂θ θ ∂Δz ⎞
= (ρθ Δx Δy Δz ) = ρ⎜ + + ⎟ Δx Δy Δz (7.14)
∂t ∂t ⎝ ρ ∂t ∂t Δz ∂t ⎠
Compression of a porous medium and water can be considered using the following
equations, respectively:
1 ∂Δz ∂p
=α (7.15)
Δz ∂t ∂t
1 ∂ρ ∂ρ
=β (7.16)
ρ ∂t ∂t
where α is the elastic compressibility coefficient of porous medium, p is the water

pressure, and β is the compressibility coefficient of water. Substitution of these
equations into Eq. (7.14) gives:
∂M ⎛ ∂p ∂θ ⎞
= ρ⎜ θ(α + β) + ⎟ Δx Δy Δz (7.17)
∂t ⎝ ∂t ∂t ⎠
Therefore, using Eqs. (7.10) and (7.13), the continuity equation can be derived as:
⎛ ∂p ∂θ ⎞ ⎛ ∂ρqx ∂ρqy ∂ρqz ⎞

ρ⎜ θ(α + β) + ⎟ = −⎜ + + ⎟ (7.18)
⎝ ∂t ∂t ⎠ ⎝ ∂x ∂y ∂z ⎠
The left side of this equation describes the change in volume of water in a porous
medium due to a change in water content or compression of the water and the
medium.
Example 7.2
Develop the steady-state flow equation in an aquifer that has the following hydraulic
conductivity function:
K = K0 ( a + becx + de fy )
where K0, a, b, c, d, and f are parameters of the system.

Solution: The continuity equation (Eq. (7.18)) in the steady-state condition can
be simplified as follows:
∂qx ∂qy
+ =0
∂x ∂y
Applying Darcy’s law, we have:
∂ ∂h ∂ ∂h
(K ) + (K ) = 0
∂x ∂x ∂y ∂y
or,
∂ ∂h ∂ ∂h
( K ( a + becx + de fy ) ) + ( K0 ( a + becx + de fy ) ) = 0
∂x 0 ∂x ∂y ∂y
Finally,
∂2 h ∂h ∂2 h ∂h
( a + becx + de fy ) + cbecx + ( a + becx + de fy ) 2 + fdecx =0
∂x 2
∂x ∂y ∂y
7.3.2 EQUATION OF MOTION IN GROUNDWATER

The equation of motion can be derived using conservation of momentum. Consid-
ering the elementary control volume (Figure 7.3), the forces that usually act on the
water in the control volume are:
• Pressure forces
• Gravity forces
• Reaction forces of solids
For example, the pressure force on the left side of the control volume is as follows:
θp( x − Δx / 2, y, z, t )ΔyΔz (7.19)
The left side also has the same force but in the opposite direction; therefore, the
resulting pressure force component in the x direction is as follows (Delleur, 1999):
∂θp
− ΔxΔyΔz (7.20)
∂x
The other components of the pressure force on the control volume in the y and z
directions can be obtained using a similar method.
The gravity force, which is equal to the total weight of water in the control
volume and acts in a direction opposite of the z direction, is:
−ρgθΔxΔyΔz (7.21)
where g is the gravity constant. The reaction forces are usually defined as average
body forces per unit of water volume and consist of the friction forces due to water
movement and the forces that act against the water pressure. The friction and reaction
forces are denoted as r = (rx, ry, rz) and f = (fx, fy, fz), respectively. The x component
of these forces is as follows:
(rx + fx )θΔxΔyΔz (7.22)
It can have similar components in the y and z directions.

Using the del operator ∇ = (∂/∂x, ∂/∂y, ∂/∂z), the effects of all forces can be
combined in one vector as follows:
[−∇(θp) − ρgθ∇z + (r + f )θ]ΔxΔyΔz (7.23)

Considering that when the fluid is at rest, the friction f is zero and the water pressure
is hydrostatic, the overall reaction force can be evaluated as follows (Delleur, 1999):
p
r= ∇θ (7.24)
θ
In the case of groundwater motion, the sum of the forces equal to the changes
of fluid momentum and the friction force is not zero. As the groundwater flow is
generally very slow, the changes in momentum are negligible and the forces that
act on the fluid in the control volume are approximately in equilibrium:
−∇p − ρg∇z + f ≈ 0 (7.25)
The friction force can be presented as:
µ
f =− q (7.26)
k
where µ is the dynamic viscosity of the fluid, k is the intrinsic permeability, and q
is the groundwater flux. The equation of motion can be expressed using Eqs. (7.20)
and (7.21) as follows:
k
q=− (∇p + ρg∇z ) (7.27)
µ
When the gradients of density are negligible, the motion equation can be sim-
plified as:
kρg
q=− (∇ψ + z ) = − K∇h (7.28)
µ
where
∫ ρg
dp
ψ=
is the pressure potential and the other variables have been defined in previous
sections. The above equation clarifies the principles and assumptions that result in
Darcy’s law. One of the most important assumptions of Darcy’s law is that the flow
of water in a porous medium is slow and the large values of friction forces balance
the driving forces. In the anisotropic porous media, Darcy’s law in Cartesian coor-
dinates becomes:
∂h
qx = − K x (7.29)
∂x
∂h
qy = − K y (7.30)
∂y
∂h
qz = − K z (7.31)
∂z
7.3.3 THE GROUNDWATER FLOW EQUATION

The groundwater equation can be derived by a combination of the continuity equation
and the equation of motion as expressed in Eqs. (7.18) and (7.28), as follows.
∂p ∂θ ⎡k ⎤
θ(α + β) + = ∇ ⎢ (∇p + ρg∇z )⎥ (7.32)
∂t ∂t ⎣µ ⎦
This equation can be written as follows:
∂p ⎡k ⎤
S = ∇ ⎢ (∇p + ρg∇z )⎥ (7.33)
∂t ⎣µ ⎦
where S is the storage coefficient and is equal to:
∂θ
S = θ(α + β) + (7.34)
∂p
S is related to water and soil characteristics such as saturated and unsaturated

conditions of soil and the water pressure. Therefore, the three-dimensional equation
of groundwater movement as a function of water pressure (not as a function of
groundwater potential) can be derived as follows:
∂p ∂ ⎛ kx ∂p ⎞ ∂ ⎛ ky ∂p ⎞ ∂ ⎛ kz ⎛ ∂p ⎞⎞
S = ⎜ ⎟+ ⎜ ⎟+ ⎜ ⎜ + ρg⎟ ⎟ (7.35)
∂t ∂x ⎝ µ ∂x ⎠ ∂y ⎝ µ ∂y ⎠ ∂z ⎝ µ ⎝ ∂z ⎠⎠
The general form of the equation of groundwater flow is usually simplified in

practice. When the density effects can be ignored, Eq. (7.32) is simplified as follows:
∂p ∂θ
θ(α + β) + = ∇.( K∇h) (7.36)
∂t ∂t
In groundwater, the density is usually considered to be constant; therefore, the

temporal variation of water pressure and groundwater potential are related as follows:
∂p ∂h
= ρg (7.37)
∂t ∂t
Therefore, the following basic groundwater flow equation is derived:
∂h ∂θ ∂h
ρgθ(α + β) + = Ss = ∇.( K∇h) (7.38)
∂t ∂t ∂t
where Ss is the specific storage coefficient. Written in Cartesian coordinates, the

equation of saturated groundwater flow becomes:
∂h ∂ ⎛ ∂h ⎞ ∂ ⎛ ∂h ⎞ ∂ ⎛ ∂h ⎞
Ss = ⎜K ⎟ + ⎜K ⎟ + ⎜K ⎟ (7.39)
∂t ∂x ⎝ x ∂x ⎠ ∂y ⎝ y ∂y ⎠ ∂z ⎝ z ∂z ⎠
In the steady-state condition, Eq. (7.39) can be simplified as follows:
∂ ⎛ ∂h ⎞ ∂ ⎛ ∂h ⎞ ∂ ⎛ ∂h ⎞
⎜K ⎟ + ⎜K ⎟ + ⎜K ⎟ =0 (7.40)
∂x ⎝ x ∂x ⎠ ∂y ⎝ y ∂y ⎠ ∂z ⎝ z ∂z ⎠
This equation shows that the difference in groundwater potential causes the move-
ment of water in porous media and the fluxes depend on the hydraulic conductivity
of the medium in different directions.
Example 7.3
Develop the groundwater flow equations for a two-dimensional, horizontal, semi-

confined or leaky aquifer that has the thickness b(x, y) at each point (x, y).
Solution: The average value of hydraulic head at each point is:
∫
1
h= h dz (7.41)
b
0
Using Eq. (7.39) and considering T = K · b, the flow equation is:
∂h ∂ ⎛ ∂h ⎞ ∂ ⎛ ∂h ⎞
Ss b = ⎜T ⎟ + ⎜T ⎟ + qz (b) (7.42)
∂t ∂x ⎝ x ∂x ⎠ ∂y ⎝ y ∂y ⎠
where qz(b) is the vertical recharge or discharge, such as leakage, pumping, or

injection. The vertical leakage from the upper semiconfining layer can be calculated
using Darcy’s law as follows:
Ka ( Ha − h )
qz ( b ) = (7.43)
ba
where Ka is the hydraulic conductivity of the upper confining layer, Ha is the head
in the upper boundary of the confining layer, and ba is the thickness of the semi-
confined layer. Considering the point injection and pumping wells in the system,
the flow equation can be written as follows (Hantush and Jacob, 1955; Willis and
Yeh, 1987):
∂ ⎛ ∂h ⎞ ∂ ⎛ ∂h ⎞ Ka ( Ha − h )
Ss b
∂h
= ⎜T ⎟ + ⎜T ⎟+
∂t ∂x ⎝ x ∂x ⎠ ∂y ⎝ y ∂y ⎠ ba
±
w∈W
Qwδ w∑ (7.44)
where –Qw is the discharge (+Qw is the recharge) from pumping (injection) well w,
and δ is a 0/1 integer variable. The variable δ = 1 for pumping or injection at well
site w.
7.4 GROUNDWATER MODELING

Prediction of subsurface flow, water table level, solute transport, and simulation of
natural or human-induced stresses is necessary for groundwater management.
Groundwater simulation models have been widely used in groundwater systems
analysis and management. These models generally require the solution of partial
differential equation. Analytical models can yield the exact solutions for some simple
or idealized problems. Numerical models provide approximate solutions by discret-
ization of time and space, and because they do not require the rigid idealized
conditions of analytical models they are more flexible. In this section, some basic
aspects of numerical groundwater modeling such as principles of numerical methods,
grid generation, and model calibration and validation, are discussed.
7.4.1 THE FLOW NET

In a two-dimensional cross-section of a porous medium, the set of equipotential
lines and flow lines is referred to as the flow net. Figure 7.4 shows the orthogonal
network of flow at a point. In this figure, ΔS denotes the distance between two
adjacent equipotential lines and ΔW is the distance between a pair of adjacent
streamlines. Flow nets can be used in graphical modeling of flow in porous media.
FIGURE 7.4 Flow at a point.
Delleur (1999) recommended the following procedure for drawing the flow net:
1. The boundaries of the flow region should be drawn to a scale such that
the streamlines and equipotential lines terminate on the figure.
2. Three or four streamlines from infinite number of possible curves should
be sketched. They must provide a smooth transition between the boundary
streamlines.
3. The equivalent lines that intersect all streamlines at a right angle are
drawn; therefore, the enclosed figures must form curve-linear rectangles
that have the same ratio of ΔW/ΔS along the flow channel.
Most of the analytical and graphical methods for solution of two-dimensional

groundwater problems are concerned with the determination of a suitable function
that can transform a problem from a geometrical domain into a domain with a more
straightforward solution algorithm. A transformation that possesses the property of
preserving angles of intersections and the approximate image of small shapes is
conformal mapping. Harr (1962) presented various functions and discussed the
manner in which these functions transform geometric figures from one complex plan
to another. He also considered the more general problem of finding a functional
relationship that will provide a specific transformation.
Example 7.4
As shown in Figure 7.5, a dam is constructed on a permeable stratum underlain by

an impermeable rock. A row of sheet pile is installed at the upstream face. The
hydraulic conductivity of the soil is 4.5 m/day. Determine (a) the rate of flow from
upstream to downstream, and (b) the uplift pressure acting in point H.
Solution: The flow net is drawn as shown in Figure 7.5. In this figure, the number
of flow channels (nf) is equal to 5 and the number of equipotential drops (nd) is equal
to 17.
35m 5m
17 A B C D E F G H I
16
60m
1
15
14 13 12 11 10 9 8 7 6 5 4 3 2
100m
FIGURE 7.5 Net flow for a dam with sheet pile.
The flow rate or seepage per unit can be calculated using the following equation:
nf
q= Kh (7.45)
nd
where h is the total head loss in the flow system:
5 m3
q= ( 4.5)(30) = 39.7
17 day
The uplift pressure at each point can be calculated using following equation:
⎛ n ⎞
u = ⎜ h + z⎟ γ w (7.46)
⎝ nd ⎠
where n is the number of equipotential lines, counting the last line on the downstream
as 0; z is the depth of base below the datum (if the base is above the datum, z is
negative), and γw is the specific weight of water.
u = ⎛ × 30 + 2⎞ × 1 = 7.29 2
3 kN
⎝ 17 ⎠ m
7.4.2 NUMERICAL METHODS

Numerical modeling of groundwater was not extensively pursued until the mid-
1960s. Since that time, numerical models have been developed and widely used in
groundwater resources management. Numerical methods transform partial differen-
tial equations of groundwater into a set of ordinary differential or algebraic equations.
State variables at discrete nodal points are determined by solving these equations.
FIGURE 7.6 Finite-difference grid.
Two major classes of numerical methods are acceptable for solving groundwater
problems: finite difference and finite element. In both methods, the study area is
discretized by grid generation into a number of subareas called elements or cells.
Figure 7.6 shows a simple discretized aquifer.
In the finite-difference method, the first derivatives in partial equations are
approximated by the difference between values of independent variables at adjacent
nodes, considering the distance between the nodes, and at two successive time levels,
considering the duration of the time-step increment. In finite-element methods,
functions of dependent variables and parameters are used to evaluate equivalent
integral formulation of partial differential equations (Delleur, 1999). Although each
approach has some advantages and disadvantages, the finite-difference methods are
generally easier to program because of their conceptual and mathematical simplicity.
A major advantage of the finite-element method is the flexibility of this method
in close spatial approximation of irregular boundaries of aquifer and parameters of
zones within the aquifer. However, mesh generation and specification, and construc-
tion of input data sets for an irregular finite-element grid is much more difficult in
comparison with a regular rectangular finite-difference grid.
Figure 7.7a shows a hypothetical aquifer with impermeable boundaries and a
water supply well. This aquifer can be discretized using finite-difference (Figure
7.7b) and finite-element (Figure 7.7c) grids. Comparing Figures 7.7b and 7.7c shows
that finer mesh can be allocated to the areas of interest to obtain a more precise
solution. As shown in Figure 7.7b, in the finite-difference method with a rectangular
grid, the boundaries are approximated in a stepwise manner, but finite-difference
grids using triangular elements can closely follow the aquifer boundary.
7.4.2.1 Finite-Difference Numerical Model Basics
Finite-difference models of groundwater flow approximate the derivatives of con-

tinuous variables in time and space by discrete linear changes over discrete intervals
Inactive cell
FIGURE 7.7 Application of finite-difference and finite-element grids for an irregularly

bounded aquifer with a well field.
of time or space. For example, forward, backward, and central first-derivative approx-
imations of head h are:
∂h
∂x x
≈
h( x + Δx ) − h( x )
Δx
(
+ O ( Δx )
2
) (7.47)
∂h
∂x x
≈
h( x ) − h( x − Δx )
Δx
(
+ O ( Δx )
2
) (7.48)
∂h
∂x x
≈
h( x + Δx ) − h( x − Δx )
2 Δx
+ O ( Δx )(
3
) (7.49)
( )
where O ( Δx ) is the truncation error with an order on n.
n
∂2 h
Similarly, a reasonable approximation for the second derivative can be
given as: ∂x 2
∂2 h
∂x 2 x
≈
h( x + Δx ) − 2h( x ) + h( x − Δx )
Δx 2
+ O ( Δx )
4
( ) (7.50)
The errors of these approximations (truncation errors) will generally be decreased

when the values of Δx and Δt are decreased.
Similar discretization should be considered for time intervals; therefore, the
approximations of ∂h/∂t using the forward and backward differences are:
∂h ht +1 − ht
= (7.51)
∂t t
Δt
∂h ht − ht −1
= (7.52)
∂t t Δt
Known head
FIGURE 7.8 (a) Explicit (forward difference) and (b) implicit (backward difference) discret-
ization of the time at node (i,j) in a two-dimensional finite-difference grid.
Figure 7.8 shows time and space discretization at node (i, j) in a two-dimensional,
finite-difference grid and illustrates the application of two simple finite-difference
methods for solving the groundwater flow equation. As shown in Figure 7.8a, all
values of head h are known at all spatial nodes at time level n. The head at point (i,
j) at time step (n + 1) can be calculated using the value of h at time step n and the
forward-difference time derivation. Therefore, a finite-difference equation exists for
each node at time step (n + 1) with only one unknown variable.
For example, in a two-dimensional groundwater flow equation for a heteroge-
neous, anisotropic aquifer, Eq. (7.39) can be written as:
hin, +j 1 − hin, j hin−1, j − hin, j hin+1, j − hin, j

Ss ( ) = Kx (i− 1 ( ) + Kx (i+ 1 ( )+
Δt 2, j) ( Δx )2 , j)
2 ( Δx )2
hin, j −1 − hin, j hin, j +1 − hin, j
Ky(i, j − 1 ) ( ) + Ky(i, j + 1 ) ( ) (7.53)
2 ( Δy)2 2 ( Δy)2
In this equation, only hin, +1

j is unknown and can be solved explicitly. Explicit finite-
difference equations are simple to solve but, when time increments are too large,
small numerical errors can result in larger errors in the next computational stages.
Figure 7.8b shows the time derivative as a backward difference from the head
at time level (n – 1), which are the known heads. Therefore, the difference equation
of each node will have five unknown variables, and for a grid that has N nodes a
system of N equations contains N unknown variables. This system of equations can
be solved simultaneously considering the boundary conditions.
The implicit finite-difference form of the two-dimensional groundwater equation

(Eq. (7.39)) is:
hin, j − hin, −j 1 hin−1, j − hin, j hin+1, j − hin, j

Ss ( ) = Kx (i− 1 ( ) + Kx (i+ 1 ( )+
Δt 2, j) ( Δx )2 2, j) ( Δx )2
(7.54)
hin, j −1 − hin, j hin, j +1 − hin, j
Ky(i, j − 1 ) ( )Ky(i, j + 1 ) ( )
2 ( Δy)2 2 ( Δy)2
Example 7.5
Develop the finite-difference approximation of the flow equation for a two-dimen-

sional, semiconfined, nonhomogeneous, anisotropic aquifer.
Solution: The flow equation in the semiconfined aquifer is given by Eq. (7.44).
The finite-difference approximation of this equation in a nonhomogeneous and
isotropic aquifer can be as follows:
1 1
{Txi +1/ 2, j (hi +1, j − hi, j ) − Txi −1/ 2, j (hi, j − hi −1, j )} + 2 {Tyi, j +1/ 2 (hi, j +1 − hi, j )
Δx 2
Δy
(7.55)
Qw,i, j K a,i, j ( Ha,i, j − hi, j )
−Tyi, j −1/ 2 (hi, j − hi, j −1 )} ± + = Si, j hi, j
Δx Δy ba
By developing Eq. (7.55) for all internal nodes (i, j; i = 1, …, n and j = 1, …, m)

of the domain, a set of linear equations can be expressed in the following form:
Ah• + Bh + g = 0 (7.56)
In these equations, which are referred to as dynamic response equations, h is a

column vector of unknown heads, where h = (h11, …, h1m, …, hn,1, …, hnm)T and h•
is the temporal derivate of vector h. The coefficient matrices A and B depend on
the flow and hydraulic properties of the porous medium. In Eq. (7.54), vector g
contains the rate of pumping and injections that can be the decision variables in the
groundwater planning models. Dynamic response equations can easily provide the
hydraulic head at the nodes as a linear function of initial condition and the pump-
ing/injection rate of well sites.
7.4.2.1.1 Alternating Direction Implicit (ADI) Method
Considering that the first space derivative of the head is calculated at time level
(n + 1) and the second at the current time level (n), the groundwater flow equation
can be written as follows:
hin−+11, j − 2hin, j+1 + hin++11, j hin, j −1 − 2hin, j + hin, j +1 n +1

S hi, j − hi, j qi, j
n n
+ = + (7.57)
Δx 2 Δy 2 T Δt T
Three unknown values exist at time level (n + 1) and other values at time level (n)
are known. Equation (7.57) can be rewritten as:
hin−+11, j + Bi hin, +j 1 + hin++11, j = Di (7.58)
where Bi and Di are constant and can be calculated as follows:
⎛ S Δx 2 ⎞
Bi = −⎜ 2 + ⎟ (7.59)
⎝ T Δt ⎠
n ⎛ S Δx 2 Δx 2 ⎞
n
Δx 2 2
Δx 2 qi, j
Di = − hin, j −1 − h ⎜ − ⎟ − h n
i , j +1 + Δx 2 (7.60)
Δy 2 i, j
⎝ T Δt Δy 2 ⎠ Δy 2 T
If Eq. (7.58) is applied for all nodes along the x-axis (jth column), we will have
(k – 2) simultaneous equations because the value of the head at the two outer nodes
is known from the boundary conditions (k is the number of nodes along the x-axis).
Therefore, by solving these equations it is possible to calculate the heads on a line
parallel to the x-axis. The coefficient matrix of this set of equations is tridiagonal.
In the next time step, ∂2h/∂y2 is approximated at time level (n + 2), while ∂2h/∂x2
is retained at the current level of (n + 1); thus (Gupta, 1989),
hin−+11, j − 2hin, j+1 + hin++11, j hin, j+−21 − 2hin, j+2 + hin, +j +21 n+2 n +1
S hi, j − hi, j qin, +j 1
+ = + (7.61)
Δx 2 Δy 2 T Δt T
Three unknown values exist at time level (n + 2) and other values at time level
(n + 1) are known. Equation (7.61) can be rewritten as follows:
hin, +j −21 + Bj hin, +j 2 + hin, j++21 = Dj (7.62)
where:
⎛ S Δy 2 ⎞
Bi = −⎜ 2 + ⎟ (7.63)
⎝ T Δt ⎠
Δy 2 n +1 ⎛ S Δy
2
2 Δy 2 ⎞ n +1 Δy
2 qin, +j 1 2
Di = − hin−+11, j − hi, j ⎜ − ⎟ − hi +1, j + Δy (7.64)
Δx 2 ⎝ T Δt Δx 2 ⎠ Δx 2 T
Applying Eq. (7.62) for all nodes along the y-axis (ith column), we will have (m – 2)
simultaneous equations because the value of the head at the two outer nodes is
known from the boundary conditions (m is the number of nodes along the y-axis).
Therefore, by solving (m – 2) equations it is possible to calculate the heads of a line
parallel to the y-axis.
300 m
Well 1
400 m
900 m
Well 2 800 m
600 m
300 m
1600 m
12 m K = 20 m/day
20 m S = 0.003
FIGURE 7.9 Plan of confined aquifer and its cross-section. (Example 7.6)
Example 7.6
Two fully penetrated wells in a confined aquifer system are shown in Figure 7.9.
Wells 1 and 2 pump at a rate of 90 L/sec and 60 L/sec, respectively. The initial
piezometric head is horizontal, in level with the stream. Determine the spatial
variation of the piezometric head after one day of pumping. The aquifer is bounded
by a stream on one side and impermeable boundaries on two sides. It is extended
semi-infinitely on the remaining side, but a width of 900 m is used for the flow
system under consideration.
Solution: The grid layout is shown in Figure 7.10. In the initial conditions, the
heads of all nodes are 20 m at t = 0, and boundary conditions are:
Recharge boundary:
h1,1 = h1,2 = h1,3 = h1,4 = 20 m

FIGURE 7.10 Finite-difference grid and its coordinates of Example 7.6.
Impermeable boundaries
h5,1 = h4,1 h2,4 = h2,3

h5,2 = h4,2 h3,4 = h3,3
h4,3 = h3,3 h4,4 = h4,3
h4,4 = h3,4
T = K × b = 20 × 12 = 240 m2/day
First time step: A time step Δt of 0.5 day is selected as the time interval.
(Maximum time step is selected based on stability requirement; see Gupta [1989]
for more details.) Because there is no pumping on the first row, piezometric heads
in this row will not change in the first time step. For row 2 (j = 2), we have:
S ( Δ x )2 0.003 ( 400)2
B2 = −(2 + × ) = −( 2 + × ) = −6.0
T Δt 240 0.5
B2 = B3 = B4 = −6.0
and
(400)2 ⎛ 0.003 ( 400 ) 2

400 2 ⎞ 400 2
D2 = − 20 × − 20 ⎜ × − 2 × ⎟ − 20 × = −80
(300)2 ⎝ 240 0.5 300 2 ⎠ 300 2
Since
0.06
q= = 0.5 × 10 −6 m / s = 0.0432 m / day
300 × 400
then
0.0432 ( 400) 2
D3 = −80 + = −51.2
240
D4 = D2 = −80
For node 2,
h10,2.5 + B2 h20,.25 + h30,.25 = D2
For node 3,
h20,.25 + B3 h30,.25 + h40,.25 = D3
For node 4,
h30,.25 + B4 h40,.25 + h50,.25 = D4
Substituting the B, D, and initial and boundary conditions, we have:
20 − 6 h20,.25 + h30,.25 = −80
h20,.25 − 6 h30,.25 + h40,.25 = −51.2
h30,.25 − 6 h40,.25 + h50,.25 = −80
or
⎡−6 1 0 ⎤ ⎡h20,.25 ⎤ ⎡ −100 ⎤

⎢1 ⎢ ⎥
⎢ −6 1 ⎥⎥ ⎢h30,.25 ⎥ = ⎢⎢−51.2 ⎥⎥
⎢⎣ 0 1 −5⎦⎥ ⎢⎣h40,.25 ⎥⎦ ⎢⎣ −80 ⎥⎦
Solving gives
h20,.25 = 19.15 m
h30,.25 = 14.89 m
h40,.25 = 18.97 m
FIGURE 7.11 Heads at the end of the first time step (at t = 0.5 day).
The value of the heads of nodes for row 3 is solved by a similar procedure. Row 4
is an impermeable boundary; therefore, it will have the same heads as in row 3. The
value of nodal heads after the first time step is shown in Figure 7.11.
Second time step: The results of the first time step are used as the initial heads
of the second time step. In this time step, computations are made column by column.
For column 2 (i = 2):
0.003 ( 400) 2
B1 = −(2 + × ) = −4.25
240 0.5
B1 = B2 = B3 = −4.25
and
(300)2 ⎛ 0.003 (300 ) 2

300 2 ⎞ 300 2
D1 = −20 × − 20 ⎜ × − 2 × ⎟ − 20 × = −45
(400)2 ⎝ 240 0.5 400 2 ⎠ 400 2
(300)2 ⎛ 0.003 (300 ) 2

300 2 ⎞ 300 2
D2 = −20 × 2 − 19.15⎜ × −2× 2⎟
− 14.89 × = −41.17
(400) ⎝ 240 0.5 400 ⎠ 400 2
Because
0.09
q= = 0.75 × 10 −6 m / s = 0.0648 m / day
300 × 400
then
(300)2 ⎛ 0.003 (300 ) 300 2 ⎞

2
D3 = −20 × 2 − 12.55 ⎜ × −2× ⎟

(400) ⎝ 240 0.5 400 2 ⎠
300 2 0.0648 (300) 2

−18.51 × + = −11.48
400 2 240
For node 1,
h21,0 + B1h21,1 + h21,2 = D1
As the aquifer in this direction extends beyond the grid boundary, the head
outside the boundary is included and is assumed to be equal to the initial head of
20 m.
For node 2,
h21,1 + B2 h22,2 + h21,3 = D2
For node 3,
h21,2 + B3 h21,3 + h21,4 = D3
Substituting the B, D, and initial and boundary conditions, we have:
20 − 4.25 h21,1 + h21,2 = −45
h21,1 − 4.25 h21,2 + h21,3 = −41.17
h21,2 − 4.25 h21,3 + h21,3 = −11.48
or
⎡−4.25 1 0 ⎤ ⎡ h21,1 ⎤ ⎡ −65 ⎤

⎢ 1 ⎢ ⎥
⎢ −4.25 1 ⎥⎥ ⎢h21,2 ⎥ = ⎢⎢−41.17⎥⎥
⎢⎣ 0 1 −3.25⎦⎥ ⎢⎣h21,3 ⎥⎦ ⎢⎣−11.48⎥⎦
FIGURE 7.12 Heads at the end of the second time step (at t = 1 day).
Solving gives
h21,1 = 19.10 m
h21,2 = 16.18 m
h21,3 = 8.51 m
The heads of the nodes in columns are solved by a similar procedure. The values
of the nodal heads after the second time step (after t = 1 day) are shown in
Figure 7.12.
7.4.2.2 Finite-Element Methods

The finite-element method is another numerical analysis technique for approximate
solutions to groundwater flow problems. The heterogeneity of porous media and
irregular boundary conditions can be handled by these methods, in which the domain
is discretized into a set of elements of different sizes or shapes. Several approaches
such as direct, weighted residual, and variational, have been proposed to formulate
the finite-element method for a problem.
In the weighted residual method, which is frequently used for groundwater
problems, the trial solution for the system of equations is written as a finite series.
If the equations of the groundwater system are expressed in terms of differential
operator L as follows:
L( h ) = 0 (7.65)
then the approximate solution in a one-dimensional problem can be written as

follows:
h ≈ hˆ( x, t ) = ∑ N ( x) h˜ (t)
i =1
i i (7.66)
where ĥ is the approximate solution, and Ni(x) are the basis or shape functions of
node i defined over the entire domain. Shape functions are defined based on assump-
tions regarding the spatial variation of the state variables in the domain; n is the
number of linearly independent shape functions, and h˜i (t ) are the unknown coeffi-
cients (heads) that are determined for each node of the finite-element grid. The
residual error, R, can be estimated by substituting the approximate solution ĥ into
R = L(hˆ ) ≠ 0 (7.67)
The unknown coefficients h̃i are determined by minimizing the residual error of the
approximate solution. This is accomplished by integrating the weighted error over
the domain (D) and setting it equal to zero as follows:
∫ W R dD = ∫ W L(hˆ)dD = 0
D
k
D
k k = 1, 2, …, n (7.68)
where Wk is the weighting function for node i. To determine the n state variables at
nodal points, the weighting function should be specified and the integral equation
is broken into n simple equations.
The type of weighted residual method is determined by the choice of the n
weighting functions. In the Galerkin method, the weighting functions are selected
to be identical to the shape functions. The modified integral equation is then given by:
∫N
D
k L(hˆ ) dD = 0 k = 1, 2, …, n (7.69)
The modified integral equation is straightforward to solve and can be converted

to n simultaneous algebraic equations. The following example illustrates the appli-
cation of Galerkin’s method to develop the dynamic response equation for a semi-
confined aquifer system. For a more comprehensive explanation of this method, see
Cooley (1992).
Example 7.7
Develop the dynamic response equation for a two-dimensional semiconfined, non-

homogeneous, anisotropic aquifer using the Galerkin procedure.
Solution: Equation (7.69) can be rewritten considering all elements of the system
as follows:
∑∫ N
e De
k L(hˆ ) dDe = 0 ∀k (7.70)
where De is the elemental domain and L(hˆ ) can be represented by Eq. (7.44).
Considering the equation hˆ = Nh˜ , Eq. (7.70) can be rewritten as follows:
∑∫ N
e De
k L( Nh˜ ) dDe = 0 ∀k (7.71)
where h̃ is the vector of unknown nodal head values and N is global shape function
which is a matrix of the nodal shape function. By substituting Eq. (7.44) in Eq.
(7.71), we have:
⎧ ∂ ⎫
∑ ∫ ⎨⎩∇.T∇( Nh˜ ) + b
Ka Ka ˜
Ha − Nh − S ( Nh˜ ) ± Q⎬N k dDe = 0 ∀k (7.72)
e a ba ∂t ⎭
De
where Q indicates the magnitude of the source or sink. For an element, each term
of this integral equation produces an elemental matrix that is dependent on the
system parameters and shape (basis) functions. For example, the fourth term for
each element can be written as follows:
−SAe h˜ • (7.73)
where h˜ = [h˜i • , h˜ j • ,..., h˜k • ]T , and matrix Ae is:

•
∫ ∫ N N dD ∫ N N dD ⎥⎥
⎡ N 2 dD ⋯ ⎤
⎢ i e i j e i k e
⎢
Ae = ⎢
⎢
∫
N j Ni dDe
∫ N dD 2
j e ⋯ ⎥
⎥
(7.74)
⎢ ⋮ ⋮ ⎥
⎢⎣ ∫
⎢ N N dD
k i e ∫ N N dD
k j e ⋯
∫ N k2 dDe ⎥
⎥⎦
Combining the various matrix terms and dropping the ~ notation, Eq. (7.72) can be
expressed as follows (Willis and Yeh, 1987):
Ah• + Bh + g = 0 (7.75)
The initial condition can be h(0) = g(x).

For more information about finite-element methods such as global matrix gen-
eration, formulation of general basis functions, and solution methods for dynamic
response equations, the reader is referred to Zienkiewicz (1971) and Willis and Yeh
(1987).
Example 7.8
Solve the differential equation of d 2 h / dx 2 − h = 0 using the Galerkin method and
considering the following conditions:
0≤ x≤3
h=0 m where x=0
h = 10 m where x=3
Solution: As shown in Figure 7.13, a set of N nodes and M elements is selected;

for each node, a partially linear-shaped function Ni(x) is defined. Therefore, the
approximate function ĥ (x, t) at node i is defined as follows:
h ≈ hˆ( x, t ) = ∑ N ( x)hˆ (t)

i =1
i i 0≤ x≤3
Considering the unknown variables (h1, h2, …, hN), Eq. (7.68) can be rewritten as
follows:
3
d 2 hˆ ˆ
∫
0
Wi (
dx 2
− h )dx = 0 i = 1, 2,…, N (7.76)
or
3
dWi dhˆ ⎡ dhˆ ⎤
∫
0
( ⋅
dx dx
+ Whˆ
i ) dx − ⎢Wi
⎣ dx ⎦
⎥=0 t = 1, 2,…, N (7.77)
FIGURE 7.13 General numbering of nodes and elements.

Equation (7.77) can be rewritten as follows using the Galerkin method (Wi = Ni and
ĥ = N · h):
Ah = g (7.78)
3
dNi dN j
Aij =
∫ ( dx ⋅ dx
0
+ Ni N j )dx 1 ≤ i, j ≤ N (7.79)
3
gi = ⎡⎢ Ni
dh ⎤
1≤ i ≤ N (7.80)
⎣ dx ⎥⎦ 0
If the above equation is used for each element e between nodes m and n, the
coefficients of the equation can be calculated using Eq. (7.71). Assuming X = x – xm,
the shape functions are as follows:
X
N m = N me = (7.81)
Le
he − X
Nn = Nne = (7.82)
Le
where Le = xn – xm. If node k is not related to element e, Nk will be zero for element e.
Therefore, the components of matrix A are calculated as follows:
Le
⎡ dNme dNne e⎤ −1 Le
A e
mn =Ae
nm =
∫
0
⎢ dx ⋅ dx + Nm Nn ⎥ dx = Le + 6
⎣
e
⎦
Le
⎡⎛ dN e ⎞ 2 2⎤
A e
=A = e
∫ ( ) 1 Le
⎢⎜ m ⎟ + Nne ⎥ dx = e +
⎢⎣⎝ dx ⎠
mm nn
0
⎥⎦ L 3
Matrix A is calculated from the above submatrices Ae, considering the position
of the element in Figure 7.14.
FIGURE 7.14 Numbering of nodes and elements for Example 7.8.

Assuming L1 = L2 = L3 = L, the matrices Ae (e = 1,2,3) and A can be calculated as:
⎡1+L −1 L
⎢L 3 + 0 0 ⎤⎥
L 6
⎢ −1 L 1 L ⎥
A1 = ⎢ + + 0 0⎥
⎢L0 6 L 3
0 0 0 ⎥⎥
⎢
⎢⎣ 0 0 0 0 ⎥⎦
⎡0 0 0 0⎤
⎢0 −1 L
0⎥
1 L
+ +
⎢ L 3 L 6 ⎥
A =⎢
2
−1 L 1 L ⎥
⎢0 + + 0⎥
⎢0 L 6 L 3
⎣ 0 0 0 ⎥⎦
⎡0 0 0 ⎤ 0
⎢0 0 0 ⎥ 0
⎢ 1 L −1 L ⎥
A = ⎢0
3
0 + + ⎥
⎢ L 3 L 6⎥
⎢0 −1 L 1 L⎥
0 + +
⎢⎣ L 6 L 3 ⎥⎦
⎡1+L −1 L ⎤
⎢L 3 + 0 0 ⎥
L 6
⎢ −1 L −1 L ⎥
2⎛ + ⎞
1 L
⎢ + + 0 ⎥
⎝ L 3⎠
A= ⎢ L 6 L 6 ⎥
−1 L −1 L ⎥
⎢ 0 2⎛ + ⎞
1 L
+ +
⎢ L 6 ⎝ L 3⎠ L 6⎥
⎢ −1 L 1 L⎥
⎢ 0 0 + + ⎥
⎣ L 6 L 3⎦
From Eq. (7.78) we have:
⎡ dhˆ ⎤
⎡ h1 ⎤ ⎢− x =0 ⎥ ⎡0 ⎤
⎢h ⎥ ⎢ dx ⎥ ⎢0 ⎥
0
A⎢ ⎥ = ⎢ ⎥=⎢ ⎥
2
⎢ h3 ⎥ ⎢ 0 ⎥ ⎢0 ⎥
⎢ ⎥ ⎢ dhˆ ⎥ ⎢ ⎥
⎣h4 ⎦ ⎢ − x =3 ⎥
⎣0 ⎦
⎣ dx ⎦
h1 and h4 are known; therefore, from the above equation set we have:
2⎛ + ⎞ h2 + ⎛ − + ⎞ h3 = 0
1 L 1 L
⎝ L 3⎠ ⎝ L 6⎠
⎛ − 1 + L ⎞ h + 2⎛ 1 + L ⎞ h = −10⎛ − 1 + L ⎞
⎝ L 6⎠ 2 ⎝ L 3⎠ 3 ⎝ L 6⎠
If L = 1, the results are as follows:
h2 = 1.08 m and h3 = 3.46 m
7.4.2.3 Grid Design
The finest mesh spacing and time steps should be selected to reduce numerical errors,
while computational time and limitations on computer memory should also be
considered. In rectangular grid design, the length-to-width ratio (aspect ratio) of
cells or elements should be kept as close as possible to one in order to prevent
numerical instability or errors (Delleur, 1999). In the case of a triangular finite
element, the angles should be less than 22.5° (Torak, 1993). To decrease the com-
putational cost, variably spaced mesh is usually used with a fine grid in the following
areas:
• Where the greatest accuracy is desirable

• Where the gradients are steepest (stress points)
• Where the data are concentrated
It is generally advisable for the increasing factor for mesh spacing in two adjacent
elements or cells to be less than two (Delleur, 1999). Similarly, in transient simulation
different time steps can be used to optimize computational resources.
7.4.2.4 Parameter Estimation and Model Calibration

Groundwater simulation models require considerable data to define all nodal param-
eters. The accuracy of simulation depends on the reliability of the estimated param-
eters as well as accuracy of model and boundary conditions. As the parameters of
groundwater models cannot be measured directly, they can be estimated from his-
torical data using an inverse parameter estimation procedure (ReVell and McGarity,
1997).
In deterministic groundwater simulation models, the objective of the calibration
procedure is to minimize the difference between observed aquifer response data
(such as water table variation) and corresponding values calculated by the model.
The comparison between observed and calculated values is usually subjective and
it should be considered that a good match does not necessarily reflect the validity
of the model.
Model calibration is usually a time-consuming, trial-and-error procedure that

modifies the model output by changing the aquifer recharges and discharges, aquifer
properties, and initial and boundary conditions. In this case, the uncertainties in
sources, sinks, and initial and boundary conditions, as well as the uncertainties in
aquifer properties, should be evaluated and considered in the calibration procedure.
Therefore, the experience and engineering judgment of the modeler are important
factors in efficient calibration of the model.
The efficiency of model calibration can be improved by automated parameter
estimation techniques that use the least-square deviation as a criterion to obtain the
estimate of system parameters.
Yeh (1992) classified the solution algorithms of parameter identification in
groundwater systems as being part of either the output error criterion method or the
equation error criterion method. In the equation error criterion method, finite-
element or finite-difference approximation of the groundwater equation is used to
obtain a set of algebraic equations in terms of unknown parameters. The unknown
parameters are determined by minimizing the sum of squares of errors for the
equation, using the observed or interpolated data at each node of the domain (ReVelle
and McGarithy, 1997). The output error criterion method is an optimization problem
that minimizes a given output error criterion during identification of the flow model
parameters.
7.4.2.5 Model Verification
The calibrated model should be verified before using it for groundwater simulation.
In this case, the observed data that have not been used in the model calibration
process are compared with the results of simulation model to evaluate the reliability
and accuracy of model prediction. As the observed data from groundwater systems
are usually limited, the model verification process usually suffers from data defi-
ciency. In this case, the unverified model should be applied only if the sensitivity
analyses for both model calibration and prediction are performed (ReVelle and
McGarity, 1997).
7.4.2.6 Predictions and Post Audits
The reliability of a calibrated model can be evaluated by the accuracy of the model
predictions. Future stresses and their uncertainties are the major sources of prediction
error. For example, boundary conditions and processes that are insignificant under
current and past stress regimes may become important when existing stresses are
increased or new stresses are added. Therefore, if a model is continuously used for
prediction of future conditions of a groundwater system, the model should be
periodically audited or calibrated using the field monitoring data (Delleur, 1999). In
this way, the model can incorporate new information such as changes in assumed
groundwater stresses.
7.5 MANAGEMENT TOOLS

7.5.1 GROUNDWATER SIMULATION MODELS
7.5.1.1 Graphic Groundwater
Graphic Groundwater is a three-dimensional groundwater model written in the C++
programming language. This is a simplified version of the U.S. Geological Survey
Three-Dimensional Modular Finite Difference Groundwater Flow Model (MOD-
FLOW; see McDonald and Harbaugh, 1988), and enhanced data input and display
features have been added (Esling et al., 1993). This program is designed to simplify
model development and data input for groundwater modeling to develop maps and
diagrams within Graphic Groundwater and to provide graphics and text files that
can be processed outside of Graphic Groundwater.
The first step toward creating a model is the development of a grid system, then
layers are added and parameters of hydraulic conductivity, initial head, storage
coefficient, top elevation, and bottom elevation are specified for each cell. The
number of stress periods over which the model will run and the time of their operation
should be set.
Results could be analyzed through contour plots, if desired. Superimposing a
grid over a graphic image or digitized line drawing is one of the most powerful
features of Graphic Groundwater. These images are generally two-dimensional plan
views representing the area that is going to be modeled.
Graphic Groundwater solves the same system of equations as MODFLOW, but
the discretization convention is slightly different. Graphic Groundwater requires the
top and bottom elevation of a cell in both confined and unconfined aquifers. It
calculates cell thickness from this information. The thickness is not implied in
various hydrologic parameters, thus Graphic Groundwater accommodates only the
vertical discretization scheme between two end members such that cell boundaries
are coincident with hydrologic boundaries.
7.5.1.2 MODFLOW
MODFLOW is a three-dimensional, finite-difference groundwater model that was
first released in 1984. It has a modular structure that allows the user to modify it
and adapt the code for a particular application. Many new capabilities have been
added to the original model in recent versions. MODFLOW-2000 simulates steady
and unsteady flow in an irregularly shaped flow system in which aquifer layers can
be confined, unconfined, or a combination of confined and unconfined. Flow from
external stresses, such as flow to wells, surface recharge, evapotranspiration, flow
to drains, and flow through river beds, can be simulated. Hydraulic conductivities
or transmissivities for any layer may differ spatially and be anisotropic (restricted
to having the principal directions aligned with the grid axes), and the storage
coefficient may be heterogeneous. Specified head and flux boundaries can be sim-
ulated by this model. It is also capable of modeling head-dependent flux across the
outer boundary of the model, which allows water to be supplied to a boundary block
in the modeled area at a rate proportional to the current head difference between a
source of water outside the modeled area and the boundary block.
In addition to simulating groundwater flow, the scope of MODFLOW-2000 has
been expanded to incorporate related capabilities such as solute transport and param-
eter estimation. In this model, the groundwater flow equation is solved using the
finite-difference approximation. The flow region is subdivided into blocks in which
the medium properties are assumed to be uniform. In plan view, the blocks are made
from a grid of mutually perpendicular lines that may be variably spaced. Model
layers can have varying thickness. A flow equation is written for each block, or cell.
Several methods are provided for solving the resulting matrix problem; the user can
choose the best one for a particular problem. Flow rate and cumulative volume,
which are balanced from each type of inflow and outflow, are computed for each
time step.
7.5.2 SAFE YIELD ESTIMATION

In groundwater management, one of the main goals is to evaluate the maximum
annual groundwater yield that can be withdrawn without producing undesirable
effects. The concept of safe yield was first defined by Lee (1915) as “the limit to
the quantity of water which can be withdrawn regularly and permanently without
depletion of storage reserves.” Early in this century, “undesirable effects” meant
violation of water laws or adverse effects on the supplies of adjacent landowners
using the same aquifer due to overdraft of the aquifer, but today undesirable effects
include any adverse effects on ecosystems. The simple concern of overdraft can be
approached using the hydrologic equilibrium equation as follows:
I = O − ΔS (7.83)
where I is inflow, O is the outflow, and ΔS is the change in storage. Inflow is the
summation of surface and subsurface inflow, precipitation, and imported water.
Outflow is the summation of surface and subsurface outflow, consumptive use, and
exported water (Todd, 1980).
To determine the safe yield in a basin, calculation of subsurface inflow and
outflow is usually the most difficult task because it is impossible to measure them
directly. Another complicating factor is that the definitions of adverse effects are
subjective and subject to temporal and spatial interpretations. Acceptable definitions
of adverse effects can change in the future due to changes in demographics, land
use, and political and social conditions.
7.5.3 OPTIMIZATION MODELS FOR GROUNDWATER MANAGEMENT

The optimal water supply to municipal, industrial, and agricultural water demands
considering the physical, socioeconomic, and environmental constraints are the
major objective of groundwater planning and management. The following planning

problems are associated with groundwater supply management (Yeh, 1992):
• Determination of an optimal pumping pattern, such as location of pumping

wells and their pumping rate, to satisfy water demands
• Timing and staging of well system development (capacity expansion) with
respect to future water demands
• Design of water transfer facilities for optimal allocation and distribution
of water to demand points in the basin
This section presents some typical optimization models for groundwater operation
management, capacity expansion, and joint use of surface and groundwater
resources.
7.5.3.1 Optimization Model for Groundwater Operation
A groundwater operation model can provide the optimal policies for groundwater
operation such as optimal extraction, allocation, and optimal recharge of the aquifer
considering physical, environmental, and socioeconomic constraints. An objective
function of the optimization model could be to minimize the total discounted oper-
ational costs as follows:
T t
Min Z = ∑t =1
⎛ 1 ⎞
⎝1+ α⎠ ∑ C (Q , h )
w∈W
w
t
w
t
w (7.84)
where:
T is the number of time steps in the planning horizon.

α is the interest rate.
W is the set of well sites.
Cw (Qwt , hwt ) is the water extraction cost from well site w and is a nonlinear
function of discharge Qwt and head hwt at each time step t.
This objective function should be minimized taking into consideration groundwater

response equations and constraints of the system.
The response equations for each planning period in a linear, distributed parameter
groundwater system can be expressed as follows (Willis and Yeh, 1987):
h t − A1h t −1 − A2 g(Q t −1 ) = 0 ∀t (7.85)
where ht and ht–1 are vectors of groundwater head at time steps t and t – 1, A1 and
A2 are coefficient matrices, and vector Q t–1 is defined as follows:
Q t −1 = (Q1t −1 , Q2t −1 ,..., Qwt−1 ) T ∀t, w ∈ W (7.86)

The following constraints can be considered in a groundwater operation optimization

model:
• Supplying demands
∑Q
w∈W
t
w ≥ Dt ∀t (7.87)
where Dt is the total water demand at time step t.
• Discharge capacity for each well
0 ≤ Qwt ≤ Qwmax ∀t, w ∈ W (7.88)
where Qwmax is the discharge capacity of well w.
• Lower bound constraints on the head levels
hwt ≥ hwmin ∀t, w ∈ W (7.89)
where hwmin is the minimum groundwater head at well site w. These con-
straints are considered to ensure that excessive depletion of the aquifer
does not occur.
• Non-negativity of the decision variables
h t , Qwt ≥ 0 ∀t, w ∈ W (7.90)
The above planning model is the basic framework for groundwater systems planning
and management.
7.5.3.2 Optimization Model for Capacity Expansion
Capacity expansion and the timing and staging of well field development are typical
problems in groundwater systems planning. In the problems presented here, it is
assumed that the well sites have been developed and operation policies are developed
for existing conditions; therefore, capital cost can be neglected when developing the
operation policies and should be considered in capacity expansion models. The
objective function of the capacity expansion model — minimization of total dis-
counted capital and operational cost — is as follows:
T t
Min Z = ∑
t =1
⎛ 1 ⎞
⎝1+ α⎠ ∑ (KC(Q
w∈W
max
w ( ) )
) Xwt + Cw Qwt , hwt Ywt (7.91)
where:

KC(Qwmax ) is the capital cost of well site w, which is considered as a function
of well capacity (Qwmax ) and Cw (Qwt , hwt ) is the water extraction costs from
well site w, which is a nonlinear function of discharge Qwt and head hwt at
each time step t.
Xwt and Ywt are 0/1 integer variables. Xwt indicates whether or not well site w
is developed in period t. Xwt is equal to 1 if well site w is developed in
time period t and equal to 0 otherwise. Similarly, Ywt is equal to 1 if well
site w is in operation and equal to 0 otherwise.
This objective function should be minimized taking into consideration ground-

water response equations and constraints of the system:
• Response equations for each planning period
ht − A1ht −1 − A2 g(Qt −1 ) = 0 ∀t (7.92)
• Supplying water demands
∑Y Q
w∈W
t
w
t
w ≥ Dt ∀t (7.93)
where Dt is the total water demand in time step t.
• Constraint of well capacity for each well site w
0 ≤ Qwt ≤ Qwmax ∀t, w ∈ W (7.94)
• Lower bound constraints on head levels
hwt ≥ hwmin ∀t, w ∈ W (7.95)

where hwmin is the minimum groundwater head at well site w. These

constraints are considered to ensure that excessive depletion of the aquifer
does not occur.
• 0/1 integer variables
Xwt , Ywt = [0, 1] (7.96)
• The maximum number of developed wells over the planning horizon is

assumed to be 1; therefore,
∑X
t =1
t
w ≤1 w ∈W (7.97)
• Limitations on total capital and operational costs at each planning period
∑ (KC(Q
w∈W
max
w ( ) )
) Xwt + Cw Qwt , hwt Ywt ≤ Ctotal ,t ∀t (7.98)
where Ctotal,t is the total capital and operational cost in planning period t.
ht , Qwt ≥ 0 ∀t, w ∈ W (7.99)
Because of uncertainties regarding future operational and capital cost functions

and projected water demands, the developed optimal policies should be revised when
more information becomes available.
7.5.3.3 Optimization Model for Water Allocation

The groundwater allocation problem provides the optimal water distribution for
various demands such as agricultural, industrial, and municipal, in a river basin with
minimal effect on the environment and water users. The objective function of this
model maximizes the net discounted benefit of the water supply as follows:
T t ⎧Qit ⎫
Max Z = ∑t =1
⎛ 1 ⎞
⎝1+ α⎠ ∑∫ ⎪
( )
⎨ Di Qi dQ −
⎪0
t
w∈W
∑ (
Cw Qwt , hwt )⎪⎬
⎪
(7.100)
i
⎩ ⎭
where:

Di (Qit ) is the groundwater aggregate demand function of subregion i, which
depends on Qit . The area beneath the demand function represents willing-
ness to pay for a specific quantity of water (Qit ) . In other words, this
function relates the price and quantity demanded of the water supply (Willis
and Yeh, 1987).
Qit is the water pumped from subregion i.
Cw (Qwt , hwt ) is the water extraction cost from well site w, which is a nonlinear
function of discharge Qwt and head hwt at each time step t.
This objective function should be maximized considering the groundwater response

equations and constraints of the system:
• Response equations for each planning period
ht − A1ht −1 − A2 g(Qt −1 ) = 0 ∀t (7.101)
• Constraint of water balance in each subregion
Qit = ∑Q
w∈W
t
w (7.102)
• Constraint of well capacity for each well site w
0 ≤ Qwt ≤ Qwmax ∀t, w ∈W (7.103)
ht , Qwt , Qit ≥ 0 ∀t, w ∈ W , ∀i (7.104)
The water allocation model can provide a schedule for optimal groundwater
extraction.
7.5.3.4 Optimization Model for Conjunctive Water Use
The conjunctive water use model provides an optimal allocation schedule for ground
and surface water resources in a region to meet different water demands. The
objective function maximizes the net discounted economic benefit as follows:
T t⎧ ⎡Qlt ⎤⎫
Max Z = ∑ ⎛ 1 ⎞ ⎪
∑∫ ( )
⎝1+ α⎠ ⎨
⎪
⎢ D Qt dQ −
⎢ l l ∑ ( ) ∑ ( )
t
gml t
Gml − fnlt Snlt ⎥ ⎬
⎥
⎪
(7.105)
t =1
⎩
l ⎢⎣ 0 m n ⎥⎦ ⎪⎭
where:
l, m, n are indexes that define different demands, groundwater resources, and

surface water resources, respectively, in the region.
Dl (Qlt ) is the demand function, which depends on the water allocated to
demand l (Qlt ) .
t t t
Ql and gml (Gml ) is the cost function of groundwater allocation from resource
m to demand l, which depends on the amount of groundwater allocated to
t
that demand, in period t, (Gml ).
t t
fnl ( Snl ) is the cost function of surface water allocation from resource n to
demand l, which depends on the amount of surface water allocated to that
demand, in period t, ( Snlt ).
The model is constrained by the response equations and physical limitations:
• Response equations for each groundwater resource
⎛ ⎞
hmt − A1m hmt-1 − A2m g ⎜
⎝
∑G m
t −1
ml ⎟ =0
⎠
(7.106)
• Surface water balance equation, assuming no hydraulic interaction

between the surface and groundwater systems:
t t −1
S n = S n + Rnt − ∑S l
t
nl (7.107)
t
where S n is the volume of water in surface storage n at the end of time
period t and Rnt is the inflow to surface storage n during period t.
• The balance equations relating the groundwater allocations and pumping

schedules
∑G = ∑Q
l
ml
w∈W
t
w ,m , ∀m (7.108)
where Qwt ,m is the pumping rate from well w in groundwater resource m,

and W defines the well sites of the region.
• Well capacity limitation
Qwt ,m ≤ Qwmax
,m , ∀m, ∀t, w ∈ W (7.109)
• Lower bound constraints on the head levels
hwt ≥ hwmin ∀t, w ∈ W (7.110)
These constraints are considered to ensure that excessive depletion of the

aquifer does not occur.
hmt , Qwt ,m , Qlt , Gml

t
, Snlt ≥ 0 w ∈W , ∀l, m, n, t (7.111)
This optimization problem can be solved using different methods such as nonlinear
or dynamic programming.
Example 7.9
Determine the average monthly allocation of groundwater and surface water to four
agricultural zones shown in Figure 7.15. As can be seen in this figure, a channel
can transfer surface water to zone 4. The water demands of zone 3 cannot be supplied
from the rivers. The monthly discharge of rivers and the volume of recharge by
precipitation during a 2-year period and the gross water demands of the agricultural
zones are presented in Tables 7.2 and 7.3, respectively. The equations of the average
groundwater table fluctuations, based on the discharge from agricultural wells and
recharge from precipitation, are presented in Table 7.4. In this table, Gt (i) is the
groundwater discharge from agricultural zone i in period t, Qt (1) and Qt (2) are the
allocated water to agricultural zones 1 and 2 in period t, Qt (3) is the surface water
transferred to zone (4) in period t, and P is the groundwater recharge from precip-
itation (million m3). The effects of other parameters such as underground inflow and
outflow on the fluctuations of groundwater table are considered to be negligible.
The acceptable (without any cost) range for cumulative groundwater table variation
in each zone is ± 4 meters. The maximum capacity of water transfer channel is 4
m3/sec, and it is assumed that 90 percent of transferred water is supplied by River
1. The initial groundwater table depths in zones 1, 2, 3, and 4 are considered to be
17, 8.1, 12.6, and 67 m, respectively. The objectives of the developed model are
supplying water to agricultural lands, minimizing pumping costs, and controlling
average groundwater table fluctuations in agricultural zones. Calculate the optimal
allocated water to agricultural zone, from surface and groundwater resources.
FIGURE 7.15 Components of surface and groundwater resources for Example 7.9.
Solution: The pumping cost can be considered as a linear function of pumping

power (Ppump) as follows:
G. H
Ppump = (7.112)
102η
where Ppump is the pump power (W), G is the pumping discharge (m3/sec), H is the
groundwater table depth (m), and η is the pump efficiency.
Because the objective function and the constraints are nonlinear, the dynamic
programming (DP) method can be used effectively in this study. The recursive
function of the DP model is developed as follows:
{
ft (Gt (i ), Qt (i )) = Min Ct (Gt (i ), Qt (i ), Ht −1 (i ), It , Qt )
(7.113)
}
+ ft −1* (Gt −1 (i ), Qt −1 (i ), Ht −2 (i ), It −2 , Qt −1 )
Groundwater Management
TABLE 7.2
Values of Monthly River Discharge and Precipitation
Recharge of Aquifer 1 Recharge of Aquifer 2
Water Discharge of River 1 Discharge of River 2 by Precipitation by Precipitation
Year Month (million m3) (million m3) (million m3) (million m3)
1 October 13.48 15.49 0.7 0.05

1 November 6.16 16.51 1.2 0.2
1 December 9.38 22.78 2 0.18
1 January 14.34 20.19 2.5 0.19
1 February 15.9 27.18 3 0.3
1 March 18.17 26.25 3 0.31
1 April 10.08 26.99 2.5 0.3
1 May 7.11 26.16 2 0.15
1 June 5.3 23.14 0.5 0.04
1 July 3.76 20.29 0 0.01
1 August 3.23 19.65 0 0.2
1 September 2.85 17.43 0.5 0.05
2 October 18.83 30.18 0.6 0.06
2 November 7.44 26.23 1.2 0.2
2 December 8.58 26.64 2.2 2.2
2 January 10.89 26.45 2.3 2.3
2 February 14.72 25.61 3.2 3.2
2 March 25.79 20.22 2.9 2.9
2 April 35.17 48.29 2.6 2.6
2 May 15.07 35.09 2 2
2 June 9.47 28.07 0.56 0.56
2 July 19.08 23.16 0.2 0.2
2 August 37.51 19.16 0 0
2 September 30.25 16.47 0.51 0.51
281
TABLE 7.3
Monthly Agricultural Water Demands and Downstream Water Rights
Downstream
Zone 1 Zone 2 Zone 3 Zone 4 Water Rights
Month (million m3) (million m3) (million m3) (million m3) (million m3)
October 8.536 10.07 0.64 3.01 1.86

November 8.494 8.59 0.48 3.82 3.43
December 3.45 2.57 0.2 1.41 1.045
January 2.448 2.03 0.153 1.24 0.965
February 2.371 1.79 0.133 0.81 0.37
March 4.5 3.33 0.353 1.32 0.54
April 11.44 9.75 1.176 6.15 6.92
May 34.64 31.76 3.44 20.58 28.57
June 36.6 35.22 4.02 19.12 27.6
July 19.45 18.83 2.53 4.93 8.92
August 23.9 25.86 2.66 4.97 5.78
September 21.31 23.49 1.68 4.26 3.76
TABLE 7.4
Equations of Average Groundwater Table Fluctuation
Zone 1 ΔL1 = 1.9 × 10 −4 G(1) + 1.2e −3G(2) − 2e −5G(3)2 + 7.8e −4G(3) − 0.02 P − 0.03
Zone 2 ΔL2 = 1 × 10 −6 G(1) + 3.7e −3 G(2) − 4e −5 G(3) 2 + 1.9e −4 G(3) − 0.018 P − 0.07
Zone 3 ΔL3 = 1 × 10 −7 G(1) + 1.1e −3 G(2) − 4.1e −4 G(3) − 0.01P − 0.02
Zone 4 ΔL4 = −0.0149G( 4) 2 + 0.41 G( 4) − 7e −5Q(3) 2 − 0.028Q(3) − 0.12
Note: Positive values represent the groundwater table drawdown.
where:
i is the index of zone i = 1, …, 4.

Gt (i) is the amount of groundwater extracted from zone i in month t (million
m3).
Qt (i) is the surface water allocated to zone i in month t (million m3).
Ct (Gt(i), Qt (i)) is the cost of operation during time period t.
ft*−1 (Gt −1 (i ), Qt −1 (i )) is the minimum total operational cost until the end of time
period t – 1.
Ht (i) is the depth of groundwater table in agricultural zone i at the end of
period t.
It /Ot is the total recharge/discharge of aquifers during the period t.
The cost of operation in each period is estimated as:
Ct (Gt (i ), Qt (i ), Ht −1 (i ), It , Ot ) = ∑ Loss (i)

i =1
t (7.114)
( )
2
Losst (i ) = α( Dt (i ) − Qt (i ) − Gt (i ))2 + β(Gt (i ) ⋅ Ht (i )) + γ Lt (i ) − Lmax (i ) (7.115)
α=0 if Dt (i ) ≤ (Qt (i ) + Gt (i )) (7.116)
γ =0 if Lt (i ) ≤ Lmax (i ) (7.117)
where:
Dt(i) is the agricultural water demand in zone i and time period t.

Lt(i) is the total variation of the water table level until the end of time period t.
Lmax(i) is the maximum allowable cumulative groundwater table fluctuation
in agricultural zone i.
Ht(i) is the depth of the groundwater table in agricultural zone i at the end
of month t.
α, β, γ are the weights of the objectives (constant).
The constraints of the model are as follows:
• Control of groundwater level fluctuations
Ht (i ) = H0 (i ) + ∑ ΔL (i)
t =1
t i = 1,..., 4 (7.118)
Lt (i ) = ∑ ΔL (i)
t =1
t i = 1,..., 4 (7.119)
ΔLt (i ) = Fi (Gt (i), Qt (i ), It , Ot ) (7.120)
where:
H0(i) is initial depth of groundwater table in agricultural zone i.

ΔLt(i) is the variation in water table level for time period t in agricultural
zone i (drawdown is considered to be positive).
Fi(Gt(i), Qt(i), It, Ot ) is groundwater table fluctuations in agricultural zone i
for time period t and are functions of total discharge (Ot ) and total recharge
(It ) of the aquifers.
• Estimation of surface water outflow from agricultural zones 1 and 2
ψ
Rt (i ) = It (i) − Qt (i ) − (1 − )Q (3) i = 1, 2 (7.121)
100 t
where ψ is the percent of transferred water to zone 4 from surface water

resources of zone 2. It(i) is the surface water inflow to zone i for time
period t (million m3).
• Minimum instream flow requirements downstream of agricultural zones
∑ R (i ) ≥ R
i =1
t t ,min (7.122)
where Rt,min is the minimum instream flow required in downstream in time

period t, and Rt(i) is the surface water outflow from zone i for time period
t (million m3).
• Maximum capacity of canal
Qt (3) ≤ Qt ,max (7.123)
Because the problem has multiple objectives, the coefficients α, β, and γ are the
relative weights of objectives which should be determined in order to make these
objectives comparable. The optimal average water allocated to agricultural zones in
July (a dry month) for α = 1, β = 0.005, and γ = 120 and the corresponding cumulative
groundwater table variations are presented in Table 7.5 and Figure 7.16 (for more
details, see Karamouz et al., 2002).
7.6 CONFLICT ISSUES IN GROUNDWATER SYSTEMS

The operation of groundwater systems is a multiple-objective problem, and some
economic, hydraulic, water quality, or environmental objectives are usually in con-
flict. For example:
• Several demand points exist, and the water supplied to one of these
demand points cannot be used by the others; therefore, a major conflict
issue in groundwater operation occurs when the groundwater storage is
not capable of supplying all of the demands.
• Extra discharge of the aquifer or high variation of the groundwater table
can cause problems, such as settlement of buildings and the ground surface.
• Recharge of the aquifer with polluted water, such as occurs due to infil-
tration of agricultural returnflows, disposal of sewage or water treatment
plant sludge, and sanitary landfills can produce environmental problems.
TABLE 7.5
Average Surface and Groundwater Allocated in July
Zone 1 Zone 2 Zone 3 Zone 4
Source (million m3) (million m3) (million m3) (million m3)
Surface water 5.8 15.6 — —

Groundwater 21.2 3.0 3.0 1.0
Transferred water — — — 3
40
Cumulative groundwater
table variation (cm)
20
0
-20
-40
-60
0 5 10 15 20
Month
FIGURE 7.16 Cumulative groundwater table variation in zone 1 for Example 7.9.
The optimal operation policies of groundwater systems can be developed using

conflict resolution models considering groundwater system hydraulic or water qual-
ity response equations and possible well capacity, hydraulic gradient, and water
demand requirements.
A common conflict issue in groundwater systems planning and operation occurs
when the aquifer should supply water to different demand points for different
purposes with constraints on groundwater table variations and groundwater quality.
Consider an aquifer system that supplies the following demands:

Even though water resources planning for operation of wells is done by a specific
agency on a regional basis (called here the Department of Water Supply), the water
allocation schemes defined and imposed by this department are greatly influenced
by other agencies, such as these generic examples:
FIGURE 7.17 Utility function for the range of average groundwater variation.
The utility function for each of these agencies should be considered when
formulating the conflict resolution problems; therefore, all of the conflict issues and
the responsible agencies should be recognized in the first step of conflict resolution
studies.
The next step is to define the utility function of each agency. For example, the
Department of Water Supply could define the utility function for a range of average
groundwater table variations, as shown in Figure 7.17, where the most favorable
range of average groundwater table variation is from 0 to lc m. This range can be
estimated based on the hydrogeologic conditions of an aquifer, as well as the
estimates of future demands. The decreasing segment on the right-hand side of this
graph also shows that the utility of the Department of Water Supply will decrease
when the groundwater table variation is increased.
All of the agencies selected in the first step should provide their favorable range
of water supply for various demands, groundwater quality, or water table fluctuation.
Besides the different priorities of these agencies, each agency has a specific level
of authority with respect to changing the water allocation schemes within the political
and institutional climate of each region. The final step is to formulate the conflict
resolution problem. For this purpose, various methods can be used, which are briefly
explained in Chapter 2.
Example 7.10
In an unconfined aquifer system, the following agencies are affected by the decision
to discharge from an aquifer to supply water demands:
• Agency 1: Department of Water Supply

• Agency 2: Department of Agriculture
• Agency 3: Department of Industries
• Agency 4: Department of Environmental Protection
• Agency 5: Department of Domestic Water Use
The Department of Water Supply has a twofold role — namely, to allocate water
to different purposes and control the groundwater table variations. The decision
makers in these agencies are asked to set their utility functions, which are shown in
Figures 7.18 to 7.20. The analyst should apply the weights presented in Table 7.6
to consider the roles and levels of authority of the agencies in the political climate
of that region.
The initial surface area, volume, and total dissolved solids (TDS) concentration
for the water in the aquifer are 600 km2, 1440 million m3, and 1250 mg/L, respec-
tively. The net underground inflow is 1100 million m3/yr, with a TDS concentration
equal to 1250 mg/l. It is assumed that 60% of the allocated water returns to the
aquifer as return flow and the average TDS concentration of the return flow is 2000
mg/L. The average storage coefficient of the aquifer is 0.06. Find the most appro-
priate water allocation scheme for this year using Nash’s bargaining theory.
Agricultural Demands
Domestic Demand
Industrial Demand
1.2
1
0.8
Utitity
0.6
0.4
0.2
0
0 100 200 300 400 500 600 700 800
Allocated Water (million m3 / year)
FIGURE 7.18 Utility function of different agencies for the range of water allocated to dif-
ferent demands.
1.2
1
0.8
Utility
0.6
0.4
0.2
0
0 1 2 3 4
Groundwater Table Variation (m)
FIGURE 7.19 Utility function of Agency 1 for the range of average groundwater variation.
1.5
1
Utility
0.5
0
0 500 1000 1500 2000
The average concentration of TDS in

groundwater (mg/l)
FIGURE 7.20 Utility function of Agency 4 for the average concentration of TDS in the
groundwater.
TABLE 7.6
Relative Weights or Relative Authority of Agencies
Relative Weight Relative Weight
Agencies Case 1 Case 2
Department of Agriculture 0.133 0.17

Department of Domestic Water Use 0.33 0.2
Department of Water Supply 0.2 0.23
Department of Environmental Protection 0.2 0.3
Department of Industries 0.133 0.1
Solution: The nonsymmetric Nash solution of problem is the unique optimal

solution of the following problem:
∏( f − d )
i =1
i i
wi
(7.124)
di ≤ fi ≤ fi∗ ∀i (7.125)
where Wi is the relative weight, fi is the utility function, di is the disagreement point,
and fi* is the ideal point of player (agency) i. This objective function should be
maximized considering the constraints of groundwater surface variation and ground-
water quality that are as follows:
1. The average groundwater table variation:
Δ l ≈ (q1 + q2 + q3 − IN ) × (1 / S) × (1 / A) (7.126)
where Δl is the average groundwater table variation (m) during the plan-
ning year, q1, q2, q3 are annual agricultural, domestic, and industrial
groundwater withdrawal (million m3), IN is net annual underground inflow
(million m3), S is storage coefficient, and A is the area of aquifer (km2).
Therefore,
(q1 + q2 + q3 − 1100)
Δl≈
0.06 × 600
2. The average groundwater quality:

Considering the TDS as the indicator water quality variable, the average
TDS concentration of groundwater at the end of the year can be simulated
as follows:
(Vin. × Cin. + IN × CIN + α1 × C1 × q1 + α 2 × C2 × q2 + α 3 × C3 × q3 − Cmean (q1 + q2 + q3 )) (7.127)

Cnew =
(Vin. + IN + (α1 − 1) × q1 + (α 2 − 1) × q2 + (α 3 − 1) × q3 )
and
Cmean = (Cin. + Cnew ) / 2 (7.128)
where:
Vin. and Cin. are the initial volume and TDS concentration of the groundwater,
respectively.
Cin is the TDS concentration of net groundwater inflow.
α1, α2, α3 are the return flows (%) of annual agricultural, domestic, and
industrial groundwater withdrawals, respectively.
C1, C2, C3 are the average TDS concentrations of agricultural, domestic, and
industrial groundwater withdrawal, respectively.
Cmean is the average TDS concentration of groundwater during the planning
year.
Cnew is the TDS concentration of the aquifer at the end of the year.
Therefore,
(1440 × 1250 + 1100 × 1250 + 0.6 × 2000 × q1 + 0.6 × 2000 × q2 + 0.6 × 2000 × q3 − Cmean (q1 + q2 + q3 ))
Cnew =
(1440 + 1100 + (0.6 − 1)(q1 + q2 + q3 )
and
Cmean = (1250 + Cnew ) / 2
This nonlinear optimization can be solved using different nonlinear programming

methods. The allocations of water considering different relative weights for the
various agencies are presented in Table 7.7. In case 1, domestic demands having the
highest priority and industrial demands having less volume have been completely
supplied but the utilities of the other agencies are less than 1. Comparing cases 1
and 2 reveals the effect of relative weights on allocated water. Increasing the relative
weights of the Department of Water Supply, the Department of Environmental
Protection, and the Department of Agriculture has improved the allocation of water
to agricultural demands and groundwater table variation, but it does not have any
significant effect on the TDS concentration of groundwater.
TABLE 7.7
Results of the Conflict Resolution Method
Variable Result (Case 1) Result (Case 2)
Water allocated to agricultural demands (million m3) 492 533

Water allocated to domestic demands (million m3) 500 460
Water allocated to industries (million m3) 80 80
Groundwater table drawdown (m) 2.64 2
Final average TDS concentration of groundwater 1440 1439
7.7 PROBLEMS
7.1 An aquifer (see figure below) is used to supply 3000 m3/day to industrial
demands; the system parameters are presented in following table:
Zone Storage Coefficient Kx (m/day) Ky (m/day)

1 0.02 20 35
2 0.04 45 40
The initial head of the aquifer is 1000 m. The nodes are located in the
southern boundary along the river are considered as constant head and
other boundaries are impermeable. Develop a simulation model for this
system using the Galerkin finite-element method. Calculate the piezomet-
ric surface after a 1-month period of operation for the following condi-
tions.
(a) The water withdrawal rates are uniform.
(b) All the water demand is withdrawn from well 3.
1000 m Well 1 * Zone 2
Well 2 *
1000 m
1000 m Zone 1
Well 3 *
River
2000 m 2000 m
FIGURE 7.1P
7.2 Solve problem 7.1 using implicit and explicit finite-difference methods
with a rectangular grid; compare the results and computational costs.
7.3 In the system shown in the following figure, the well is 0.5 m in diameter
and fully penetrates a confined aquifer of 40-m thickness and has a
hydraulic conductivity of 22 m/day. If the pumping rate is 4 m3/sec,
determine the drawdown at point x and at the well face.
7.4 In the confined aquifer system shown in the following figure, the pumping
rates in wells 1 and 2 are 0.02 and 0.03 m3/sec, respectively. The initial
piezometric head is 20 m everywhere. Applying the alternate-direction
implicit procedure, determine the head distribution after one day of
pumping.
FIGURE 7.3P
FIGURE 7.4P
7.5 Find the position of the water table, height of the seepage face, and flow
rate through a dam with a width of 6 m (see figure below) using a finite-
element method. The hydraulic conductivity in x and y directions are
0.001 m/day.
1
5
1 9
7
13 13
2 8 14
4m y=2m 2 6 10
14
3 9 15
4 10 16 3m
y=1m 3 7 11 15
5 11 17
6 12 18
4 8 12 16
x=2m x=4m x=6m
free node x
fixed node
FIGURE 7.5P
7.6 Solve Example 7.9 considering the quality of allocated water. Assume
that the average TDS concentration of the groundwater in zones 1 to 4
are 500, 745, 700, and 500 mg/l, respectively. The TDS concentration in
Rivers 1 and 2 are provided in the following table:
TDS Concentration in River TDS Concentration in River 2

1 (mg/l) (mg/l)
4340 1440
2000 2040
1790 720
370 390
120 450
610 310
2250 1540
2590 1860
3290 1350
1850 2680
3660 1710
2760 1780
The allowable concentration of allocated water is considered to be equal

to 1100 mg/l.
7.7 Find the most appropriate water allocation scheme for Example 7.10 using
Nash’s bargaining solution, considering that 40% of the allocated water
returns to the aquifer as return flow and the average TDS concentration
of the return flow is 2400 mg/l.
REFERENCES
Cooley, R. L., A Modular Finite-Element Model (MODFE) for Areal and Axisymmetric
Groundwater Flow Problems, Part 2: Derivation of Finite-Element Equations and
Comparisons with Analytical Solutions, Book 6, Techniques of Water Resources
Investigations series of the U.S. Geological Survey, 1992, chap. A4.
Delleur, J. W., Ed., The Handbook of Groundwater Engineering, CRC Press/Springer-Verlag,
Boca Raton, FL, 1999.
Esling S. P., Larson, T. A., and Sharpe, D. M., Graphic Groundwater User’s Manual, Micro-
Innovation, Inc., Carbondale, IL, 1993.
Freeze, R. A. and Cherry, J. A., Groundwater, Prentice-Hall, Englewood Cliffs, NJ, 1979.
Gupta R. S., Hydrology and Hydraulic Systems, Prentice-Hall, Englewood Cliffs, NJ, 1989.
Hantush, M. S. and Jacob, C. E., Non-steady radial flow in an infinite leaky aquifer, Trans
Am. Geophys. Union, 36(1), 95–100, 1955.
Harr, M. E., Groundwater and Seepage, McGraw-Hill, New York, 1962.
Istok, J., Groundwater Modeling by the Finite Element Method, AGU Water Resources
Monograph 13, Washington DC, 1989.
Karamouz, M., Kerachian, R., Zahraie, B., and Zahraie, B., Conjunctive use of surface and
groundwater resources, in Proc. ASCE Environmental and Water Resources Institute
Conf., Roanoke, VA, May 19–22, 2002.
Kashef, A. A. I., Groundwater Engineering, McGraw-Hill, New York, 1987.
Lee, C. H., The determination of safe yield of underground reservoirs of the closed basin
type, Trans. Am. Soc. Civ. Eng., 78, 148–151, 1915.
McDonald, M. G. and Harbaugh, A. W., A Modular Tree-Dimensional Finite-Difference
Groundwater Flow Model, Book 6, Techniques of Water Resources Investigations
series of the U.S. Geological Survey, 1988, chap. Al.
ReVelle, C. and McGarity, A. E., Eds., Design and Operation of Civil and Environmental
Engineering Systems, John Wiley & Sons, New York, 1997.
Todd, D. K., Groundwater Hydrology, John Wiley & Sons, New York, 1980.
Torak, L. J., A Modular Finite-Element Model (MODFE) for Areal and Axisymmetric Ground-
water Flow Problems, Part 1: Model Description and User’s Manual, Book 6, Tech-
niques of Water Resources Investigations series of the U.S. Geological Survey, 1993,
chap. A3.
Willis R. and Yeh, W. W.-G., Groundwater Systems Planning and Management, Prentice-
Hall, Englewood Cliffs, NJ, 1987.
Yeh, W. W.-G., System analysis in groundwater planning and management, J. Water Resources
Plan. Manage., 118(3), 224–237, 1992.
Zienkiewicz, O. C., The Finite-Element Method in Engineering Science, McGraw-Hill, New
York, 1971.
8 River–Reservoir Systems
Modeling
8.1 INTRODUCTION
This chapter discusses elements of river–reservoir systems modeling, including data
and information processing, objectives, constraints identification, and details of the
modeling process. Also presented are the objectives of river–reservoir systems plan-
ning and management modeling, including water supply, flood control, power gen-
eration, and supplying instream flow requirements, as well as formulation of simu-
lation and optimization models for different objectives.
8.2 DATA COLLECTION AND PROCESSING

8.2.1 INFLOW
Inflow to reservoirs is usually measured at hydroclimatic stations located on tribu-
taries entering the river and reservoirs and other control points. In some of the
reservoirs, the inflow is measured by the water balance in the reservoir using the
actual release figures on an hourly or daily basis. Streamflow has to be estimated
when there is no direct observation is lacking. Inflow to the reservoir is one of the
most important sources of uncertainty in development of operating policies for a
system. In a systems approach to reservoir operation and modeling, it is preferable
to have a long record of streamflows that includes worst-case scenarios of droughts
and floods experienced during the historical record.
Deterministic models for reservoir operation and modeling for expansion of the
system do not implicitly consider inflow uncertainties; however, they have been
widely used for stochastic modeling of complex and large-scale water resources
systems. In these models, the uncertainty is explicitly incorporated using generated
synthetic time series of uncertain inflows. Statistical models such as ARMA (autore-
gressive moving average) have been widely used by investigators to generate long
records of data. In stochastic modeling of reservoirs, the statistical behavior and
distribution of historical records is used in order to model the uncertainty of inflows.
Seasonal stochastic processes are frequently applied to hydrologic time series
in order to model the data affected by stochastic periodic components. This approach
allows one to fit many records that exhibit seasonal patterns caused by the periodicity
in the data (Montanari et al., 2000).
Defining hydrologic seasons is an important step when estimating forecast
parameters and defining heuristics for seasonal and monthly forecasts. Statistical
modeling of hydrologic time series can be improved significantly by defining seasons
295
in a region (Karamouz, 1999a). Seasons could be defined based on a number of

factors, such as:
• Estimation of the snow budget over the watershed and analysis of the
correlation between the average snow water equivalent for different months
of snow accumulation or snowmelt seasons and seasonal stream flows
• Similarity in basic statistical characteristics of streamflow data, such as
mean and coefficient of variation for different months of a season
• General hydrologic characteristics of the study area, such as variations in
precipitation, snow budget, and groundwater contribution to surface water
recharge
• Other factors, such as period of snowmelt and streamflow associated with
specific climatic regimes such as monsoon and tropical storms
The main objective of defining seasons is to capture the coupled effect of climatic
and hydrologic factors. It also allows modelers to reduce the number of estimated
parameters (compared with monthly time steps). Models with fewer parameters to
estimate tend to perform better and have a smaller tendency for error accumulation
(parsimony of parameters) (Salas et al., 1982).
Some of the stochastic reservoir operation models and their structures are
explained in the following text. In these models, inflow forecast time series are
utilized to incorporate uncertainty in inflows and the seasonal forecast model itself.
In addition to modeling uncertainties, inflow forecasts are necessary for applying
optimal operation policies to real-time operation of reservoirs.
A number of forecast models have been used for reservoir inflow forecasting
for real-time operation. These methods follow approaches ranging from pure statis-
tical methods to those based primarily on conceptual modeling of the hydrologic
cycle. Development of statistical models started with some very simple regression
models and significantly improved to incorporate seasonal and nonseasonal corre-
lations. Details of some of the most widely used statistical models are presented in
Chapter 5.
Figure 8.1 shows the data required and steps that should be taken for reservoir
inflow modeling and forecasting. As can be seen in this figure, the algorithm utilizes
a wide range of hydrologic and climatic data. Using such data will significantly
improve the accuracy of a forecast model (Karamouz et al., 2001). Different forecast
modification schemes are also considered in the algorithm and are used to reduce
the errors of statistical models.
8.2.2 WATER DEMANDS

Different water demands and their priorities are the main constraints in reservoir
operation optimization problems. Water-use circulars prepared by the U.S. Geolog-
ical Survey (USGS) identify the following water uses in the overall water accounting
system (Solley et al., 1993):
River–Reservoir Systems Modeling 297
Different Precipitation Historical Monthly Average Snow Budget

Patterns: Inflow Timeseries Estimation:
• Monsoon and
Tropical Storms • SWE Gradient in
Statistical Analysis Different Parts of Basin
• Mid-Latitude
Systems • Hypsometric
Characteristics of Basin
Monthly Variation of
Precipitation
Snowmelt Temperature
Modeling Data
Surface and Subsurface Water

Resources Interactions
Forecast Modification:
• Weekly update
• Large-scale
Defining Identification of climatic signals
Seasons Model Type • User-specified
heuristics
Estimation of Model
Parameters
FIGURE 8.1 An algorithm for reservoir inflow modeling.
• Water withdrawal for offstream purposes

• Water deliveries at point of use or quantities released after use
• Consumptive use
• Conveyance loss
• Reclaimed wastewater
• Return flow
• Instream flow
Water demands in river–reservoir systems resulting from offstream uses include:
• Domestic or municipal
• Industrial
• Agricultural
These uses require withdrawal of water from the surface or groundwater

resources. Part of the water withdrawn may return to the system, perhaps in another
location with different quality and over a different period of time. Water can also
be allocated to instream uses that alter the distribution of flow in time and space
(Loucks et al., 1981). Such uses include:
• Reservoir storage, possibly for recreational uses or navigation

• Flow augmentation for water quality control or for navigation
• Hydroelectric power production
There is a lag time between water demands and inflows. The peak of water
demand may occur in months during which inflows are at a minimum such as in
the middle of summer. The main purpose of water resources modeling is to derive
policies for water allocation in a way that demands can be met with reasonable
certainty. Thus, to manage a system, it is important to be able to describe the current
and projected water use fluctuations and variations.
Water demands change from year to year and month to month. Many physical,
economical, social, and political reasons for these variations can be identified. In
recent years, significant climatic changes have been observed in many parts of the
world, including more intense floods, greater precipitation, and even unusual
droughts in some regions. These changes have significantly affected the water
demands in many parts of the world.

8.2.3.1 Types of Dams
Dams are usually classified in terms of materials and forms. Common types are
homogenous or zoned earthfills, rockfills with an earth core or concrete face, and
concrete dams. Concrete dams are classified as gravity, arch, and buttress resistance
(Figure 8.2). Topography and geomorphology are primary factors in weighting the
comparative merits of dam types.
Embankment dams are constructed of earth and/or rock with a provision for
controlling seepage by means of an impermeable core, concrete face, grouting
curtain, or upstream blanket. Embankment dams have relatively poor resistance to
overflow; therefore, the spillway capacity must be determined conservatively.
Arch dams can carry large loads, but their integrity depends inherently on the
strength of abutments. A gravity dam is an essentially solid concrete structure that
resists imposed forces principally by its own weight. Although they are usually
straight in plan, these dams are sometimes curved or angled to accommodate site
topography.
A buttress dam is a gravity structure that, in addition to its own weight, utilizes
the weight of water over the upstream face to provide stability. The simplest type
of these dams is the slab type, which consists of sloping flat slabs supported at
intervals by buttresses.
Storage capacity is the most important physical characteristic of reservoirs. The
storage can be determined for each level of water from topographic map of the site.
An area–volume–elevation curve can be constructed by implementing the area
enclosed with each topographic contour in the reservoir site and summation of the
increments of storage below each level. This curve can be used in selection of total
capacity for reservoir and reservoir operation optimization. Figure 8.3 shows an
area–volume–elevation curve.
Material of
Type Construction
Typical cross section Plan view
Concrete,
Gravity rubble masonry
Arch Concrete
Sidewalls
of canyon
Buttress
Slab
Concrete
Buttress (also timber
and steel)
Rock-fill toe
Embankment Earth or
rock
Impermeable core
FIGURE 8.2 Basic types of dams. (From Linsley, R. K. et al., Water Resources Engineering,
McGraw-Hill, New York, 1992. With permission.)
8.2.3.2 Ancillary Facilities
A spillway is a safety valve for the dam and should have enough capacity to discharge
major floods without damaging the dam and ancillary structures and also keep the
reservoir level below the maximum water level. A spillway that has gates to adjust
the rate of outflow is referred to as a controlled spillway. An uncontrolled spillway
will discharge water based on the water level in the reservoir.
The majority of the active storage of reservoirs is below the spillway crest, and
outlets must be built in order to withdraw water from various parts of the reservoir
storage. Release from these outlets may be discharged into a channel below the dam
or may be transported in pipes or canals to some distant point. The basic structural
components of outlets are sluiceways and intakes. A sluiceway is a pipe or tunnel
that passes through a dam or hillside at one end of the dam and discharges into the
stream. An intake structure is required at the entrance to a conduit through which
Area (Km2)
26 24 22 20 18 16 14 12 10 8 6 4 2 0
1100
1080
Maximum operating level
1060
1040 – Spillway crest –

Elevation (meters above msl)
Volume Area
– Flood control –
1020 level
1000
980 – Minimum operating level –
960
– Centerline of penstock –
940 – Centerline of sluiceways –
– Centerline of turbines –
920 – Zero pool –
900
0 20 40 60 80 100 120 140 160 180 200 220 240 260
Volume (million m3)
FIGURE 8.3 An example of area–volume–elevation curve of dams.
water is withdrawn from the reservoir. Large dams may have sluiceways and/or
intakes in different levels (Linsley et al., 1992).
8.2.3.3 Hydropower Facilities
Listed here are the basic elements of a hydropower plant (see Figure 8.4) (Mays,
1996):
• Reservoir for creating the necessary head that will provide energy needed
for deriving turbines
• Intake structure that directs water from the reservoir into the penstock or
conduit; gates or valves are used to control water discharge to the power
plant, and racks or screens prevent trash or debris from entering the
turbines
• Conduit that conveys water from the intake structure to the powerhouse
• Power plant, including turbines, generators, and control and auxiliary
equipment
Energy loss in
Energy Line penstock
Gen Net head Gross head

H
Ve2
2g
Turbine
Ve
Draft Tube
FIGURE 8.4 Element of a typical hydropower plant.
FIGURE 8.5 Reservoir storage zoning.
8.3 OBJECTIVES
Large dams are usually multipurpose structures. Besides providing water for domes-
tic, agriculture, and industrial uses (the main objectives of reservoir planning and
operation), hydropower electric production is another objective of development of
many river–reservoir systems. High efficiency, lower costs, and the specific capa-
bilities of hydropower plants for controlling the frequency of power networks have
made hydropower plants a necessary component of power systems. Flood control
and damage reduction is another objective for dam construction. A reservoir reduces
the peak flow of a flood hydrograph to an amount lower than the river carrying
capacity. In the following sections, these objectives are briefly explained and the
structure of reservoir optimization models for satisfying demands is demonstrated.
Navigation requirements and environmental water rights are also presented. Other
purposes, such as quality control in the reservoir and downstream river, are
thoroughly explained in Chapter 9. Modeling power generation systems is also

explained in detail in Chapter 10.
8.3.1 WATER SUPPLY: RESERVOIR STORAGE ZONES

As shown in Figure 8.5, the total capacity of the reservoir can be divided into three
major parts:
• Active storage
• Dead storage
• Flood control storage
Active storage is the volume between minimum and normal water levels. Normal
water level is the maximum elevation of a reservoir during normal operation. For
many reservoirs, the normal water level is the elevation of the spillway crest, and
the minimum water level is the lowest elevation of the reservoir during normal
operation and may be fixed by elevation of the lowest outlet or minimum head
required for hydropower generation. Active storage is required for conservation
purposes, including water supplies, navigation, etc.
Flood control storage is reserved for storing excessive flood volumes to prevent
overtopping of the dam and to reduce the potential flood damage. The flood control
storage zone is between normal water level and maximum water level in the reser-
voir. During floods, discharge over the spillway may cause the water level to rise
above normal water level. This excess storage is surcharge storage and is normally
uncontrolled.
8.3.2 POWER GENERATION

Energy production is a major objective of water resources development in many
river–reservoir systems. In hydropower plants, energy is produced when water of
adequate head runs electric turbines and generates electricity without actually con-
suming water. As can be seen in Figure 8.4, the dam provides the required head.
The efficiency of power generation in the power plant shown in Figure 8.4 depends
upon:
• Plant efficiency
• Volumetric water flow through the turbine
• Net hydraulic head of water above the turbine
The energy production of a hydropower plant could be estimated using the following
equation (Mays, 1996):
∫
EO = e f γ Q(t ) H (t ) dt (8.1)
where EO is the electrical output of power plant, ef is the overall efficiency of power
generation, and γ is the specific weight of water. As can be seen in this equation,
without reservoir
Discharge (cms)
Target outflow
with reservoir
Time (hour)
FIGURE 8.6 Impact of storage on flood peak discharge reduction.
increasing head, H(t), and discharge, Q(t), through the turbines increases the energy
production. Efficiency is also a function of net hydraulic head above the turbines.
8.3.3 FLOOD CONTROL

To reduce flood damage, part of the active storage of the reservoirs is reserved for
flood control. As it can be seen in Figure 8.6, the volume of water represented by
the shaded area is stored and then released gradually at a rate that does not exceed
the carrying capacity of the river. The total volume of the inflow and outflow
hydrographs is almost the same (less evaporation) but the time distribution is changed
by the reservoir.
8.3.4 NAVIGATION REQUIREMENTS

Rivers and lakes have long been used for public and commercial transportation.
Recreational boating is also another concern in the operation of some river–reservoir
systems. The main constraint for potentially navigable parts of the river downstream
of a reservoir is maintaining a minimum acceptable depth at various sites s ( ELmin
s ).
Assuming the known river flow–stage relationship ( FSts ), the constraint that should
be considered in the reservoir operation optimization model is as follows (Loucks
et al., 1981):
( )
FSts ELst ≥ ELmin
s (8.2)
where ( ELst ) is the depth in time step t.

A more detailed analysis of navigation benefits and operation policies may be
necessary, especially when locks are required. If capital investment in locks is
necessary, their fixed costs as a function of the design depth of the channel, any
expected operating costs at each lock site s, and benefits of the transportation system
should be included in the objective function of the planning models.
8.3.5 ENVIRONMENTAL WATER RIGHTS

Providing adequate water to protect the aquatic, biological, and aesthetic values of
a stream and to preserve existing fisheries is an important constraint in river–reservoir
systems planning and operation. Because of the variety of habitats and streamflow
quality and quantity conditions, estimation of minimum instream flow requires
comprehensive studies. The simplest constraint that can be considered for satisfying
necessary environmental conditions is to supply the minimum monthly flow required
NM
in each reach of the river (Qmin ) after supplying demands:
Rt − ∑ RA
j =1
j ,t ≥Qmin
NM
(t = 1,⋯, n) ( NM = 1,…,12) (8.3)
where Rt is release from reservoir in month t and RAj,t is water allocated to demand
j in month t. NM is the number of months and m is the total number of water demands.
8.4 CAPACITY EXPANSION

Capacity expansion is mainly focused on increasing water supply by structural
methods. Construction of a new dam, increasing the height of an existing reservoir,
and construction of tunnels or channels for interbasin water transfers are some
common examples of capacity expansion; however, they are somewhat limited by
site-related, geophysical, geographic, economic, and institutional factors.
8.4.1 INTERBASIN WATER TRANSFER

Historically, many attempts have been made to divert water from areas of water
surplus to areas of water deficit. Interbasin water transfer refers to water withdrawal
and removal by means of ditches, canals, tunnels, or pipelines from a source con-
taining excess water to an adjacent area experiencing a shortage. The shortage
capacity in the importing basin is the main constraint in determining the design
capacity of water transfer structures, but economic and geophysical constraints are
also important limitations. The following model is the general form of an optimi-
zation model that can be used for optimizing the design capacity based on economic
analysis and the projected life of the project:
Minimize z= ∑ (C − B )
t =1
t t (8.4)
Subject to: Qd ≤ QM (8.5)
Ct = ft (Qd ) + Ci′,t + Ce,t ∀t (8.6)
Bt = Bi,t − Be′,t ∀t (8.7)

where:
n is the planning horizon (time period).

Qd is the design capacity of water transfer system.
QM is the maximum capacity of the water transfer system (based on geophys-
ical characteristics of the study area and supply limitations).
Ct is the present value of the total cost of the system after implementation of
water transfer system in time period t.
ft(Qd) is the present value of the total cost of the water transfer system
including capital investment, operation, and maintenance costs in time
period t.
Ci′,t is the present value of the cost of importing water from basin i after
implementation of the water transfer system in time period t.
Ce,t is the present value of the costs of exporting water to basin e due to water
transfer in time period t.
Bt is the present value of benefits of the system after implementations of the
water transfer system in time period t.
Bi,t is the present value of the net benefits of importing basin i in time period t.
Be′,t is the decrease in the present value of the net benefits of exporting basin
e in time period t.
In order to estimate the present value of an investment in year t of the planning

time horizon, the following equation can be used:
F
P= (8.8)
(1 + i ′)t
where P and F are the present and future value, respectively, of the investment in
year t; i′ is the interest rate (for more details, see Chapter 4).
Example 8.1
Based on a population growth analysis, the water demand of a city is estimated to

change over time as shown in Figure 8.7. Four interbasin water transfer projects are
studied for supplying future demands of this city over a 20-year planning time
horizon. The capacity and initial investment of these projects are shown in Table 8.1.
The interest rate is considered to be 6%. Find the optimal sequence of implementing
projects.
Solution: Table 8.2 shows the present value of the initial investment in the
projects which has been estimated using Eq. (8.8) for the years in which water
demand has increased. In order to find the best timing for construction of interbasin
water transfer projects, minimization of the present value of the costs is considered
as the objective function of a linear programming model as follows:
Minimize C = 1.78 X11 + 1.12 X12 + 0.88 X13 + 0.62 X14

+ 2.22 X21 + 1.40 X22 + 1.11 X23 + 0.78 X24
+ 1.60 X31 + 1.01 X32 + 0.80 X33 + 0.56 X34
+ 3.56 X41 + 2.23 X42 + 1.77 X43 + 1.25 X44
Subject to: 15 X11 + 20 X21 + 10 X31 + 30 X41 ≥ 20

15 X12 + 20 X22 + 10 X32 + 30 X42 ≥ 30
15 X13 + 20 X23 + 10 X33 + 30 X43 ≥ 20
15 X14 + 20 X24 + 10 X34 + 30 X44 ≥ 10
X11 + X12 + X13 + X14 ≤ 1
X21 + X22 + X23 + X24 ≤ 1
X31 + X32 + X33 + X34 ≤ 1
X41 + X42 + X43 + X44 ≤ 1
⎪⎧1 if the project i constructed in the j time period

th
Xij = ⎨
⎪⎩0 otherwise
FIGURE 8.7 Variation in water demand of a city over a 30-year period.
TABLE 8.1
Initial Investment and Capacity of Interbasin Water Transfer
Interbasin Water Initial Investment Capacity
Transfer Project ($106) (million m3)
A 2 15
B 2.5 20
C 1.8 10
D 4 30
TABLE 8.2
Present Value of Initial Investment for Interbasin Water Transfer Projects
Project t = 2 years t = 10 years t = 14 years t = 20 years
A 1.78 1.12 0.88 0.62
B 2.22 1.40 1.11 0.78
C 1.60 1.01 0.80 0.56
D 3.56 2.23 1.77 1.25
In this problem, the constraints are defined in order to supply demands in different
years and to ensure that each project can only be constructed once. Optimal timing
for implementation of the projects is: build projects C at year 2, D at year 10, B at
year 14, and A at year 20. The present value of the costs of construction of the
projects is estimated to be 5.56 million dollars.
8.4.2 OPTIMAL SIZING OF A SINGLE RESERVOIR SYSTEM

If the water demand of the river–reservoir system remains constant in different months,
then a basic planning model searches for the minimum reservoir capacity required for
supplying water with a steady release, which could be formulated as follows:
Minimize z = Cap (8.9)
Subject to: St +1 = St + It − R − Et − Lt (t = 1, …, n) (8.10)
Smin ≤ St ≤ Cap (t = 1, …, n) (8.11)
D ≤ R ≤ Rmax,t (t = 1, …, n) (8.12)
St , Et , Lt , R ≥ 0 (t = 1, …, n) (8.13)
where Cap is the total capacity of the reservoir, and St and It are storage at the
beginning of month t and inflow to the reservoir during month t, respectively. The
main constraint of the model is the continuity equation. As shown in Eq. (8.10),
storage at the end of each month is equivalent to the storage at the beginning of the
month plus inflow minus release, evaporated water (Et ), and leakage from the
reservoir (Lt ). The volume of water evaporated is a function of the evaporation rate
and the surface area of the reservoir. Equation (8.11) shows the upper and lower
bound on the volume (Smin). Equation (8.12) limits any release from the reservoir to
less than the maximum allowable release (Rmax,t ) (such as the carrying capacity of
the river) downstream of the reservoir, depending on the capacity of the hydraulic
structures (e.g., penstock or spillway). The minimum release is assumed to be a
constant demand, D, but a fraction of demand could be used as minimum release if
water allocation could be relaxed.
Reservoir Surface Area A2

a2
A1
Ad
Sd S1 St St+1 S2
Reservoir Storage
FIGURE 8.8 Reservoir storage–area relationship and piecewise approximation.
As shown in Figure 8.8, the evaporation losses can be estimated using piecewise
approximation of the reservoir area–volume curve. The evaporation losses could be
calculated using the information from this figure:
⎛ S + St +1 ⎞
Et = a2 ⋅ et ⎜ t ⎟ + A1 ⋅ et (8.14)
⎝ 2 ⎠
where et is the rate of evaporation (m) in period t and a2 is the slope of segment 2.
Equations (8.9) to (8.14) are linear; therefore, this problem can be solved with linear
programming. Based on the above formulation, the optimization algorithm will
search for the minimum capacity required to supply the total demand all the time.
But, in many real cases, such a scenario is not economically feasible. In other words,
based on economic analysis of the costs and benefits of the system, the planner may
forego meeting the demands and choose a smaller capacity that could cause shortages
in critical periods of operation. For this purpose, the operating policy is sought in
some conjunction with the search for optimal reservoir capacity (Mays and Tung,
1992). The objective function of simultaneous optimization of operation and sizing
of the reservoir can be formulated as follows with the same constraints as for Eqs.
(8.10) to (8.13):
Minimize z = ∑ (C − B )
t =1
t t (t = 1, …, n) (8.15)
As seen in Eq. (8.15), the objective function is considered to be the net revenue of
the system. The costs of operation can be classified as follows:
• Initial investment
• Operation and maintenance costs
• Costs due to shortages in supplying water demand
• Costs due to damages incurred during emergency situations, such as floods
and droughts
Linear programming techniques cannot be used for solving this problem because
the costs and benefits of operating the dams are usually nonlinear functions of
parameters such as reservoir release compared with water demands. Nonlinear opti-
mization and dynamic programming techniques are frequently used to solve this
problem. Details of these optimization techniques were explained in Chapter 2, and
application of these methods for optimization of reservoir operation and design is
explained later in this chapter.
In real-life situations, release from reservoir varies from month to month (Rt)
and reservoirs supply water to varying demands. In these cases, formulation of the
reservoir sizing problem would be as follows:
Minimize z = ∑ (C − B )
t =1
t t (8.16)
Subject to: St +1 = St + It − Rt − Et − Lt (t = 1, …, n) (8.17)
Smin ≤ St ≤ Cap (t = 1, …, n) (8.18)
0 ≤ Rt ≤ Rmax,t (t = 1, …, n) (8.19)
St , Et , Lt , Rt ≥ 0 (t = 1, …, n) (8.20)
In this case, the release from reservoir would be allocated to various demands:
Rt = ∑ RA
j =1
j ,t + Qt (t = 1,…, n) (8.21)
where RAj,t represents water allocated to demand j in month t and Qt represents water
remaining in the downstream river. The allocation policies depend on contractual,
legal, and institutional requirements for various purposes.
8.4.3 RELIABILITY-BASED DESIGN

Usual design practice is based on recommended factors of safety, which should
allow the system to be subject to only a limited number of failures over its projected
lifetime. This approach is usually based on a maximum probability of failure within
a given lifetime (n) of the system.
Based on this approach and taking into account economic aspects such as project
costs, benefits, or benefit–cost ratio, the optimal size of a reservoir can be estimated.
In reservoir sizing problems, it is assumed that the probability distribution of inflows
is known. As the design return period increases, the initial cost of the dam and
accompanying facilities increases while the expected damages associated with
unsatisfactory operation of the system decreases. For each return period, the total
cost can be estimated by adding the expected annual damage costs and the capital
costs of constructing a reservoir that can control the floods with that return period.
Then, the design return period can be found by estimating the minimum total cost.
Suppose X is the variable for maximum annual flood of the reservoir. If the design
return period of a reservoir is considered to be T, then a flood with magnitude x will
incur no cost if x ≤ x(T). The expected annual cost can be estimated as (Kottegoda
and Rosso, 1997):
+∞
dT =
∫
x (T )
d ( x ) f x ( x ) dx (8.22)
where fx(x) is the probability density function of floods and d(x) is the cost function
and x(T) is the magnitude of a flood with return period T.
Example 8.2
The design return period T of a dam that is facing floods can be estimated using a
reliability-based approach. As can be seen in Table 8.3, the initial cost or capital
investment required for a dam and the facilities increases as the design return period
increases. It is assumed that the probability distribution of the floods and cost of
damages can be estimated. As shown in the table, the damage cost increases as the
flood return period increases. Find the optimal design return period using the reli-
ability-based approach. A finite-difference approximation of Eq. (8.22) is used as
follows:
d (Ti −1 ) + d (Ti ) ⎛ 1 1⎞
Incremental expected damage = ⎜T − T ⎟
2 ⎝ i −1 i⎠
where d(Ti) is the estimated damage if the flood with return period Ti occurs. The
risk of damage for each return period is then computed using partial summation of
the relevant incremental values. The total costs are computed by adding the corre-
sponding capital investment. As shown in Table 8.3, the minimum cost is estimated
for a 500-year return period. The risk of failure for the design based on this return
period is 0.002.
8.4.5 SEDIMENTATION AND DREDGING

The sedimentation rate for a reservoir is an important factor in determining the active
storage and useful life of the reservoir. The projected rate of reduction in useful
storage of an existing reservoir is a significant constraint in capacity expansion
problems. The main issue is to decide whether to construct a new dam, modify the
land use and vegetation upstream of the dam to reduce sediment transport to the
reservoir, or to remove the accumulated sediments from the reservoir bed by mechan-
ical means such as dredging.
TABLE 8.3
Calculation of Optimum Design Capacity of a Reservoir and Facilities
Annual Incremental Expected
Return Probability Expected Damage
Period of Capital Cost Damage Damage Cost Total Cost
(years) Exceedance ($106/year) ($106) ($106/year) ($106/year) ($106/year)
100 0.01 20.1 0 — 53.8 73.9

200 0.005 21.2 500 1.25 52.55 73.75
500 0.002 32.1 10,000 15.75 36.8 68.9
1000 0.001 45.4 15,000 12.5 24.3 69.7
10,000 0.0001 70.5 39,000 24.3 0 70.5
Sum 53.8
8.5 RESERVOIR OPERATION OPTIMIZATION

8.5.1 WATER SUPPLY WITH MONTHLY VARYING RELEASE
The basic model for reservoir operation when the releases from the reservoir are
varying from month to month is as follows:
Minimize z = ∑ loss ( R , D , S )
t =1
t t t t (8.23)
Subject to: St +1 = St + It − Rt − Et − Lt (t = 1, …, n) (8.24)
Smin ≤ St ≤ Cap (t = 1, …, n) (8.25)
0 ≤ Rt ≤ Rmax,t (t = 1, …, n) (8.26)
St , Et , Lt , Rt ≥ 0 (t = 1, …, n) (8.27)
As it can be seen in Eq. (8.23), the objective function of the model is to minimize
total losses in the historical period of operation. Loss functions have been proposed
to be a function of shortages in supplying demands and the difference between water
storage in the reservoir and target storage in each month.
Based on the above formulation, total losses are estimated by comparing total
releases with total water demands of the system. The optimal releases for allocating
water to different demands are determined by utilizing a simulation model; in this
approach, the details of supplying shortages are not considered within the optimi-
zation routine. In a more elaborate algorithm for multipurpose reservoir operation,
the allocation policy is applied in the optimization loop. In other words, each time
the optimization algorithm searches for the optimal solution, the losses are estimated
based on the details of water allocation to the different demands. The objective
function of this model would be as follows:
n m
Minimize z= ∑ ∑ loss ( R
t =1 j =1
t j ,t , Dj ,t , St ) (8.28)
where:
Rj,t is water allocated to demand j in month t.

Dj,t is the volume of water demand j in month t.
St is reservoir storage in month t.
m is the number of water demands.
The constraints are the same as before.
Example 8.3
A reservoir is constructed on a river for supplying the power load of a city located
near the river and water demands of downstream agricultural lands. The water that
is released for power generation can also be used for supplying other demands.
The required instream flow is estimated to be 1 million m3 each month. The total
capacity of the reservoir is 10 million m3. Maximum monthly release from the
reservoir is limited to 7 million m3. Table 8.4 shows the monthly inflows to the
reservoir and benefits of power generation and water supply. The reservoir storage
on January 1 is considered to be 5 million m3. Find the optimal releases using
linear programming.
Solution: The following formulation can be used for finding the optimal monthly
volume of water that should be released from the reservoir:
Maximize B = 2.6 R1 + 2.9 R2 + 3.6 R3 + 3.9 R4 + 4.2 R5 + 4.2 R6 + 4.5 R7 + 4.1 R8

+ 3.6 R9 + 3.1 R10 + 2.7 R11 + 2.5 R12
Subject to: S1 = 5
Si + Ii − Ri = Si +1 (i = 2, ⋯, 12)
1 ≤ Ri ≤ 7 (i = 1, ⋯, 12)
Si ≤ 10 (i = 1, ⋯, 13)
By solving this problem using a linear programming package, the total maximum
benefit is estimated as $128,000 for the two sets of optimal monthly releases shown
in Table 8.5.
TABLE 8.4
Monthly Inflows to the Reservoir and Benefits of Power Generation
and Water Supply in Example 8.3
Benefits of Power Benefits of Water
Inflow Generation Supply
Month (million m3) ($103/million m3) ($103/million m3)
January 2 1.6 1.0

February 2 1.7 1.2
March 3 1.8 1.8
April 4 1.9 2.0
May 3 2.0 2.2
June 2 2.0 2.2
July 2 2.0 2.5
August 1 1.9 2.2
September 2 1.8 1.8
October 3 1.7 1.4
November 3 1.6 1.1
December 2 1.5 1.0
TABLE 8.5
Optimal Releases from the Reservoir in Example 8.3
Jan. Feb. Mar. Apr. May June July Aug. Sept. Oct. Nov. Dec.
1 1 1 3 7 3 7 1 2 3 3 2
1 1 1 3 3 7 7 1 2 3 3 2
8.5.2 FLOOD CONTROL

The flood storage capacity of a reservoir is one of the important components of
many flood control systems. For this purpose, part of the active storage of a reservoir
is kept empty to store potential floods and gradually release excess water at rates
not exceeding the capacity of the downstream river. Methods of assessing the volume
required for flood storage are based on minimizing expected damages. To incorporate
the flood control objective, the following constraint should be considered in reservoir
sizing or operation models:
St ≤ Cap − Vt (t = 1, …, n) (8.29)
where Vt is the flood control storage in month t. Flood routing procedures, along with
knowledge of flood control operating policies and channel storage characteristics,
can be used to predict the impact of flood peaks on downstream of the reservoir.
Various discrete flood storage capacities in an upstream reservoir could be utilized
to provide a buffer for extreme flood situations. The important step for determining
flood control storage for different months is to define the hydrographs that represent
the response of the watershed to floods with varying return periods. The following
steps may be taken (Karamouz and Zahraie, 1998):
• Identification of observed flood hydrographs of significant magnitude

• Selection of representative hydrographs showing flood characteristics of
the watershed
• Estimation of flood peaks of different return periods using comprehen-
sive statistical analysis of observed values and fitting the best statistical
distribution
• Extrapolation of flood peaks on the representative flood characteristics
The next step would be to determine the ability of a reservoir to withstand floods
of various magnitudes. Physical characteristics of the reservoir, discharge outlets,
and the carrying capacity of the river downstream of the reservoir should be taken
into account. Figure 8.9 shows the steps that should be taken to determine flood
storage necessary for floods of various return periods. As seen, the final result of
this algorithm is that the reservoir will be graduated to flood control storage required
for floods with varying probabilities of occurrence.
Example 8.4
Consider a dam constructed near a city as shown in Figure 8.10. Reservoir capacity
is considered to be 70 million m3. Control point A, considered to be a critical point
within the city, is located 10 km downstream of the reservoir and has a relatively
lower carrying capacity than the rest of the reaches of the river. The river carrying
capacity at this control point is estimated to be 200 m3/sec. A local stream discharges
to the river upstream of the city. Figure 8.11 shows the inflow to the reservoir and
the local flow hydrographs estimated for a 100-year flood. As can be seen in this
figure, the peak flow of the inflow to the reservoir is about 280 m3/sec, which is
higher than the river carrying capacity. For finding the flood control volume, variation
of the peak reservoir discharge with respect to the reservoir storage level at the
beginning of the flood should be investigated.
The HEC-5 package (Hydrologic Engineering Center, 1982) is used to estimate
the required flood control storage for reducing a 100-year flood peak to the carrying
capacity of the river at the control point. The Maskingum method is used for routing
the reservoir outflow and local flows in the river between the dam and control point
A. It is assumed that the storage of the reservoir is 67 million m3 at the beginning
of the flood. Reservoir outflow and the control point discharge hydrographs for this
scenario are shown in Figure 8.12. The discharge at control point A is higher than
the carrying capacity at this point between hours 29 to 57. To estimate the maximum
flood control level at the reservoir to ensure that point A will not be flooded, the
flood control level of the reservoir is changed. Table 8.6 shows the results of the
simulation of the system with different flood control storages in the reservoir, and
the required flood control volume is about 5 million m3.
FIGURE 8.9 Flowchart for finding flood storage required for controlling floods with different
return periods.
A
Dam
City
Local Stream
FIGURE 8.10 Schematic of the system explained in Example 8.4.
300
250
Discharge (CMS)
200
150
100
50
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140
Time (Hour)
Reservoir Inflow Local Flows
FIGURE 8.11 100-year hydrograph of reservoir inflow and local flows.
8.6 APPLICATION OF DETERMINISTIC DYNAMIC

PROGRAMMING IN RESERVOIR OPERATION: DPR
MODEL
The deterministic dynamic programming (DPR) model was developed by Karamouz
et al. (1992) for optimization of multiple reservoir systems. The objective function
of this model is expressed as:
T
⎛ X
⎞
Min Z = ∑
t =1
Loss ⎜
⎝
∑R
s =1
s ,t ⎟
⎠
(8.30)
where T is the time horizon, X is the total number of reservoirs in the system, and
Loss is the cost of operation based on the ratio of supplied water (Rs,t) from reservoir
s in month t to the monthly demand. As discussed earlier, this loss function is used
350
300
Discharge (CMS)
250
200
150
100
50
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140
Time (Hour)
Reservoir Outflow Discharge at Control Point A
FIGURE 8.12 Reservoir outflow and discharge at control point A (the reservoir storage had
been 67 million m3 when the flood started).
TABLE 8.6
Results of Flood Routing in a River–Reservoir System (Example 8.4)
for Various Flood Control Volumes
Reservoir Storage at Peak Discharge at Duration of Flooding
Beginning of Flood Flood Control Volume Control Point A Condition at A
(106 m3) (106 m3) (m3/sec) (hr)a
69 1 297.7 30
68 2 286.3 26
67 3 254.1 21
66 4 235.3 11
65 5 200.0 —
a Total number of hours that the discharge at the control point has been higher than 200 m3/sec.
to prevent or minimize operation outside of a safe range. The continuity or mass

balance of the contents of the reservoir, considering regulated and unregulated release
(seepage and spill) from the beginning of the month to the next, is also included in
the model:
Ss,t +1 − Ss,t + Rs,t = Is,t (8.31)
Rs,t = RsR,t + RsU,t (8.32)
where Rs,t is release from reservoir s in month t, and It is inflow volume to the
reservoir in month t. RsR,t and RsU,t are regulated and unregulated releases, respectively,
from reservoirs. Additional constraints on maximum and minimum allowable release

and storage during any season for all sites can be stated as:
,t ≥ Rs ,t ≥ Rs ,t
Rsmax t = 1, ..., T s = 1,..., X
min
(8.33)
,t ≥ Ss ,t ≥ Ss ,t
Ssmax t = 1, ..., T s = 1,..., X
min
(8.34)
Ss,t +1 − Ss,t ≤ SCs t = 1, ..., T s = 1,..., X (8.35)
where SCs is the maximum allowable change in storage of reservoir s within each
month considering dam stability and safety conditions of the reservoir. This separable
mathematical program is solvable as a discrete dynamic program. The recursion
relation when Eq. (8.30) is used can be written as:
⎛ ⎛ X
⎞ ⎞
ft +1 ( S1,t +1,..., SX ,t +1 ) = min ⎜ Loss⎜
⎜
⎝ ⎝
∑
s =1 ⎠
(
Rst ⎟ + ft S1,t ,..., SX ,t )⎟⎟
⎠
S1,t ∈ Ω1,t
… (8.36)
…
SX ,t ∈ Ω X ,t
The initial conditions are:
f1 ( S1,1 ,..., SX ,1 ) = 0 S1,1 ∈ Ω1,1 ,..., SX ,1 ∈ Ω X ,1 (8.37)
where ft(S1,t, …, SX,t) is the total minimum losses of operation from the beginning
of month 1 to the beginning of month t, when the storage volume at the beginning
of month t is S1,t at site 1, …, SX,t at site X. Ωs,t is the set of discrete storage volumes
that will be considered for the beginning of month t at site s.
The optimal release ( Rˆ s,m ), optimal storage ( Sˆs,m ), and inflows for each site can
be regressed to estimate the general operating rules for different months in a simple
form:

Rs,m = as,m Is,m + bs,m Sˆs,m + cs,m m = 1,..., 12 s = 1,..., X (8.38)
where as,m, bs,m, and cs,m are regression coefficients that should be estimated in each
reservoir s for each month m. Initially, no upper bound on releases exists, and the
lower bound on release is zero. The dynamic program is solved and the optimal
policies are used in a multiple regression to find the general operating rules. These
rules are then used to simulate the operation of reservoirs over a long time horizon.
Also, the losses associated with simulated operation are determined. The procedure
concludes the first iteration of the model. For the second iteration, the allowable
release in the dynamic program is limited to some percentage (bound) of the previ-
ously defined general operating rule, R̂s,t :

(1 − bound ) Rs,t ≤ Rs,t ≤ (1 + bound ) Rs,t s = 1,…, X (8.39)
Anytime the refined rule or best rule for a specific value of the bound is found, the
algorithm can be further restricted by reducing the bound value, and the dynamic
programming, regression analysis, and simulation procedure will continue. If the
bound is very small (0.01), then the algorithm is stopped because further restrictions
on the release in the dynamic program will probably not result in better operating
rules.
8.7 STOCHASTIC RESERVOIR OPERATION MODELING

So far, the deterministic modeling approach has been discussed, where the state at
the next stage is completely determined by the state and policy decision at the current
stage. In other words, in the deterministic approach, it is assumed that what has
happened in the past (e.g., historical inflows to the reservoirs) will be fully repre-
sentative of what will happen in the future.
Considering the uncertainties inherent in the prediction of hydrologic, economic,
and other factors affecting the performance of water resources systems, deterministic
planning models may result in under- or overestimated results; systems benefits are
generally overestimated, while systems costs are usually underestimated. Two types
of uncertainty may affect reservoir operation models:
• Uncertainty in objective function. Models are subject to uncertainty in

economic parameters and imprecise knowledge about future benefits and
costs resulting from applying different policies. Such uncertainty can often
be handled by substitution of the expected value for the uncertain objective
value. Use of the expected value in the objective function usually results
in satisfactory policies when the future states of the system are not too
extreme (Loucks et al., 1981).
• Uncertainty in constraints. If the uncertainty is small (small variation in
constraint variables), substitution of an expected value might be useful.
When only the right-hand side of one or more inequality constraints is
random, chance constraint programming can be used, as explained in the
next section.
Two approaches have been developed in order to incorporate the stochastic nature
of streamflows in the reservoir operation models:
• Incorporate the stochastic nature of streamflows implicitly in the reservoir

operation models.
• Generate several synthetic streamflow records as inputs to deterministic
models.
In the discussion that follows, linear and dynamic programming reservoir operation
models in which the stochastic nature of streamflows is implicitly incorporated are
explained.
8.7.1 STOCHASTIC LINEAR PROGRAMMING

As mentioned in Chapter 3, two approaches are considered in stochastic modeling
of reservoir operation:
• Stochastic programming
• Chance constraint programming
The primary distinction between these types is that stochastic programming requires
all constraints to hold with probability 1 and the objective function has probabilistic
terms, whereas in chance constraint programming a small probability of violating
any functional constraint is permitted but the objective function is usually considered
to be deterministic. The general approach for dealing with the second type is to
reformulate constraints as new equivalent linear programming problems for which
the certainty assumption is satisfied.
In the first approach, the random variable (such as reservoir inflow) is usually
represented by its discrete probability distribution and all the constraints associated
with this random variable are repeated for values of the discretized variables. This
results in stochastic programming models becoming computationally large (more
details about stochastic optimization models are presented in Chapter 3). Chance
constraint programming can be a suitable alternative approach, as explained in more
detail in the next section.
8.7.1.1 Chance Constraint Reservoir Operation Model
Assume that reservoir storage at the end of each month (St+1) is a decision variable
(dt) that should be determined:
St +1 = dt (t = 1, …, n) (8.40)
The mass balance constraint would be as follows:
Rt = It − dt + dt −1 (t = 1, …, n) (8.41)
Other constraints of the reservoir operation optimization model are the same as the
deterministic model except for an additional constraint that represents the required
water supply reliability:
[
Prob Rt ≥ Dt ≥ λ t ] (t = 1, …, n) (8.42)
This equation shows that the probability of releases being greater than the demands
(Dt ) in the planning horizon should be greater than the reliability level (λt). This
equation can be rewritten as:
[
Prob It − dt + dt −1 ≥ Dt ≥ λ t] (t = 1, …, n) (8.43)
Each complementary probability (1 – λt ) represents the allowable risk that the

random variables will take on values such that:
It − dt + dt −1 < Dt (8.44)
Then, the deterministic equivalent of the probabilistic constrain can be written as:
Dt − dt −1 + dt ≤ it1−λt (8.45)
where it1−λt is the inverse of the cumulative distribution function of reservoir inflow
at level λt such that:
( )
P It ≤ it1−λt = 1 − λ t (8.46)
Example 8.5
Consider the following chance constraint problem assuming that b̃1 , b̃2 , and b̃3 are
normal variables with known mean and variances. Is X = (5, 10, 15) a feasible
solution?
Minimize Z = 10 x1 + 20 x 2 + 30 x3
Subject to: [ ]
Prob 3 x1 + 2 x 2 + x3 ≥ b˜1 ≥ 0.96
Prob [2 x + 4 x + 2 x ≥ b˜ ] ≥ 0.98
1 2 3 2
Prob [ x + 3 x + 5 x ≥ b˜ ] ≥ 0.99
1 2 3 3
b˜1 = N ( 45, 4), b˜2 = N (70, 9), b˜3 = N (90, 25)
The deterministic equivalent of chance constraints can be computed as follows:
⎧3 x1 + 2 x 2 + x3 ≥ 45 + 2(1.75) = 48.5
⎪⎪
⎨2 x1 + 4 x 2 + 2 x3 ≥ 70 + 3(2.054) = 76.16
⎪
⎪⎩ x1 + 3 x 2 + 5 x3 ≥ 90 + 5(2.327) = 101.64
where the value in the parentheses are the z value from standard normal table for
0.96, 0.98, and 0.98 probability levels.
By substituting X = (5, 10, 15) in the above deterministic equivalent of chance
constraints, it can be seen that this point is a feasible solution:
⎧3 × 5 + 2 × 10 + 15 = 50 ≥ 48.5
⎪
⎨2 × 5 + 4 × 10 + 2 × 15 = 80 ≥ 76.16
⎪
⎩5 + 3 × 10 + 5 × 15 = 110 ≥ 101.64
8.7.2 STOCHASTIC DYNAMIC PROGRAMMING MODEL

Stochastic dynamic programming (SDP) differs from deterministic dynamic pro-
gramming in that the state at the next stage is not completely determined by the
state and policy decision at the current stage; rather, a probability distribution is
used to determine what the next state will be. Figure 8.13 shows the basic structure
of stochastic dynamic programming for reservoir operation. As can be seen in this
figure, the state of the system in stage t can be represented by reservoir storage.
Assuming a specific value for the decision variable (reservoir release) results in
different states in stage t + 1 with probabilities that can be estimated based on
associated probabilities with reservoir inflows in that time step.
The recursive function of a backward-moving stochastic dynamic programming
algorithm can be written as follows:
⎡ ⎤
[ ]
ft Stc (k ), Itc (l ) = Min
c
St +1∈Ω
[
⎢⎣
] ∑P
⎢Ct Stc (k ), Rt + t
f *
l ,ll t +1 [Sc
t +1 (kk ), Rt +1 ]⎥
⎥⎦
(8.47)
ll
FIGURE 8.13 The basic structure for stochastic dynamic programming.

where:
Ω is the set of possible storage characteristics.

Stc (k ) is the reservoir storage characteristic value in state k in time t.
Itc (l ) is the reservoir inflow characteristic value in class interval l in time t.
[ ]
ft Stc (k ), Itc (l ) is the minimum value of expected cumulative losses of res-
ervoir operation from the present period to the end of the planning horizon,
provided that the kth and lth scenario occurs in time t.
[ ]
ft*+1 Stc+1 (kk ), Itc+1 (ll ) is the minimum value of expected losses during periods
T to t + 1 associated with the optimal solution obtained when the kkth and
llth scenario occurs in time t + 1.
[ ]
Ct Stc (k ), Itc (l ), Stc+1 (kk ) is the value of losses during period t, associated
with the release R t during the same time period, where R t =
Stc (k ) + Itc (l ) − Stc+1 (kk ) .
t
Pl ,ll is the probability of inflow in class interval l in period t, when the inflow
in period t + 1 is in class interval ll.
The recursive function shown in Eq. (8.47) can be used in a procedure that moves
backward. It is based on the principle that, no matter in what state of a stage, to find
the best policy we should proceed from that state and stage in an optimal manner.
Similar formulations can be developed for forward-moving procedures. This princi-
ple, first defined by Bellman (1957), is referred to as the Bellman optimality principle
(Loucks et al., 1981).
The conditions for obtaining the steady-state optimal policy of the SDP model
are:
• The expected value of the recursive function does not change from one
period to the next (stationary solution).
• The optimal policies do not change.
For example, in reservoir operation, the optimal solution is reached when the optimal
releases associated with each initial storage are the same as the corresponding
releases and storage in the previous time step. Inflow and storage discretization
significantly affect the performance of stochastic dynamic programming models for
reservoir operation optimization.
8.7.2.1 Reservoir Storage Discretization
Due to the nature of dynamic programming, the storage volume must be discretized
into representative storage zones, with a corresponding number of characteristic
storage levels (SDN). In the classical scheme, Z is simply defined as the number of
zones equal to SDN and the storage increment is:
Cap
ΔS = (8.48)
SDN
where Cap is the total permissible storage capacity of the reservoir. For each state
k at time t, the characteristic storage states can be calculated as:
ΔS 2k − 1
Stc (k ) = + (k − 1) ΔS = ΔS ∀k ∈[1, ..., SDN ] (8.49)
2 2
In this approach, the zero and full storage levels are not included as separate
characteristic storage values. Savaranskiy (1940) developed another scheme in which
each Z zone is regarded as a class interval, and the corresponding storage state is
defined as the center point of this interval. Empty and full reservoir storages are
also defined as separate states with zero class intervals:
Cap
Z = SDN − 1 and ΔS = (8.50)
SDN − 2
So, the characteristic values can be estimated as follows:
ΔS 2k − 3
Stc (k ) = + (k − 2) ΔS = ΔS k = 2, ⋯, Z (8.51)
2 2
Moran (1945) presented another method in which the characteristic storage

values are defined on the boundaries between the zones:
Cap
Z = SDN − 1 and ΔS =
SDN − 1
and storage stages can be calculated as
Stc (k ) = (k − 1) ΔS k = 1, ⋯, Z (8.52)
EXAMPLE 8.6
The capacity of a reservoir is 16.8 million m3. Find the storage zones using the
discretization schemes explained previously.
Solution: Figure 8.14 shows storage zones obtained from discretization schemes
for SDN = 8.
8.7.2.2 Reservoir Inflow Discretization
The governing idea for flow discretization consists of dividing the entire flow range
into a specific number of class intervals (DN) and calculating a characteristic flow
value ( Itc (l ) ) for each interval. This value represents all the flows of that class
FIGURE 8.14 Discrete representation of storage in 1000 m3 (for capacity equal to 16.8
million m3 and SND = 8). (From Karamouz and Vasiliadis, 1992.)
interval. The classical scheme that has been used by many researchers consists of
dividing the entire flow range (ΔI), from Imin to Imax, into DN class intervals wit
constant flow increment δI, where Imin and Imax are minimum and maximum inflows,
respectively. In this scheme:
ΔI
δI = (8.53)
DN
The characteristic flow ( Itc (l ) ) for each class interval is defined as the mean of the
lower and upper bounds of the corresponding class interval. In other words, the
entire discretization model is based on the minimum and maximum values of each
interval, and any changes of these values will result in completely different charac-
teristic values.
An improved approach is to divide class interval into a number of subintervals

and define a probability distribution of inflows occurring in each class interval using
frequency analysis for each class subinterval. The characteristic value for (represen-
tative of) each interval is then defined as the expected value of inflows within that
interval. This and the previous approach have some shortcomings. First, the variance
of the characteristic flows could be very different from those of the original series.
Furthermore, some intervals might have fewer data, which could result in a poor
estimation of transition probabilities from these intervals to others.
To overcome these shortcomings, another method was developed by Karamouz
and Vasiliadis (1992), where the intervals are unequally spaced. In this model, the
amount of data (N(i)) for each class interval is made the same (so the intervals are
equally probable) by adjusting the width of each class interval (see Figure 8.15),
and the variance of data within each class interval is kept small by increasing the
number of class intervals. If N(i) is the number of data points in the ith class interval,
n(i, j) is the number of data points in the ith class interval and the jth subinterval
(where n(i, j) < N(i)), fc(i, j) is the average of the lower and the upper bound of the
jth subinterval, and SI(i) is the number of subintervals in the ith class, then the
characteristic value for each class interval, Fc (i), is calculated using the following
formula:
SI ( i )
⎡ n(i, j ) ⎤
F c (i ) = ∑ ⎢⎣ N(i) ⋅ f
j =1
c
(i, j )⎥
⎦
∀i ∈[1, ⋯, DN ]
(8.54)
n(i, j ) ∈[1, ⋯, N (i )], N (i ) ∈[1, ⋯, N ]
An alternative expression for Eq. (8.54) is:
N (i )
⎡ x (i, j ) ⎤
F c (i ) = ∑ ⎢⎣ N(i) ⎥⎦
j =1
∀i ∈[1, ⋯, DN ], N (i ) ∈[1, ⋯, N ] (8.55)
where x(i, j) is the jth flow in the ith class interval. This method of selecting
characteristic flows results in better transition matrices because an adequate number
of flows is available for each interval, zero entries in the transition matrices are
eliminated, and the value of each data point receives equal weight in the construction
of the transition matrices. Furthermore, the characteristic flows preserve the mean
and variance of the original time series, as long as the number of class intervals is
sufficiently large.
Example 8.7
The monthly inflows to a reservoir have been:
q = {5, 1, 6, 3, 4, 7, 8, 2, 10, 7}
1(F)
DF(1) DF(2) DF(3) DF(4)
fc or Fr
Fmax
df(2) df(2) df(2)
Fc or FR
0.5
fc (1,3)
fc (1,2)
Fc (1)
fc (1,1)
N(1) N(2) N(3) N(4)

0.25
FR(1) FR(2) F c(2) FR(3) n(3,1) F c(3) n(3,2) FR(4) F c(4) FR(5)
fr(1,1) fr(1,4) i.e. DN=4
fr(1,2) fr(1,3)
LEGEND EQUATIONS:
Fmn = FR(1) = fr(1,1)

Characteristic Flow
Fmax = FR(DN+1) = fr(DN,dn+1)
N(1)=INT[N/DN] (Constant)
Bounds of Class Intervals
note: FR(i) is the [N(i) x(i-1)+1]th flow,
after sorting
Characteristic Flow of Class dn(i) = INT[1+3.3 xlog(N(i))] (Constant)
Sub-Intervals
df(i) =[FR(i+1)- FR(i) ] / dn(1) (Constant)
fr(i,j) = df(i)x(j-1)+F(i)
Frequency Histogram
fc(i,j) = [rf(i,j) + fr (i,j+1)] / 2
Frequency Histogram within
each Class Interval
Fc(i) =
Σ
all j
n(i,j)
N(i)
c
* f (i,j)
all i in [1,...,DN]
all j in [1,...,dn(i)]
FIGURE 8.15 Discrete representation of flow. (From Karamouz and Vasiliadis, 1992.)
The transition probabilities of inflows considering two intervals for q ≤ 5 and q > 5
are shown in the following matrix:
q≤5 q>5
q ≤ 5 ⎡⎢ 0.4
( )
0.6 ⎤⎥
P qt +1 qt = ⎢ ⎥
q > 5 ⎢⎣ 0.5 0.5⎥⎦
Consider the mass balance equation for the reservoir (Rt = St – St+1 + qi) and
two storage intervals with the characteristic values St = {1, 2}. The cost of operation
can be estimated as ft = Rt2 . The characteristic volumes of the streamflows are
considered to be 3 and 7. Find the optimal releases using the stochastic dynamic
programming technique.
Solution: The release from the reservoir for each month can be estimated by
considering different storage levels at the beginning and end of the month. The
stochastic dynamic programming solution for finding the optimal releases is shown
in Table 8.7. Using the backward-moving procedure in the first iteration, which is
assumed to be the 10th time interval in the planning horizon, the optimal storage
level at the beginning of the month and releases are estimated considering possible
inflow and ending storage states. In the next iterations, the cumulative cost of
operation is calculated considering the inflow transition probabilities.
For example, in the first row of results for the 9th time step, storage at the end
of the month is considered to be 1; therefore, the cost of the best policies obtained
in the 10th time step for starting storage equal to 1 should be added to the cost of
operation in the 9th time step. The least costs are estimated as 4 and 36 when the
starting storage is 1 and inflows in the 10th time step are 3 and 7, respectively. As
can be seen for the second and third iterations, the optimal releases and storages
obtained are the same and the expected cost in one time step remains constant;
therefore, the stationary solution has been obtained.
8.7.3 BSDP AND DDSP MODELS

The Bayesian Stochastic Dynamic Programming (BSDP) model was developed by
Karamouz and Vasiliadis (1992). In this model, the release from a reservoir in each
time period t depends on the state of the system, which is defined by the following
state variables:
• Characteristic value of reservoir storage at the beginning of time period

t ( Stc (k ))
• Characteristic inflow to the reservoir during time period t ( Itc (l ))
• Characteristic forecast for the next time period, t + 1 ( Htc+1 (ii))
For every state [ Stc (k ), Itc (l ), Htc+1 (ii)], the optimal release of Rt* is obtained by solving
the following DP recursion equation:
∑ [φ[ I (ll ) Htc+1 (ii), Itc (l )]

IDN
[
ft S (k ), I (l ), H
t
c
t
c c
t +1 (ii)] = min{Bt [S (k ), I (l ), S (kk )] + α
t
c
t
c c
t +1
c
t +1
ll =1
∑ [π[ H }
FDN
⋅ c
t +2 (iv) Itc+1 (ll ), Htc+1 (ii), Itc (l ) ] ⋅ ft*+1[Stc+1 (kk ), Itc+1 (ll ), Htc+2 (iv) ]]]
iv =1
(8.56)
where:
Bt [ Stc (k ), Itc (l ), Stc+1 (kk )] is the loss during time period t.

φt +1 ( Itc+1 (ll ) | Htc+1 (ii), Itc (l )) is the posterior flow transition (mass function)
from the forecast in the next month and the current actual flow to the next
actual flow.
π[ Htc+2 (iv) | Itc+1 (ll ), Htc+1 (ii), Itc (l )] is the posterior forecast transition probabil-
ity function from the next actual and predicted flow in time period t + 1
and the current actual flow to future predicted flow in time period t + 2.
In this model, the posterior probabilities are estimated considering the Bayesian
process. The prior flow transition probabilities ρt +1[ It +1 (l ) | It (lm)] and likelihoods
λ t +1[ Ht +1 (i) | It +1 (l )] can be calculated from the historical and predicted inflow time
series. Then, the posterior probabilities are estimated as follows:
( )
φ t +1 Itc+1 (ll ) Htc+1 (ii), Itc (l ) =
[ ]
λ t +1 Ht +1 (i) It +1 (l ) ⋅ ρt +1 It +1 (l ) It (lm) [ ] (8.57)
IDN
∑ λ [H
l =1
t +1 t +1 (i) It +1 (l )] ⋅ ρt +1[ It +1 (l ) It (lm)]
[ ] ∑ {λ [ H
IDN
π Htc+2 (iv) Itc+1 (ll ), Htc+1 (ii), Itc (l ) = t +1 t +2 (ii) It +2 (ll )]

ll =1
∑ [φ [ I
FDN
⋅ t +1 t +2 (ll ) Ht +2 (ii), It +1 (l )]
ii =1
⎧⎡ IDN ⎤⎫
IDN ⎪⎢ ∑ [ ] [
λ t +1 Ht +1 (i) It +1 (l ) ⋅ ρt +1 It +1 (l ) It (lm) ⎥ ⎪ ]
⋅∑ ⎪⎢
⎨⎣ l =1
⎪
⎥⎦ ⎪⎬
⎪
( )
l =1
⎪⎩⋅φ t +1 It +1 (ll ) Ht +1 (ii), It (l ) ⎪⎭
c c c
(8.58)
330
TABLE 8.7
SDP Solution for Reservoir Operation Problem (Example 8.6)
t St qt St+1 qt+1 Rt ft P(qt|qt-1) ft(St,Rt ) Rt* S*t+1
First Iteration
10 1 3 1 3 3 9 0.4 9 2 2
7 0.6
2 3 2 4 0.4 4
7 0.6

7 1 3 7 49 0.5 49 6 2
7 0.5
2 3 6 36 0.5 36
7 0.5
2 3 1 3 4 16 0.4 16 3 2
7 0.6
2 3 3 9 0.4 9
7 0.6
7 1 3 8 64 0.5 64 7 2
7 0.5
2 3 7 49 0.5 49
7 0.5
River–Reservoir Systems Modeling
Second Iteration
9 1 3 1 3 3 9 0.4 9 + 0.4 × 4 + 0.6 × 36 = 32.2 3 1
7 0.6
2 3 2 4 0.4 4 + 0.4 × 9 + 0.6 × 49 = 37
7 0.6
7 1 3 7 49 0.5 49 + 0.5 × 4 + 0.5 × 36 = 69 6 2
7 0.5
2 3 6 36 0.5 36 + 0.5 × 9 + 0.5 × 49 = 65
7 0.5
2 3 1 3 4 16 0.4 16 + 0.4 × 4 + 0.6 × 36 = 39.2 4 1
7 0.6
2 3 3 9 0.4 9 + 0.4 × 9 + 0.6 × 49 = 42
7 0.6
7 1 3 8 64 0.5 64 + 0.5 × 4 + 0.5 × 36 = 84 7 2
7 0.5
2 3 7 49 0.5 49 + 0.5 × 9 + 0.5 × 49 = 78
7 0.5
(continued)
331
332
TABLE 8.7 (CONTINUED)
SDP Solution for Reservoir Operation Problem (Example 8.6)
t St qt St+1 qt+1 Rt ft P(qt|qt-1) ft(St,Rt ) Rt* S*t+1
Third Iteration
8 1 3 1 3 3 9 0.4 9 + 0.4 × 32.2 + 0.6 × 65 = 60.88 3 1
7 0.6
2 3 2 4 0.4 4 + 0.4 × 39.2 + 0.6 × 78 = 66.48
7 0.6
7 1 3 7 49 0.5 49 + 0.5 × 32.2 + 0.5 × 65 = 97.6 6 2
7 0.5
2 3 6 36 0.5 36 + 0.5 × 39.2 + 0.5 × 78 = 94.6
7 0.5

2 3 1 3 4 16 0.4 16 + 0.4 × 32.2 + 0.6 × 65 = 67.88 4 1
7 0.6
2 3 3 9 0.4 9 + 0.4 × 39.2 + 0.6 × 78 = 71.4
7 0.6
7 1 3 8 64 0.5 64 + 0.5 × 32.2 + 0.5 × 65 = 112.6 7 2
7 0.5
2 3 7 49 0.5 49 + 0.5 × 39.2 + 0.5 × 78 = 107.6
7 0.5
The Demand-driven Stochastic Dynamic Programming (DDSP) model was

developed by Vasiliadis and Karamouz (1994) to generate operating policies. It
accounts not only for the natural and forecast uncertainties of the system (as does
BSDP) but also for the seasonal (monthly) variation of various parameters (such as
inflow and demand) in the system. The major characteristics of the DDSP model
are as follows:
• Flow transition probabilities associated with seasonal variations are con-

tinuously updated on a monthly basis using Bayesian decision theory
(BDT).
• The model is constructed to accept variable and uncertain demands.
• Each iteration is considered to represent 12 months of a year at a time,
thus the computational efforts are reduced.
• The cost function includes release and expected storage.
A flow discretization scheme developed by Karamouz and Vasiliadis (1992) is

used in this model. With this scheme, the characteristics flow values preserve the
mean and standard deviation of the flow time series. In DDSP, the decision variable
(release) depends on the state of the system, which is defined by the following
variables:
M, number of month (M = 1, …, 12)

DM, corresponding demand in month M
SM, reservoir storage at the beginning of month M (SM = 0, …, Cap)
IM, inflow to the reservoir during month M (IM = I1, …, IIDN)
HM+1, forecast of flow for the next month (M + 1), where HM+1 = H1, …, HFDN
The recursive function at year Y and month M should be written in terms of char-
acteristic values of the above variables:
fY , M ( DMc , SMc , IMc , HMc +1 ) = Min{E( DMc +1 ).CRM + CSM +2 + ( E( IMc +1 HMc +1, IMc )
(8.59)
.{E( HMc +2 IMc +1 ).[ f *Y , M +1 ( DMc +1, SMc +1, IMc +1, HMc +2 )]})]}
where:
Y is the index of years/iterations.

CRM is the cost related to release during month M.
CSM+2 is the cost related to expected ending storage in month M + 2.
E is the expected value.
DMc is the characteristic value of water demand in month M.
SMc is the characteristic value of reservoir storage at the beginning of month M.
I Mc is the characteristic value of inflow to the reservoir during month M.
H Mc is the characteristic value of inflow forecast for month M.
In Eq. (8.59), where the demand uncertainty is incorporated in the expected value
of demands (E(DM+1)), it is assumed that inflow and demand are independent sto-
chastic variables. However, if the water supply (inflow) and demand are site specific,
the expectation of E(DM+1⏐IM) can be used in this equation. Formulation of the loss
function and the optimal operation policies for this model are explained in the next
sections.
8.8 MULTIPLE RESERVOIR OPERATION MODELING

In this section, the single reservoir operation model is extended to a system of
cascade and/or parallel reservoirs. The basic issue examined here is how much water
can be supplied to different demand points from which sources. As an example, the
framework of an optimization model for a system of reservoirs consisting of both
cascade and parallel reservoirs, taking into consideration priority-based heuristics
for water allocation, is also presented in the following sections.
8.8.1 CASCADE RESERVOIRS

Cascade systems have more than one reservoir on the same river (see Figure 8.16).
It is assumed that demand points are located along the river downstream of the
reservoirs. Also, local rivers join the main river at various reaches. The optimization
model for this system is as follows:
n NR mi
Minimize z= ∑ ∑ ∑ loss ( RA
t =1 i =1 j =1
i , j ,t i , j ,t , Di, j ,t , Si,t ) (8.60)
Subject to: S1,t +1 = S1,t + I1,t − R1,t − E1,t − L1,t (t = 1,…, n) (8.61)
mi
Si,t +1 = Si,t + ∑(R

j =1
i −1,t )
− RAi −1, j ,t + qi −1,t − Ri,t − Ei,t − Li,t
, (8.62)
(t = 1,… n) (i = 2,…, NR)
mi
Ii,t = Ri −1,t − ∑ RA
j =1
i , j ,t + qi −1 (t = 1,…, n)(i = 2,…, NR) (8.63)
Si,min ≤ Si,t ≤ Capi (t = 1,…, n)(i = 1,…, NR) (8.64)
0 ≤ Ri,t ≤ Ri,max,t (t = 1,…, n)(i = 1,…, NR) (8.65)
Si,t , Ei,t , Li,t , Ri,t ≥ 0 (t = 1,…, n)(i = 1,…, NR) (8.66)

q1 qNR
i=1 i=2 i=NR

I
D1,1 D1,2 D1,m DNR,1 DNR,2 DNR,m
FIGURE 8.16 Cascade reservoir system.
where:
RAi,j,t is the water allocated to demand j from reservoir i in month t (volume

of water allocated to the demand j minus volume of return flows in reach i).
Ri,t is the release from reservoir i in month t.
Ri,max,t is the maximum allowable release from reservoir i in month t.
qi,t is the local flow downstream of reservoir i in month t.
Ei,t is the volume of water evaporated from reservoir i in month t.
Li,t is the volume of water leakage from reservoir i in month t.
Si,t is the volume of water storage in reservoir i at the beginning of month t.
NR is the number of river reaches.
Di,j,t is the water demand j downstream of reservoir i in month t.
Capi is the total capacity of reservoir i.
Si,min is the minimum allowable storage in reservoir i.
Ii,t is the inflow to reservoir i in month t.
mi is the number of demands in reach i.
Using this model, the contribution of each reservoir to supplying the different
demands can be determined.
8.8.2 PARALLEL RESERVOIRS

Parallel reservoirs have a junction point with other rivers in the system. It is assumed
that demand points are located downstream of each reservoir and after the junction
point. Reservoirs in parallel require special modeling when demand points down-
stream of the junction point should be supplied from upstream reservoirs; the state
of each reservoir for supplying direct demands must then be considered. Formulation
of the optimization problem can be as follows:
n NR +1 mi
Minimize z= ∑ ∑ ∑ loss ( RA
t =1 i =1 j =1
i , j ,t i , j ,t , Di, j ,t , Si,t ) (8.67)
Subject to: Si +1,t +1 = Si,t + Ii,t − Ri,t − Ei,t − Li,t (t = 1,…, n)(i = 1,…, NR) (8.68)
i=1
I1
i=2 D1,1 D1,2 D1,m

I2
D2,1 D2,2 D2,m

DNR+1,1 DNR+1,2 DNR+1,m
i=NR
INR
DNR,1 DNR,2 DNR,m
FIGURE 8.17 Parallel reservoir system.
NR mi
I NR+1,t = ∑ R − ∑ RA
i =1
i ,t
j =1
i , j ,t (t = 1,…, n) (8.69)
Si,min ≤ Si,t ≤ Capi (t = 1,…, n)(i = 1,…, NR) (8.70)
0 ≤ Ri,t ≤ Ri,max,t (t = 1,…, n)(i = 1,…, NR) (8.71)
Si,t , Ei,t , Li,t , Ri,t ≥ 0 (t = 1,…, n)(i = 1,…, NR) (8.72)
By solving the above formulation, the contribution of each reservoir toward supply-
ing the demand points along the NR + 1 reach is determined.
Example 8.8
Consider a system of three reservoirs as shown in Figure 8.18. In this system, two
parallel reservoirs are in series with a single reservoir; demand points are considered
downstream of each reservoir. As shown in Figure 8.18, agricultural lands are
downstream of dam 1. Also, towns A and B are located downstream of dam 2 and
dam 3; navigation and the instream flow rights exist downstream and upstream of
dam 3, respectively. A third river joins the releases of dams 1 and 2 in the junction
point of the two downstream rivers. Formulate the optimization model for finding
optimal outflows from these reservoirs.
Solution: The following formulation may be used for optimization of this
multiple reservoir system:
n 3
Minimize z= ∑ ∑ loss ( RA , D , S )
t =1 i =1
i ,t i ,t i ,t i ,t (8.73)
Subject to: Si +1,t +1 = Si,t + Ii,t − Ri,t − Ei,t − Li,t (t = 1,…, n) (i = 1, 2) (8.74)
S3,t +1 = S3,t + I3,t + ∑(R

i =1
i ,t − RAi,t ) −E3,t − L3,t (t = 1,…, n) (8.75)
Si,min ≤ Si,t ≤ Capi (t = 1,…, n)(i = 1, 2, 3) (8.76)
0 ≤ Ri,t ≤ Ri,max,t (t = 1,…, n)(i = 1, 2, 3) (8.77)
∑(R
i =1
i ,t ) (
− RAi,t + I3,t ≥ Qu,m (t = 1,…, n) m = t − 12 [(t − 1) / 12] (8.78) )
(
( R3 − RAi,3 ) ≥ Qd ,m ∀t m = t − 12 [(t − 1) / 12] ) (8.79)
Si,t , Ei,t , Li,t , Ri,t ≥ 0 (t = 1,…, n)(t = 1,…, 3) (8.80)
where:
Capi is the total capacity of dam i.

D1,m is the water demand of agricultural lands downstream of dam 1 in
month m.
D2,m is the water demand of town A downstream of dam 2 in month m.
D3,m is the water demand of town B downstream of dam 3 in month m.
Qu,m is the minimum instream water required upstream of dam 3 in month m.
Qd,m is the minimum water required for navigation downstream of dam 3 in
month m.
RAi,t is the water allocated to demand downstream of dam i in month t.
8.9 RESERVOIR OPERATING RULES

Real-time operation of river–reservoir systems requires specific operating rules.
These rules are guides for water conservation and release policies prepared for
reservoir operators. The several types of rules range from very simple and static to
dynamic for considering the varying states of inflow and physical characteristics of
a reservoir in each time period. One of the simplest rules for reservoir operation is
the rule curve, which specifies the target storage at the end of each month
FIGURE 8.18 Example of a multiple reservoir system.
(Figure 8.19). Rule curves are static forms of operating policies that do not get any
feedback from reservoir storage and current hydrologic situations such as predicted
inflows to reservoirs in the following months. Rule curves are not very efficient
policies, particularly when inflows and demands are highly varied, but they have
been widely used because of their simplicity.
Application of deterministic and stochastic optimization models for reservoir
operation has led investigators to define nonstatic types of operating rules. Karamouz
et al. (1992) have used the following form of operating rule in the DPR model for
optimization of multiple reservoir systems:
( )
Rt = am It + bm St + cm m = t − 12 [(t − 1) / 12] (8.81)
where am, bm, and cm are rule constants in month m. In DPR models, the optimal
releases, optimal storages, and historical inflows are regressed to estimate the above
operating rule. More complex nonlinear forms of operating rules are also tested by
different researchers. Results of these studies have shown that the simple linear rules
are as good as or sometimes better than nonlinear rules (Bhaskar and Whitlatch,
1980). This type of operating policy takes into account the state of the system in
each month and has shown a better reliability in satisfying different objectives of
the reservoir operation.
Another type of operation policy is produced by stochastic dynamic program-
ming models. In these policies, the decision variable (release) depends on variables
representing the state of the system for each month, including:
• Inflow to the reservoir

• Water demand
• Storage at the beginning of a month
• Inflow forecast for the next month
Water level at the end of month (m)
Oct. Nov. Dec. Jan. Feb. Mar. Apr. May June July Aug. Sep.
FIGURE 8.19 Typical reservoir rule curve.
TABLE 8.8
DDSP Model Optimal Operation Policies
Demand Storage at First Inflow Forecast Inflow Forecast Release
No. of in Month t of Month t in Month t in Month t + 1 in Month t
Month (106 m3) (106 m3) (106 m3) (106 m3) (106 m3)
1 20 150 35 35 15
1 20 150 35 30 15
1 20 150 35 25 10
1 20 150 30 35 15
1 20 150 30 30 10
1 20 150 30 25 10
1 20 150 25 35 15
1 20 150 25 30 15
1 20 150 25 25 5
Therefore, the operation policies in these models are like the contents of a table in
which various combinations of characteristic values for a state variable and the
optimal release for that selection are presented in each row. Table 8.8 shows a part
of the optimal operating policies derived from the DDSP model as developed by
Vasiliadis and Karamouz (1994). The DDSP model is a demand-driven stochastic
dynamic programming model that takes into account all of the above state variables.
Neural networks and fuzzy systems have been widely used for reservoir opera-
tion modeling. These techniques estimate input/output functions, and all are trainable
dynamic systems. Unlike statistical estimators, they estimate a function without a
mathematical model of how outputs depend on inputs (Hasebe and Nagayama, 1996).
Application of these techniques for defining reservoir operation policies has been
considered by many investigators, including Raman and Chandramouli (1996) and
Shrestha et al. (1996).
Artificial neural networks consist of numerous, simple processing units or “neu-
rons” that can be globally programmed for computation. In reservoir operation
Fuzzy membership
5 15 25 5 10 15 90 110 130 10 15 20
Forecast inflow Demand Storage Release
FIGURE 8.20 An example of fuzzy reservoir operating rules.
modeling, historical records of inflow to the reservoirs, demands, and optimal values
of release and storage in the reservoir obtained from any optimization algorithm can
be used for training a neural network system. This trained neural network can be
used as an operating policy in real-time operation.
In a similar case, these inputs can be used for defining fuzzy rules. The following
steps should be taken to develop fuzzy rules:
1. Divide input/output space into fuzzy regions.

2. Generate fuzzy rules from the given data.
3. Calculate the degree of fulfillment for each rule.
4. Assign a weight to each rule.
Various types of fuzzy membership functions, such as triangular or trapezoidal

shapes, have been used by investigators (Figure 8.20). Selection of the proper shape
and domain of the membership functions can be done based on the variability of
each input variable and the sensitivity of the rules to these variations. The degree
of fulfillment is a threshold for selecting more effective rules in real-life situations.
Figure 8.20 shows an example of fuzzy reservoir operating rules. Details of neural
networks and fuzzy set theory are discussed in Chapter 3.
Lund and Guzman (1999) classified the conceptual operating rules for reservoirs
in series and in parallel into the categories shown in Tables 8.9 and 8.10. As seen
in these tables, some of the conceptual rules for multiple reservoir systems show
compatibility between different reservoir purposes, such as flood control, water
supply, and energy storage for refill periods on reservoirs in series. However, in
many complex systems, it is necessary to use optimization models and methods such
as regression, neural network, and fuzzy set theory for defining optimal operating
policies.
8.10 COST FUNCTIONS

Reservoir operation cost functions can be classified into two categories:
• Relative cost functions

• Economic based cost functions
TABLE 8.9
Conceptual Rules for Reservoirs in Series
Purpose Refill Period Drawdown Period
Water supply Fill upper reservoirs first. Empty lower reservoirs first.
Flood control Fill upper reservoirs first. Empty lower reservoirs first.
Energy storage Fill upper reservoirs first. Empty lower reservoirs first.
Hydropower production Maximize storage in reservoirs with Maximize storage in reservoirs
greatest energy production. with greatest energy production.
Recreation Not applicable. Equalize marginal recreation
improvement of additional
storage among reservoirs.
Source: Lund, J. R. and Guzman, J., ASCE J. Water Resources Plan. Manage., 125(3), 143–-153, 1999.
With permission.
TABLE 8.10
Conceptual Rules for Reservoirs in Parallel
Purpose Refill Period Drawdown Period
Water supply Equalize probability of seasonal spill Equalize probability of emptying among
among reservoirs. reservoirs.
Flood control Leave more storage space in reservoirs Not applicable.
subject to flooding.
Energy storage Equalize expected value of seasonal For last time step, equalize expected value
energy spill among reservoirs. of refill season energy spill among
reservoirs.
Water quality Equalize expected value of marginal For last time step, equalize expected value
seasonal water quality spill among of refill season water quality spill among
reservoirs. reservoirs
Hydropower Maximize storage in reservoirs with Maximize storage in reservoirs with
production greatest energy production. greatest energy production.
Recreation Equalize marginal recreation Equalize marginal recreation
improvement of additional storage improvement of additional storage
among reservoirs. among reservoirs.
Source: Lund, J. R. and Guzman, J., ASCE J. Water Resources Plan. Manage., 125(3), 143–153, 1999.
With permission.
200000 OPERATIONS:
STORAGE
180000 IDEAL
RELEASE
160000 FEASIBLE
CS CR
140000 INFEASIBLE
120000 LEGEND
CR: Release Cost Function
100000 CS CR CS: Storage Cost Function
R: Release during month M
D: Demand for month M
80000 S: Available expected
storage at the beginning
of month M+2
60000 CAP: Capacity of the
reservoir
40000
20000
0
XR = R / D
-1 -0.5 0 0.5 0.8 11.2 1.5 2 2.5 3 XS = S / CAP
CR = 1.58 × 104 × [10(0.8–XR) –1] CR = 3.88 × 103 × [10(XR–1.2) –1]

CS = 1013.867 × (10 [0.4–XS )×2.51]
–1) (
CS = 307.228× 10[(XS–0.6)×2.45] –1 )
FIGURE 8.21 A typical relative cost function. (From Karamouz, M. and Vasiliadas, H., Water
Resources Research, 28(5), 1221–1232, 1992. With permission.)
In the relative case, the structure of the cost function reflects the cost of deviation
from targets. These targets can be in terms of release, storage, or both. Figure 8.21
shows an example of the ratio of release to demand and storage cost functions. As
shown in this figure, the cost functions are convex exponential curves and are
developed for comparison purposes. In both sets of cost functions, the x-axis is
dimensionless (either release over demand or expected storage over capacity). The
cost functions include a specific release safe range (e.g., 80 to 120% of the demand),
and a specific storage safe level (e.g., 60 to 80% reservoir capacity for conservation
and a higher storage level and perhaps 70 to 85% for generating hydropower). No
loss is associated with being in the safe range because:
• Sufficient water is released to satisfy all consumptive needs.

• No flooding occurs because the release is kept at acceptable rates.
• Reservoir storage is kept at a relatively high pool level for conservation
and hydropower generation.
Outside the safe range, the losses increase exponentially as the deviations from
targets increase.
In the real cost function, the prices of water and power are considered. In
estimating the price of power, the following steps should be taken (Karamouz et al.,
2002):
• The present value of initial investment is estimated based on the age of

the dam and its related facilities, the cumulative interest rate, and rate of
depreciation.
• The present value is then distributed over the expected remaining life of
the dam on a monthly basis.
• The cost of maintenance and operation should be added to the monthly cost.
The price of power can also be taken into account when considering the duration
of power generation in peak and off-peak hours. Attempts to apply actual cost
functions usually encounter data insufficiencies, but this type of cost function gives
a better picture of the capabilities of operating tools for managers and decision
makers in terms of monetary benefits and costs.
8.11 EFFICIENCY OF OPERATING POLICIES

Evaluating the performance of a measuring system is the final step in the application
of simulation and optimization models for river–reservoir systems planning and
management. The operational status of water resources systems can be classified as
satisfactory, unsatisfactory, or failed. Failure can be structural or operational; struc-
tural failure of a dam can occur because of flood, earthquake, etc. In this section,
operational failure of the system is of concern and we want to evaluate how well
the objectives have been satisfied. In other words, the system status in time period
t expressed by the random variable Xt can be in set S, which is the set of all
satisfactory outputs, or in set F, which is the set of all unsatisfactory outputs.
First, the performance criteria or indices that help the planner to classify the
status of a system as S or F should be defined. Even though appropriate definitions
for the performance indices depend on the problem and objectives of planning, some
basic concepts are similar.
Hashimoto et al. (1982) described systems performance from three different
viewpoints:
• How often the system fails (reliability)

• How quickly the system returns to a satisfactory state once a failure has
occurred (resiliency)
• How significant the likely consequences of failure may be (vulnerability)
In the following sections, some of the basic performance measures in each of

the above categories are presented in order to assess the performance of operation
policies for a river–reservoir system.
8.11.1 RELIABILITY
Reliability is the probability that no failure occurs within a fixed period of time:
[
α = Prob Xt ∈ S ] ∀t (8.82)
Based on this definition, reliability is the opposite of risk, in which the probability
of system failure is expressed. Another definition of reliability is the probability that
no failure will occur within the planning horizon. This indicator is a measure of
system performance in meeting target values, which is an important indicator for
analyzing the performance of water resources systems in normal conditions.
In the case of a multipurpose river–reservoir system, reliability can be defined
in different ways. For example, if the reservoir supplies water and or energy demands,
reliability can be defined by the probability that a specific percentage of water and
energy demands is supplied within the planning horizon.
Example 8.9
The monthly reservoir releases in 3 years are shown in Table 8.11. The monthly
water demand downstream of the reservoir is 10 million m3. Estimate the reliability
of supplying 100% and >80% of demands.
Solution: The reliability of supplying 100% of the monthly demand (α100) can
be estimated as the fraction of time that system performance is satisfactory:
No. of months that 100% of demands are supplied 19

α100 = = = 52.7%
Total No. of Months 36
The reliability of supplying at least 80% of the monthly demands can also be
estimated as follows:
No. of months that more than 80% of demands are supplied 28

α 80 = = = 77.7%
Total No. of Months 36
It should be noted that the above approach can be used only when the time period is
long enough that the frequency of occurrence of a sample approaches the probability.
8.11.2 RESILIENCY
Resiliency describes how quickly a system recovers from failure once failure has
occurred. The resiliency of a system can be considered in the planning horizon as
follows:
{
β = Prob Xt +1 ∈ S Xt ∈ F} (8.83)
TABLE 8.11
Reservoir Releases for a 3-Year Period (Example 8.9)
Year 1 Year 2 Year 3
Month (million m3) (million m3) (million m3)
January 10 8 9
February 12 10 11
March 11 11 10
April 8 8 8
May 7 6 5
June 7 7 6
July 8 9 5
August 10 10 7
September 11 11 9
October 10 9 10
November 11 10 12
December 13 12 11
Annual deficit 10 13 21
As can be seen, resiliency is basically a measure of the duration of an unsatisfactory

condition. This indicator is important in drought and flood conditions because
damages and costs associated with floods and droughts are affected greatly by the
duration of unsatisfactory operations.
Example 8.10
Estimate the resiliency for the river–reservoir system of Example 8.9.

Solution: As shown in Table 8.11, the transition from unsatisfactory to satisfac-
tory states has occurred in the following months:
Year 1: July
Year 2: January, July, and October
Year 3: January and September
Therefore, the probability shown in Eq. (8.83) can be estimated as:
6
β= = 35%
17
If the duration of unsatisfactory condition is needed, then the inverse of β is the

average duration of the unsatisfactory condition.
8.11.3 VULNERABILITY
Vulnerability measures the possible magnitude of a failure if one occurs. For mea-
suring vulnerability, a severity index should be defined. For example, when the main
objective of a river–reservoir system is satisfying water demands, the severity index

(sj) can be defined as the volume of shortage in each time interval. Hashimoto et al.
(1982) defined overall system vulnerability as follows:
υ= ∑s e
j ∈F
j j (8.84)
where ej is the probability that xj, corresponding to sj, is the most unsatisfactory and
severe outcome that occurs among a set of unsatisfactory states. Another definition
for vulnerability was presented by Datta and Burges (1984), who defined the vul-
nerability of a system as the total deficit in the planning time horizon.
Example 8.11
Estimate the vulnerability for the river–reservoir system of Example 8.9.

Solution: The vulnerability can be estimated based on the total deficit occurring
within the planning horizon. As shown in Table 8.11, the total deficit of the system
can be estimated as:
υ = 10 + 13 + 21 = 44 million m3
8.12 CONFLICT ISSUES IN RIVER–RESERVOIR SYSTEMS

Reservoirs usually serve multiple objectives, many of which may be in conflict at
any given time. For example:
• Several consumptive demand points are located downstream of the reser-

voirs, and water supplied to one of these points cannot be used by others;
therefore, the major conflict issue in reservoir operation in this case occurs
when the reservoir is not capable of supplying all of the demands.
• Keeping enough flood control storage conflicts with water conservation
strategies for storing water in high-flow seasons.
• Supplying instream requirements conflicts with water conservation and
supply objectives.
• Monthly variations of power loads (energy demand) and water demands
do not follow the same pattern; therefore, supplying one of these demands
in a specific period of time might conflict with supplying another in the
future.
In the first item, the main conflict issue in river–reservoir systems planning and
operation happens when the reservoir should supply water to various demand points
for different purposes. In order to formulate this problem, consider a river–reservoir
system that supplies the following demands:

• Instream flows and environmental water rights for health care
Even though water resources planning for reservoir operation is done by a

specific agency (called here the Department of Water Supply) on a regional basis,
the water allocation schemes defined and imposed by this department are highly
affected by other agencies, such as:
The priorities and favorable ranges of water supply for each of these agencies
should be considered in formulating the conflict resolution problems; therefore, the
first step in conflict resolution studies is to recognize all of the conflict issues and
the responsible agencies.
Each of these departments has its own set of priorities for allocating water to
different demands. In the second step, the relative priority of supplying water
demands for all related agencies should be defined. For example, irrigation demands
have the highest priority for the Department of Agriculture. Other demands, such
as industrial water demand, might be a second priority for this department because
of the water required for agro-industrial setups.
The next step is to define the acceptable range of demand based on the priority
of demands for each agency. For example, the Department of Agriculture could
define the utility function for a range of water allocation for irrigation purposes as
shown in Figure 8.22. The most favorable annual volume of water allocation for
this organization ranges from Vb to Vc million m3. This range can be estimated based
on the variability of irrigation demands. Variations in irrigation demands are a
function of variations of net crop water demand due to climate change and uncer-
tainties associated with market price shifts and other socioeconomic factors, as well
as the prospect of future development of the agricultural sector. More details about
water demand variations are presented in Chapter 10.
1.2
1
0.8
Utility
0.6
0.4
0.2
0
Va Vb Vc Vd
Volume (million m )3
FIGURE 8.22 Utility function of different agencies for water allocated to irrigation demands.
The increasing segment on the left-hand side of this figure shows the range of
least favorable to most favorable volume of water that might be allocated to irrigation
projects. The points on this line are not totally rejected by this organization because
the shortage can be supplied from other sources, such as more expensive groundwater
resources. The decreasing segment on the right-hand side of this graph shows that
even though the water allocated to irrigation projects supplies all the demands, these
conditions are also less favorable because of probable flood damages and violation
of water conservation concerns, which might cause shortages in the future.
All the agencies selected in the first step should provide their favorable range
of water supply for their various demands. Besides different water demands and
priorities of agencies in supplying those demands, each agency has a specific level
of authority with respect to changing the water allocation schemes and imposing a
favorable range of water supply in the political and institutional climate of each
region. For example, the Department of Agriculture has an interest in supplying a
relatively higher volume of water to the Department of Industries than the Depart-
ment of Water Supply. The Department of Industries is comprised of major agencies
having higher authority in industrial water allocation schemes. Analysts can also
define a set of relative weights based on the authority of these agencies with regard
to the final decision on water allocation. Different agencies can also affect the final
decision based on the social and political climate of each region.
The final step is to formulate the conflict resolution problem. For this purpose,
different methods can be used (as briefly explained in Chapter 2). The steps that
should be taken for conflict resolution in river–reservoirs systems modeling can be
summarized as follows:
1. The conflict issues should be recognized based on the operation objectives

of the river–reservoir systems.
2. The agencies engaged in water resources allocation and consumption
should be recognized.
3. The agencies considered in step 2 should provide their own sets of favor-
able ranges of water supply to different demands.
4. The analyst might define the relative authority of agencies in imposing
the allocated water to each demand.
The formulation of a conflict resolution problem in river–reservoir systems is as

follows:
4
⎛ 4
⎞
Maximize U ( x1 , x 2 , x3 , x 4 ) = ∑ ∑ (µ
j =1
⎜
⎝ i =1
i, j ( xi ) × Wi, j )⎟
⎠
(8.85)
Subject to: x1 + x 2 + x3 + x 4 = R (8.86)

where:
x1, x2, x3, x4 are the annual water allocation to domestic, agricultural, indus-
trial, and environmental demands, respectively (million m3).
Wi,j is the assigned weight to agency i in imposing allocated water to demand j.
µi,j(x) is the degree of acceptance of agency i if x units of water are allocated
to demand j.
U(X1, X2, X3, X4) is the utility function of the system if x1, x2, x3, and x4 units
of water are supplied to demands 1, 2, 3, and 4, respectively.
R is the annual release from the reservoir (million m3)
Example 8.12
In a river–reservoir system, the following agencies are affected by the decisions
made to release water from the reservoir:

• Agency 4: Department of Environmental Protection
The main objective is to meet the water demand that is of concern to these
agencies. The Department of Water Supply has a twofold role — namely, to allocate
water to different purposes as well as to supply water for domestic purposes (DD).
The irrigation demand (RD), industrial demand (ID), and instream flow (environ-
mental) water demand (IF) should also be provided. The decision makers in these
agencies are asked to set the most favorable range of water allocation for each
demand by assigning a 1 (most favorable) or 0 (least favorable) (see Figures 8.23
to 8.26). The analyst sets the following weights to the role and authority of each
agency keeping in mind the political climate of that region:
Agency 1 2 3 4
DD ⎡5 1 1 3⎤
ID ⎢5 2 5 2⎥
MATA = ⎢ ⎥
RD ⎢5 5 2 2⎥
⎢ ⎥
IF ⎣4 2 2 5⎦
where MATA is the matrix of relative authority of agencies.

The average annual inflow to the reservoir is estimated as 1400 million m3. Find
the most appropriate water allocation scheme for a normal year and years with the
projected inflows of 1000, 1200, and 1300 million m3.
Solution: In order to use Eq. (8.85), the values of Wi,j are estimated by normal-
izing the numbers in the rows of matrix MATA. The normalized values are as follows:
350
FIGURE 8.23 Utility
function of different
agencies for water allo-
cated to domestic
demand.
FIGURE 8.24 U t i l i t y
cated to agricultural
demand.
351
352
FIGURE 8.25 Utility
cated to industrial
demand.
FIGURE 8.26 U t i l i t y
cated to environmental
water demand.
353
Agency: 1 2 3 4
DD ⎡ 0.5 0.1 0.1 0.3 ⎤
ID ⎢0.357 0.142 0.357 0.143⎥
MATA = ⎢ ⎥
RD ⎢0.357 0.357 0.143 0.143⎥
⎢ ⎥
IF ⎣0.308 0.154 0.154 0.385⎦
µi,j can also be computed using Figures 8.23 to 8.26. For example, allocating 1200
million m3 to irrigation demand is the most favorable option for agency 2 (Depart-
ment of Agriculture), while it is the least favorable option for agency 4 (Department
of Environmental Protection). The linear programming technique is used for solving
this problem assuming that release R is equal to inflows of 1000, 1200, 1300, and
1400 million m3. The results of water allocation are shown in Table 8.12.
For this particular system, the annual water demand for the above purposes has
been estimated by an independent party to be, at a minimum, about 150, 960, 110,
and 70 million m3 for the four agencies, respectively, and the agencies are aware of
these estimates. As can be seen in the table, in the dry year scenario (inflow = 1000),
the domestic and environmental demands have been supplied, while the shortage is
imposed on industrial and agricultural demands. By increasing the annual inflow to
1200 and 1300 more water is allocated to the agricultural demands while some
shortages are still imposed on industrial sector. When the inflow is increased to 1400
million m3 all of the demands are supplied.
Another application of MCDM methods in solving conflict resolution problems
can be formulated by minimizing the distance from the most favorable point asso-
ciated with each objective. In case of water supply to different demands from the
reservoir, fi can be estimated based on the authority of each decision maker (respon-
sible agency) as follows:
fi ( xi ) = ∑ (µ
j =1
i, j ( xi ) × Wi, j ) (8.87)
TABLE 8.12
Results of Linear Programming for Finding the Most Appropriate Water
Allocation Scheme (Example 8.12)
Demand R = 1000 R = 1200 R = 1300 R = 1400
Domestic 150 150 150 155

Agricultural 690 890 990 1040
Industrial 90 90 90 130
Environmental 70 70 70 75
where:
Wi,j is the assigned weight to agency i in imposing allocated water to demand j.

µi,j(xi) is the degree of acceptance of agency i if xi units of water are allocated
to demand j.
m is the total number of decision makers.
The minimum distance from the most favorable points can be found as follows:
Minimize d (8.88)
(
Subject to: w1 ⋅ f1* ( x1 ) − f1 ( x1 ) ≤ d ) (8.89)
(
wn ⋅ fn* ( x n ) − fn ( x n ) ≤ d ) (8.90)
where:
d is the distance from the most favorable points associated with different
objectives.
fi* ( xi ) is the favorable point for objective i if xi units of water are allocated
to demand i.
fi(xi) is the value of objective function i if xi units of water are allocated to
demand i.
wi is the relative weight of objective i.
By solving the above problem, the optimal values for decision variables (x1, ⋯, xn),
which are the volume of water allocated to different demands, can be found.
Example 8.13
In the river–reservoir system explained in Example 8.12, find the optimal allocation
of 1400 and 1000 million m3 released from the reservoir using the most favorable
point method. In this problem, the four agencies decide on water allocation schemes.
Therefore, Eqs. 8.88 to 8.90 can be used as follows:
Minimize d
Subject to: (f 1
*
( x1 ) − f1 ( x1 )) ≤ d
(f 2
*
( x2 ) − f2 ( x2 )) ≤ d
(f 3
*
( x3 ) − f3 ( x3 )) ≤ d
(f 4
*
( x4 ) − f4 ( x4 )) ≤ d
TABLE 8.13
Results of the Most Favorable Point Method for
Finding the Most Appropriate Water Allocation
Scheme (Example 8.13)
R = 1400 R = 1000
Demand (million m3) (million m3)
Domestic 176 140

Agricultural 1008 708
Industrial 151 86
Environmental 65 66
Because no specific priority for each demand exists, the wi values are set to 1. The
fi values are estimated using Eq. (8.87), the utility function for different demands
shown in Figures 8.23(a), 8.24(b), 8.25(c), and 8.26(d), and the normalized value
of the relative authority of the agencies as shown in the normalized MATA. The
results of water allocation are shown in Table 8.13.
As mentioned in the previous example, the annual water demand for the above
purposes has been estimated by an independent party to be, at a minimum, about
150, 960, 110, and 70 million m3, respectively, and the agencies are aware of these
estimates. As shown in the table, in the dry year scenario (inflow = 1000), none of
the demands has been completely supplied but a more balanced level of shortage is
achieved compared with the previous method. Higher levels of shortages (22 to 27%)
are applied to industrial and agricultural demands, while the shortage in supplying
domestic and environmental demands is limited to 7%.
8.13 PROBLEMS
8.1 Consider reservoir A, which supplies water for domestic, irrigation, and
industrial purposes with monthly demands of Dd(m), Dir(m), and Di(m)
million m3 (m = 1, …, 12), respectively. The carrying capacity of the
downstream river is also important in the high flow months. The damage
due to reservoir releases (Rt, t = 1, …, N) higher than the carrying capacity
of the river (RT) is estimated as the difference between releases and the
river carrying capacity. The benefits of supplying demands are estimated
based on the price of water allocated to different demands, which are Pd,
Pir, and Pi for domestic, irrigation, and industrial purposes, respectively.
Formulate the optimization model for finding the optimal monthly releases
in T months when the reservoir inflow is It.
8.2 In problem 8.1, if a hydropower plant is constructed to produce electricity
from the reservoir releases, formulate the optimization model for finding
the optimal release from the reservoir. Consider the benefits of power
generation based on the price of power (Pp) and the discharge capacity
of the power plant (Qp).
8.3 In problem 8.1, if the minimum discharge required downstream of the

reservoir (after supplying water to demand points) is estimated to be Qmin
and the minimum reservoir storage is limited to Smin for recreation pur-
poses, show how the constraints of the optimization model change.
8.4 The variation of annual water demand of a city over a 30-year planning
time horizon is shown in the following table:
Year
2010 2020 2025 2035 2040
Water demand (million m3) 100 120 130 145 170
The current demand of this city is about 90 million m3. Three projects
are studied to supply the demands of this city. Considering the cost of
each project and a 3% rate of return, find the optimal sequence of imple-
menting projects using linear programming:
Capital Investment Capacity

Project ($ × 106) (million m3)
Dam A 5 35
Interbasin water transfer tunnel B 8 20
Dam C 7 35
8.5 Suppose you want to design the spillway of a dam. In the following table,
the variation between the flood return period and capital investment esti-
mated based on the required capacity of the spillway is shown. The total
damages associated with floods during various return periods are also
shown in this table.
Return Period Annual Probability of Capital Cost Damage

(years) Exceeding Capacity (units/year) (units)
100 0.01 10.05 0

200 0.005 10.6 250
500 0.002 16.05 5000
1000 0.001 22.70 7500
10000 0.0001 30.05 19500
(a) Find the design return period of the spillway using the reliability-
based approach.
(b) What is the risk of failure for your design?
8.6 Solve problem 6-4 by using dynamic programming.
8.7 Consider a reservoir that supplies water to an industrial complex. The
monthly water demand of this complex is 10 million m3. The total capacity
of the reservoir is 30 million m3. Let St (reservoir storage at the beginning
of month t) take the discrete values of 0, 10, 20, and 30. The cost of
operation (Loss) can be estimated as a function of the difference between
release (Rt) and water demand as follows:
⎧⎪0 Rt ≥ 10
Losst = ⎨
⎩⎪(10 − Rt ) Rt > 10
2
(a) Formulate a forward-moving deterministic dynamic programming

model for finding the optimal release in the next 3 months.
(b) Formulate a backward-moving deterministic dynamic programming
model for finding the optimal release in the next 3 months.
(c) Solve the DP model developed in part (a), assuming that the inflows
to the reservoir in the next three months (t = 1, 2, 3) are forecast to
be 10, 50, and 20, respectively. The reservoir storage at the current
month is 20 million m3.
8.8 The cumulative density function (CDF) of inflow to a reservoir (Q) is
FQ(q) = q/(q +1). Find the deterministic equivalent of the following con-
straints:
(a) Prob[x ≤ Q] ≥ 0.90
(b) Prob[x ≤ Q] ≤ 0.95
(c) Prob[x ≥ Q] ≥ 0.75
(d) Prob[x ≥ Q] ≤ 0.80
8.9 The annual inflow to a reservoir (q) is classified as low (5 ≤ q < 15),
medium (15 ≤ q < 25), or high (25 ≤ q < 35). The reservoir inflow transition
probabilities are estimated as follows:
low med. high

low ⎡0.5 0.3 0.2 ⎤
( )
Prob qt +1 qt = med. ⎢⎢0.3 0.3 0.4 ⎥⎥
high ⎢⎣ 0.1 0.5 0.4 ⎥⎦
(a) If the inflow in the current year is low (5 ≤ q < 15), how much is
the probability of having high inflow (25 ≤ q < 35) 2 years from
now?
(b) How much is the probability of having a low, medium, and high
flow year a few years from now?
8.10 The historical record of monthly inflows to a reservoir has been as follows:
q = {50, 10, 60, 30, 40, 70, 80, 20, 100, 70}
(a) Consider two intervals for the inflow, q ≤ 50 and q > 50, with
characteristic values of 30 and 70, respectively. Calculate the inflow
transition probability matrix.
(b) Suppose two storage levels for the reservoir are 40 and 60 million
m3. The target volume of monthly release from the reservoir is 20
million m3. If the cost of operation is estimated to be equal to the
difference between the release and the target release (Rt – 20),
formulate a backward-moving stochastic dynamic programming
model for finding the optimal releases from the reservoir.
(c) Find the stationary policy.
8.11 In problem 8.1, suppose reservoir B is constructed upstream of reservoir
A. A stream discharges to the river between reservoirs A and B with
monthly discharges of Lt (t = 1, …, N).
(a) Formulate the optimization model for finding optimal releases from
reservoirs A and B.
(b) Solve the problem using linear programming and the following
assumptions:
Dir(m) Di(m)
Dd(m) (million (million Inflow to Dam B Lt
Month (million m3) m3) m3) (million m3) (million m3)
January 7 10 5 18 5
February 7 20 5.5 20 6.5
March 7 22 5.5 22 7
April 7.3 25 5.5 24 5
May 7.5 15 5 20 4
June 8 10 5 19 4
July 8.2 10 4.5 19 3.5
August 8.1 5 4.5 18 3
September 7.4 0 4.5 17 3
October 7.5 0 4.5 15 3
November 7.1 0 3 15 3.5
December 7 0 3 17 4
RT = 11.4 cms, P d = 0.8 $/m3, P ir = 0.4 $/m3, and Pi = 1.1 $/m3
8.12 In a river–reservoir system, the following agencies are affected by the

decisions made to release water from the reservoir:
• Agency 2: Department of Energy
The main objective is to meet the water and energy demands that are of
concern to these agencies. The Department of Water Supply has a twofold
role — namely, to release water from the reservoir for different purposes
as well as to supply water for domestic purposes (DD). The main concern
for the Department of Energy is to supply power loads (PL). A run-of-
river power plant downstream of the reservoir, which in turn is located
downstream from the diversion point for domestic demands, is operated
for power generation. Therefore, that portion of the release from the
reservoir that is not supplied to the domestic demand can be used for
power generation. The decision makers in these agencies are asked to set
the most favorable range of water and energy supply for each purpose by
assigning 1 as most favorable and 0 as least favorable. The following
ranges are assigned by the decision makers (see Figure 8.22):
The utility function for the range of water allocation to domestic demand
⎧Va = 100 ⎧Va = 80

⎪ ⎪
⎪Vb = 110 ⎪Vb = 100
Department of Water Supply ⎨ Department of Energy ⎨
⎪Vc = 140 ⎪Vc = 130
⎪ ⎪
⎩Vd = 160 ⎩Vd = 140
The utility function for the range of water allocation for energy supply
⎧Va = 200 ⎧Va = 250

⎪ ⎪
⎪Vb = 250 ⎪Vb = 300
Department of Water Supply ⎨ Department of Energy ⎨
⎪Vc = 300 ⎪Vc = 350
⎪ ⎪
⎩Vd = 320 ⎩Vd = 450
The following weights are set for the role and authority of these agencies
in allocating water to different purposes:
Agency: 1 2
DD ⎡5 1⎤
MATA =
PL ⎢⎣4 5⎥⎦
The average annual inflow to the reservoir is estimated to be 400 million

m3. Find the most appropriate water allocation scheme.
REFERENCES
Bellman, R., Dynamic Programming, Princeton University Press, Princeton, NJ, 1957.
Bhaskar, N.R. and Whitlatch, E. E., Derivation of monthly reservoir release policies, Water
Resources Res., 16(6), 987–993, 1980.
Datta, B. and Burges, S. J., Short-term single multiple purpose reservoir operation: impor-
tance of loss function and forecast errors, Water Resources Res., 20(9), 1167–1176,
1984.
Hasebe, M. and Nagayama, Y., Application of fuzzy set theory and neural networks to the
control system of reservoir operation from the view point of water resource, in
Stochastic Hydraulic ’96: Proc. of the Seventh IAHR Int. Symp., Mackay, Queen-
sland, Australia, July 29–31, 1996.
Hashimoto, T., Stedinger, J. R., and Loucks, D. P., Reliability, resiliency, and vulnerability
criteria for water resources performance evaluation, Water Resources Res., 18(1),
14–20, 1982.
Hydrologic Engineering Center, HEC-5 Simulation of Flood Control and Conservation
Systems, U.S. Army Corps of Engineers, Davis, CA, 1982.
Karamouz, M., Houck, M. H., and Delleur, G. W., Optimization and simulation of multiple
reservoir systems, ASCE J. Water Resources Plan. Management, 118(1), 71–81,
1992.
Karamouz, M. and Vasiliadis, H., A Bayesian stochastic optimization of reservoir operation
using uncertain forecast, J. Water Resources Res., 28(5), 1221–1232, 1992.
Karamouz, M. and Zahraie, B., Nonstructural means of flood control for protection of
Esfahan metropolitan area in Iran, in Proc. of Int. Workshop on Non-Structural Flood
Control in Urban Areas, Sao Paulo, Brazil, 1998.
Karamouz, M., Zahraie, B., Torabi, S., and Shahsavari, M., Integrated water resources
planning and management for Tehran metropolitan area in Iran, in Proc. of the 26th
ASCE Annual Water Resources Planning and Management Conf., Tempe, AZ, June
6–9, 1999.
Karamouz, M., Zahraie, B., Shahsavari, M., Torabi, S., and Araghi-Nejhad, Sh., An
integrated approach to water resources development of Tehran region in Iran, J. Am.
Water Resources Assoc., 37(5), 1301–1311, 2001.
Karamouz, M., Zahraie, B., and Sorooshian, S., Seasonal streamflow forecasting using
snow budget and climate signals: application to Salt River Basin in Arizona, ASCE
J. Hydrologic Eng., 2003 (submitted).
Karamouz, M., Zahraie, B., and Araghi-Nejhad, Sh., Operating hydropower reservoirs: a
decision support system framework for Dez and Karoon reservoirs in Iran, ASCE J.
Computing Civil Eng., 2003 (submitted).
Kottegoda, N. T. and Rosso, R., Statistics, Probability, and Reliability for Civil and Envi-
ronmental Engineers, McGraw-Hill, New York, 1997.
Linsley, R. K., Franzini, J. B., Freyberg, D. L., and Tchobanoglous, G., Water Resources
Engineering, McGraw-Hill, New York, 1992.
Lund, J. R. and Guzman, J., Derived operating policies for reservoirs in series or in parallel,
ASCE J. Water Resources Plan. Manage., 125(3), 143–153, 1999.
Mays, L. W. and Tung, Y. K., Hydrosystems Engineering and Management, McGraw-Hill,
New York, 1992.
Montanari, A., Rosso, R., and Taqqu, M., A seasonal fractional ARIMA model applied to
the Nile River monthly flows at Aswan, Water Resources Res., 36(5), 2000.
Moran, P. A. P., A probability theory of dams and storage systems, Aust. J. Appl. Sci., 5,
116–124, 1954.
Nash, J., Noncooperative games, Ann. Math., 54, 1951.
Raman, H. and Chandramouli, V., Deriving a general operating policy for reservoir using
neural networks, ASCE J. Water Resources Plan. Manage., 122(5), 342–347, 1996.
Salas, J. D., Boes, C. D., and Smith, R. A., Estimation of ARIMA models with seasonal
parameters, Water Resources Res., 18, 1006–1010, 1982.
Savarenskiy, A. D., A method for streamflow control computation (in Russian), Gidrotekh,
Stroit., 2, 24–28, 1940.
Shrestha, B. P., Duckstein, L. E., and Stokhin, Z., Fuzzy rule-based modeling of reservoir
operation, ASCE J. Water Resources Plan. Manage., 122(4), 262–269, 1996.
Solley, W. B., Pierce, R. R., and Perlman, H. A., Estimated Use of Water Resources in the
United States in 1990, U.S. Geological Survey Circular 1081, Washington, D.C.,
1993.
Vasiliadis, H. V. and Karamouz, M., Demand-driven operation of reservoirs using uncer-
tainty-based optimal operating policies, ASCE J. Water Resources Plan. Manage.,
120(1), 101–115, 1994.
9 Water Quality
Management
INTRODUCTION
The quality of surface and groundwater resources can significantly affect water use
in many regions, especially in those that are arid and semi-arid. In regions where
water pollutants from human activities have seriously degraded water quality, the
main issue in water quality management is to control pollution sources. Definition
of the control levels depends on the water quality standards defined for the various
water uses. The term water quality management implies that water should be man-
aged so that no uses at any location will be detrimental to its use at another location.
Water quality management is thereby distinguished from water quantity manage-
ment, which is the engineering of water resources systems so that enough water will
be provided to all potential users within a region (Krenkel and Novotny, 1980).
Section 9.1 discusses the principles of water quality systems analysis, including the
major types of pollutants and their resources, water quality criteria, and water quality
monitoring. Water quality management in rivers and reservoirs is discussed in Sec-
tion 9.2, and Section 9.3 addresses groundwater pollution sources and the principles
of groundwater quality management.
9.1 PRINCIPLES OF WATER QUALITY SYSTEMS

ANALYSIS
9.1.1 WATER POLLUTANTS AND THEIR PRINCIPAL SOURCES
When the discharge of wastes disturbs the natural ecological balance of a water
body, water pollution occurs. The wide range of water pollutants can be classified
into major pollutant categories as shown in Table 9.1, which also includes the
principal sources of each category. Point sources include domestic sewage and
industrial wastes because they are collected and discharged into receiving surface
and groundwaters from a single point. Sanitary sewage from homes, commercial
establishments, and public institutions is referred to as domestic sewage. The
pollutant sources are non-point sources if the pollutants are discharged to water
from multiple points. The major non-point sources can be classified as agricultural
return flows and urban runoffs. Reduction or elimination of point sources of pollution
can be implemented by proper treatment processes before discharging to receiving
waters, but treatment of non-point effluents usually is not economically feasible.
363
TABLE 9.1
Major Pollutant Categories and Principal Sources of Pollutants
Point Sources Non-Point Sources
Domestic Industrial Agricultural Urban
Pollutant Category Sewage Wastes Runoff Runoff
Oxygen-demanding material X X X X
Nutrients X X X X
Pathogens X X X X
Suspended solids/sediments X X X X
Salts X X X
Toxic metals X X
Toxic organic chemicals X X
Heat X
Source: Davis, M. L. and Cornwell, D. A., Environmental Engineering, McGraw-Hill, New York,
1991. With permission.
9.1.1.1 Oxygen-Demanding Material
This category of pollutants consists of materials that can be oxidized, thus consuming
dissolved oxygen (DO) during the oxidation process. Human waste and food residues
are sources of oxygen-demanding materials in domestic sewage. Depletion of DO
can occur in a water body by discharging industrial wastes and agricultural runoffs
that consist of organic matters.
9.1.1.2 Nutrients
Nitrates and phosphates, derived from municipal wastewater, are inorganic nutri-
ents. These nutrients are necessary for the growth of all living organisms but are
classified as pollutants because excessive amounts of them can promote plant and
algae growth. Fertilizers and phosphorus-based detergents and domestic wastes are
the major sources of nutrients.
9.1.1.3 Pathogens
Bacteria, viruses, and other microorganisms excreted by diseased animals or persons
and found in wastewater are pathogenic organisms. When pathogens containing
sewage are discharged into a water body used for drinking and recreation, they pose
a dangerous health hazard for the public.
9.1.1.4 Suspended Solids

The total solids content of a wastewater can be classified as dissolved or suspended
solids. Suspended solids consist of organic or inorganic particles, which are trans-
ported by wastewater into a receiving water body. Suspended solids can change the
qualitative characteristics of water such as turbidity and oxygen demand.
Water Quality Management 365
9.1.1.5 Salts
Whenever salt concentrations exceed the level acceptable for a particular use of
water, they are classified as pollutants. Groundwater resources usually have higher
levels of salt concentration in comparison to surface water resources because of the
accumulation of minerals from soil and rocks.
9.1.1.6 Toxic Metals and Organic Compounds
Many heavy metals and chemicals are toxic to living organisms. Pesticides in
agricultural runoff, lead and zinc in urban runoff, and toxic metals and toxic organic
substances in industrial wastewater are some examples of toxic pollutant sources.
9.1.1.7 Heat
An increase in water temperature can have a negative impact on the ecosystem. The
rate of oxygen depletion increases as the temperature increases. This is important
where the oxygen-demanding wastes are discharged to the water bodies. Many
industries, such as thermal power plants, are the major sources of heat pollution of
water bodies.
9.1.2 WATER QUALITY CRITERIA AND STANDARDS FOR WATER USE

McKee (1960) aptly described the differences between standards and criteria as
follows:
The term standard applies to any definite rule, principle, or measure established by
authority. The fact that it has been established by an authority makes a standard
somewhat rigid, official, or quasi-legal, but this fact does not necessarily mean that
the standard is fair, equitable, or based on sound scientific knowledge, for it may
have been established somewhat arbitrarily on the basis of inadequate technical data
tempered by a caution factor of safety. Where health is involved and where scientific
data are sparse, such arbitrary standards may be justified. A criterion designates a
mean by which anything is tried in forming a correct judgment concerning it. Unlike
a standard, it carries no connotation of authority other than of fairness and equity nor
does it imply an ideal condition. Where scientific data are being accumulated to serve
as yardsticks of water quality, without regard for legal authority, the term criterion
is most applicable.
The early water quality requirements and standards were defined based on the safety
of drinking water. Later, pollution control and health authorities gave more attention
to the protection of water resources. The surface water quality standards that are
used by water pollution control agencies in various countries are classified as stream
standards or effluent standards, or as a combination of both. These standards specify
numerical effluent limits for designated uses of water resources. The effluent limits
allow discharge of a specific amount of a conventional pollutant and either limit or
prohibit emission of toxic pollutants. Although effluent standards are emphasized
more than other standards, stream standards are enforced when the capacity of water
to assimilate waste is insufficient for the level of effluent loading. In addition to

stream and effluent water quality standards, water quality requirements for various
beneficial water uses are also important and are presented in the next section. Water
quality requirements for domestic, industrial, and agricultural water supplies and for
fish and other aquatic life are also presented in the following discussion.
9.1.2.1 Drinking Water Supply
Quality criteria for drinking water have been presented in many documents. Current
guidelines for evaluating the suitability of surface or groundwater resource for public
water supply are the regulations mandated by the U.S. Environmental Protection
Agency (EPA), Title 40, parts 141 and 143, and by the Safe Drinking Water Act
(1974) (Mays, 1996). Based on these guidelines, primary and secondary standards
should be established. The primary standard is for human health protection and the
secondary standard implies a regulation that specifies the maximum allowable
contamination levels that protects the public welfare and may adversely affect the
appearance or odor of water. Table 9.A1 (at the end of the chapter) presents a
summary of the EPA’s national primary standards.
9.1.2.2 Industrial Water Supply
Because of the large range of industrial processes, the water quality requirements
for industrial water supply are industry dependent. The water quality requirements
can even be different for various parts of a single industry. Water quality character-
istics that exceed those given in Table 9.A2 (at the end of the chapter) would probably
not be acceptable to industry. For a detailed description of water quality requirements
for industries, the reader is referred to Corbitt (1990).
9.1.2.3 Agricultural Water Supply
Water quality is critical to most agricultural crops. Plant growth and soil character-
istics such as permeability can be affected by water impurities. Dispersion of clay
soils reduces the size of soil pores and decreases the soil permeability. In most water
resources used for irrigation, Ca++, Mg++, and Na+ significantly affect the soil dis-
persion. A widely used relationship for estimation of permeability is the sodium
adsorption ratio (SAR), which can be estimated using the following equation:
(Na + )
SAR = (9.1)
(Ca +2 ) + (Mg +2 )
2
where Na+, Ca++, and Mg++ are concentrations of sodium, calcium, and magnesium
(in eq/m3), respectively. Acceptable SAR values for irrigation water are related to
soil characteristics and electrical conductivity of water. Representative values are
reported in Table 9.A3 (at the end of the chapter).
TABLE 9.2
Summary of Chemical Characteristics of Sampled Water
Equivalent Mass Concentration Concentration
Constituent (g/eq) (g/m3) (eq/m3)
Ca++ 20.0 1.5 0.075

Mg++ 12.15 1.1 0.09
Na++ 23.0 141 6.13
Example 9.1
Table 9.2 shows the chemical characteristics of a sample of water taken from a well.
Determine the SAR for this sample. If the concentration of total dissolved solids
(TDS) of this sample is measured to be 300 g/m3, can the water be used for irrigation?
Solution: The SAR value can be computed using Eq. (9.1) as follows:
6.13
SAR = = 21.34
0.075 + 0.09
( )
2
This is a high SAR value. The TDS value is moderate, but according to Table 9.A3,
the water would be generally unacceptable for irrigation.
9.1.2.4 Aquatic Life
The main objective of water quality standards for aquatic life is to preserve essential
environmental conditions for their survival, reproduction, and growth. Based on
water quality investigations, the DO content is usually the most important water
quality variable. The DO concentration required for warmwater biological systems
generally should be more than 5 mg/L, but for coldwater biota it should not be below
6 mg/L. It is important to note that the establishment of water quality criteria for
aquatic life is quite difficult because of the different effects of various pollutants;
site-specific research may be required depending upon the complexity of aquatic
systems.
9.1.3 POLLUTANT MATERIAL BALANCE IN SURFACE WATER

The mass balance equation for pollutants is the basis for surface water quality
modeling. From Reynolds’ transport theorem in fluid mechanics, the pollutant mate-
rial mass balance equation is derived as follows:
Dc ∂c ∂c ∂c ∂c
= +u +v +w (9.2)
Dt ∂t ∂x ∂y ∂z
where c is the pollutant concentration; x, y, and z are the Cartesian coordinates; and
u, v, and w are the respective components of velocity vector in the x, y, and z
directions. Concentration c consists of the dissolved pollutant concentration (cd)
and sorbed pollutant concentration (csp); cd is expressed as the solute mass per
unit volume of water, and csp is defined as the mass of solute per unit mass of the
suspended solids or sediments. The concentration of suspended solids maybe defined
as ms, the mass of solids per unit volume of water. csp and cd are related as:
csp = Ksp . cd (9.3)
where Ksp is the partition coefficient. As the sorption phenomenon is related to the
carbon content of particles, the Ksp and Ksc are related as:
Ksp = fcp ⋅ Ksc (9.4)
where fcp is the mass fraction of carbon in the particles and Ksc is the partition
coefficient of the solute between water and carbon. Therefore, sorption into the solid
is due to the carbon fraction that it contains (Sincero and Sincero, 1996).
The partition coefficient is the ratio of pollutant concentration in the solid phase
to its concentration in the liquid phase; it reflects the distribution of pollutants
between water and solids. The total pollutant concentration can be estimated as
follows:
c = cd (1 + fcp . Ksc . ms ) (9.5)
Pollutants can be classified as:
• Conservative pollutants
• Nonconservative pollutants
• Growing pollutants
Conservative pollutants (e.g., TDS) do not decay, but nonconservative pollutants

decay in a control mass (e.g., biochemical oxygen demand, or BOD). Growing
pollutants grow inside the control mass (e.g., algae). If distributed pollutant sources
are located along the travel path of the control mass, these sources will diffuse into
the voids and can be considered as direct input, G, in unit of mass per unit time per
unit volume of space. A common example of G is the sediment oxygen demand
(SOD). Using Eqs. (9.2) to (9.5), the complete material balance equation can be
derived as follows (Sincero and Sincero, 1996):
∂cd ∂ ⎛ ∂c ⎞ ∂ ⎛ ∂c ⎞ ∂ ⎛ ∂cd ⎞
+ Exx d + Eyy d + Ezz
∂t ∂x ⎝ ∂x ⎠ ∂y ⎝ ∂y ⎠ ∂z ⎝ ∂z ⎠
(9.6)
1 ⎛ ∂cd ∂c ∂c ⎞
+ ⎜u +υ d +ω d⎟ = 0 conservative substances
1 + fcp . Ksc . ms ⎝ ∂x ∂z ∂z ⎠
∂cd ∂ ⎛ ∂c ⎞ ∂ ⎛ ∂c ⎞ ∂ ⎛ ∂cd ⎞
− Exx d − Eyy d − Ezz
∂t ∂x ⎝ ∂x ⎠ ∂y ⎝ ∂y ⎠ ∂z ⎝ ∂z ⎠
1 ⎛ ∂cd ∂c ∂c ⎞
+ ⎜u +υ d +ω d⎟ (9.7)
1 + fcp .Ksc .ms ⎝ ∂x ∂y ∂z ⎠
+ kcd − G = 0 nonconservative substances
∂cd ∂ ⎛ ∂c ⎞ ∂ ⎛ ∂c ⎞ ∂ ⎛ ∂cd ⎞
+ Exx d + Eyy d + Ezz
∂t ∂x ⎝ ∂x ⎠ ∂y ⎝ ∂y ⎠ ∂y ⎝ ∂z ⎠
(9.8)
1 ⎛ ∂cd ∂c ∂c ⎞
+ ⎜u + υ d + ω d ⎟ − µcd = 0 growing substances
1 + fcp .Ksc .ms ⎝ ∂x ∂y ∂z ⎠
where µ is the growth coefficient, k is the lumped decay coefficient, and Exx, Eyy,
and Ezz are dispersion coefficient in the x, y, and z directions, respectively. It should
be noted that in Eqs. (9.7) and (9.8), dispersion is going out of the control mass;
therefore the positive sign has been prefixed.
9.1.4 WATER QUALITY MONITORING

The purpose of water quality monitoring is to obtain quantitative information about
the qualitative characteristics of water using statistical sampling. In other words, the
purpose of water quality monitoring is to define the physical, chemical, and biological
characteristics of water. The growing concern for maintaining the quality of surface
and groundwater has put significant demands on optimal design of monitoring net-
works and extraction of suitable information from collected data. Monitoring is one
of the most important steps of water quality management, as decisions to be made
are based on available information related to water quality characteristics. Houlihan
and Lucia (1999) categorized the purposes of water quality monitoring as follows:
1. Detection monitoring programs are used to detect an impact to surface

and groundwater quality.
2. Assessment monitoring programs are used to assess the nature and extent
of detected contaminants and to collect data that may be needed for
remediation of contaminants.
3. Corrective action monitoring programs are used to assess the impact of
remediation or pollution control programs on contaminant concentrations.
4. Performance monitoring programs are used to evaluate the effectiveness
of each element of a remediation system in meeting its design criteria.
Sanders et al. (1987) proposed general guidelines for the establishment of a moni-
toring system (Table 9.3). Identifying the information to be obtained and the statis-
tical methods for converting the data to useful information are considered in the first
two steps, and the third step deals with the monitoring network design. Monitoring
operation plans are specified in the fourth step, while in the fifth step information
related to the original monitoring objectives is developed.
TABLE 9.3
Major Steps in the Design of a Water Quality Monitoring
System
Step 1: Evaluate Information Expectations

Water quality concerns
Information goals
Monitoring objectives
Step 2: Establish Statistical Design Criteria

Development of hypotheses
Selection of statistical methods
Step 3: Design Monitoring Network

Where to sample
What to measure
How frequently to sample
Step 4: Develop Operation Plans and Procedures

Sampling procedures
Laboratory analysis procedures
Quality control procedures
Data storage and retrieval
Step 5: Develop Information Reporting Procedures

Types and timing of reports
Reporting formats
Distribution of information
Monitoring program evaluation
Are information expectations being met?
Source: Sanders, T. G. et al., Design of Networks for Monitoring Water Qual-

ity, Water Resources Publications, Littleton, CO, 1987. With permission.
9.1.4.1 Water Quality Variable Selection
Selection of the water quality variables is one of the most important steps in the
design and operation of water quality monitoring networks. The quality of a water
body is usually characterized by interrelated sets of physical, chemical, and biological
variables. Therefore, the relationships between water quality and quantity variables
are often complex. Many variables are required to completely describe the quality
of a water body, but it is not economically feasible to measure all of them. Therefore,
the variables should be ranked, and a minimum number of water quality variables
should be selected for sampling. The variables can be scored and ranked based on
multiple-criteria–decision-making methods. The variables should be selected in both
time and space. One problem with the economical selection of water variables is that
the water quality variables are time–space stochastic processes that cannot be
recorded continuously by an instrument at a site. Establishing the relationship
between water quality and quantity variables and analyzing the dependencies between
water quality variables are necessary to select representative variables. These corre-
lations can be used to produce information for variables that are not regularly
monitored.
Sanders et al. (1987) proposed a hierarchical ranking of water quantity and
quality variables that can be efficiently used in variable selection for a monitoring
network (Figure 9A.1 at the end of the chapter). In this hierarchical classification
of variables, the optimal cutoff point for selecting suitable variables is determined
by considering the monitoring objectives, sample collection laboratory analysis
costs, and the correlations among variables.
9.1.4.2 Location of Water Quality Monitoring Stations
The location of a permanent monitoring station is very important; if the collected

samples are not representative of the water body, other activities in the water quality
monitoring program can be of no consequence. The factors that can affect the
location of a water quality monitoring station are different for groundwater and
surface water resources; length is the dominant space coordinate in rivers, while
depth and width have the same role in groundwater. Therefore, in the following
discussion, the principles of river and groundwater sampling location are discussed
separately.
9.1.4.2.1 Location of River Water Quality Sampling Station
The location of river quality sampling stations is related to the monitoring objectives,
number of samples to be collected at each station, and monitoring budget. When
the monitoring network has more than one objective, samples and stations should
be classified based on the different objectives. When important areas (for example,
those with a large population) require more extensive water quality monitoring, the
monitoring resources can be allocated according to the spatial variation of land use.
Sanders et al. (1987), defined three levels of design criteria for river sampling station
location:
1. Macro-location — selection of river reaches that will be sampled

2. Micro-location — selection of a station location within a selected river
reach
3. Representative location — selection of points in the river cross-section
that provide the best lateral water quality profile of the stream
Selection of water quality sampling reach (macro-location). When an entire

river or river basin is monitored, some suitable reaches in the river should first be
selected for sampling based on the specific objectives of the monitoring. In the
classical methods, macro-location of sampling stations is based on the percentage of
basin area that belongs to each reach and on land-use variations such as distributions
of industries, population, and agricultural areas. Sharp (1971) proposed a systematic
procedure for locating sampling stations in a river system based on the number of
tributaries. In this method, which identifies and isolates sources of pollution, the river
FIGURE 9.1 Location of sampling stations using Sharp’s procedure based on the number of
tributaries. (From Sanders, T. G. et al., Design of Networks for Monitoring Water Quality,
Water Resources Publication, Littleton, CO, 1987. With permission.)
system is subdivided into reaches that have relatively equal contributing tributaries.
A magnitude (number) of 1 is assigned to the exterior tributaries, and it is assumed
that an exterior tributary has a minimum mean discharge. As shown in Figure 9.1,
the magnitude of each section of river is equal to the sum of the magnitudes of the
intersecting streams. Therefore, the magnitude of the final stretch of the river will
be equal to the total number of exterior tributaries. By dividing the magnitude of the
final stretch of the river by 2, the centroid of the basin can be determined. The first
centroid, which is referred to as the first-hierarchy sampling reach, provides two
approximately equal portions. Again by dividing by 2, a new centroid can be found
for each part and second-hierarchy sampling reaches are defined. More levels of
hierarchy are defined by successive subdivisions. As shown in Figure 9.1, the link
with the closest magnitude to the centered magnitude is specified as the centroid.
Once a link is chosen, a sampling location is selected at its downstream junction.
For example, in Figure 9.1, which has three hierarchies, sampling may be done at
eight stations: two samples from the first hierarchy, two samples from the second
hierarchy, and four samples from the third hierarchy. To identify and isolate a
pollution source, Sharp (1971) proposed that samples should be drawn at one hier-
archy and analyzed to select the next section of the river that should be sampled (at
a sampling station of the next hierarchy), and so on, until all the pollution sources
are identified. The magnitudes of tributaries and the main stem can be assigned based
on the number of outfalls that discharge into them or by their pollutant load. When
two or more variables are measured, more than one water quality monitoring network
might be established, and it can be costly and time consuming.
Sampling Station Location within a River Reach (Micro-Location). After
the macro-location process and selection of the sampling reaches, sampling sites
within each reach should be identified to provide representative samples. This kind
of longitudinal sampling location is referred to as micro-location. Micro-location
should take place in a reach of river in which the water is completely mixed — that
is, where the concentration of water quality variables is independent of lateral
location and depth in a cross-section. The distance from an outfall to a mixing zone
is a function of the physical characteristics of the river, average stream flow velocity,
lateral and vertical dispersion coefficients, and exact location of pollutant source
(outfall) in the lateral transect of the river. Sanders et al. (1987) defined mixed
distances from a point source for both complete lateral and vertical mixing as
follows:
σ2 y u
Ly = .
0.46d u*
(9.9)
7.5σ 2 z u
Lz = . *
d u
u * = gRSe (9.10)
where:
Ly is the mixing distance between the point source and complete lateral
mixing.
Lz is the mixing distance between the point source and complete vertical
mixing.
σy is the distance between the point of injection and the farthest lateral
boundary of river.
σz is the distance between the point of injection and the farthest vertical
boundary of river.
d is the depth of flow.

u is the mean stream velocity.
u* is the shear velocity.
g is the gravity acceleration.
R is the hydraulic radius.
Se is the slope of the energy gradient.
The calculated mixing distances are valid only for the considered set of hydrologic
and hydraulic variables such as stream flow, hydraulic radius, and slope; therefore,
the uncertainty of the variables and the estimated mixing distances should be con-
sidered and a mixing distance with an acceptable reliability should be determined.
Sometimes the river reach does not include a complete mixed zone, for example,
when temperature or concentration stratification is present. In this case, determining
the sampling location is difficult and water quality concentration variation within a
river cross-section should be analyzed using statistical methods such as analysis of
variance. When the water quality variation, which can be time dependent, is insig-
nificant, the section is considered to be completely mixed; otherwise, more than one
sampling point is required within the cross-section. For a detailed description of the
complete mixing assessment of a river cross-section, the reader is referred to Sanders
et al. (1987).
Example 9.2
Compute the minimum distance between a pollutant discharge point in a river and
the complete mixing zone. The outfall point is located in mid-depth and mid-width
in the river cross-section. Assume that the average stream velocity is 0.9 m/sec, the
average width is 100 m, and the average depth is 4 m. The slopes of the stream bed
and the energy gradient are assumed to be the same and equal to 0.002.
Solution: u* can be computed using Eq. (9.10):
m
u ∗ = gRSe = 9.8 × 4 × 0.002 = 0.28
s
Using Eq. (9.9) and assuming that R ≈ 4 m, it can be written that:
50 2 0.9
Ly = × = 4367.23 m
0.46 × 4 0.28
7.5 × 2 2 0.9
Lz = × = 241 m
4 0.28
As the Lz is less than L y , the mixing distance from the discharge point is equal to
4367.2 m.
9.1.4.2.2 Groundwater Sampling Location

Groundwater monitoring objectives should be defined before monitoring because
they can specify what procedures, techniques, and analyses are required. The main
objectives of groundwater monitoring are:
1. Ambient trend monitoring

2. Source monitoring
Ambient trend monitoring is used to discover the spatial and temporal trends of
groundwater quality variables. The cost of ambient trend monitoring is very much
related to the number of sampling stations and the sampling frequency.
Source monitoring determines the qualitative characteristics and the rate of
pollutant movement. The flow rate can be determined using flow net. The mass flow
rate of the pollutant can be determined as:
Rtotal = ∑ q .c
i =1
i i (9.11)
where:
Rtotal is the total mass flow rate.

n is the total number of sections.
qi is the flow rate in section i.
ci is the water quality variable concentration in section i.
A suitable groundwater monitoring network should provide the spatial and temporal
characteristics of groundwater pollution for various sources of pollution such as
plumes.
Hudak and Loaiciga (1993) proposed a quantitative, analytically based hydro-
logic approach for monitoring network design (sampling location) in a multilayered
groundwater flow system. In this method (described below), the susceptibility to
contamination of points in the model domain is quantified using weight values. This
network design methodology consists of three steps:
1. Domain definition and discretization

2. Determination of relative weights for candidate monitoring sites
3. Selection of an optimal configuration for monitoring network
Definition and Discretization of the Model Domain. The model domain

includes the contaminant source and surrounding areas that could be affected by the
pollutant source and within which monitoring wells are selected. Zones that encom-
pass the pathways of contaminant migration are defined by advection envelopes.
Advection envelopes are extended from hydrogeologic outlets of pollutant sources
based upon the known or inferred variation of the hydraulic head (Figure 9.2). The
FIGURE 9.2 Advection envelope extended from hypothetical contaminant source. (From Hudak,
P. F. and Loaiciga, H. A., Water Resources Res., 29(8), 2835–2845, 1993. With permission.)
two flow lines that originate from separate ends of the hydrogeological outlet can
be considered as boundaries of advection envelopes. Monitoring well sites can be
classified as background and detection monitoring sites. Detection monitoring wells
are constructed for early detection of contamination by pollutant sources; back-
ground monitoring sites determine the background water quality and are located in
the upgradient zone of the model domain. The background monitoring wells should
not be located adjacent to the contaminant source and advection envelope. They also
should not intersect an outward normal line extended from an advection envelope
(Hudak and Loaiciga, 1993).
Calculation of Weights for Candidate Monitoring Sites. The relative weight
of each monitoring site should be calculated based on the location of the site relative
to the pollutant source and advection envelope. The following model can be used
for designing a monitoring network for an uncontaminated aquifer (before the
pollutant is released from the pollutant source). Therefore, the existing contamination
concentrations are not used in the monitoring well configuration. The relative weight
of each candidate monitoring well in a multilayer aquifer is determined as (Hudak
and Loaiciga, 1993):
1
Wjk = (9.12)
D(s) jk D(e) jk
where:
j is a real index of the potential well site

k is the hydrostratigraphic interval (HSI) index (HSI is defined as a layer that
has relatively uniform hydraulic conductivity).
D(s)jk is the horizontal distance from a contaminant source boundary to node
j in HSI k.
D(e)jk is the perpendicular distance from node j in HSI k to the closest
boundary of the advection envelope boundary (equal to D(e)(max) if node
j cannot be intersected by a perpendicular line from an advection envelope).
D(e)jk(max) is the maximum value of D(e)jk among nodes for which the
quantity can be measured.
Based on the above equation, nodes located near an area with higher probability
of contamination have higher weights. To reduce the computational efforts, the
boundaries of pollutant source and advection envelopes are approximated by nodes
within a grid of candidate groundwater monitoring wells. The value of D(e)jk is
calculated only for nodes that are susceptible to contamination from plume spreading
and are intersected by an outward normal line extended from an advection envelope.
Other nodes that are not intersected by an outward perpendicular line from an
advection envelope are assigned the maximum value of D(e)jk calculated for all other
nodes. Nodes located within the advection envelope are assigned the minimum value
of D(e)jk to avoid division by zero. For example, in Figure 9.2, nodes A and B are
located an equal distance from the advection envelope, but node B has the greater
probability of being contaminated because node A is located in an upgradient area
and should be assigned the maximum of D(e)jk (D(e)(max)).
Optimal Configuration of Monitoring Well Network. Optimal configuration
of monitoring sites can be obtained using a mathematical model. The objective
function of this model is the preferential location of monitoring wells at points with
high probabilities of contamination. The model formulation can be as follows (Hudak
and Loaiciga, 1993):
Max Z = ∑ ∑ W x −∑∑W x
k ∈K j ∈Jk − uk
jk jk
k ∈K j ∈Juk
jk jk (9.13)
Subject to ∑x
j ∈Jk
jk ≥ Pk (min) for each k ∈ K (9.14)
∑x
j ∈Juk
jk = Puk for each k ∈ K (9.15)
∑∑ x
k ∈K j ∈Jk
jk =P (9.16)
⎧0 otherwise
x jk = ⎨ for each j ∈ Jk , k ∈ K (9.17)
⎩1 if a well is installed at node j in HSI k
where:
j is the areal index of a potential well site.

Jk is the set of potential well sites in HSI k.
Jk–uk is the set of potential well sites (excluding sites in the upgradient zone)
in HSI k.
Juk is the set of potential well sites in the upgradient zone in HSI k.
k is the HSI index.
K is the set of HSIs.

Wjk is the weight for node j in HSI k.
Pk(min) is the minimum number of wells to be located in HSI k.
Puk is the number of wells allocated to the upgradient zone in HSI k.
P is the total number of wells.
The second term of Eq. (9.13) ensures that the nodes with the lowest weight will
be selected for background monitoring in the upgradient zone. Equations (9.14) and
(9.15) constrain the minimum and total number of wells in the upgradient zones in
each HSI. Equation (9.16) ensures that all the wells are located throughout the model
domain, and Eq. (9.17) shows that variable xjk is a binary integer.
The total number of monitoring wells in each layer can be determined by
budgetary constraints or regulatory requirements. The budget constraint can be
written as:
∑∑C
kεK jεJk
jk x jk ≤ R (9.18)
where Cjk is the construction cost of a well at monitoring site j in HSI k, and R is
the total available budget. The proposed formulation can be solved using integer
programming techniques. This model can define the location of the P monitoring
wells, including Puk upgradient wells. For more details, the reader is referred to
Hudak and Loaiciga (1993) and Loaiciga et al. (1992).
9.1.4.2.3 Sampling Frequency
Sampling frequency shows how often water quality samples should be gathered.
Detection of stream standards violations and determination of temporal water quality
variations are the main criteria for sampling frequency analysis. The selection of
sampling frequency requires some background information on the behavior of the
random variables that are important in water quality. The analyst may face three
situations:
• Water quality data are available for the stream being studied.
• Water quality data are available only from a stream in a region that has
characteristics similar to those of the study area.
• No data exist for the stream under consideration or other similar streams.
When no data are available, a two-stage monitoring program should be utilized. In

this case, design data collected in the first stage can be used for suitable design of
the monitoring network and to estimate the sampling frequency. When the water
quality data are available only for similar rivers in the region, data should be
generated for the stream that is under study. In this case, similarities between their
geological and hydrologic characteristics, distribution of the pollutant sources, and
pollutant characteristics are usually considered for water quality data generation.
9.1.4.2.4 Methods of Selecting Sampling Frequencies

Sampling frequency design can vary from a single-station, single water quality
variable to a multiple-station, multiple water quality variable. Sanders et al. (1987)
proposed statistical methods for each case which are presented next.
Single-Station and Single Water Quality Variables. In this case, the selected
sampling frequency for a specified water quality variable at a station should provide
the desired confidence interval around the annual mean. If the collected samples can
be assumed to be independent, the variance of sample mean is:
σ2
Var x = () n
(9.19)
where σ2 is the population variance, and n is the number of samples. The confidence
interval for the population mean, µ, is as follows:
1 1
⎛ σ2 ⎞ 2
⎛ σ2 ⎞ 2
x − Zα .⎜ ⎟ ≤ µ ≤ x + Zα .⎜ ⎟ (9.20)
2 ⎝ n ⎠ 2 ⎝ n ⎠
where Zα/2 is the standard normal deviation corresponding to the probability of α/2.
By rearranging terms, the number of samples (n) required to estimate the mean with
a known confidence level (α) is:
⎡ Zα .σ ⎤
2
n≥⎢ 2 ⎥ (9.21)
⎢⎣ µ − x ⎥⎦
When the sample standard deviation(s) is used instead of σ, the Zα/2 is replaced with
Student’s t statistic, tα/2:
2
⎡ tα .s ⎤
n≥⎢ 2 ⎥ (9.22)
⎢⎣ µ − x ⎥⎦
In this case, the computation of n becomes an iterative problem because when the
t distribution is used the number of degrees of freedom (n – 1) must be known, but
in this case it is not known. Therefore, for solving the problem, an initial value for
n is estimated, the degrees of freedom are determined, and a new estimation of n is
obtained using Eq. (9.22). By repeating this procedure, the values of n converge to
a desired value. When the number of samples is large (more than 30), σ can be
estimated with s, without any significant error in determination of the sampling
frequency.
Example 9.3
Select the sampling frequency for a station that monitors BOD concentration in an
important control point on a river. The annual mean and variance of the BOD
concentration based on historical data are 6.01 mg/l and 8.1 (mg/l)2, respectively.
The desired confidence interval width is assumed to be 3 mg/l with a 95% confidence
level.
Solution:
σ
The confidence interval width = 2 × 1.96 ×
n
σ
2 × 1.96 × =3
n
2 × 1.96 ⎤ 2 ⎡ 2 × 1.96 ⎤
2 2
n = ⎡⎢ σ =⎢ × 8.1
⎣ 3 ⎦⎥ ⎣ 3 ⎦⎥
n = 13.8 ≈ 14 samples / year
Single Station and Multiple Water Quality Variables. If the problem of sam-
pling frequency selection includes more than one water quality variable, a compro-
mise program should be used to obtain the most acceptable confidence interval that
considers all water quality variables. A simple way to do this is to compute a weighted
average of confidence interval widths for several water quality variables. The relative
weights of water quality variables can be selected by engineering judgment so that
the contributions of all water quality variables are comparable.
Example 9.4
Select the sampling frequency for a station that has four water quality variables.
Assume that the 95% confidence interval width about the mean of the water quality
variables will be equal to one fourth of the average of these variables. The historical
population statistics of the variables are as follows:
Mean
Variable (mg/l) Variance ([mg/l]2)
BOD 41 286
Total Nitrogen (TN) 10 18.1
Ca++ 40 38
Mg++ 34 60
Solution: The width of 95% confidence interval should be one fourth the average
of the means. Thus:
R1 + R2 + R3 + R4 1 ⎡ µ1 + µ 2 + µ 3 + µ 4 ⎤
= ⎢ ⎥⎦
4 4⎣ 4
where Ri is the width of confidence interval in the station i that can be computed
using the following equation:
σi
Ri = 2(1.96) i = 1, 2, 3, 4
n
1 2 × 1.96
(µ1 + µ 2 + µ 3 + µ 4 ) = (σ1 + σ 2 + σ 3 + σ 4 )
4 n
2
⎡ (16.91 + 4.25 + 6.16 + 7.75) ⎤
n = ⎢4 × 2 × 1.96 ⎥ = 19.35 ≈ 20 samples / year
⎣ ( 41 + 10 + 40 + 34) ⎦
Multiple Stations and Single Water Quality Variable. Because the designer
usually wants to obtain the same information from each sampling station of the
monitoring network, this requirement can be a basis for selection of the sampling
frequency. When based on historical data, the average concentrations of water quality
variables at the stations are approximately equal, the sampling frequencies should
be selected so they provide equal confidence intervals about the means for the
different stations. In a simple procedure proposed by Ward et al. (1979), the total
number of samples (N) is allocated to stations based on their relative weights using
the following equation:
ni = wi . N (9.23)
where ni is the number of samples at station i, wi is the relative weight of station i,

and N is the total number of samples. The relative weights show the relative impor-
tance (priority) of stations and can be estimated based on historical data. For exam-
ple, relative weights based on historical means or historical variances can be com-
puted using:
µi
wi = Ns (9.24)
∑µ
i =1
i
σi2
wi = Ns (9.25)
∑σ
i =1
i
2
where µi and σ i 2 are the historical mean and variance of the water quality variable
at station i and Ns is the total number of stations in the water quality monitoring
network.
Multiple Stations and Multiple Water Quality Variables. The design problem
encountered most frequently in practice involves more than one water quality vari-
able as well as more than one sampling station. In this case, a separate sampling
TABLE 9.4
Means and Variances for Three Water Quality Variables (Example 9.5)
TN BOD Total Phosphorus (TP)
Mean Mean Mean
Station (mg/l) Variance (mg/l) Variance (mg/l) Variance
1 2.4 0.50 4.51 3.41 0.21 0.01

2 3.3 0.90 5.01 8.04 0.22 0.01
3 5.4 1.31 3.50 1.52 1.28 0.01
4 3.8 1.20 4.41 3.66 0.23 0.02
frequency should be selected for each water quality variable at each station and a
weighted average of them can be considered as each station’s sampling frequency.
Sanders et al. (1987) suggested that a weighted average variance for the water quality
variables can be computed for each station and used in Eq. (9.25). In the case of
multiple variables, it is not possible to obtain equal confidence interval widths for
water quality variables in all stations if all variables at a given station are sampled
at the same frequency.
Example 9.5
The historical means and variances for three water quality variables in four stations
have been calculated and are shown in Table 9.4. Select the sampling frequency for
each station using the weighting factors based on historical variances. The total
number of samples per year is considered to be equal to 40.
Solution: The total number of samples should be allocated to each station for
each water quality variable using weighting factors. The relative weight of each
station for each variable is as follows:
σi2
wi = 4 i = 1, 2, 3, 4 (9.26)
∑σi =1
2
i
The weighting factors for the water quality variables are based on the variances and
are provided in Table 9.5. Table 9.6 shows the average number of samples estimated
for each station.
9.2 SURFACE WATER QUALITY MANAGEMENT

9.2.1 RIVER WATER QUALITY MANAGEMENT
The protection of public health in large cities and near industries located along rivers
is the main purpose of river pollution control; however, protection of water resources
TABLE 9.5
The Relative Weights of Stations for Example 9.5
Station TN BOD TP
1 0.13 0.21 0.2

2 0.23 0.48 0.2
3 0.33 0.09 0.2
4 0.31 0.22 0.4
TABLE 9.6
Allocation of Samples via Proportional Sampling Based on Variances
Station Number of Samples Allocated Average

Number TN BOD TP (TN, BOD, and TP)
1 5 8 8 7
2 9 19 8 12
3 13 4 8 8
4 13 9 16 13
and ecosystems and preserving the natural environment related to these resources
have become more important in recent years. Therefore, the main objective of river
water quality management today is to control pollutants discharged to the river so
that river water quality at critical sections is not reduced to an unacceptable extent
below the natural background level. Important steps in river water quality manage-
ment include measurement of discharged pollutants, prediction of the impact of a
pollutant on water quality using suitable river water quality simulation models, and
determination of water quality criteria.
Pollution of rivers can occur directly from sewer outfalls or industrial discharges
as point sources and agricultural or urban runoffs as non-point sources or indirectly
from air pollution. Impacts of a pollutant on river water quality depend on the
pollutant type and load and river characteristics. The following sections provide an
overview of the derivation of descriptive models for the transport of pollutants and
the formulation of some important theoretical optimization models for water quality
management in rivers and streams.
9.2.2 LOW FLOW DETERMINATION

Estimates of low flow quintiles, such as the 7-day, 10-year low flow, are used widely
in steady-state water quality modeling and management. Low flow quintiles are
usually obtained by statistical modeling of observed data series. For this purpose,
several methods have been presented in the literature. In general, the low flow
determination can be stated as the estimation of the quantity qT in the following
equation.
qT
FQ (qT ) =
∫ f (q)dq = T
1
Q (9.27)
0
where T is the recurrence interval, fQ(q) is a model of the probability density function
of the random variable Q. In low flow analysis, T usually is considered to be equal
to 10 when Q is the annual minimum 7-day average discharge and q10 is the 7-day,
10-year low flow. For details, the reader is referred to Durrans (1996).
The low flow can also be computed using the empirical frequency analysis
method. In this method, the smallest flow that occurs for 7 consecutive days in each
year is computed using the long-term flow record (n years) for the location being
modeled. In the next step, the n flows are tabulated in ascending order and a rank
m is assigned to them. Then, the cumulative values of occurrence and recurrence
intervals are calculated using the following equations:
m
p= (9.28)
n +1
1
T= (9.29)
p
Application of this method is illustrated in the following example.
Example 9.6
The ranked values of 20 years of data for the 7-day minimum average flows of a
river are given in Table 9.7. Prepare the low flow frequency curve and determine
the 10-year low flow.
Solution: As can be seen in Figure 9.3 and Table 9.7, a value of about 1.65
m3/sec corresponds to the 7-day, 10-year low flow.
9.2.3 COMPUTER MODELS FOR STREAMWATER QUALITY

SIMULATION
Computer models for streamwater quality simulation have been used widely for
water quality management of complex stream systems. These models enable water
resources engineers to predict the effects of numerous pollutant discharges, and they
often can be used to preserve water quality by managing human activities as the
main pollutant source. Tables 9.A5 and 9.A6 at the end of the chapter present and
compare the important attributes of some surface water quality models. The follow-
ing discussion summarizes the characteristics of two useful water quality simulation
models.
Water Quality Management
TABLE 9.7
Annual 7-Day Minimum Average Flow in Example 9.6
7-Day Low Recurrence 7-Day Low Recurrence
Rank Flow (m3/sec) Probability Interval Rank Flow (m3/sec) Probability Interval
1 1.49 4.76 21.00 11 3.10 52.38 1.91

2 1.64 9.52 10.50 12 3.15 57.14 1.75
3 1.80 14.29 7.00 13 3.31 61.90 1.62
4 1.90 19.05 5.25 14 3.66 66.67 1.50
5 2.20 23.81 4.20 15 4.01 71.43 1.40
6 2.30 28.57 3.50 16 4.11 76.19 1.31
7 2.50 33.33 3.00 17 4.23 80.95 1.24
8 2.70 38.10 2.63 18 4.50 85.71 1.17
9 2.75 42.86 2.33 19 5.10 90.48 1.11
10 2.80 47.62 2.10 20 5.38 95.24 1.05
385
Recurrence interval
25
20
15
10
5
0
0 1 2 3 4 5 6
Q (cms)
FIGURE 9.3 Frequency curve corresponding to the minimum 7-day flow.
9.2.3.1 QUAL2E Model
The enhanced streamwater quality model (QUAL2E) is a comprehensive and ver-

satile model that can simulate several water quality factors in any combination
desired by the user. QUAL2E is a one-dimensional model and uses a finite-difference
solution of the mass transport and reaction equation. It allows multiple waste dis-
charges, withdrawals, tributaries, and incremental inflow and outflow. QUAL2E can
operate either as a steady-state or as a quasi-dynamic model. Dynamic operation of
the model allows study of the effects of diurnal variation of meteorological data on
water quality and the effects of diurnal variation of respiration and algal growth on
dissolved oxygen. The QUAL2E model considers the major interactions and effects
of nutrient cycles, benthic and carbonaceous oxygen demand, atmospheric aeration,
and algal production on the dissolved oxygen concentration in a stream. It also can
model coliforms and nonconservative as well as conservative constituents that do
not decay or interact with other constituents.
QUAL2E-UNCAS is an enhancement to the QUAL2E model that allows users
to perform uncertainty analyses on steady-state streamwater quality simulations.
Three uncertainty analysis techniques can be employed in this model:
• Sensitivity analysis
• First-order error analysis
• Monte Carlo simulation
For detailed information, the reader is referred to Brown and Barnwell (1987).
9.2.3.2 WASP5
The water quality analysis simulation program (WASP5) is a dynamic compartment
modeling framework that can be used to simulate a variety of water quality contam-
inants in a diverse set of water bodies. The WASP5 has been developed by the EPA’s
Center for Exposure Assessment Modeling (EPA CEAM). It can be applied to one-,
two-, and three-dimensional simulation of transportation and transformation of
conventional and toxic pollutants in ponds, streams, lakes, reservoirs, rivers, estu-
aries, and coastal waters. TOXI5 and EUTRO5 are two major subcomponent models
that can be linked to WASP5. TOXI5 is a dynamic model of the transport and fate
of organic chemicals and metals in all types of aquatic systems that can simulate
the transport and transformation of up to three chemicals and up to three types of

particulate materials. TOXI5 uses a simple sediment balance and organic chemical
process kinetics to predict dissolved and sorbed chemical concentrations. EUTRO5
simulates the transport and transformation reactions of a wide range of water quality
variables, which can be considered as four interactive systems: phytoplankton kinet-
ics, the phosphorus cycle, the nitrogen cycle, and the dissolved oxygen balance. The
transport process can be simulated by coupling this model with external hydrody-
namic models (see Ambrose, 1987, for more details).
9.2.4 SIMPLE STREAMWATER QUALITY MODELS

9.2.4.1 Oxygen Sag Model
An oxygen-consuming pollutant, either organic or inorganic, discharged into a

stream causes depletion of the dissolved oxygen. In this case, DO is initially con-
sumed faster near the discharge point, which poses a danger for aquatic life if the
concentration of oxygen falls below a critical level. Determining the amount of
discharged waste and how much oxygen will be required to degrade the waste is
necessary for predicting the extent of oxygen depletion.
Oxygen is continuously being replenished from the atmosphere and consumed
by organisms, so the oxygen concentration in a river is determined using the relative
rates of these competing processes. Downstream of the discharge point, the river
recovers and the DO concentration goes back up again. This profile of DO concen-
tration along the river is referred to as DO sag. Figure 9.4 shows a plot of DO sag.
DO deficit, D (mg/L)
Dc
Do
0
0 Distance, x
time of travel
(a)
DO saturation
DO (mg/L)
Co Dc
0
0 Xc Distance, x
time of travel
(b)
FIGURE 9.4 Dissolved oxygen sag.

The one-dimensional mass balance equation for DO can be derived using Eq. 9.7:
Dc ∂ ⎛ ∂c ⎞
= ⎜ Exx. ⎟ + k2 (Os − c) + P − R − kc . Lc − kc . Ln + Go (9.30)
Dt ∂x ⎝ ∂x ⎠
where:
Go is the distributed dissolved oxygen source.

c is the dissolved oxygen concentration.
k2 is the reaeration coefficient.
Os is the dissolved oxygen saturation.
P is the rate of photosynthesis.
R is the rate of respiration.
kc is the carbonaceous biochemical oxygen demand (CBOD) deoxygenation
coefficient.
Lc is the concentration of CBOD.
kn is the nitrogenous deoxygenation coefficient.
Ln is the nitrogenous biochemical oxygen demand (NBOD) concentration.
Exx is dispersion coefficient in the x direction.
As originally derived by Streeter and Phelps in 1925, if the effects of dispersion,

P, R, NBOD, and Go are neglected, then Eq. (9.30) could be simplified as:
Dc
= k2 (Os − c) − kc . Lc (9.31)
Dt
Lc = Lc 0 .e − kc .t
'
(9.32)
where kc' = kc + ks , ks is the rate of settling of CBOD, and Lc0 is Lc at t = 0. Using

Eqs. (9.31) and (9.32) and defining the DO deficit as D = Os – c, D = D0 at t = 0,
D = 0 at t = ∞, and kc' = kc , the Streeter–Phelps equation is derived as:
kc . Lc 0 ⎛ − kc .t − k .t − k .t
D= e − e 2 ⎞ + D0 .e 2 (9.33)
k2 − kc ⎝ ⎠
If the deficit calculated from the above equation is greater than the saturation
DO, all the dissolved oxygen has been depleted and the DO is zero. The lowest
point on the DO sag curve indicates the worst water quality condition and is of
major interest. The time to this minimum DO point, the critical time (tc), can be
found by differentiating Eq. (9.33), setting it equal to 0, and solving for t:
1 ⎡k ⎛ k − k ⎞⎤
tc = ln ⎢ 2 ⎜1 − D0 × 2 c ⎟ ⎥ (9.34)
k2 − kc ⎢⎣ kc ⎝ kc − Lc 0 ⎠ ⎥⎦
TABLE 9.8
Typical Values for the BOD Rate Constant
Sample k (20°C) (day–1)
Raw sewage 0.15–0.30
Well-treated sewage 0.05–0.10
Polluted river water 0.05–0.10
The critical deficit (Dc) is then found by using tc in Eq. (9.33). The bio-oxidation
rate (kc) is dependent on the waste characteristics, the ability of organisms to oxidize
the waste, and the water temperature. Temperature increases the speed of most of
the biological processes. Laboratory testing is usually done at a standard temperature
of 20°C, and the BOD rate constant at any temperature T (°C) is:
kc,T = kc,20 .(θ)T −20 (9.35)
where kc,T is the BOD rate constant (expressed in 1/day) at temperature T (°C), kc,20
is the BOD rate constant determined under laboratory conditions at 20°C (1/day),
and θ is the temperature coefficient. The value of θ is considered to be 1.135 for
4 < T < 20°C and 1.056 for 20 < T < 30°C. Typical values for the BOD rate constant
under laboratory conditions are presented in Table 9.8.
The deoxygenation rate constant (kc) in a river differs from the BOD rate constant
(k) due to physical and biological differences between a bottle and a river. In river
systems, the BOD is exerted more rapidly because of turbulent mixing and BOD
removal by suspended and sediment organisms. kc can be estimated from the BOD
rate constant (k) by the following equation:
v
kc = k + η (9.36)
h
where:
kc is the deoxygenation rate constant at 20°C (1/day).

v is the average velocity of streamflow (m/sec).
k is the BOD rate constant determined in laboratory conditions at 20°C (1/day).
η is the bed activity coefficient (0.1 for deep water; 0.6 or greater for a rapidly
flowing stream).
h is the average depth (m).
When the stream temperature is not 20°C, the calculated kc from the above equation
should be adjusted using Eq. (9.35). The value of the reaeration rate constant (k2)
depends on the degree of turbulent mixing, which is related to river flow velocity
TABLE 9.9
Range of Reaeration Constants for Various Water Bodies
Water Body Ranges of k2 at 20°C (1/day)
Small ponds 0.05–0.1

Small streams of low velocity 0.10–0.15
Large lakes 0.10–0.15
Large streams 0.15–0.30
Swift streams >0.30
Note: For other temperatures, use k2,T = k2,201.024T–20.
and the surface and depth of the water body (surface area exposed to the atmosphere
per unit volume of water). The reaeration rate constant is also related to temperature
and should be adjusted to river temperature using Eq. (9.35) but with a temperature
coefficient θ, which is equal to 1.024.
Some researchers have developed formulas for calculating k2 directly from
stream physical parameters. One of these formulas was developed by O’Connor and
Dobbins (1958) as follows:
1
3.9 × v 2
k2 = 3 (9.37)
h 2
The Langbein and Duram (1967) formula can be used to calculate k2 as follows:
2.208 × v
k2 = 4 (9.38)
3
h
where:
k2 is the reaeration rate constant at 20°C (1/day).

v is the average stream velocity (m/sec).
h is the average depth (m).
Table 9.9 shows the typical reaeration constants for various surface water bodies.
Example 9.7
A treatment plant discharges 0.15 m3/sec of partially treated sewage into a free-
flowing stream. The stream characteristics upstream of the point of discharge at
summer low flow conditions are as follows:
Flow discharge = 0.5 m3/sec.

Temperature = 22°C.
BOD5 = 3 mg/l.
DO = 8 mg/l.
Velocity = 0.1 m/sec.
Average depth = 2.2 m.
Bed activity coefficient = 0.2.
The effluent characteristics have been measured as follows:
Temperature = 25°C.
BOD5 = 40 mg/l.
DO = 2 mg/l
k2 at 20°C = 0.4 1/day.
k (laboratory test) = 0.22 1/day.
Calculate the DO sag curve for a section located 120 km downstream of the discharge
point.
Solution: The characteristics of mixed flow can be computed as:
m3
Qmix = 0.15 + 0.5 = 0.65
s
(0.5 × 3) + (0.15 × 40)
BODmix = = 11.54 mg / L
0.65
(0.5 × 8.0) + (0.15 × 2.0)
DOmix = = 6.62 mg / L
0.65
(0.5 × 22) + (0.15 × 25)
Tmix = = 22.7 mg / L
0.65
The deoxygenation and reaeration coefficients are calculated using Eqs. (9.36) and
(9.37), respectively, as follows:
0.1 × 0.2 1
kc,20 = 0.22 + = 0.23 at 20° C
2.2 day
3.9 × (0.1)0.5 1
k2,20 = 3 = 0.38 at 20° C
2
day
(2.2)
kc and k2 should be adjusted to 22.7°C as follows:
1
kc,22.7 = kc,20 (1.056 22.7−20 ) = 0.23 × 1.158 = 0.266
day
1
k2,22.7 = k2,20 (1.024 22.7−20 ) = 0.38 × 1.066 = 0.41
day
The ultimate BOD (Lc0) can be calculated using the following equation:
− kc ,20 × 5
BOD5 = Lc 0 − Lc 0 .e
BOD5 11.54 (9.39)
Lc 0 = −k ×5 = = 16.88 mg / L
1 − e c ,20 1 − e −0.23×5
As the DO saturation (Os) at 22.7°C is 8.7 mg/L, the initial DO deficit (Do) can be
computed as follows:
Do = 8.7 − 6.62 = 2.08 mg / L
Now we have all the information we need to calculate the time to the critical point
and maximum DO deficit. Using Eqs. (9.33) and (9.34), it can be written that:
1 ⎡ 0.41 ⎛ 0.41 − 0.26 ⎞ ⎤

tc = ln 1 − 2.08 ×
0.41 − 0.26 ⎢⎣ 0.26 ⎝ 0.26 × 16.88 ⎠ ⎥⎦
tc = 2.54 day
0.266 × 16.88
Dc =
0.41 − 0.266
( )
× e −0.266×2.54 − e −0.41×2.54 + 2.08 × e −0.41×2.54 = 3.56 mg / L
This critical condition occurs at the following distance downstream of the discharge
point:
xc = 0.1 × 2.54 × 86, 400 = 21, 946 m = 21.9 km
By the same method, oxygen deficits at 20, 50, 75, 100, and 120 km downstream
of the point of discharge can be calculated as follows:
x (m)
t (day) =
86, 400 × v ⎛ ⎞
m
⎝ s⎠
t20 = 2.315 day t50 = 5.787 day
t75 = 8.680 day t100 = 11.57 day
t120 = 13.88 day
9
DO saturation
8
7
DO (mg/L)
6
5
4
3
2
1
0
0 20 40 60 80 100 120
Distance x(km)
FIGURE 9.5 Dissolved oxygen sag for Example 9.7.
The DO deficit at these times are calculated using Eq. (9.33) as follows:
0.26 × 16.88 −0.26×2.315 −0.41×2.315

D20 =
0.41 − 0.26
e ( −e )
+ 2.08e −0.41×2.315
D20 = 5.51 mg / L
D50 = 3.96 mg / L D75 = 2.29 mg / L
D100 = 1.21 mg / L D120 = 0.7 mg / L
The dissolved oxygen concentrations at these points are also computed as:
c20 = 3.19 mg / L c21.6 = 3.17 mg / L

c50 = 4.74 mg / L c75 = 6.41 mg / L
c100 = 7.49 mg / L c120 = 8 mg / L
Figure 9.5 shows the DO sag curve based on these computations.
9.2.4.2 Simple Water Quality Model for Multiple-Discharge

Streams
The Streeter–Phelps equation can be used for water quality modeling of streams
having several reaches with different flows, reaeration and bio-oxidation parameters,
and pollutants discharged from point sources. A typical section of river is presented
in Figure 9.6. For an important inflow and withdrawal or where sections of river
have different velocities and reaeration coefficients, a separate reach is considered.
Each reach is indexed by i and each discharge by j. Each discharge can have an
associated withdrawal that changes the stream flow. Each reach has three important
points. The first point (ti,0), located downstream of discharge point j, is the first
boundary condition of reach i. The next important point in each reach is the critical
point (ti,c) having the greatest oxygen deficit or minimum oxygen concentration. The
third important point is located just before the next discharge point. The pollutant
FIGURE 9.6 Schematic diagram of multireach/multidischarge stream model. (From ReVelle,

C. and McGarity, A. E., Design and Operation of Civil and Environmental Engineering
Systems, John Wiley & Sons, New York, 1997. With permission.)
concentration is calculated at points ti,0 and ti,e to evaluate dilution effects at the
discharge points. The pollutant concentration at critical point ti,c is compared with
the water quality standard to determine whether any water quality standards are
being violated.
Assume that Qi is the flow rate in reach i (m3/sec), qj-out is the withdrawal flow
rate at point j (m3/sec), and qj-in is the discharged flow rate of discharge point j. It
should be noted that in this section only the CBOD is considered; therefore, notation
L is used instead of Lc. For BOD pollutants and considering complete mixing at
discharge points, we have:
Qi −1 − q j ,out q j ,in
Li,0 = . Li −1,e + lj (9.40)
Qi Qi
Qi −1 − q j ,out q j ,in
Di,0 = . Di −1,e + dj (9.41)
Qi Qi
where:
Li,s and Di,s are the BOD concentration and DO deficit, respectively, at point
s in reach i.
lj is the BOD concentration of discharge j (mg/l).
dj is the DO deficit of discharge j (mg/l).
The BOD concentration at points s = c and e on reach i can be calculated using

the BOD decay equation:
− kci .ti ,s
Li,s = Li,0 × e (9.42)
And, from the Streeter–Phelps equation, the DO deficit at points s = c and e in

reach i are:
Di,s =
kci . Li,0
k2 i − kci
( − k ⋅t −k t
)− k ⋅t
. e ci i ,s − e 2 i i ,s + Di,0 .e 2 i i ,s (9.43)
The critical DO deficit in reach i can be derived from Eqs. (9.31) and (9.32) as
follows:
kci − k ⋅t
Di,c = ⋅ Li,0 ⋅ e ci i ,c (9.44)
k2 i
The time t at each point can be calculated based on the distance the pollutant
has traveled divided by the velocity (t = x/v), where x is the distance traveled and v
is the average streamflow velocity. Therefore, when considering the average velocity
in each reach, Eqs. (9.42) and (9.43) are simplified to linear equations and can easily
provide dissolved oxygen and BOD concentrations at each point of the stream.
9.2.5 WATER QUALITY MANAGEMENT MODELS FOR RIVERS AND

STREAMS
In this section, we describe deterministic and stochastic models that have been
developed for river and streamwater quality management and waste load allocation.
Linear, nonlinear, and dynamic programming techniques have been used to develop
optimal management policies and to identify optimal solutions for waste load allo-
cation problems. The study of waste load allocation as an important element of river
water quality management can be necessary in several situations. The most common
problem arises when the amount of pollutant removal should be determined at a
number of discharge points along a water body in order to achieve or maintain an
acceptable level of water quality. Another situation could arise from the capacity
expansion problem, where in one or more point sources influent loads are increased.
In these cases, the appropriate increase in the size and capacity of treatment facilities
must be determined. Another situation may occur when an additional discharger
wishes to locate on a water body that could necessitate reallocation of the assimilative
capacity of the water body among the existing discharges (Burn, 1989). Minimization
of treatment costs has been considered by various investigators as the objective
function of water quality management models developed for streams. An ideal
strategy can be obtained by the combined use of optimization and simulation models,
but the optimization process can be more difficult.
9.2.5.1 Deterministic Approach

9.2.5.1.1 Linear Programming Models for Dissolved Oxygen
Management
In this section, a management model is developed based on the multireach/multi-
discharge system presented in the previous section. The objective function of this
model is minimization of treatment costs. The general form of the wastewater
treatment cost function (see Figure 9.7) is a nonlinear function of the treatment
efficiency. The cost curve shows three classes of treatment: primary, secondary, and
tertiary. According to Figure 9.7, the treatment costs can be represented by the cost
function c(x), where x is the treatment plant efficiency. The cost function has both
1.4
1.2
1
annual cost ($)
0.8
0.6
0.4
0.2
0.2 0.4 0.6 0.8 1 1.2

treatment plant efficiency - x
FIGURE 9.7 Annual costs of treatment vs. treatment efficiency. (From ReVelle, C. and
McGarity, A. E., Design and Operation of Civil and Environmental Engineering Systems,
John Wiley & Sons, New York, 1997. With permission.)
convex and concave sections. By assuming that the treatment efficiency is greater
than the primary treatment efficiency, the cost function can be changed to convex
form by linearly extending the secondary treatment section to the origin (dashed
line in Figure 9.7). The linear programming technique can also be used; for this
purpose, piecewise linear approximation of the cost function should be used. The
total cost of wastewater treatment can be written as follows (ReVelle and McGarity,
1997):
Total Cost = ∑ (c
j =1
j0 + c j1 . x j1 + c j 2 . x j 2 + c j 3 . x j 3 ) (9.45)
where m is the number of treatment plants; cj0 is the intercept of the convex
approximation with the cost axis ($/year); cj1, cj2, and cj3 are the slopes of the linear
approximation for the secondary, first tertiary, and second tertiary treatment seg-
ments, respectively ($/year/efficiency fraction); xj1, xj2, and xj3 are the efficiency
segments in the secondary, first tertiary, and second tertiary treatment segments,
respectively; and xj = xj1 + xj2 + xj3.
In general, Eq. (9.45) can be written as:
m Pj
Z= ∑ (c + ∑ c . x
j =1
j0
k =1
jk jk ) (9.46)
where Pj is the number of linear segments that are required to present the treatment
cost function. The relationship between l uj and lj, the BOD concentration in untreated
and treated wastewater discharge j, respectively, can by presented as:
l uj − l j
xj = j = 1,..., m (9.47)
l uj
where l uj and lj are expressed in mg/L, and xj is the efficiency of BOD removal. As
described before, the BOD concentration and oxygen deficit at the end of each reach
can be expressed as:
Li,e = ρi,e . Li,0 i = 1, 2,..., m (9.48)
Di,e = α i,e . Li,0 + βi,e . Di,0 i = 1, 2,..., m (9.49)
Using the average velocity of stream flow, the ti,e can be calculated; therefore, ρii,e ,
αii,e , and βii,e are constants for discharger j = i into reach i. The DO deficit at point
s in each reach i can be estimated as:
Di,s = α i,s Li,o + βi,s . Di,o ( i = l , …, m ) (9.50)
In order to estimate where the critical points are, each reach i is divided into ni
segments; the DO deficits are calculated at the end of each segment using the flow
travel times ti,s (i = 1, …, m; s = 1, …, ni) and the parameters α i,s and βi,s . The DO
deficit is limited to maximum allowable deficit at each reach i (Di,max). The number
of segments at each reach (ni) can be calculated based on the reach length and
characteristics of the oxygen sag curve. Following is the complete linear formulation
for DO and BOD concentration control in streams using treatment cost minimization
as the objective function (ReVelle and McGarity, 1997):
n Pj
Minimize z= ∑∑c
j =1 k =1
jk x jk (9.51)
Subject to:
Pj
∑x
1
+ l =1 j = 1, 2,..., m (9.52)
k =1
jk
l uj j
η j ,min ≤ x j1 ≤ η j1 j = 1, 2,..., m (9.53)

0 ≤ x jk ≤ ( η jk − η j ,k −1 ) k = 2, 3,..., Pj
(9.54)
j = 1, 2,..., m
(Qi −1 − q j ,out ) Li −1,e + q j ,in l j − Qi Li,0 = 0 i = 1, 2,..., m

(9.55)
j=i
(Qi −1 − q j ,out ) Di −1,e + q j ,in d j − Qi Di,0 = 0 i = 1, 2,..., m

(9.56)
j=i
Li,e − ρi,e Li,0 = 0 i = 1, 2,..., m (9.57)
Di,e − α i,e Li,0 − βi,e Di,0 = 0 i = 1, 2,..., m (9.58)
α i,s Li,0 − β i,s Di,0 ≤ Di,max i = 1, 2,..., m

(9.59)
s = 1, 2,..., ni
Di,0 ≤ Di,max i = 1, 2,..., m (9.60)
lj ≥ 0 j = 1, 2,..., m (9.61)
Li,e ≥ 0 i = 1, 2,..., m (9.62)
where:
ηj,min is the lowest treatment efficiency allowed at wastewater discharge j.

Pj is the number of linear segments necessary to represent the treatment cost
curve for wastewater discharge j.
ηj,k–1 and ηj,k are the upper and lower limits, respectively, of the kth segment.
qj,in and qj,out are the volumetric flow rates of discharge and withdrawal,
respectively, at point j.
lj and dj are the BOD and DO concentrations, respectively, of wastewater
discharge j.
9.2.5.1.2 General Cost-Minimization Model for Water Quality

Management
In the previous section, a cost-minimization model was presented for dissolved
oxygen management in rivers and streams. This model can be extended for an
arbitrary water quality parameter as follows (for a detailed description, see ReVelle
and McGarity, 1997):
Minimize Z= ∑ C (x )
j =1
j j (9.63)
Subject to: ∑a
j =1
i, j x j ≥ bi i = 1, 2,..., n (9.64)
lj < x j < uj j = 1, 2,..., m (9.65)
where:
xj is the treatment efficiency at discharge j.

Cj (xj) is the treatment cost function at discharge j.
m is the number of dischargers on the stream.
ai,j is the transfer coefficient between the pollutant source or discharger j and
checkpoint i indicating water quality improvement at point i as a result of
pollution removal at discharge j.
n is the number of checkpoints in the stream.
bi is the required water quality improvement at checkpoint i to satisfy water
quality standards.
lj and uj are the lower and upper limits, respectively, of the treatment efficiency
of discharge j.
Example 9.8
Many small communities and some small industries are located on a river. In a
stretch of the river two major polluters are located: Industry A at river mile 310 and
City B at river mile 302 with some industries. For this example, it can be assumed
that the water quality of the river is important from river mile 310 to river mile 275
and is particularly important at river mile 294, where a lot of swimming and boating
take place. The requirement at river mile 294 is that the dissolved oxygen concen-
tration should be equal to or exceed 5 mg/L. At other points between river miles
310 and 275, the dissolved oxygen concentration must be not less than 4 mg/L. Both
Industry A and City B already have some treatment facilities, so their treatment
efficiencies already exceed 30% (i.e., x1, x3 ≥ .30). In addition, due to the nature of
the wastes and current treatment practices, it would be impossible to achieve greater
than 95% efficiencies (i.e., x1, x3 ≤ .95). River C is a small tributary and has no
treatment plant associated with it (x2 = 0).
The data provided in Table 9.10 are from laboratory analyses of water just below
river miles 310, 307, and 302. The water samples were enclosed in airtight bottles,
and the BOD5 concentrations of the water samples were measured every 3 hours.
The water was collected on a summer day when the streamflows were equal to the
7-day, 10-year low flow; the water temperature was kept at 20°C during the analysis.
The flow rate at river mile 310.1 (upstream of Industry A discharger) was measured
as 3 (m3/sec). The results are presented in Table 9.10.
TABLE 9.10
Results of Measurement of BOD5 Concentrations (Example 9.8)
Time (hr) At Mile 309.9 (mg/l) At Mile 306.9 (mg/l) At Mile 301.9 (mg/l)
0 13.4 11.4 16.7

3 13.1 11.1 16.3
6 12.7 10.8 15.9
9 12.4 10.6 15.5
12 12.1 10.3 15.1
15 11.8 10.1 14.7
18 11.5 9.8 14.4
21 11.2 9.6 14.0
24 11.0 9.3 13.7
TABLE 9.11
Qualitative and Quantitative Characteristics of River and Wastewater
Discharge
Reach or Streamflow Discharge Influent Influent Average Stream
Discharge Velocity Rate BOD5 DO Depth
Point (m/day) (m3/sec) (mg/L) (mg/L) (m)
1 16062.96 0.0566 1020 8 0.732

2 10789.92 0.339 3 7 0.543
3 12923.52 0.283 210 7 0.564
1. For each reach of the river, estimate the value of the bio-oxidation rate
constant (kc) from the data provided. Some field measurements were made
last summer during what was considered to be the 7-day, 10-year low
flow for this river. The collected data are presented in Table 9.11.
2. For each reach of the river, estimate the value of the reaeration rate
constant (k2) using the Langbein–Duram formula.
The quantitative and qualitative characteristics of wastewater discharge
points are given in Table 9.11 and will be assumed to be constant in the
remainder of this example. Last summer, the data in Table 9.12 were
collected at the same time as the data of Table 9.11. The treatment effi-
ciencies of both Industry A and City B were 50%. The actual BOD5
concentration in the treated effluents of Industry A and City B were 510
mg/l and 105 mg/l, respectively. The water temperature was 27°C through-
out the river, and the DO saturation concentration was 8.0 mg/l.
3. Use the values of kc and k2 determined in parts 1 and 2 and the
Streeter–Phelps equation to simulate the BOD and DO for river miles 301
to 280. Be sure to adjust the values of kc and k2 for changes in temperature
(the temperature coefficient, θ, is considered to be 1.047 and 1.016 for kc
and k2, respectively). You should find that, between river miles 310 and
302, the estimated BOD and DO values closely match the data of
TABLE 9.12
BOD and DO Concentration (mg/l) along the River (Example 9.8)
BOD5 DO
(Computed (Computed
River BOD5 BOD5 with DO DO with
Mile (Measured) (Computed) Correction) (Measured) (Computed) Correction)
(1) (2) (3) (4) (5) (6) (7)
310.2 4.0 — — 7.0 — —

309.9 13.4 13.33 — 7.0 6.99 —
309.0 13.0 13.0 — 6.7 6.73 —
308.0 12.7 12.46 — 6.5 6.47 —
307.2 12.3 12.32 — 6.3 6.26 —
306.9 11.4 11.32 — 6.3 6.28 —
306.0 11.0 10.90 — 6.1 6.05 —
305.0 10.5 10.46 — 5.9 5.82 —
304.0 10.1 10.04 — 5.7 5.64 —
303.0 9.7 9.64 — 5.5 5.49 —
302.2 9.3 9.28 — 5.4 5.38 —
301.9 16.7 16.56 16.45 5.5 5.46 5.39
301.0 15.5 16.05 15.42 4.6 5.19 4.59
300.0 14.4 15.51 14.31 3.9 4.96 3.86
296.0 10.7 13.52 10.61 2.2 4.31 2.19
292.0 7.9 11.79 7.87 1.9 4.10 1.86
288.0 5.9 10.28 5.84 2.2 4.13 2.21
284.0 4.4 8.96 4.33 2.9 4.32 2.87
280.0 3.2 7.81 3.21 3.7 4.58 3.64
Table 9.12. Between river miles 301 and 280, the simulated BOD and DO
should vary significantly from the data presented in Table 9.12.
4. Assume that the value of kc for river miles 301 to 275 equals the value
found in part 2. Estimate a new value of k2 for river miles 301 to 275 that
reproduces the BOD and DO data of Table 9.12.
5. By solution of a linear program, find the minimum percentage removal for
Industry A and City B (i.e., the treatment efficiencies of both polluters are
equal) that ensures a DO concentration of at least 4 mg/L along the river.
Solution:
1. The bio-oxidation rate constant for each station can be calculated using
⎛ L ⎞
− ln⎜ c ⎟
L ⎝ Lc 0 ⎠
Lc = Lc 0 × e − kc .t , c = e − kc .t , and kc =
Lco t
TABLE 9.13
Values of –ln(Lc /Lc0) for Stations
Time (hr) Mile 309.9 (mg/L) Mile 306.9 (mg/L) Mile 301.9 (mg/L)
0 — — —
3 0.0226 0.0267 0.0242
6 0.0537 0.0541 0.0491
9 0.0776 0.0728 0.0746
12 0.1020 0.1015 0.1007
15 0.1272 0.1211 0.1276
18 0.1529 0.1512 0.1482
21 0.1793 0.1719 0.1764
24 0.1974 0.2036 0.1980
The values of –ln(Lc/Lc0) are presented in Table 9.13. If we draw the value
of –ln(Lc/Lc0) vs. t, the slope of this line will give the value of kc. Using
this method, the value of kc and the regression coefficient are calculated
for each station as follows:
x = 309.9 mile: kc1 = 0.2023 R = 0.99

x = 306.9 mile: kc 2 = 0.2002 R = 0.99
x = 301.9 mile: kc 3 = 0.1998 R = 0.99
The values of the regression coefficients (near 1) show that the values of
kc are acceptable.
2. The values of k2, in different reaches using Eq. (9.38), can be calculated as:
2.2208 × 16062.96
Reach 1: k2 = 4 = 0.626
86400(0.732) 3
Reach 2: k2 = 0.626
Reach 3: k2 = 0.713
3. The coefficients of kc and k2 should be adjusted using Eq. (9.35) as

follows:
Reach 1:
kc,27 = 0.2023(1.047)( 27−20 ) = 0.2790
k2,27 = 0.626(1.016)( 27−20 ) = 0.7

Reach 2:
kc,27 = 0.2761 and k2,27 = 0.7
Reach 3:
kc,27 = 0.2756 and k2,27 = 0.797
It is assumed that correction of kc based on Eq. (9.36) is negligible.

Considering the treatment efficiency of each treatment plant, the BOD
concentration of each wastewater discharge point is estimated as follows:
j = 1: x1 = 50%, and l1 = 0.5 × 1020 = 510 mg / L

j = 2: x2 = 0%, and l1 = 3 mg / L
j = 3: x1 = 5 = 50%, and l1 = 0.5 × 210 = 105 mg / L
The mixed BOD and DO concentrations just downstream of discharge

point 1 are calculated as:
(3 × 7) + (0.0566 × 8)
DOmix = = 7.019 mg / L
3.056
(3 × 4) + (0.0566 × 510)
BODmix = = 13.36 mg / L
3.056
The DO and BOD concentrations in each discharge point are calculated

using Eqs. (9.42) and (9.43). The travel time (ti,s) for each distance (s)
from initial point of reach i is computed using reach average velocity. For
example, in reach 1, Eqs. (9.42) and (9.43) are used as follows:
BOD concentration at point s:
Lc,s = Lc,0 e − kcts = 13.36e −0.279ts
DO concentration at point s can be calculated considering Cs = Os – Ds

as follows:
( )
Cs = Os 1 − e − k2ts + Doe − k2ts (
kc Lc 0 − kcts
k2 − kc
e − e − k2ts )
(9.66)
0.279 × 13.36 ⎞
(
Cs = 8 1 − e −0.7 ts
) + 7.019e −0.7 ts
−⎛
⎝ 0.7 − 0.279 ⎠ (
× e −0.279ts − e −0.7ts )
At river mile 309.9, ts = 0.01 day and the BOD and DO concentrations are:
L = 13.36e −0.279×0.01 = 13.3 mg / L

0.279 × 13.36 ⎞
( )
C = 8 1 − e −0.7×0.01 + 7.019e −0.7×0.01 − ⎛
⎝ 0.7 − 0.279 ⎠
×
(e −0.279× 0.01
)
− e −0.7×0.01 = 6.99 mg / L
The computed BOD and DO concentrations for different stations along

the river are presented in columns 3 and 6 of Table 9.12.
4. The new value of k2 for reach 3 can be calculated based on the data
(column 2 of Table 9.12). Using this method (which was presented in the
first part of this example), a new value of kc for reach 3 is computed as
0.6. The new values for BOD and DO concentrations at stations in reach
3 can be found in columns 4 and 7 of Table 9.12.
5. Linear programming models can be used to calculate optimum treatment
efficiencies. The objective function of this model is considered as:
Minimize z = x1
or
Minimize z = x3
where x1 and x2 are treatment efficiencies for treatment plant 1 and 3 at

river miles 310.0 and 302.0. The constraints of this optimization model
are as follows:
Constraints related to treatment efficiencies
x1 = x3, x2 = 0, and 0.3 ≤ x1, x3 ≤ 0.95
l1 = 1020(1 – x1)
l2 = 3(1 – x2)
l3 = 210(1 – x3)
Constraints related to BOD and DO concentrations at the initial and
endpoints of each reach using Eqs. (9.32), (9.40), (9.41), and (9.66)
C1,0 = 7.019
L1,0 = 3.926 + 0.0185 L1
C1,e = 7.2048 – 0.0723 L1,0
L1,e = 0.93 L1,0
where C1,0 and C1,e are the DO concentrations and L1,0 and L1,e are the
BOD concentrations at the initial and endpoints of reach 1, respectively.
C2,0 = 0.9 C1,e + 0.7
L2,0 = 0.9 L1,e + 0.3
TABLE 9.14
BOD and DO Concentrations at River Checkpoints (Example 9.8)
River Travel Time from DO Concentration BOD Concentration
Mile River Mile 310.0 (mg/L) (mg/L)
309.9 0.010 DO1 = 0.0558 + 6.970 – 0.00276 L1 BOD1 = 0.997 L1

307.1 0.290 DO2 = 1.470 + 5.728 – 0.0702 L1 BOD2 = 0.922 L1
306.9 0.015 DO3 = 0.0836 + 0.990 C2 – 0.0041 L2 BOD3 = 0.996 L2
302.1 0.731 DO4 = 3.204 + 0.599 C2 – 0.1418 L2 BOD4 = 0.817 L2
300 0.012 DO5 = 0.0761+ 0.990 C3 – 0.00714 L3 BOD5 = 0.993 L3
301.9 0.249 DO6 = 1.440 + 0.820 C3 – 0.1256 L3 BOD6 = 0.861 L3
298 0.498 DO7 = 2.621 + 0.672 C3 – 0.2111 L3 BOD7 = 0.742 L3
296 0.747 DO8 = 3.589 + 0.551 C3 – 0.2662 L3 BOD8 = 0.639 L3
294 0.996 DO9 = 4.383 + 0.452 C3 – 0.2985 L3 BOD9 = 0.550 L3
290 1.494 DO10 = 5.568 + 0.304 C3 – 0.3169 L3 BOD10 = 0.408 L3
286 1.992 DO11 = 6.364 + 0.204 C3 – 0.2992 L3 BOD11 = 0.303 L3
282 2.490 DO12 = 6.900 + 0.137 C3 – 0.2651 L3 BOD12 = 0.224 L3
278 2.988 DO13 = 7.261 + 0.092 C3 – 0.2256 L3 BOD13 = 0.166 L3
274 3.486 DO14 = 7.503 + 0.062 C3 – 0.1868 L3 BOD14 = 0.123 L3
C2,e = –0.1437 L2,0 + 0.593 C2,0 + 3.254

L2,e = 0.814 L2,0
C3,0 = 0.923 C2,e + 0.539
L3,0 = 0.831 L2,e + 0.0769L3
C3,e = 0.0686 C3,0 – 0.196 L3,0 + 7.451
L3,e = 0.4 L3,0
Other constraints that give the DO and BOD concentrations at other river
miles are presented in columns 3 and 4 of Table 9.14. The proposed linear
model was solved using a linear programming package. The optimum
treatment efficiency for both treatment plants is 70%. The BOD and DO
concentrations along the river based on these treatment efficiencies are
shown in Figure 9.8.
9.2.5.1.3 Management Model To Control Water Quality
Violation
This model minimizes the worst water quality response within the river or stream
system. In this formulation, it is assumed that the treatment cost of dischargers may
be expressed as piecewise linear cost functions. The model is of the form (Burn and
Lence, 1992):
11
10 BOD
9 DO
BOD, DO (mg/lit)
8
7
6
5
4
3
2
1
0
275 280 285 290 295 300 305 310
Distance (mile)
FIGURE 9.8 Dissolved oxygen and BOD concentrations along the river (Example 9.8).
Minimize Z = MV (9.67)
Subject to: ∑ W (1 − R )a
j
j j ij + BGi + Ui − Vi = Os,i − DOi (9.68)
Pj
Rj − ∑x k
jk = Rjmin j = 1, 2,..., m (9.69)
m Pj
∑ ∑ c .x
j =1 k
jk jk ≤B (9.70)
η j ,min ≤ x j1 ≤ η j ,1 j = 1, 2,..., m (9.71)
0 ≤ x jk ≤ η j ,k − η j ,k −1 j = 1, 2,..., m (9.72)
MV − Vi ≥ 0 ∀i (9.73)
where:
MV is the maximum violation at any checkpoint (mg/L).

Wj is the influent waste load of BOD for discharge point j (kg/day).
Rj is the removal fraction of source j.
aij is the transfer coefficient between source j and checkpoint i (milligram
per liter of DO deficit per kilogram per day of BOD loading).
BGi is the background DO deficit concentration at checkpoint i.
Ui is the amount by which the DO concentration exceeds the standard level
at checkpoint i (mg/L).
Vi is the violation of a water quality standard at checkpoint i (mg/L).
Csi is the saturation DO concentration at checkpoint i.
DOi is the DO standard at checkpoint i.
xjk is the removal fraction at source j, which is associated with treatment on
the kth linear segment of piecewise linear cost function.
Rjmin is the minimum allowable removal fraction at discharge j.
cjk is the slope of the kth linear segment of the treatment cost curve for
discharge j.
B is the total budget for treatment ($/yr).
Pj is the number of linear segments necessary to represent the treatment cost
curve for wastewater discharge j.
The violation depends on the treatment efficiency at the sources and the transfer
coefficients and background dissolved oxygen deficits. Equation (9.68) defines the
violation of the DO standard at location i, Eq. (9.70) constrains the amount of money
required for treatment to be less than the budgeted amount, and Eq. (9.73) states
that the maximum violation is equal to or greater than each of the violations.
9.2.5.1.4 Deterministic Dynamic Programming Model for
One of the important advantages of dynamic programming for water quality man-
agement is that it permits the use of more realistic water quality simulation models.
In contrast to many water quality simulation models, linearity is inherent in the
Streeter–Phelps equation. In this section, a deterministic dynamic programming is
presented that minimizes the total treatment cost such that water quality standards
are met at boundaries. The recursive function of this model is:
ν(s n)k = k ′ = r k ′ + ν( s* n − 1) (9.74)
where v is treatment cost; r k′ is the treatment cost of control option k′; and s* is the
resultant water quality state in stage n – 1 if control option k′ is applied in stage n.
Water quality state s at stage n is mapped onto water quality state s* at stage
n – 1 by control action k through the transition function T as follows:
s ∗ = T (s, k )
FIGURE 9.9 River reaches and dischargers. (From Fujiwara, O., Puangmaha, W., and
Hanaki, K., ASCE J. Environ. Eng., 114(4), 1988. With permission.)
The objective function of the dynamic programming model minimizes the overall
cost of all feasible control actions (Cardwell and Ellis, 1993):
ν(s n) = min (r k + ν( s* n − 1)) (9.75)

k ∈K
The feasible control set K consists of all control options k that give acceptable water
quality at the end of the reach:
K = [ k s* ≥ S ] (9.76)
where S is the standard level of water quality variable. One of the problems of
deterministic dynamic programming models is that, when a water quality standard
must be attained (i.e., treated as an inviolate lower bound), control actions that
implicitly or explicitly create transitions to an unacceptable downstream state are
declared not feasible (Cardwell and Ellis, 1993).
9.2.5.2 Stochastic Approach
Streamwater quality management is complex because the rivers and streams are
polluted from multiple sources, and water quality depends on uncertain streamflow
conditions. Therefore, it is necessary to use optimization models that incorporate
both uncertainty and complex characteristics of the pollution problem. Stochastic
optimization, in which streamflow and/or wastewater flow rate and also various river
parameters are assumed to be random variables, have been formulated by many
researchers (e.g., Lohani and Thanh, 1978; Cardwell and Ellis, 1993; Takyi and
Lence, 1999). In this section, an optimization model with an ability to reflect
important uncertainties of streamwater quality problems is demonstrated. This
model, which was developed by Fujiwara et al. (1988), optimizes wastewater treat-
ment efficiencies at source points and considers various random parameters of the
river. Quintiles of random variables concerning DO concentration are calculated

using a Monte Carlo simulation.
As shown in Figure 9.9, the river has n pollutant sources, and the distance between
two sources is considered as a reach. Each pollutant source has a wastewater treatment
plant, and mi water quality control points are considered at each reach i. The objective
function of the model is to minimize the total sum of weighted DO deficits at all
checkpoints. Based on the Streeter–Phelps model, the DO deficit at the jth checkpoint
of reach i can be defined as follows (Loucks et al., 1967; Fujiwara et al., 1988):
⎡⎛ ⎞ ⎤ ⎛ ⎞
i
⎢⎜ k ⎟ ⎥ ⎜ ⎟
⎢⎜ aij ⎟ ⎥ ⎜ dij ⎟
Dij = ∑ ⎢⎜ i b
⎟ k
(1 − x )
k ⎥ +
⎜ i ⎟
(9.77)
k =1 ⎢⎜
⎢⎝ ∑ql ⎟
⎣ l =0 ⎠
⎥ ⎜
⎥ ⎝
⎦ l =0
∑
ql ⎟
⎠
where:
DDij is the DO deficit at the jth checkpoint in reach i.

aikj and dij are functions of the deoxygenating and reaeration constants, sed-
imentation coefficients, BOD additions by resuspension of sediments, oxy-
gen reductions by benthal demand, oxygen productions by photosynthesis,
and flow travel time.
qk is the wastewater discharge at the kth pollutant source.
bk is the BOD loading before entering the kth treatment plant.
xk is the wastewater treatment efficiency of the kth discharger.
The objective function of the model that minimizes the weighted sum of DO
deficits can be written as follows:
⎧ ⎡⎛ ⎞ ⎤ ⎫
mi
⎪ i ⎢⎜ k ⎟ ⎥ ⎪
⎪ dij ⎪
n
⎢⎜ aij ⎟ ⎥
Minimize ∑∑ ∑ wij ⎨ ⎢⎜ i b (1 − xk )⎥ + i
⎟ k ⎬ (9.78)
i =1 j =1 ⎪ k =1
⎪
⎢⎜
⎢⎝∑
⎣ l =0
ql ⎟
⎠
⎥
⎥ ∑
⎦ l =0 ⎭
ql ⎪
⎪
⎩
where n is total number of pollutant sources and wij are the relative weights of Dij.
Equation (9.78) can be rewritten by dropping the terms that are independent of
the decision variables (xk):
Maximize ∑a b x
k =1
k k k (9.79)
where:
⎛ ⎞
n mi
wij aijk
ak = ∑∑
i=k j =1
⎜ q + q +⋯+ q ⎟
⎝ 0 1 k⎠
(9.80)
For a piecewise linear BOD removal cost function, the optimization model has the
following linear form:
Maximize gT x (9.81)
Subject to: Bx ≤ h (9.82)
x≥0 (9.83)
where:
g is a column vector of n random components (n × 1).

x is the vector of decision variables (n × 1).
B is an (m × n) constant matrix.
h is an (m × 1) constant matrix.
T is the transposition of a matrix or vector.
This linear programming model has a dual form:
Minimize hT λ (9.84)
Subject to: BT λ ≥ g (9.85)
λ≥0 (9.86)
where λis the vector of dual variables. Equation 9.85 can be replaced by the chance
constraint to develop a stochastic optimization model, as follows:
⎡ m
⎤
Pr ⎢
⎢⎣
∑b λ
i =1
ij i ≥ gi ⎥ ≥ e j
⎥⎦
for j = 1, 2,..., n (9.87)
which have the deterministic equivalent as follows (for more details, see Chapter 3):
∑ b .λ ≥ u
i =1
ij i j j = 1, 2,…, n (9.88)
where ej is the probability of exceeding (ej = Pr[gj ≤ uj]), and uj is the ejth quantile
of gj. Therefore, the random vector g is replaced by a quantile vector u of g.
Taking the dual problem of this modified LP, deterministic model proposed by
Burn and McBean (1985) and Fujiwara (1988) is derived as:
Maximize z=u x= T
∑u xk =1
k k (9.89)
Subject to: ∑∑c x

k =1 i
i
k
i
k ≤c (9.90)
Pk
xk − ∑x i
i
k = xkl ∀k (9.91)
0 ≤ x ki ≤ zki ∀i, k (9.92)
xkl ≤ xk ≤ xku ∀k (9.93)
where:
uk is the ekth quintile of akbk.

ek is the probability of exceeding akbk.
cki is the ith slope of the piecewise linear treatment cost function at the kth
treatment plant.
c is the budget limit.
zki is the upper limit of x ki .
xkl is the lower limit of xk.
x ku is the upper limit of xk.
Pk is the number of linear segments necessary to represent the treatment cost
curve for wastewater discharge k.
It is difficult to determine the probability distribution function of akbk when the

variables such as river flow, deoxygenating and reaeration constants, and flow travel
time are random. The Monte Carlo simulation can be employed to calculate the
desired quantities uk. The Monte Carlo simulation procedure to determine uk as
proposed by Fujiwara et al. (1988) is as follows:
1. Select the random variables.

2. Generate random numbers and calculate m independent values for akbk.
3. Sort the calculated values of akbk, (Y1, Y2, …, Ym) in ascending order (Y1
≤ Y2 ≤ … ≤ Ym).
4. Determine Yi where i is the first i for which i > mek. Therefore, Yi is the
ekth quintile of akbk.
5. For Z1 = Yi, replicate the above procedure N times to determine (Z2, …,
ZN).
6. Calculate the mean and variance of sample Zi:
∑Zj =1
j
ZN = (9.94)
N
∑ (Z )
2
j − ZN
j =1
SN2 = (9.95)
( N − 1)
7. The relative precision (RP) is calculated as follows:
δ( N , α )
RP = (9.96)
ZN
where δ(N, α), which is half the length of the 100(1 – α)% confidence
interval of the expected value of Zj, is:
⎛ S2 ⎞
δ( N , α ) = t α .⎜ N N ⎟ (9.97)
N −1,1 −
2
⎝ ⎠
and tN–1,1–(α/2) is the (1 – (α/2))th quintile of the distribution with (N – 1)

degree of freedom.
8. If the relative precision is smaller than a given value β (0 < β < 1),
terminate the Monte Carlo simulation with Z N being the desired quintile
uk; otherwise, increase the N to N + 1 to obtain the desired precision.
For more details, the reader is referred to Fujiwara et al. (1988).
9.2.6 LAKE AND RESERVOIR WATER QUALITY MANAGEMENT

Water quality management in lakes and reservoirs is entirely different from rivers
and streams. Reservoirs can experience dramatic changes in the physical, chemical,
and biological characteristics of inflows. Phosphorus is the most significant pollutant
of lakes as it can seriously affect the overall water quality. Oxygen-demanding wastes
can also be important, especially when domestic or industrial wastes are discharged
to a lake or upstream river. The water in reservoirs and lakes can be stratified due
to a difference in density caused by temperature, dissolved substances, and sus-
pended solids; therefore, the water quality can vary with depth, and water of varying
quality can be obtained from a reservoir by withdrawing it from various levels. Water
Epilimnion (warm,
Thermocline acrobic, well mixed)
Hypolimnion (cool,
poorly mixed,
anaerobic)
(a)
Thermocline
(poorly defined)
Ice
0°C
Hypolimnion
4°C
(b)
FIGURE 9.10 Stratification of a lake during (a) summer and (b) winter.
withdrawal from reservoirs can be managed in this way to supply water that satisfies
water quantity and quality requirements.
9.2.6.1 Stratification and Turnover
Most lakes and reservoirs in the tropical areas are stratified during the summer and
overturn in autumn, but in cold regions reservoirs can undergo winter stratification
and spring overturn. Heat transfer changes the temperature at the water surface and
depends on a number of factors, including air temperature, wind, humidity, and the
magnitude of solar radiation. During the summer, warm water has less density and
remains near the surface. The upper layer, which is nearly mixed by wind and other
forces (the epilimnion), floats on the lower layer (the hypolimnion), which is cooler
and poorly mixed. Figure 9.10a shows the hypolimnion and epilimnion layers in
summer and winter. As shown in this figure, the boundary of these two layers is
called thermocline, which has a high temperature and density gradient.
In the fall, as the surface layer becomes cool and winter storms arrive, the surface
water that is denser than the hypolimnion sinks and the lake or reservoir experiences
a fall turnover. During the turnover, the lake is isothermal from top to bottom. The
turnover process stops when the temperature reaches 4°C, because water has the
highest density at this condition. If the temperature of the surface water decreases,
the winter stratification will occur as shown in Figure 9.10b. A spring turnover occurs
when the water warms and becomes isothermal and well mixed again.
Vertical stratification affects the chemical and biological characteristics of water
in the lakes and reservoirs. The thermocline restricts the diffusion of dissolved
substances; therefore, the hypolimnion and the atmosphere do not contact, and
dissolved oxygen may become depleted in the hypolimnion. This condition can result
in the production of hydrogen sulfide (H2S) and the release of methane from iron-
bound phosphorous of sediments, which can dramatically change the biological
environment.
9.2.6.2 Eutrophication
Eutrophication is a significant factor of lakes and reservoirs that can change their
ecological conditions. The introduction and cycling of nutrients can make lakes and
reservoirs more productive and nutrient rich. High levels of nutrients, large popula-
tions of phytoplanktons (algae), and low transparency are characteristics of eutrophic
lakes. Eutrophication time depends on the size of the lake or reservoir and the rate
and characteristics of introduced nutrients. In highly eutrophic lakes, blue–green
algae can result in undesirable odors and tastes and may change the appearance of
the water. In deep and stratified lakes, algal production in the epilimnion can result
in oxygen depletion and anaerobic conditions in the hypolimnion. Nitrogen and
phosphorus are the main factors contributing to algal growth and lake eutrophication.
Therefore, a major concern of water quality management in lakes and reservoirs is
to limit the introduction of nutrients and to slow the entrophication rate, which is
commonly accomplished by limiting phosphorus.
9.2.7 RIVER–RESERVOIR SIMULATION MODEL PACKAGES

Several computer simulation models have been developed to help water managers
to develop appropriate policies to meet qualitative and quantitative objectives of the
systems. These models can simulate the effects of pollution discharges into surface
water bodies and evaluate temperature changes resulting from the thermal discharges
or thermal stratification of reservoirs. In this section, we present the capabilities and
restrictions of HEC-5Q, which has been widely used for river–reservoir quality
management by different investigators.
9.2.7.1 HEC-5Q
The HEC-5Q is a simulation model for flood control and conservation systems. It
has useful abilities to accept user-specified water quality and quantity objectives and
to operate the network of reservoirs. The decision criteria are programmed to con-
sider flood control, hydropower, instream flow (municipal, industrial, and irrigation
water supply, and fish habitat needs), and water quality requirements (Willey et al.,
1996). The first version of HEC-5Q was presented in 1979 as a modified version of
HEC-5 model to evaluate reservoir operation for water quality control in large
reservoir systems. The sixth phase of development of this model (1992) added the
capability of simulating new qualitative parameters such as chemicals adsorbed onto
organic and inorganic particles.
Deep reservoirs are conceptually modeled as a series of one-dimensional hori-
zontal layers. Each layer is characterized by a surface area, thickness, and volume,
such as those shown in Figure 9.11. Within each layer, the water is considered
completely mixed, and only the vertical gradient is retained during the water quality
Evaportation
Tributary
inflow
Tributary
inflow
Vertical advection
& effective diffusion
Volume Element
Outlet
Outflow
FIGURE 9.11 Geometric representation of stratified reservoir and mass transport mecha-
nism. (From U.S. Army Corps of Engineers, HEC5: Appendix on Water Quality Analysis,
U.S. Army Corps of Engineers, Washington, D.C., 1986.)
FIGURE 9.12 Geometric representation of stream system and mass transport mechanism.
(From U.S. Army Corps of Engineers, HEC5: Appendix on Water Quality Analysis, U.S.
Army Corps of Engineers, Washington, D.C., 1986.)
simulation. Withdrawals and external inflows are considered as sources or sinks of

each layer.
The stream system is conceptually modeled as a linear network of volume
elements. The flow simulation module requires hydraulic characteristics of a stream
such as length, width, cross-section area, Manning’s number, hydraulic radius, and
flow and depth relationships (see Figure 9.12). The flow simulation module calculates
streamflow rates at control points using one of the several programmed hydrologic
routing methods. The water quality simulation module can simulate temperature and
selected conservative and nonconservative constituents that might be included in
planning studies. This module computes the vertical distribution of temperature and
other water quality parameters in the reservoir and downstream control points.
HEC-5Q can select the gate opening of the selective withdrawal structure to
satisfy the water quality limits at downstream checkpoints. The water quality module
has three modes:
1. Calibration mode
2. Annual simulation mode
3. Long-term simulation mode
In the calibration mode, decay rates and dispersion coefficients are determined
based on the historical flows and water quality monitoring data. The annual simu-
lation mode is executed on a daily basis and can evaluate the effects of reservoir
operation on water quality of reservoir and downstream reaches. The time steps of
the long-term simulation mode are longer (generally 30 days) and can show the
effects of reservoir operation on water quality over a long-term planning horizon
(e.g., more than 10 years). In the calibration and annual simulation modes, the time
interval must be one day, and simulations are limited to periods contained in one
calendar year (U.S. Army Corps of Engineers, 1986).
9.2.8 RESERVOIR WATER QUALITY MANAGEMENT

The water quality in reservoirs often degrades over time because of thermal strati-
fication and accumulation of various kinds of pollutants; consequently, reservoir
releases can change the quality of downstream water bodies. The water in reservoirs
usually becomes stratified due to differences in the density gradient caused by
temperature or suspended and dissolved substances. The use of a selective with-
drawal structure is useful for controlling the quality of releases. This structure allows
the release of water from various layers of the reservoir. In this way, reservoirs can
be operated to satisfy downstream water quality and quantity requirements. Numer-
ous methods for optimal operation of reservoirs have been presented in literature,
but relatively few studies have considered the qualitative and quantitative operation
of reservoirs (e.g., Fontane and Labadie, 1981; Dandy and Crawly, 1992; Nandalal
and Bogardi, 1994).
In the next section, a stochastic optimization model is presented for qualitative
and quantitative operation of reservoirs. In this model (developed by Nandalal and
Bogardi, 1994), a nonlinear optimization model at each time interval optimizes a
qualitative objective function using the vertical distribution of reservoir water quality.
This objective function is based on the quality of reservoir releases from outlets that
are situated at different levels. Figure 9.13 shows the reservoir segments and the
related outlet of each segment (layer). The volume of each layer (each smaller
hypothetical reservoir) is determined based on the elevation of outlets.
FIGURE 9.13 Division of reservoir and simplified configuration. (From Nandalal, K. D. W.

and Bogardi, J. J., Reservoir management for water quality control, in Proc. of the Ninth
Congress of Asian and Pacific Division of the International Association for Hydraulic
Research, Singapore, 1994. With permission.)
It is assumed that each outlet has two openings, and the outflow of one of them
should satisfy the quantitative downstream demands while the second one releases
the extra water for scouring and flashing the reservoir. The following equations
address the releases, scour volumes and salinity during the time period j (j = 1, 2,
…, N) (Nandalal and Bogardi, 1994):
Rel j = ∑ rel
i =1
i, j (9.98)
Sco j = ∑ sco
i =1
i, j (9.99)
∑C
i =1
i, j .reli, j
Crel , j = n (9.100)
∑ rel
i =1
i, j
∑C
i =1
i, j .scoi, j
Cscour , j = n (9.101)
∑ i =1
scoi, j
where:
Relj is the total release during period j (million m3).

Scoj is the total scour volume during period j (million m3).
Crel,j is the release salinity during period j (mg/L).
Cscour,j is the scour salinity during period j (mg/L).
Ci,j is the outflow salinity during period j from outlet i (mg/L).
N is the total number of periods.
reli,j is the release from outlet i during period j (million m3).
n is the number of outlets.
scoi,j is the volume of scour from outlet i during period j (million m3).
Satisfying the downstream water quality requirements and avoiding salinity buildup
in the reservoir are the main qualitative objectives of reservoir operation, and the
qualitative objective function can be written as:
N ⎡ ⎛C − C ⎞2 ⎛ Cscour , j − Cˆ trg, j ⎞ ⎤
2
Minimize Z= ∑ ⎢W rel , j
⎢ 1 ⎜⎝ Ctrg, j
trg , j
⎟ + W2 ⎜
⎠ ⎝ C ˆ ⎟ ⎥⎥
⎠ ⎦
(9.102)
j =1
⎣ trg , j
Subject to: Rel j = ∑ rel

i =1
i, j ≥ Dem j (9.103)
Rel j + Sco j = ∑ (rel

i =1
i, j )
+ scoi, j ≤ Aj (9.104)
reli, j + scoi, j ≤ Bi ≤ Vi, j i = 1, 2,…, n (9.105)
Crel , j − Ctrg, j = 0 if Crel , j ≤ Ctrg, j (9.106)
Cscour , j − Cˆ trg, j = 0 if Cscour , j ≥ Cˆ trg, j (9.107)
where:
Aj is the allowable stochastic outflow (release + scour) during period j (million

m3), as determined from a quantitative reservoir operation optimization
model.
Bi is the total allowable outflow from outlet i (million m3).
Ctrg,j is the target release water quality (salinity) during period j (mg/L).
Cˆtrg, j is the target score salinity during period j (mg/L).
Demj is the downstream demand of water in period j (million m3).
Vi,j is the initial volume of water in segment i in period j (million m3).

W1 and W2 are the relative weights.
In the above nonlinear optimization model, the inflows during each time interval are
neglected. The nonlinear optimization model can result in several different local
optima of possible sets of releases from outlets. Some of these solutions that give
the least objective function should be selected for further investigation. A reservoir
water quality simulation (e.g., HEC-5Q) can be used to compare these potential sets
of selective withdrawals to select the best release policy.
9.2.9 CONFLICT RESOLUTION IN SURFACE WATER QUALITY

MANAGEMENT
Environmental protection is commonly in conflict with other uses of water resources.
In other words, environmental management is a multiple-objective problem that can
be in conflict with economic, hydraulic, water quality, or environmental objectives.
For example:
• Allocation of water to domestic, industrial, and agricultural demands is

in conflict with instream flow and the amount of water that can be allocated
to lakes and wetlands to protect their aquatic life.
• Extra discharges of water resources such as rivers and reservoirs can cause
some environmental problems such as dissolved oxygen depletion.
• Agricultural return flows and industrial effluents discharged to water
resources reduce the quality of water; the extent of such a reduction
depends on the assimilative capacity of a particular system. Dischargers
want to reduce their treatment costs, but this is not the concern of envi-
ronmental agencies.
Identification of conflict issues and engaged agencies, defining their utility functions
for different objectives, and defining the relative authority of agencies are the key
steps for defining the required information for conflict resolution models.
The application of Nash bargaining theory (NBT) and multiple-criteria decision-
making (MCDM) methods to conflict resolution of water resources has been pre-
sented in previous chapters. These methods can help to provide a better understand-
ing of the mechanisms of cooperation and the roots of and reasons for conflict and
its resolution in various political and social settings. They also can promote a better
understanding of the issues, and help stakeholders more fully comprehend the
potential consequences of their preferences and choices. This chapter presents
another example of the application of the Nash conflict resolution method to illustrate
the conflict issues and their resolution in water quality management problems.
Example 9.9
Determine the monthly allocated water to domestic, industrial, agricultural, and

recreation demands in a river system shown in Figure 9.14. The river discharge is
FIGURE 9.14 Components of the river system (Example 9.9).
1.2
Agriculture
1 Domestic
Industry
0.8
Utility
0.6
0.4
0.2
0
0 20 40 60 80 100
Allocated water (cms)
FIGURE 9.15 Utility function of water allocated to different demands.
40 m3/sec, and the return flows of domestic and industrial sectors are 20% of the
allocated water. The travel time in reaches 1 and 2 is assumed to be 8 and 4 hours,
respectively. The BOD5 concentration of headwater, domestic, and industrial return
flows are 1, 30, and 50 mg/L, respectively. The BOD5 concentration of the river at
the entrance point of the lake should be less than 10 mg/L, the BOD5 decay rate is
0.15, and the initial volume of the lake is 30 million m3. The utility functions for
agencies allocating water to different sectors are presented in Figures 9.15 and 9.16.
Solution: The nonsymmetric Nash solution of this problem is the unique optimal
Maximize ∏( f − d )
i =1
i i
wi
Subject to: di ≤ fi ≤ fi∗ ∀i

FIGURE 9.16 Utility function of the volume of a lake.
where:
wi is the relative weight.

fi is the utility function.
di is the disagreement point.
fi* is the ideal point of player (agency) i.
This objective function should be maximized taking into consideration the con-
straints of river water quality. The BOD5 decay can be considered using Eq. (9.32).
Results of the model, with and without the water quality constraint, are presented
in Table 9.15. The relative weights of agencies that are related to agricultural,
domestic, industrial, and recreational demands are assumed to be 0.047, 0.465, 0.023,
and 0.465, respectively, and their disagreement points are 0.0, 0.9, 0.9, and 0.0,
respectively. Considering that the environmental constraint and the inflow equal to
40, the BOD constraint cannot be met, and, therefore, we cannot find any conflict
resolution until a hard environmental constraint is considered. Increases in upstream
river discharges can greatly change the water allocated to the various sectors. Results
of the conflict resolution model for river discharge of 50 m3/sec are also presented
in Table 9.15.
TABLE 9.15
Water Allocation Based on the Conflict Resolution Method
With Without Without
Environmental Environmental Environmental
Constraint Constraint Constraint
Demands (Inflow = 50) (Inflow = 40) (Inflow = 50)
Agriculture (m3/sec) 15 8.43 14.49

Domestic (m3/sec) 20 20 20
Industry (m3/sec) 14 17 19.57
Final volume of lake 46 35 39.78
(million m3)
9.3 GROUNDWATER QUALITY MANAGEMENT

9.3.1 INTRODUCTION
Groundwater is a reliable source because it usually does not have high quality and
quantity fluctuations. Subsurface environment is a complex system subject to con-
tamination from numerous sources. Furthermore, the extremely slow movement of
pollutants in groundwater results in a little diffusion of pollutants and a longer
residence time. The restoration of groundwater quality is difficult and expensive. An
understanding of the contaminants transport and degradation process is essential for
groundwater quality planning and management. For this purpose, the groundwater
quality simulation models that can describe how the groundwater quality system
responds to, or is affected by contaminants is needed . From a planning perspective,
the main issues in groundwater quality management are (Willis and Yeh, 1987):
1 To ensure that surface and subsurface waste disposal have minimal impact
on the groundwater environment.
2 To optimize the waste treatment capacity of the subsurface system.
This section discusses the main groundwater pollutant sources and contaminant
transport in groundwater systems. Groundwater quality management models are
described in the last section of this chapter.
9.3.2 SOURCES OF GROUNDWATER CONTAMINATION

As water moves through the hydrologic cycle, its quality varies due to changes in
physical, chemical, and biological characteristics of environment. The changes can
be natural or manmade. For example, the quality of groundwater is commonly
affected by waste disposal, such as storage of waste materials in excavations.
Table 9.16 and Figure 9.17 show a summary of human-influenced groundwater
pollutant sources. These sources can be categorized as:
1. Water-soluble products or liquid contaminants that are placed on the land

or in surface water.
2. Deposited or stored substances in the ground above the water table.
3. Pollutants that are discharged in or extracted from the ground, below the
water table.
For a detailed explanation of groundwater pollutants sources, the reader is referred

to EPA (1985).
9.3.3 CONTAMINANT TRANSPORT IN THE SATURATED ZONE

Groundwater quality management requires anticipating the temporal and spatial
changes of contaminant concentrations in geological material. In this section, the
equations of pollutant transport in surface water are extended for groundwater quality
modeling. The transport and kinetic processes in groundwater are very similar to
those for surface water. For groundwater quality modeling, ms is defined as the solids
TABLE 9.16
Sources of Groundwater Quality Degradation
Groundwater Quality Problems that Originate on the Land Surface

Infiltration of polluted surface water
Land disposal of either solid or liquid wastes
Dumps
Disposal of sewage and water-treatment plant sludge
Deicing salt usage and storage
Animal feedlots
Fertilizers and pesticides
Accidental spills
Particulate matter from airborne sources
Groundwater Quality Problems that Originate in the Ground above the Water Table
Septic tanks, cesspools, and privies
Holding ponds and lagoons
Sanitary landfills
Waste disposal in excavations
Leakage from underground storage tanks
Leakage from underground pipelines
Artificial recharge
Groundwater Quality Problems that Originate in the Ground below the Water Table
Waste disposal in well excavations
Drainage wells and canals
Well disposal of wastes
Underground storage
Exploratory wells
Groundwater development
Source: EPA, Protection of Public Water Supplies from Ground-Water Contamination, Center for Envi-
ronmental Research Information, U.S. Environmental Protection Agency, Washington, D.C.
content per unit volume of water in groundwater, and md is defined as the mass of
solids per unit volume of space; ms and md are related to each other as ηms = md,
where η is the soil porosity.
The Reynolds transport equation for the dissolved pollutant concentration (cd)
in a mass control can be written as follows (Sincero and Sincero, 1996):
For conservative substances
∂( ηcd ) ∂ ⎛ ∂ηcd ⎞ ∂ ⎛ ∂ηcd ⎞ ∂ ⎛ ∂ηcd ⎞

+ ⎜E ⎟ + ⎜E ⎟ + ⎜E ⎟
∂t ∂x ⎝ xx ∂x ⎠ ∂y ⎝ yy ∂y ⎠ ∂z ⎝ zz ∂z ⎠
(9.108)
η ⎛ ∂cd ∂c ∂c ⎞
+ ⎜u +υ d +ω d⎟ = 0
1 + fcp K sc md η ⎝ ∂x ∂y ∂z ⎠
424
FIGURE 9.17 Sources of groundwater contamination. (From Delleur, J. W., The Handbook of Groundwater Engineering,
CRC Press/Springer-Verlag, Boca Raton, FL, 1999. With permission.)
For nonconservative substances
∂( ηcd ) ∂ ⎛ ∂ηcd ⎞ ∂ ⎛ ∂ηcd ⎞ ∂ ⎛ ∂ηcd ⎞

− ⎜E ⎟ − ⎜E ⎟ − ⎜E ⎟
∂t ∂x ⎝ xx ∂x ⎠ ∂y ⎝ yy ∂y ⎠ ∂z ⎝ zz ∂z ⎠
η ⎛ ∂cd ∂c ∂c ⎞
+ ⎜u +υ d +ω d⎟ (9.109)
1 + fcp K sc md η ⎝ ∂x ∂y ∂z ⎠
+ kηcd + (kcsp md = kcd fcp K sc md ) − G = 0
In Eqs. (9.108) and (9.109), kcspmd represents the decay of substance sorbed onto
solids. The other variables have been defined in Section 9.1.3 of this chapter.
The piezometric head h in groundwater can be calculated using Laplace
equations:
∂h ∂2h ∂2h ∂2h

sc + K xx 2 + K yy 2 + K zz 2 = 0 (9.110)
∂t ∂x ∂y ∂z
where sc is the storage coefficient. Then, u, ν, and ω can be computed using Darcy’s
law:
K xx ⎛ ∂h ⎞
u= ⎜ ⎟ (9.111)
η ⎝ ∂x ⎠
K yy ⎛ ∂h ⎞
υ= ⎜ ⎟ (9.112)
η ⎝ ∂y ⎠
K zz ⎛ ∂h ⎞
ω= ⎜ ⎟ (9.113)
η ⎝ ∂z ⎠
where Kxx, Kyy, and Kzz are the soil hydraulic conductivities in the x, y, and z directions,
respectively. Some useful equations that can be used to estimate Ksc are as follows
(Sincero and Sincero, 1996):
Log Ksc = 1.00 log K0,w − 0.21 (10 polyaromatic hydrocarbons)

Log Ksc = 0.544 log K0,w + 1.377 (45 organics, mostly pesticides)
Log Ksc = 1.029 log K0,w − 0.18 (13 pesticides)
If an organic compound is shaked in a mixture of n-octanol and water, the ratio of

its concentration in the octanol phase to that in the water phase is defined as K0,w .
For further discussion on nonreactive and reactive contaminant transport in saturated
and unsaturated zones, the reader is referred to Delleur (1999).
9.3.4 OPTIMIZATION MODEL FOR GROUNDWATER QUALITY

MANAGEMENT
Quantitative and qualitative groundwater operation is inherently a multiple-objective
problem with conflicting objectives. Groundwater system optimization models can
provide the optimal operational policy by considering various objectives, constraints,
and their uncertainties. To satisfy the given water demand, limitation of groundwater
table variation, optimal allocation of water to demands, minimization of possible
contamination of the aquifer, and minimizing water or wastewater treatment costs
can be the primary objectives of groundwater management models. In the next
section, a typical optimization model for groundwater quality management is pre-
sented.
This model can determine the optimal waste injection concentration and pump-
ing and injection rates considering groundwater quality standards and water
demands. The objective function of the model is a weighted sum of the individual
objectives:
T n
Maximize Z = ∑ ∑ w . f (Q , Q , h , C , C )
t =1 l =1
l l
t
p
t
r i
t t
p
t
r (9.114)
where:
fl (Qpt , Qrt , hit , C pt , Crt ) is the lth objective.

Qpt is the pumping discharge from pumping site p, in planning time period t
(m3).
Qrt is the injection rate at injection site r in planning time period t (m3).
wl is the relative weight of the lth objective.
hit is the groundwater head at the end of planning time period t and at control
point i (m).
Cpt and Crt is the pollutant concentration at pumping well/injection sites, dur-
ing time period t (mg/m3).
T is the number of time periods of the planning horizon.
n is the number of objectives.
The model constraints are as follows:

Water supply from groundwater
∑Q
p =1
t
p ≥ Dt ∀t (9.115)
Qpt ≤ QpMax ∀t, p (9.116)

where:
D t is the total water demand in time period t (m3).

P is the total number of pumping sites.
QpMax is the maximum pumping rate in each time period (m3).
Groundwater recharge
∑Q ≥ W
r =1
t
r
t
∀t (9.117)
Qrt ≤ QrMax ∀t, r (9.118)
where:
W t is the total waste load disposal in period t (m3).

R is the total number of injection sites.
QrMax is the maximum injection volume in each time period (m3).
Variations of water table
hit ≥ hiMin ∀t (9.119)
hit ≤ hiMax ∀t (9.120)
where hiMin and hiMax are the minimum and maximum head levels, respectively, at
control points i (m).
Groundwater quality affecting water supply quality and resulting in clogging
problems
C pt ≤ C pMax ∀t, p (9.121)
Crt ≤ CrMax ∀t, r (9.122)
where C pMax and CrMax are the maximum mass concentration in the pumped and
injected water (mg/m3).
The head and groundwater quality response equation can be derived from qual-
itative and quantitative groundwater simulation models. They can have a typical
form as follows:
(
hit = fi Qpt , Qrt , hit −1 ) ∀t, i, p, r (9.123)
(
Cit = gi Qpt , Qrt , Crt , Cit −1 ) ∀t, i, p, r (9.124)
When nonlinear Eqs. (9.114), (9.123), and (9.124) or groundwater quality and
quantity simulation models are used, the discrete dynamic programming method can
be used effectively to determine operational policies and the conflict resolution for
groundwater quality management. A simple example of the application of Nash’s
bargaining theory has been presented in Chapter 7.
9.4 PROBLEMS
9.1 If the BOD5 of wastewater discharge by an industry is 300 mg/L and the
ultimate BOD is 380 mg/L, what is the constant rate of decay? If the
temperature increases to 25°C, calculate the new constant rate and ultimate
BOD.
9.2 Evaluate the water quality of a river that has the following characteristics
and comment on supplying agricultural and domestic demands from this
river.
TDS = 364 mg / L Ca 2+ = 40 mg / L Mg+2 = 34 mg / L Na + = 25 mg / L

K + = 25 mg / L HCO3– = 300 mg / L SO −4 = 33 mg / L Cl – = 8 mg / L
NO3– = 3.3 mg / L PH = 7.3 EC, 25°C = 540 µs / cm
9.3 Select seven sampling stations to monitor a river basin that has the fol-
lowing stream network shown here.
FIGURE 9.3P
9.4 Determine the mixing distance downstream of an outfall in a straight river

reach that has an average width equal to 80 m and average depth equal
to 6 m. The mean river velocity and the average slope of the river bed are
1 m/sec and 0.0008, respectively.
9.5 In Example 9.5, select the sampling frequency of each station using the
weighting factors based on the historical means.
9.6 The statistical characteristics of total organic carbon (TOC) concentrations
in five sampling stations on a river are as follows:
Station Mean (mg/L) Standard Deviation
1 10.0 7.8
2 11.0 3.2
3 8.0 4.5
4 9.5 2.5
5 10.1 7.0
(a) Determine the annual sampling frequencies for the stations when
the width of the confidence interval of the mean at each station at
the 95% confidence level is 3 mg/L.
(b) Allocate 100 samples per year to each station for a 90% confidence
level.
9.7 A treatment plant plans to discharge 20,000 m3/day partially treated waste-
water into a river. Calculate the maximum BOD5 concentration that can
be discharged into the stream and the partially treated wastewater have
the following characteristics:
River flow is 5 m3/sec.
Stream temperature is 18°C.
Background BOD and DO are 2.5 and 8.4 mg/L, respectively.
Average river velocity after mixing is 0.30m/sec.
kc = 0.4 and k2 = 2.1 at 20°C, which is the effluent temperature.
Minimum allowable DO = 5 mg/L.
9.8 Repeat Example 9.8 considering the following items:
(a) Treatment efficiencies of Industry A and City B treatment plants
can be unequal, but they have the same treatment cost.
(b) Treatment efficiencies of Industry A and City B treatment plants
can be unequal, but the treatment cost in Industry A is 1.5 times the
treatment cost for City B’s treatment plant.
9.9 Consider a river with two point sources. Consider two reaches of lengths
20 km and 30 km and 26 checkpoints at 2-km intervals on this river (see
Figure 9.9). The flow rates in a dry season have the following character-
istics: The main stream flow at upstream of point source 1, qo (m3/sec),
is assumed to be log-normally distributed and the kth wastewater flow
rate, qk (m3/sec), is considered to be a function of qo, as follows:
q k = fq ( qo ) + ε k
where εk is normally distributed with zero mean. The temperature is

assumed to be constant and equal to 20°C. The O’Connor and Dobbins
(1958) formula and Manning equation are used to calculate the reaeration
constant in each reach. The cross-section of the river is considered to be
rectangular, and the river width is assumed to be large compared with

flow depth. The BOD concentration of the river before the first reach, l0,
is assumed to be normally distributed. The deoxygenation constant in
reach i, kc,i, and the BOD loading of the kth discharger, bk, are assumed
to be independent and normally distributed, but the DO deficit of the kth
discharger, dk, is considered to be constant. Other information is presented
in Tables 9.17 through 9.21. Calculate the optimal BOD removal rates,
x1 and x2, for sources 1 and 2, respectively. The maximum allowable DO
deficit at each control point is considered to be 3 mg/L, and the total
annual budget for treatment costs is 14 units. (This problem has been
adapted from Fujiwara et al., 1988.)
TABLE 9.17
Streamflow Characteristics
Standard
Variables Unit Mean Deviation Lower Limit Upper Limit
qo m3/sec 45.0 20.0 20.0 —

lo mg/L 2.0 0.4 1.2 —
do mg/L 0.0 — — —
ε1 m3/sec 0.0 2.0 –5.1 6.01
ε2 m3/sec 0.0 1.0 –2.5 2.50
b1 tonnes/day 2000.0 300.0 1000.0 2500.0
b2 tonnes/day 500.0 45.0 70.0 710.0
d1 mg/L 4.0 — — —
d2 mg/L 4.0 — — —
kc,1 1/day 0.24 0.033 0.15 0.34
kc,2 1/day 0.19 0.028 0.13 0.36
TABLE 9.18
Physical Parameters of the River
Parameters Unit Value
Width (reach 1) m 65
Width (reach 2) m 65
n1 −
1 2.5 × 10–2
m 3s
n2 −
1 2.5 × 10–2
m 3s
S1 — 2.0 × 10–4
S2 — 2.0 × 10–4
TABLE 9.19
Cumulative Treatment Cost Data
Monthly Cumulative BOD Removal Cost (Unit)
BOD Removal Range (%) Treatment Plant 1 Treatment Plant 2
35–50 4.17 5.13

51–65 4.52 5.42
66–75 5.23 5.98
76–80 6.32 6.76
81–85 8.23 8.17
86–90 8.76 12.57
91–95 19.92 20.94
96–98 29.13 27.73
TABLE 9.20
Monte Carlo Simulation Data
Variable Value
m 300
No (minimum number of iterations) 40
α 0.10
β 0.15
e1 0.9
e2 0.9
TABLE 9.21
Functional Relationship for Discharge Flows,
Reaeration Constant, and Flow Travel Time
Variable Value
q1 f1 (q0) = 0.26 q0 + ε1
q2 f2 (q0) = 0.12 q0 + ε2
k21 33.22(q0 + q1 ) –0.7
k22 34.87(q0 + q1 + q2) –0.7
t1j 9.62 × 10 –2 s1j (q0 + q1 ) –0.4
t2j 9.62 × 10 –2 s2j (q0 + q1 + q2) –0.4
Note: sij : Distance from the upstream boundary of the reach

i to the checkpoint j.
TABLE 9.A1
Summary of EPA’s National Primary Drinking Water Standards
(Maximum Contaminant Level Goals [MCLG], Maximum Contaminant
Levels [MCL])
MCLG MCL
Contaminant (mg/l) (mg/l)
Inorganics
Antimony 0.006 0.006
Asbestos (>10 µm) 7 MFLa 7 MFLa
Barium 2 2
Beryllium 0.004 0.004
Cadmium 0.005 0.005
Chromium (total) 0.1 0.1
Copper 1.3 TTb
Cyanide 0.2 0.2
Fluoride 4 4
Lead 0 TTb
Mercury (inorganic) 0.002 0.002
Nitrate 10 10
Nitrite 1 1
Selenium 0.05 0.05
Thallium 0.0005 0.002
Coliform and surface water treatment
Giardia lambia 0 detected TT b
Legionella 0 detected TT b
Standard plate count N/A TT b
Total coliform 0 detected —c
Turbidity N/A TT b
Viruses 0 detected TT b
Radionuclides
Radium 226 + radium 228 0 5 pCi/l
Gross alpha particle activity 0 15 pCi/l
Beta particle + photon radioactivity 0 4 mrem/yr
a MFL = million fibers per liter.
b TT = treatment techniques in lieu of a numerical standard.
c No more than 5% of the samples can be total coliform positive.
Source: Delleur, J. W., The Handbook of Groundwater Engineering, CRC Press/Springer-Verlag,

Boca Raton, FL, 1999. With permission.
TABLE 9.A2
A Summary of Quality Tolerance For Industrial Process Waters
Hardness as Manganese
Turbidity Color CaCO3 Iron as Fe as Mn Total Solids
Industry or Use (mg/L) (mg/L) (mg/L) (mg/L) (mg/L) (mg/L) Other Requirements
Confectionery — — — 0.2a 0.2 100 Potable

Food, general 10 — — 0.2a 0.2 — Potable
Ice 5 5 — 0.2a 0.2 — Potable, SiO2 < 10 mg/L
Laundering — — 50 0.2a 0.2 — No grit or corrosiveness
Plastic, clear, uncolored 2 2 — 0.02a 0.02 200 No slime formation
Paper and pulp ground wood 50 20 180 1.0a 0.5 — Al2O3 < 8 mg/L, SiO2 < 25 mg/L, Cu <
Kraft pulp 25 15 100 0.2a 0.1 300 5 mg/L, pH 7.8 to 8.3
Soda and sulfite 15 10 100 0.1a 0.05 200
High-grade light papers 5 5 50 0.1a 0.05 200
Rayon (Viscose) pulp production 5 5 8 0.05a 0.03 100
a Limit given applies to both iron alone and the sum of iron and manganese.
Source: Corbitt, R. A., Standard Handbook of Environmental Engineering, McGraw-Hill, New York, 1990. With permission.
433
TABLE 9.A3
Selected Guidelines for Assessing the Suitability of Water for Irrigation
Purposes
Degree of Restriction of Use
Slight to
Potential Problem Variablesa Unitsb None Moderate Severe
Salinity (affects ECw dS/m <0.7 0.7–3.0 >0.3

availability of crop water) TDS g/m3 <450 450–2000 >2000
Permeability (affects SAR and ECw SAR 0–3 0–3 0–3

infiltration rate of water dS/m >0.7 0.7–0.2 <0.2
into soil, evaluated using SAR and ECw SAR 3–6 3–6 3–6
SAR and ECw together) dS/m >1.2 1.2–0.3 <0.3
SAR and ECw SAR 6–12 6–12 6–12
dS/m >1.9 1.9–0.5 <0.5
SAR and ECw SAR 12–20 12–20 12–20
dS/m >2.9 2.9–1.3 <1.3
Specific ion toxicity Sodium SAR <3 3–9 >9

(affects sensitive crops) Chloride Meq/l <4 4–10 >10
Boron mg/L <0.7 0.7–2.0 >2.0
a ECw is the electrical conductivity of irrigation water, SAR is the sodium absorption ratio.
b dS/m = deciSiemens per meter.
Source: Tchobanoglous, G. and Schroeder, E. D., Water Quality: Characteristics, Modeling, and
Modification, Addison–Wesley, Reading, MA, 1985. With permission.
Primary, basic Water Quantity First Level

Variables carrier Variables Variables
-------------------------------------------------------
Quantity Discharge , Water Level,
Small Volume in a Body of Water
-------------------------------------------------------
Secondary Associated Quality Variables of Second Level

Quality Variables Aggregated Effects Variables
-------------------------------------------------------
Examples Temperature, PH, Turbidity, BOD, DO, Cations
Anions, Conductivity, Chlorides, Radioactivity
-------------------------------------------------------
Aggregate-Producing Quality Variables that Third Level

Quality Variables Produce Aggregated Effects Variables
-------------------------------------------------------
Example 1 Radioactivity-Producing Variables:
Strontium, Cesium, Tritium, etc.
-------------------------------------------------------
-------------------------------------------------------
Turbidity-Producing Variables:
Example 2 Suspended Matter, Colloids, Biota Groups,
Dissolved Minerals, etc.
-------------------------------------------------------
Specific Compounds
Or Species Producing Most Detailed Classification Fourth Level
Effects in Aggregation of Quality Variables Variables
-------------------------------------------------------
Minerals Affecting Turbidity:
Examples Iron Oxides, Mangenese
Compounds, Alumina, etc.
-------------------------------------------------------
FIGURE 9.A1 A hierarchical ranking of water quantity and water quality random variables
for the purpose of procuring information on water quality processes. (From Sanders, T. G.,
Ward, R. C., Loftis, J. C., Steele, T. D., Ardin, D. D., and Yevjevich, V., Design of Networks
for Monitoring Water Quality, Water Resources Publications, Littleton, CO, 1987. With
permission.)
TABLE 9.A4
Surface Water Quality Models: Basic Information
CEQUALRIV1
CEQUALICM
CEQUALW2
SALMONQ
SMPTOX3
QUAL2E
MIKE11
HEC5Q
WASP5
EXAMS
HSPF
Model
Water body type Stream, river Y Y Y Y Y Y Y Y Y Y Y

Lake, reservoir N N N Y Y N Y Y Y N N
Estuary Y N N Y N Y Y Y N N N
Dimensions One-dimensional, Y Y Y Y Y Y Y Y Y Y Y
branching
Two-dimensional, X/Y N N N Y Y N N Y N N N
Two-dimensional, X/Z N N N Y Y N N Y N N N
Three-dimensional, box N N N Y Y N N Y N N N
Time Steady Y Y N N Y N N N N N N
Quasi-dynamic Y N Y Y Y N N N N Y Y
Dynamic N N Y Y N Y Y Y Y Y Y
Hydrodynamics Input Y Y Y Y Y N N N N N Y
Simulated N N Y Y N Y Y Y Y Y Y
Control structure N N Y N N Y Y Y Y Y Y
Transport Advection Y Y Y Y Y Y Y Y Y Y Y
Dispersion Y Y N Y Y Y Y Y Y Y Y
Benthic exchange N Y Y Y Y N Y Y N Y Y
Loading Input, steady Y Y Y Y Y Y Y Y Y Y Y
Input, variable N N Y Y Y Y Y Y Y Y Y
Simulated N N Y Y N N N N N Y Y
Other Preprocessor Y Y N Y Y N N N N Y Y
Postprocessor Y Y Y Y Y N N N N Y Y
Source: Mays, L. W., Water Resources Handbook, McGraw-Hill, New York, 1996. With permission.
TABLE 9.A5
Surface Water Quality Models: Variables and Processes
CEQUALRIV1
CEQUALICM
CEQUALW2
SALMONQ
SMPTOX3
QUAL2E
MIKE11
HEC5Q
WASP5
EXAMS
HSPF
Processes Model
Chemical process First-order decay Y Y Y Y Y Y Y Y Y Y Y

Process kinetics N N Y Y Y N N N N N N
Daughter products N N Y Y Y N N N N N N
Sorption N Y Y Y Y N N N N Y N
Sediment process Input rates N Y Y Y N Y Y Y Y Y Y
Noncohesive N N Y N N N N N N Y N
processes
Cohesive processes N N Y N N N N N N Y N
Water quality Temperature Y N Y N N Y Y Y Y Y Y
processes Salinity Y N Y N N Y Y Y Y Y Y
Bacteria N N Y N N Y Y Y Y Y Y
DO, BOD Y N Y Y N Y Y N Y Y Y
DO, carbon balance N N N N N N Y Y N N N
Nitrogen cycle Y N Y Y N Y Y Y Y Y Y
Phosphorus cycle Y N Y Y N Y Y Y Y Y Y
Silicon cycle N N N N N N Y Y N Y Y
Phytoplankton Y N Y Y N Y Y Y Y Y Y
Zooplankton N N Y N N N N Y N Y N
Benthic algae N N Y N N N N N N Y Y
Simulate SOD N N N Y N N Y Y N N N
Source: Mays, L. W., Water Resources Handbook, McGraw-Hill, New York, 1996. With permission.
REFERENCES
Ambrose, R. B., Wool, T. A., Connolly, J. P., and Schanz, R.W., WASP4: A Hydrodynamic
and Water Quality Model, EPA/600/3-87/039, U.S. Environmental Protection Agency,
Athens, GA, 1987.
Brown, L. C. and Barnwell, T. O., The Enhanced Stream Water Quality Models QUAL2E
and QUAL2E-UNCAS Documentation and User Manual, 1987.
Burn, D. H., Water quality management through combined simulation optimization approach,
ASCE J. Environ. Eng., 115(5), 1011–1024, 1989.
Burn, D. H. and Lence, B. J., Comparison of optimization formulation for waste load allo-
cation, ASCE J. Environ. Eng., 118(4), 1992.
Burn, D. H. and McBean, E. A., Optimizing modeling of water quality in an uncertain
environment, Water Resources Res., 21(7), 934–940, 1985.
Cardwell, H. and Ellis, H., Stochastic dynamic programming models for water quality man-
agement, Water Resources Res., 29(4), 803–813, 1993.
Corbitt, R. A., Standard Handbook of Environmental Engineering, McGraw-Hill, New York,
1990.
Dandy, G. and Crawley, P., Optimum operation of multiple reservoir system inducing salinity
effects, Water Resources Res., 28(4), 979–990, 1992.
Davis, M. L. and Cornwell, D. A., Environmental Engineering, McGraw-Hill, New York, 1991.
Delleur, J. W., The Handbook of Groundwater Engineering, CRC Press/Springer-Verlag, Boca
Raton, FL, 1999.
Durrans, S. R., Low flow analysis with a conditional Weibull tail model, Water Resources
Res., 32(6), 1749–1759, 1996.
EPA, Protection of Public Water Supplies from Ground-Water Contamination, Center for
Environmental Research Information, U.S. Environmental Protection Agency, Wash-
ington, D.C., 1985.
Fontane, D. G., Labadie, J. W., and Loftis, B., Optimal control of reservoir discharge quality
through selective withdrawal, Water Resources Res., 17(6), 1594–1602, 1981.
Fujiwara, O., Puangmaha W., and Hanaki, K., River basin water quality management in
stochastic environment, ASCE J. Environ. Eng., , 114(4), 864–877, 1988.
Houlihan, M. F. and Lucia, P. C., Groundwater monitoring, The Handbook of Groundwater
Engineering, Edited by J. W. Delleur, CRC Press and Springer-Verlag, Chapter 24,
1999.
Hudak, P. F. and Loaiciga, H. A., An optimization method for monitoring network design in
multilayered groundwater flow systems, Water Resources Res., 29(8), 2835–2845,
1993.
Krenkel, P. A. and Novotny V., Water Quality Management, Academic Press, San Diego, CA,
1980.
Langbein, W. B. and Durum, W. H., The Aeration Capacity of Streams, U.S. Geological
Survey Circular No. 542, U.S. Department of Interior, Washington, D.C., 1967.
Loaiciga, H. A., Charbeneau, R. J., Everett, L. G., Fogg, G. E., Hobbs, B., and Rouhani, S.,
Review of ground water quality monitoring network design, ASCE J. Hydraulic Eng.,
118(1), 11–37, 1992.
Lohani, B. N. and Thanh, N. C., Stochastic programming model for water quality management
in a river, J. Water Pollution Control Fed., 50, 2175–2182, 1978.
Loucks, D. P., ReVelle, C. S., and Lynn, W. R., Linear programming models for water pollution
control, Manage. Sci., 14(4), B166–B181, 1967.
McKee, J. E., The need for water quality criteria, in Proc. of a Conf. on the Physiological
Aspects of Water Quality, Division of Water Supply and Pollution Control, U.S. Public
Health Service, Washington D.C., 1960.
Nandalal, K. D. W. and Bogardi, J. J., Reservoir management for water quality control, in
Proc. of the Ninth Congress of Asian and Pacific Division of the International
Association for Hydraulic Research, Singapore, 1994.
O’Connor, D. J. and Dobbins, W. E., Mechanisms of reaeration in natural streams, Trans.
Am. Soc. Civil Eng., 123, 655, 1958.
ReVelle, C. and McGarity, A. E., Design and Operation of Civil and Environmental Engi-
neering Systems, John Wiley & Sons, New York, 1997.
Sanders, T. G., Ward, R. C., Loftis, J. C., Steele, T. D., Ardin, D. D., and Yevjevich, V., Design
of Networks for Monitoring Water Quality, Water Resources Publications, Littleton,
CO, 1987.
Sharp, W. E., A topologically optimum water-sampling plan for rivers or streams, Water
Resources Res., 7(6), 1641–1646, 1971.
Sincero, A. P. and Sincero, G. A., Environmental Engineering: A Design Approach, Prentice-
Streeter, H. W., and Phelps, E. B., A study of the pollution and natural purification of Ohio
River, U.S. Public Health Service Bull., 146, 1925.
Takyi, A. K. and Lence, B. J. (1999), Surface water quality management using multiple
realization chance constraint method, Water Resources Res., 35(5), 1999.
Tchobanoglous, G. and Schroeder, E. D., Water Quality: Characteristics, Modeling, and
Modification, Addison-Wesley, Reading, MA, 1985.
U.S. Army Corps of Engineers, HEC5: Appendix on Water Quality Analysis, U.S. Army Corps
of Engineers, Washington, D.C., 1986.
Ward, R. C., Loftis, J. C., Nielsen, K. S., and Anderson, R. D., Statistical evaluation of
sampling frequencies in monitoring networks, J. Water Pollution Control Fed., 51(9),
2292–2300, 1979.
Willey, R. G., Smith, D. J., and Duke, J. H., (1996), Modeling Water Resources Systems for
Water Quality Management, ASCE J. Water Resources Planning Manage., 122(3),
171–179, 1996.
Willis, R. and Yeh, W. W.-G., Groundwater Systems Planning and Management, Prentice-
10 Hydroelectric Systems
Analysis
10.1 INTRODUCTION
At the United Nations Conference on Environment and Development (UNCED) held
in Rio de Janeiro, Brazil, in June of 1992, various countries presented their plans
for sustainable development, and most of these plans had energy sustainability as
the nucleus of other activities. Definitions of sustainable development depend upon
the subject matter, but the definition we will use here is “development by using
renewable natural resources in a manner that does not eliminate or degrade them or
otherwise diminish their ‘renewable’ usefulness for future generations…” (Kar-
amouz, 1993).
Among all of the energy production methods, electricity, the most environmen-
tally sound energy production method, has been given greater attention in recent
years. Development of hydropower electricity generation, which has high efficiency
(compared to thermal and hydrothermal units) and contributes negligible pollution
to the environment, is a major attribute to the sustainable development.
Electric energy production is estimated to account for about 10% of the world’s
energy production (World Resources Institute, 1998), and hydropower electricity
production contributes about 20% of that total (Mays, 1996). Table 10.1 shows the
variability of electric energy production throughout the world. Europe, with electric
energy production of 15.1%, has the highest ranking among all continents.
The published figures of electric energy production show that in developed
countries, hydropower electric energy production accounts for about 3% of total
energy production, whereas in oil importing and exporting developing countries,
hydropower electricity production has been about 4.2 and 0.5% of total energy
production, respectively. This chapter presents the concerns of water resources
engineers regarding the design and operation of hydroelectric power plants and the
framework of long-term and short-term operation optimization models.
10.2 BASIC DEFINITIONS

Some of the basic definitions in the area of hydroelectric systems analysis can be
Gross head (H) is the difference in elevation between the water surface in at
the intake and the tailwater surface.
Net head (Hn) is the head available for energy production after deducting
hydraulic losses.
441
TABLE 10.1
Total Energy and Electric Energy Production
Commercial Energy Primary Electricity Percent of Primary
Productiona Productionb Electricity
Region (peta-joulesc ) (peta-joules) (%)
Africa 22,667 324 1.4

Europe 94,496 14,302 15.1
North America 89,924 11,147 12.4
Central America 8992 554 6.2
South America 17,937 1784 9.9
Asia 122,438 6957 5.7
Oceania 8436 237 2.8
World 364,891 35,305 9.7
a Commercial energy production includes total energy production: solid, liquid, and gases fuels and
primary electricity production.
b Primary electricity refers to electricity generated by noncombustible energy sources and includes
nuclear, wind, tidal, wave, solar, geothermal, and hydroelectric power sources.
c 1 peta-joule = 10 joules.
Source: Adapted from World Resources Institute (1998).
Hydraulic efficiency (eh) is the ratio of net head to gross head:
Hn
eh = H (10.1)
Overall efficiency (et) is the hydraulic efficiency multiplied by the efficiencies

of turbines (etur) and generators (eg):
et = etur × eg × eh
Capacity is the maximum amount of power that a generator or power plant

can deliver at any given time (in kW = 1000 watts).
Installed capacity is the nominal capacity of a power plant and is equal to
aggregate rating of all generating equipment installed in the plant.
Generation is the energy output of a power plant.
Power is the rate of energy production, P (in kW), which can be estimated
as follows:
P( kW ) =
( ) ( )
e ⋅ γ N m 3 ⋅ Q m3 s ⋅ Hn (m )
(10.2)
1000
where γ is the specific weight of water (9810 N/m3) and Q is discharge

through turbine (in m3/sec) and e = etur × eg.
Hydroelectric Systems Analysis 443
Energy is capable of doing work. It is exchange of the potential energy of

water stored in a reservoir with the turbo generation. The energy production
of a hydropower plant can be estimated using the following equation:
E=
∫ P ⋅ t =∫ e ⋅ γ ⋅ Q(t) ⋅ H(t) ⋅ dt
t (10.3)
where E is the electrical output of the power plant (kWh) and et is the
overall efficiency of the power generation. As can be seen in this equation,
increasing head H(t) and discharge Q(t) through turbines increases the
energy production. If the head and discharge are measured in N discrete
time intervals, the energy production can be estimated as follows:
E= ∑ e ⋅ γ ⋅ Q(t) ⋅ H(t)
t =1
t (10.4)
where Q(t ) and H (t ) are the average head and discharge through turbine
in discrete time interval t.
Load is the demand for electricity which can be expressed in terms of energy
demand (average power demand) or capacity demand (peak power
demand). It can also be defined as the sequence of instantaneous power
levels over a period that a power system must meet.
Load factor is the ratio of average power demand to peak power demand for
a specific period that can be computed on a daily, weekly, monthly, or
annual basis.
Firm or primary power is the electric energy that should be made available
to a customer to meet any or all of the agreed-upon portion of that
customer’s load requirements. For hydropower plants, it usually can be
estimated based on the energy output of the plant in the most critical period
(low flow) of the historical records.
Secondary power is all power available in excess of firm power.
Plant factor is the ratio of the average load on a plant for a particular time
period to its installed capacity.
Base load is the minimum load of a power system in a specific period of time.
Peak load is the maximum load in a specific period; the peaking portion of
the load is usually defined as the load that occurs at the peak demand hours,
which is about 8 hours a day.
Intermediate load is that portion of the load between the base load and the
peaking portion of the load.
10.3 COMPONENTS OF HYDROPOWER PLANTS

Hydroelectric power plants can be classified in a number of different ways (Linsley
et al., 1992):
• Interms of capacity
• Microhydro
• Macrohydro
• Interms of head
• Low head — usually operated under heads of less than 20 m
• Medium head — usually operated under heads between 20 and 30 m
• High head — usually operated under heads of more than 30 m
• Interms of layout and operative mode
• Run-of-river —very limited storage capacity; head varies with discharge
• Storage — reservoir sufficient to provide seasonal regulation for car-
ryover storage from the wet seasons to the dry seasons to increase
energy generation in the dry season
• Pumped storage — designed to convert low-value, off-peak energy to
high-value, on-peak energy; water is pumped at off-peak hours from
the tailwater pool to the headwater pool for energy generation in peak
hours
Basic elements of a hydropower plant are as follows (see Figure 10.1) (Mays, 1996):
• The reservoir creates the necessary head that will provide the energy
required for driving turbines.
• The intake structure directs water from the reservoir into the penstock or
conduit. Gates or valves are used to control water discharge to the power
plant. Racks or screens are also used to prevent trash or debris from
entering the turbines.
• The conduit conveys water from the intake structure to the powerhouse.
• The power plant includes the turbines, generators, control, and auxiliary
equipment.
Energy production is a major application of water in many river–reservoir sys-

tems. In hydropower plants, the energy in falling water runs electric turbines and
generates electricity without actually consuming water. As shown in Figure 10.1,
Energy loss in
Energy Line penstock
Gen Net head Gross head

H
Ve2
2g
Turbine
Ve
Draft Tube
FIGURE 10.1 Element of a typical hydropower plant.

the dam provides the required head. The efficiency of power generation in the power
plant shown in Figure 10.1 depends upon:
• Plant efficiency
• Volumetric water flow through the turbine
• Net hydraulic head of water across the turbine
• Type of turbine
10.4 DATA REQUIREMENTS

The following sections cover the data required to estimate the power potential of a
new hydroplant or to develop operating policies for an existing hydropower plant.
10.4.1 HYDROLOGIC DATA

Historical streamflow records at the diversion point or inflows to a reservoir are the
basic information required for hydropower plant development or operation studies.
The proper time interval for the hydrologic data depends on factors such as the type
of plant and the hydrologic characteristics of streamflows. Monthly average stream-
flows are often used for preliminary or even advanced studies if the streamflow does
not change significantly from day to day. The degree of at-site and upstream regu-
lation is also important in the selection of time intervals. Water losses due to
evaporation from the reservoir and leakage through or around the dam and other
appurtenant facilities should also be considered. In addition to water losses from the
reservoir, water that is diverted for consumption or water that is reserved for oper-
ation of navigational locks or fish passage facilities should also be taken into account
when estimating losses of the hydropower system.

Data needed about physical characteristics of
the plant facilities depends on the type of
Tailwater elevation
project. For storage projects, the eleva-

tion–area–volume curve and the minimum and
maximum allowable reservoir elevations should
be determined, whereas for run-of-river
projects, elevation–discharge curves are usually
helpful. Elevation–area–volume curves repre-
sent the head and volume of water stored in the Discharge
reservoir that could be used for power genera-
tion. Tailwater data are also important when FIGURE 10.2 Typical tailwater rat-
estimating the electricity generation of a hydro- ing curve.
power plant. A typical tailwater rating curve is
shown in Figure 10.2 which can be used for estimating tailwater elevation for
different rates of plant discharge. Tailwater elevation depends on downstream chan-
nel geometry, project discharge, and downstream backwater effects. In hydropower
plants with considerable head, the variation of tailwater is negligible.
Relevant data about the physical characteristics of power plants can be summa-
rized as follows:
• Installed capacity, which represents the upper limit of power that can be
generated by the plant
• Plant factor, which represents the peak and firm energy production of the
plant
• Maximum and minimum discharge
• Head range that the plant operation is satisfactory
• Turbine characteristics, including shape of efficiency curve and minimum
discharge
Head losses due to friction in the intake structures and penstock should also be taken
into account in order to determine the net head of the system. The relation between
net head and head losses can be written as follows:
Ve2
Hn = H − H f − 2g (10.5)
where Hf represents the friction losses in the trash-rack, intake structure, and pen-
stock; Ve2 /2g is the velocity at the draft tube exit.
10.4.3 POWER LOAD AND RELIABILITY OF POWER SYSTEMS

Estimation of future power loads of a specific system is a primary step for hydro-
power development planning or operation management studies. Load predictions
are usually based on the rate at which the energy consumption has increased over
the past few years. Three methods have been used for estimating future power loads
(American Society of Civil Engineers, 1989):
• Trend analysis, based on extending historical trends and modifying the

projections to reflect expected changes
• End-use analysis, based on the expected use of electricity by different
users
• Econometric analysis, based on the relationships between electricity
demand and various factors that influence demand
In systems expansion planning, a load–resources analysis should be utilized to

determine the need and timing of the output of a proposed hydropower project. Need
is a measure of deficits that may occur in the future due to increased power loads
compared with the current supply capacity. Timing refers to the periods when the
need for additional generation occurs.
In addition to annual and monthly variation of power loads, weekly and daily
load configuration are also necessary in order to determine the type of the load that
a hydropower project could carry and to estimate the benefits of the system expansion
FIGURE 10.3 Typical daily and weekly load curves for an electric power system. (From
Lindsley et al., 1992.)
or operation management. Figure 10.3 shows typical daily and weekly load curves.
As can be seen in this figure, the demand for electricity varies from a minimum in
the early hours of the morning to peak loads in the late morning or early evening.
Two samples of weekly summer and winter loads for the southeastern United States
are shown in Figure 10.4, where the load shape in summer has a closer correlation
with air temperature which makes it easier to predict load fluctuations.
Power system operators usually divide the load into three segments (Mays,
1996):
20
Load
18
System load (GW)
Air temperature (°C)

16
35
14
30
12
25
10 Air temperature 20
Sun Mon Tue Wed Thu Fri Sat
(a) Example of Summer load
22
20 Load
System load (GW)
18
Air temperature (°C)

16
10
14
Air temperature 5
12
0
10 -5
Sun Mon Tue Wed Thu Fri Sat
(b) Example of Winter load
FIGURE 10.4 Sample of weekly loads for a utility located in the southeastern United States.
(From Gulliver, J. S. and Arndt, R. E. A., Hydropower Engineering Handbook, McGraw-Hill,
New York, 1991. With permission.)
• Base load, which is continuous, 24 hours a day

• Peak load, which is the highest portion of the load, occurring only for a
few hours a day
• Intermediate load, which is the portion of the load between base and peak
Besides supplying the power loads, the system should have enough capacity to
supply the expected peak load plus additional capacity to take care of breakdowns
and necessary maintenance shutdowns which is usually defined in terms of reliability.
Reliability in a hydropower system refers to the adequacy and security of a system.
Adequacy relates to the existence of sufficient energy within the system to satisfy
the power loads or system operational constraints. Security refers to the ability of
the system to respond to disturbances within the system (Rangarajan et al., 1999).
10.4.4 NON-POWER OPERATING CRITERIA

As mentioned in Chapter 6, a reservoir usually supplies water for various instream
and offstream uses. Different criteria can be used in order to determine the system
efficiency in satisfying other objectives such as flood control, supplying the water
demands of users, and providing the required instream flow. For more details about
objectives and criteria, see Chapter 8.
10.5 METHODS FOR ESTIMATING POWER POTENTIAL

Two methods have been widely used for primary estimation of power potential:
• Nonsequential or flow–duration curve

• Sequential streamflow routing (SSR)
The flow–duration curve method is the better method for all preliminary or screening
studies. This method is also the best choice for high-head, run-of-river projects where
head is generally fixed or even for low-head projects where head varies with
discharge. For multipurpose storage projects, the SSR method is more appropriate
and also can be used for examining the feasibility of including power at new water
conservation or flood control projects. For peaking and pumped storage projects,
hourly SSR routings are required (American Society of Civil Engineers, 1989). These
methods are explained in the following sections.
10.5.1 THE FLOW–DURATION METHOD

In this method, a flow–duration curve is developed based on the observed data.
Figure 10.5 shows an example of a flow–duration curve. Streamflows are related to
500
without storage effect

400
Discharge (cms)
300
with storage effect
200
100
0
0 10 20 30 40 50 60 70 80 90 100
Percent of time equaled or exceeded
FIGURE 10.5 Typical flow–duration curve with and without storage effect.
the percentage exceedance values, which show the percent of time that different
levels of streamflows are equaled or exceeded. To develop the flow–duration curves,
all of the daily flow data should be ranked according to discharge, not the sequence
in which they occurred. The use of daily data for development of flow–duration
curves is recommended (Gulliver and Arndt, 1991). The flow–duration curve can be
converted to a power–duration curve through application of the following power
equation:
ei ⋅ γ ⋅ Qi ⋅ Hi
Pi ( kW ) = (10.6)
1000
where:
Pi is the power production (in kW) when the turbine discharge is at exceed-
ance percentage i.
ei is the overall plant efficiency with turbine discharge equal to Qi and net
head equal to Hi.
Qi is the turbine discharge at percentage exceedance i (Q10, Q20, …); turbine
discharge is assumed to be equal to river discharge except when river
discharge exceeds turbine capacity or other constraints on turbine discharge
are encountered.
Hi is the net head available with river flow at exceedance percentage i.
Figure 10.6 shows the flow–duration and power–duration curves developed for St.
Cloud Dam (Gulliver and Arndt, 1991). The area under the power–duration curve
is the average annual energy production.
40
River discharge in 1000 ft3/s = 0.0283 m3/s
30
Generator output MW
20 flow duration curve

10
power duration
curve
10
5
7.1
0.2
0
20 40 60 80 100
% of time river discharge is exceeded
FIGURE 10.6 Flow–duration curve at St. Cloud Dam. (From Gulliver, J. S. and Arndt, R. E. A.,
Hydropower Engineering Handbook, McGraw-Hill, New York, 1991. With permission.)
TABLE 10.2
Monthly Streamflow Data for Examples 10.1 and 10.2
Streamflow Data (m3/sec)
Month 1999 2000 2001
January 105 440 102

February 108 275 860
March 645 337 640
April 1000 515 690
May 1308 1968 1330
June 419 1965 930
July 89 360 155
August 28 71 92
September 32 50 50
October 40 95 88
November 80 100 305
December 210 88 300
Example 10.1
Table 10.2 provides the monthly streamflow data for a gauging station for the years
1999 to 2001. Plot the flow–duration curve for this river.
Solution: To develop the flow–duration curve, the observed streamflows should
be arranged in descending magnitude, as shown in Table 10.3. The streamflows have
been equal to or exceeded 28 m3/sec. As can be seen in Table 10.3, the data are
ranked from 1 to 36. The probability of exceedence is then estimated by the following
relation:
i
Pei =
N +1
where i is the rank of the data and N is the total number of data (36 in this example).
The flow duration curve in Figure 10.7 shows the values in column 2 of Table 10.3
vs. the values in column 3 of this table.
Example 10.2
A run-of-river power plant is proposed at the site for which the monthly flow data
are presented in Table 10.2. The head available at the site is about 10 m and plant
efficiency is about 70%. Assume that the turbine discharge capacity is 500 m3/sec.
Plot the power–duration curve and find the firm energy that is expected with 90%
probability of exceedance.
Solution: Using Eq. (10.6), the power–duration curve can be estimated. For
example, the power output of the proposed plant for a flow of 419 m3/sec, as in June
1999, can be determined as follows:
TABLE 10.3
Monthly Streamflow Data in Descending Order of Magnitude
for Example 10.1
Rank Sorted Streamflows Probability of Exceedance
1 1968 0.03
2 1965 0.05
3 1330 0.08
4 1308 0.11
5 1000 0.14
6 930 0.16
7 860 0.19
8 690 0.22
9 645 0.24
10 640 0.27
11 515 0.30
12 440 0.32
13 419 0.35
14 360 0.38
15 337 0.41
16 305 0.43
17 300 0.46
18 275 0.49
19 210 0.51
20 155 0.54
21 108 0.57
22 105 0.59
23 102 0.62
24 100 0.65
25 95 0.68
26 92 0.70
27 89 0.73
28 88 0.76
29 88 0.78
30 80 0.81
31 71 0.84
32 50 0.86
33 50 0.89
34 40 0.92
35 32 0.95
36 28 0.97
2500
2000
Discharge (cms)
1500
1000
500
0
0 0.25 0.5 0.75 1
Percent of time equaled or exceeded
FIGURE 10.7 Flow–duration curve for Example 10.2.
P = e ⋅ γ ⋅ Q ⋅ H = 0.7 × 9.81 × 419 × 10 = 28, 773 kW
The power generation in each of the months is estimated using the above relation
and shown in Table 10.4. It should be noted that in the months when the discharge
is higher than the maximum turbine discharge (500 m3/sec), the maximum power
generation can be estimated as:
Pmax = e ⋅ γ ⋅ Q ⋅ H = 0.7 × 9.81 × 500 × 10 = 34, 335 kW
Figure 10.8 shows the resulting power–duration curve. The minimum power gener-
ation has been 1922.8 kW. The firm energy with 90% reliability is then estimated
as 2967 kW using the numbers in Table 10.4.
Example 10.3
Assume that we are sizing the turbines in the proposed power plant for Example
10.2 to run full 30% of the time. Find the turbine design discharge.
Solution: The turbine design discharge can be found by taking the intercept of
the 30% exceedance ordinate on the flow–duration curve and moving horizontally
across to the river discharge abscissa at 515 m3/sec. This is the river discharge that
is met or exceeded 30% of the time and will be the turbine design discharge.
10.5.2 SEQUENTIAL STREAMFLOW ROUTING (SSR) METHOD

Sequential streamflow routing, which was primarily developed for evaluating storage
projects, is based on the continuity equation:
ΔS = I − O − L (10.7)
TABLE 10.4
Monthly Power Generation (kW) Based on the Streamflow Data in
Table 10.3 (Example 10.1)
Power
Sorted Probablity of Generation
Rank Streamflows Exceedance (kW)
1 1968 0.03 34335

2 1965 0.05 34335
3 1330 0.08 34335
4 1308 0.11 34335
5 1000 0.14 34335
6 930 0.16 34335
7 860 0.19 34335
8 690 0.22 34335
9 645 0.24 34335
10 640 0.27 34335
11 515 0.30 34335
12 440 0.32 30215
13 419 0.35 28773
14 360 0.38 24721
15 337 0.41 23142
16 305 0.43 20944
17 300 0.46 20601
18 275 0.49 18884
19 210 0.51 14421
20 155 0.54 10644
21 108 0.57 7416
22 105 0.59 7210
23 102 0.62 7004
24 100 0.65 6867
25 95 0.68 6524
26 92 0.70 6318
27 89 0.73 6112
28 88 0.76 6043
29 88 0.78 6043
30 80 0.81 5494
31 71 0.84 4876
32 50 0.86 3434
33 50 0.89 3434
34 40 0.92 2747
35 32 0.95 2197
36 28 0.97 1923
40000.0
Power generation (kW)

35000.0
30000.0
25000.0
20000.0
15000.0
10000.0
5000.0
0.0
0 50 100
% of time power generation is exceeded
FIGURE 10.8 Powerñ

dur ation curve for Example 10.2
50
STORAGE
40
ENERGY (GIGAWATT-HOURS)
RELEASES
STORAGE
30
20
10
FIRM
ENERGY
0
OUTPUT
1937 1938 1939 1940 1941 1942
WATER YEAR
FIGURE 10.9 Effect of storage on firm energy production of a hydropower plant. (From
American Society of Civil Engineers, Civil Engineering Guidelines for Planning and Design-
ing Hydroelectric Developments, Vol. 1, Planning, Design of Dams, and Related Topics, and
Environmental, American Society of Civil Engineers, New York, 1989. With permission.)
where I and O are inflow to and outflow from the reservoir; L represents the losses
due to evaporation, seepage, etc.; and ∆S is the change in reservoir storage. This
equation is applied sequentially to all historic records in order to obtain a continuous
record of project operation. Hourly, daily, weekly, or monthly data can be used
depending on the nature of the study and the type of data available (American Society
of Civil Engineers, 1989). Energy output can then be estimated by applying the
reservoir outflow values to the power equation, Eq. (10.2).
The SSR method can be used for maximizing firm energy production. In run-
of-river plants, the firm energy that can be produced is equal to the energy production
in the most critical low-flow periods in the historical records. However, in storage
projects water can be stored in the high-flow periods in order to increase plant
discharge in low-flow periods. Figure 10.9 shows the effect of storage on firm energy
production of a typical storage power plant.
Example 10.4
Assume a reservoir with 8 billion m3 of storage is constructed at the site discussed
in Example 10.2. The firm yield of the reservoir is considered to be 400 m3/sec,
which is also equal to the discharge capacity of the turbine. Assume that the total
hydraulic losses are negligible. The storage-water elevation relation at the reservoir
site is given as follows:
h = 4 × ( S + 7)
where h is the water elevation in the reservoir (in meters) and S is the reservoir
storage (in billion cubic meters). Use the following relation for estimating the
tailwater elevation:
( )
hT = R + R s × 0.0009
where R and Rs are the regulated and spilled releases (in m3/sec), respectively, from
the reservoir, and hT is the tailwater head (in meters). Consider that the reservoir is
full at the beginning of the first month. Use the SSR method to estimate the expected
monthly power generation in the reservoir.
Solution: Table 10.5 shows the results of the SSR method. For example, examine
the data for the first month. The reservoir storage was assumed to be at full capacity,
or 8 billion m3. When taking into consideration the 400-m3/sec constant release and
105-m3/sec inflow to the reservoir, the reservoir storage at the end of this month can
be estimated as:
⎛ hrs ⎞
31⎛
days ⎞ ⎛ mins ⎞ × 60⎛ secs ⎞
× 24⎜ ⎟ × 60⎝
⎝ mon ⎠ ⎝ day ⎠ hr ⎠ ⎝ min ⎠
Storage at the end of Jan. = 8 + (105 − 400) ×
10 9
= 7.21 Billion cubic meters
The elevation at the beginning and the end of January can be found as:
Elevation at the beginning of Jan. = 4 × (8 + 7) = 60 m

Elevation at the end of Jan. = 4 × (7.21+ 7) = 56.84 m
Hydroelectric Systems Analysis
TABLE 10.5
Results of the SSR Method in Example 10.4
Tail-
Begin. Ending Ending water Net
Storage Elev. Inflow Release (controlled) Release (Spilled) Stor. Elev. Elev. Head Energy
Month (109 m3) (m) (cms) (MCM) (cms) (MCM) (cms) (MCM) (109 m3) (m) (m) (m) (MWh)
Jan. 8.00 60.00 105 281.23 400 1071.36 7.21 56.84 0.36 58.06 114,824.52
Feb. 7.21 56.84 108 270.60 400 1002.24 6.48 53.91 0.36 55.02 108,805.35
Mar. 6.48 53.91 645 1727.57 400 1071.36 7.13 56.54 0.36 54.87 108,507.01
Apr. 7.13 56.54 1000 2505.60 400 1002.24 637.80 0.25 8.00 60.00 0.36 57.91 114,525.73
May 8.00 60.00 1308 3503.35 400 1071.36 2431.99 908.00 8.00 60.00 1.18 58.82 116,333.62
Jun. 8.00 60.00 419 1086.05 400 1036.80 49.25 19.00 8.00 60.00 0.38 59.62 117,915.97
Jul. 8.00 60.00 89 238.38 400 1071.36 7.17 56.67 0.36 57.97 114,655.02
Aug. 7.17 56.67 28 75.00 400 1071.36 6.17 52.68 0.36 54.32 107,419.23
Sep. 6.17 52.68 32 82.94 400 1036.80 5.22 48.87 0.36 50.41 99,705.34
Oct. 5.22 48.87 40 107.14 400 1071.36 4.25 45.01 0.36 46.58 92,118.59
Nov. 4.25 45.01 80 207.36 400 1036.80 3.42 41.69 0.36 42.99 85,023.94
Dec. 3.42 41.69 210 562.46 400 1071.36 2.91 39.66 0.36 40.31 79,730.30
Jan. 2.91 39.66 440 1178.50 400 1071.36 3.02 40.09 0.36 39.51 78,141.18
Feb. 3.02 40.09 275 689.04 400 1002.24 2.71 38.83 0.36 39.10 77,326.12
Mar. 2.71 38.83 337 902.62 400 1071.36 2.54 38.16 0.36 38.14 75,419.86
Apr. 2.54 38.16 515 1290.38 400 1002.24 2.83 39.31 0.36 38.37 75,892.15
May 2.83 39.31 1968 5271.09 400 1071.36 7.03 56.11 0.36 47.35 93,643.46
(continued)
457
458
TABLE 10.5 (CONTINUED)
Tail-
Begin. Ending Ending water Net
Storage Elev. Inflow Release (controlled) Release (Spilled) Stor. Elev. Elev. Head Energy
Month (109 m3) (m) (cms) (MCM) (cms) (MCM) (cms) (MCM) (109 m3) (m) (m) (m) (MWh)
Jun. 7.03 56.11 1965 5093.28 400 1036.80 3083.79 1189.73 8.00 60.00 1.43 56.62 111,984.78
Jul. 8.00 60.00 360 964.22 400 1071.36 7.89 59.57 0.36 59.43 117,526.02
Aug. 7.89 59.57 71 190.17 400 1071.36 7.01 56.05 0.36 57.45 113,616.79
Sep. 7.01 56.05 50 129.60 400 1036.80 6.10 52.42 0.36 53.87 106,543.00
Oct. 6.10 52.42 95 254.45 400 1071.36 5.29 49.15 0.36 50.42 99,723.46
Nov. 5.29 49.15 100 259.20 400 1036.80 4.51 46.04 0.36 47.24 93,416.54
Dec. 4.51 46.04 88 235.70 400 1071.36 3.67 42.70 0.36 44.01 87,035.46
Jan. 3.67 42.70 102 273.20 400 1071.36 2.88 39.50 0.36 40.74 80,573.04
Feb. 2.88 39.50 860 2154.82 400 1002.24 4.03 44.11 0.36 41.45 81,974.89

Mar. 4.03 44.11 640 1714.18 400 1071.36 4.67 46.69 0.36 45.04 89,076.36
Apr. 4.67 46.69 690 1728.86 400 1002.24 5.40 49.59 0.36 47.78 94,493.04
May 5.40 49.59 1330 3562.27 400 1071.36 7.89 59.56 0.36 54.21 107,219.65
Jun. 7.89 59.56 930 2410.56 400 1036.80 1262.82 487.20 8.00 60.00 0.80 58.98 116,643.93
Jul. 8.00 60.00 155 415.15 400 1071.36 7.34 57.38 0.36 58.33 115,354.23
Aug. 7.34 57.38 92 246.41 400 1071.36 6.52 54.08 0.36 55.37 109,495.68
Sep. 6.52 54.08 50 129.60 400 1036.80 5.61 50.45 0.36 51.90 102,644.36
Oct. 5.61 50.45 88 235.70 400 1071.36 4.78 47.10 0.36 48.42 95,750.66
Nov. 4.78 47.10 305 790.56 400 1036.80 4.53 46.12 0.36 46.25 91,471.32
Dec. 4.53 46.12 300 803.52 400 1071.36 4.26 45.05 0.36 45.22 89,437.93
TABLE 10.6
Rule Curve Policy for Reservoir Operation for Example 10.5
Month
St a 5 6 7 8 8 7 6 5 6 6 6 6
a St is the reservoir storage at the beginning of month t (109 m3).
The average water elevation in the reservoir is (60 + 56.84)/2 = 58.42 m. The tailwater
elevation in this month can be found as:
hT = ( 400 + 0) × 0.0009 = 0.36 m
Therefore, the net head in this month would be 58.42 – 0.36 = 58.06 m. The power
generation in this month can then be estimated as:
e⋅ γ ⋅Q⋅ H 0.7 × 9.81 × 400 × 58.06

E= ×t = × 720 = 114,825 MWh
1000 1000
Example 10.5
In Example 10.4, assume that the reservoir should be operated based on the rule
curve shown in Table 10.6. Assume also that the maximum turbine discharge is 700
m3/sec. Determine the monthly power generation of the reservoir.
Solution: Table 10.7 shows the results of the SSR method. For example, examine
the first month. The reservoir storage was 5 billion m3 at the beginning of the month.
Based on the rule curve policy, the reservoir storage at the end of January should
be 6 billion m3. The inflow volume in this month is 281 million m3, and the storage
at the end of this month is estimated to be 5000 + 281 = 5281 million m3. Therefore,
the release is set to 0 for this month. The estimated monthly power generation values
for the rest of the operation period are shown in Table 10.7.
10.6 HYDROPOWER RESERVOIR OPERATION

Firm energy generation in most runoff river hydropower plants is very low due to
seasonal variations of streamflows (low discharge in dry seasons) and low head.
Storage power plants provide more energy with a relatively high head and less
variable discharge. In general, through storage and pondage in hydropower projects
the following objectives can be attained:
• Increasing firm energy

• Maximizing average annual energy
• Maximizing dependable capacity
460
TABLE 10.7
Tail-
Begin. Total Release Ending Ending water Net
Storage Elev. Inflow Release (Controlled) Release (Spilled) Stor. Elev. Elev. Head Energy
Month (109 m3) (m) (cms) (MCM) (MCM) (cms) (MCM) (cms) (MCM) (109 m3) (m) (m) (m) (MWh)
Jan. 5.00 48.00 105 281.23 0.00 0.00 0.00 0.00 0.00 5.28 49.12 0.00 49.12 0
Feb. 5.28 49.12 108 270.60 0.00 0.00 0.00 0.00 0.00 5.55 50.21 0.00 50.21 0
Mar. 5.55 50.21 645 1727.57 0.00 0.00 0.00 0.00 0.00 7.28 57.12 0.00 57.12 0
Apr. 7.28 57.12 1000 2505.60 1785.00 688.66 1785.00 0.00 0.00 8.00 60.00 0.62 59.38 202,184

May 8.00 60.00 1308 3503.35 4503.35 700.00 1874.88 981.36 2628.47 7.00 56.00 1.51 54.49 188,577
Jun. 7.00 56.00 419 1086.05 2086.05 700.00 1814.40 104.80 271.65 6.00 52.00 0.72 51.28 177,463
Jul. 6.00 52.00 89 238.38 1238.38 462.51 1238.38 0.00 0.00 5.00 48.00 0.42 47.58 108,813
Aug. 5.00 48.00 28 75.00 0.00 0.00 0.00 0.00 0.00 5.07 48.30 0.00 48.30 0
Sep. 5.07 48.30 32 82.94 0.00 0.00 0.00 0.00 0.00 5.16 48.63 0.00 48.63 0
Oct. 5.16 48.63 40 107.14 0.00 0.00 0.00 0.00 0.00 5.27 49.06 0.00 49.06 0
Nov. 5.27 49.06 80 207.36 0.00 0.00 0.00 0.00 0.00 5.47 49.89 0.00 49.89 0
Dec. 5.47 49.89 210 562.46 1034.90 0.00 0.00 386.39 1034.90 5.00 48.00 0.35 47.65 0
Jan. 5.00 48.00 440 1178.50 178.50 0.00 0.00 66.64 178.50 6.00 52.00 0.06 51.94 0
Feb. 6.00 52.00 275 689.04 0.00 0.00 0.00 0.00 0.00 6.69 54.76 0.00 54.76 0
Mar. 6.69 54.76 337 902.62 0.00 0.00 0.00 0.00 0.00 7.59 58.37 0.00 58.37 0
Apr. 7.59 58.37 515 1290.38 882.04 0.00 0.00 340.30 882.04 8.00 60.00 0.31 59.69 0
Hydroelectric Systems Analysis
May 8.00 60.00 1968 5271.09 6271.09 700.00 1874.88 1641.36 4396.21 7.00 56.00 2.11 53.89 186,521
Jun. 7.00 56.00 1965 5093.28 6093.28 700.00 1814.40 1650.80 4278.88 6.00 52.00 2.12 49.88 172,648
Jul. 6.00 52.00 360 964.22 1964.22 700.00 1874.88 33.36 89.34 5.00 48.00 0.66 47.34 163,842
Aug. 5.00 48.00 71 190.17 0.00 0.00 0.00 0.00 0.00 5.19 48.76 0.00 48.76 0
Sep. 5.19 48.76 50 129.60 0.00 0.00 0.00 0.00 0.00 5.32 49.28 0.00 49.28 0
Oct. 5.32 49.28 95 254.45 0.00 0.00 0.00 0.00 0.00 5.57 50.30 0.00 50.30 0
Nov. 5.57 50.30 100 259.20 0.00 0.00 0.00 0.00 0.00 5.83 51.33 0.00 51.33 0
Dec. 5.83 51.33 88 235.70 1069.11 0.00 0.00 399.16 1069.11 5.00 48.00 0.36 47.64 0
Jan. 5.00 48.00 102 273.20 0.00 0.00 0.00 0.00 0.00 5.27 49.09 0.00 49.09 0
Feb. 5.27 49.09 860 2154.82 428.01 0.00 0.00 176.94 428.01 7.00 56.00 0.16 55.84 0
Mar. 7.00 56.00 640 1714.18 714.18 0.00 0.00 266.64 714.18 8.00 60.00 0.24 59.76 0
Apr. 8.00 60.00 690 1728.86 1728.86 0.00 0.00 667.00 1728.86 8.00 60.00 0.60 59.40 0
May 8.00 60.00 1330 3562.27 4562.27 700.00 1874.88 1003.36 2687.39 7.00 56.00 1.53 54.47 188,508
Jun. 7.00 56.00 930 2410.56 3410.56 700.00 1814.40 615.80 1596.16 6.00 52.00 1.18 50.82 175,872
Jul. 6.00 52.00 155 415.15 1415.15 0.00 0.00 528.36 1415.15 5.00 48.00 0.48 47.52 0
Aug. 5.00 48.00 92 246.41 0.00 0.00 0.00 0.00 0.00 5.25 48.99 0.00 48.99 0
Sep. 5.25 48.99 50 129.60 0.00 0.00 0.00 0.00 0.00 5.38 49.50 0.00 49.50 0
Oct. 5.38 49.50 88 235.70 0.00 0.00 0.00 0.00 0.00 5.61 50.45 0.00 50.45 0
Nov. 5.61 50.45 305 790.56 402.27 0.00 0.00 155.20 402.27 6.00 52.00 0.14 51.86 0
Dec. 6.00 52.00 300 803.52 1803.52 0.00 0.00 673.36 1803.52 5.00 48.00 0.61 47.39 0
461
Optimal Monthly Optimal Long-term

Policy Optimization (Strategy)
Optimal Weekly
Optimal Mid-term Optimization
Policy
Optimal Hourly Optimal Short-term

Policy Optimization
Real-time
Operation
Operator
FIGURE 10.10 Flowchart comparing long-, mid-, and short-term operation optimization
models for a hydropower reservoir.
Firm energy can be increased in storage power plants because the streamflows stored
in high-flow season can be discharged in dry seasons at a rate higher than the normal
flow rates in those seasons (see Figure 10.9). To optimize hydropower reservoir
operations, maximizing the firm and/or average annual energy can be considered as
the objective function. Hydropower storage projects are usually multipurpose (see
Chapter 8 for discussion on various objectives of reservoir operation). Considering
the variety of these objectives and the temporal variation of power loads, a time
decomposition approach is necessary for hydropower reservoirs (see Figure 10.10).
In the flowchart shown in Figure 10.10, the higher level policies impose constraints
on the lower time sequence models (Yeh et al., 1992). As seen in this figure,
optimization of the hydropower reservoirs operation consists of four steps:
1. Long-term planning (strategic): optimization of reservoir operation on a

monthly scale within the planning time horizon
2. Mid-term planning (strategic and tactical): optimization of reservoir oper-
ation on a weekly scale within the 1-year time horizon
3. Short-term planning (tactical): optimization of reservoir operation on an
hourly scale within the 1-week time horizon
4. Real-time operation: optimization of load dispatching and unit commit-
ment scheduling on an hourly scale
Hydropower units often operate as a part of a larger system. Even though perfor-
mance of a reservoir in supplying water demands might affect only local users
downstream of the reservoir, the power generation of a hydropower plant might have
more far-reaching effects on the power network of the region. In the following
sections, the bases of the long-term and short-term planning and real-time operation
models for multipurpose hydropower storage projects are discussed.
10.6.1 LONG-TERM PLANNING MODELS

The main objective of long-term planning of hydropower storage projects is to
maximize the net benefits of systems operation. Reservoirs usually should satisfy
such objectives as:
• Supply of different water demands

• Flood control
• Supply of power loads
With respect to supplying instream flows, a conflict arises between the water supply
objectives and maintaining enough head in the reservoirs for efficient power gener-
ation. A general formulation of a long-term planning model can be summarized as
follows:
Minimize Z= ∑ (loss ( R , D ) + loss (P , D ))

t =1
1
t t t
W 2
t t t
L
(t = 1,…, n) (10.8)
Subject to: St +1 = St + It − Rt − Lt (t = 1,…, n) (10.9)
(
Pt = f Rtp , Et ) (t = 1,…, n) (10.10)
Et = g( St ) (t = 1,…, n) (10.11)
Rt = Rtp + Rts (t = 1,…, n) (10.12)
Smin ≤ St ≤ Cap (t = 1,…, n) (10.13)
0 ≤ Rt ≤ Rmax,t (t = 1,…, n) (10.14)
St , Et , Lt , Rt , Pt , Rtp , Rts ≥ 0 (t = 1,…, n) (10.15)
where:
It is the reservoir inflow in month t.

Rt is the total reservoir release in month t.
St is the reservoir storage at the beginning of month t.
DtW is the water demands in month t.
DtL is the power load in month t.
Et is the average net head level in month t, which is a function of reservoir
storage and release.
Lt is the reservoir losses in month t due to evaporation and seepage.
Cap is the total reservoir storage.

Pt is the power generation in month t.
Rtp is the reservoir release used for energy production in month t.
Rts is the reservoir release from non-power outlets in month t.
losst1 and losst2 are the operation losses associated with performance of the
reservoir in supplying water demands and generation of power, respectively.
n is the number of months in the planning horizon.
As seen in Eq. (10.8), the objective function of the monthly optimization model
consists of two parts. The first part represents operation losses based on the water
supply objectives. The second part of the objective function represents losses asso-
ciated with power generation and is usually estimated based on a comparison
between the generated power and power load of the system or a target power
generation defined by system operators and decision makers. Transmission losses
should also be taken into account. A suitable way of considering transmission losses
in preliminary studies is to incorporate them into a plant efficiency factor (Yeh et
al., 1992).
Example 10.6
For the first year of the data given in Example 10.4, consider that a reservoir supplies
the domestic demands of a city located downstream of the reservoir. It also supplies
part of the energy demands of the city. Formulate an optimization model to determine
the optimal releases for only the first year of data. Use the following loss functions
to estimate the losses associated with shortages in supplying water demands and
energy loads:
(
⎧⎪1.56 × DW − R )
3
if DtW > Rt
loss = ⎨
1
t
t t
⎩⎪0 otherwise
losst2 = ⎨
(
⎧⎪7.35 × DtL − Pt ) if DtL > Pt
⎩⎪0 otherwise
where losst1 and losst2 are losses associated with water supply and energy generation
objectives, respectively. DtW and DtL are water and energy demands (expressed in
m3/sec and MWh, respectively). Rt and Pt are water release and power generation,
respectively, in month t.
Solution: The objective function is to minimize the losses:
12
Minimize ∑ [loss + loss ]

t =1
1
t
2
t
Subject to: S2 = S1 + 281.23 − R1 ⎞

S3 = S2 + 270.6 − R2 ⎟
⎟
S4 = S3 + 1727.57 − R3 ⎟
Constraints for monthly
S5 = S4 + 2505.60 − R4 ⎟
⎟ reservoir water balance
S6 = S5 + 3503.35 − R5 ⎟
S7 = S6 + 1086.05 − R6 ⎟
⎟
S8 = S7 + 238.38 − R7 ⎟⎠
S9 = S8 + 75 − R8
S10 = S9 + 82.94 − R9
S11 = S10 + 107.14 − R10
S12 = S11 + 207.36 − R11
S13 = S12 + 562.44 − R12
S1 = 8000
Si ≤ 8000 ∀i
Si , Ri ≥ 0
10.6.2 MID-TERM PLANNING MODELS

A mid-term planning model has a structure similar to the long-term model. The only
difference between these models is in constraints and forecasts accuracy. Because
the planning horizon consists of a smaller number of periods, the model is capable
of further decompositions and closer simulation. It can be on a weekly or daily scale
within a monthly or weekly time horizon, respectively. The long-term operation
policy that defines the monthly release from reservoir and power generation should
be considered as a constraint for a mid-term model.
10.6.3 SHORT-TERM PLANNING MODELS

The purpose of the short-term optimization model is to determine the optimal
operating policies on an hourly time scale within a weekly or daily time horizon.
The hourly model is considerably more detailed than the long-term and mid-term
planning models. Different types of generating units in each power plant and their
respective input–output performance relationships should be considered. The eco-
nomic objective functions could be taken into account in the short-term optimization
models by incorporating the value of peak and offpeak generation. Figure 10.11
shows a schematic diagram of the value of power generation within a month. In the
first two classes (k = 1, 2), covering about 220 hours each month, the power generated
would have a high value (peak generation). The third class (k = 3) would have a
FIGURE 10.11 Schematic classification of benefits of power generation on an hourly time

scale for a month. (Karamouz, M., Zahraie, B., and Araghi-Nejhad, Sh., ASCE J. Computing
Civil Eng., 2003 [submitted].)
medium value, and the last two classes (k = 4, 5) would have a very low value (firm
generation). Therefore, the main objective of the short-term optimization model
could be defined to maximize the peak power generation of the plant considering
the optimal policies obtained from the mid-term optimization model as a constraint.
The objective function of this model can be defined as:
5 720
Maximize ∑∑ P
k =1 t =1
k ,t × Vk (10.16)
where Vk is the value of power and Pk,t is the power generation of the power plant
in class k.
10.6.4 REAL-TIME OPERATION MODELS

The real-time operation models for hydropower units optimize the unit commitment
problem, which is a complex decision-making process involving the integrated
hourly scheduling of generators in a multiple-objective hydropower system. The
purpose is to satisfy power loads, water demands, reliability constraints, operational
restrictions, and security requirements on an hourly scale and within each day.
Constraints of the unit commitment problem can be formulated as follows:
• Unit generation (Pt,i ) should be less than the unit capacity:
Pt ,i ≤ Pi cap (t = 1,…, 24) (10.17)
where Pt,i is the power generated by unit i in time interval t and Pi cap is
the power generation capacity of unit i.
• Total daily releases of the units of a hydropower plant should be equal to

releases determined by the short-term optimization model:
24 NU
R =
t
os
∑∑ R
t =1 i =1
t ,i (10.18)
where Rtos is the optimal daily release obtained from the short-term opti-
mization model, NU is the number of units in the hydropower plant, and
Rt,i is the hourly release from unit i in time t.
• System security requirements should be satisfied:
NU
∑P
i =1
t ,i ≤ Pt max (t = 1,…, 24) (10.19)
where Pt max is the maximum allowable rate of power generation in time t.

• Release from each unit should be less than the maximum discharge capac-
ity of the generators:
Rt ,i ≤ Rimax (t = 1, …, 24) (10.20)
where Rimax is the maximum discharge capacity of unit i.

The objective function of the unit commitment problem in a system of
multiple power plants can be defined in two different ways:
• Maximize the operation efficiency of the system, as follows (Yi et al.,

1997):
24 NU
∑∑ P
t =1 i =1
t ,i
Etotal = 24 NU (10.21)
∑∑ R E
t =1 i =1
t ,i t ,i
where Et,i is the net head on the unit i in time t.

• Minimize the total cost of operation of the system considering costs of
operation of each of the units.
10.7 HYDROTHERMAL COORDINATION

Optimal coordination of hydropower systems offer significant benefits, including
reductions in fossil fuel consumption and the investment required for a possible
expansion of the power plants (Yeh et al., 1992; Yeh and Becker, 1983). The
hydrothermal coordination problem can be classified into two categories (Wood and
Wollenberg, 1996):
• When the hydroelectric component is by far the largest of the hydrother-

mal system, scheduling of these systems can be optimized by minimizing
the costs of the thermal plants.
• Systems with a closer balance between the hydro and thermal plants and
those for which the hydroelectric system is a small fraction of the total
capacity can be optimally scheduled by minimizing the thermal generation
production costs and considering all of the hydraulic constraints that may
exist.
The hydrothermal optimization problem is of considerable difficulty because of the

following factors:
• High dimensionality of the problem with regard to:

• Seasonal state variables for reservoir operation, including storage and
historical and forecasted inflows
• Hourly state variables for short-term optimization of power plant oper-
ation
• Nonlinearity of the problems, including:
• Power transmission losses
• Generator input–output relationships
• Cost functions of hydro and thermal units
• Stochastic nature of the problem variables, including:
• Reservoir inflows
• Power loads
• Large number of constraints for considering:
• Supply and demand balance
• Flow balance or continuity equation
• Operational limits on hydro and thermal generation, including thermal
plant loading and unloading limitations
• Bounds on water release
• Equipment maintenance schedule
• Spinning reserve requirements
Because of these factors, a time decomposition approach is necessary. The

weekly and daily cyclic nature of power loads has led researchers to specify oper-
ational periods in terms of weeks, days, and hours. The following provides details
regarding short-term optimization of the hydrothermal power generation problem.
The primary objective of hydrothermal scheduling is to determine the amount
of hydro and thermal generation required to meet the power loads such that the costs
of operation are minimized. Operating costs of hydroplants are usually invariant and
negligible compared to the costs of thermal power plants; therefore, minimization
of thermal production costs, which are primarily a function of the fuel costs required
to run the thermal generators, can be considered as the objective function of this
problem.
For this purpose, thermal plant cost functions, also called lambda (λ) curves,
can be used. Figure 10.12 shows the typical lambda curve for the southeastern United
States (Georgakakos et al., 1997). This curve shows that at higher system loads, the
production cost of an additional unit of energy is higher. Therefore, hydropower
plants should be scheduled to replace the more expensive thermal generation with
hydrogeneration. Figure 10.13 shows the typical weekly system generation require-
ment and hydrothermal coordination for supplying power loads. For example, for
an hourly interval k, the lambda curve can be approximated by:
λ = aL + b (10.22)
where a and b are constant coefficients and L is the power load. As shown in Figure
10.14, the cost savings can be estimated as follows:
CSk =
[aL
k ] [ ]
+ b + a Lk − Pk + b
Pk =
−a 2
[ ]
P + aLk + b Pk (10.23)
2 2 k
where Pk is the displaced power generation by hydropower units and CSk is the cost
savings for hourly time interval k. This approximation is based on assuming the λ
curve is linear within specific intervals.
Specific constraints for optimization of hydrothermal systems should be consid-
ered. Among the different constraints for operation of thermal units, constraints for
the maximum allowable change in power generation of these systems within short
time intervals (e.g., hourly intervals) are the most common:
Tk − Tk −1 ≤ ΔT
(10.24)
Tk −1 − Tk ≤ δT
Tk ≤ Tmax (10.25)
where ΔT and δT are the maximum allowable increase and decrease, respectively,
in the thermal unit output at hourly time intervals; Tk is the power generation of a
thermal unit at hour k; and Tmax is the maximum capacity of the thermal unit. If
λk(Tk) represents the costs of a thermal unit operating at power level Tk in time
interval k, the objective function of the hydrothermal coordination problem in a
scheduled horizon including M intervals can be written as:
Minimize C = ∑ n λ (T )
k =1
k k k (10.26)
FIGURE 10.12 Typical lambda curve. (From Georgakakos, A. P. et al., ASCE J. Water
Resources Planning Manage., 123(1), 30–38, 1997. With permission.)
16
System capacity required
14 System generation required

capacity and generation requirement (1000 MW)
12 Hydro generation
10 Residual
thermal
generation
required
8
6
Hydrocapacity
4
Hydro load duration curve
FIGURE 10.13 Typical weekly
system generation requirement 2
and hydrothermal coordination.
(From Gulliver, J. S. and Arndt,
R. E. A., Hydropower Engineering 0
0 40 80 120 160
Handbook, McGraw-Hill, New
York, 1991. With permission.) Hours
Thermal plant cost function (λ)
λ =aL+b
System load (L)

Lk
Pk
FIGURE 10.14 Approximation of thermal cost savings. (From Georgakakos, A. P. et al.,

ASCE J. Water Resources Planning Manage., 123(1), 30–38, 1997. With permission.)
where nk is the length of interval k in the schedule horizon. The total power load in
the kth interval is:
Lk = Pk + Tk − Pkl (10.27)
where Pkl is the electric loss between the hydroplant and the load and can be estimated
as a function of hydropower generation:
Pkl = h( Pk ) (10.28)
By combining Eqs. (10.27) and (10.28), we can write:
h( Pk ) − Pk + ( Lk − Tk ) = 0 (10.29)
Assuming that the hydrogeneration is a function of water discharge rate, the water
discharge rate in interval k can be estimated as a function of (Lk – Tk):
QK = ∑ n g( L
k =1
k k − Tk ) (10.30)
where QK is the total water discharge in schedule horizon M (Wong and Wong,
1994). It should be mentioned that in the above formulation, the transmission losses
are estimated only for hydropower generation. In more sophisticated formulations,
the transmission losses are considered for both hydro and thermal units (see Yeh et
al., 1992 for more details).
Example 10.7
Suppose you want to optimize the energy-scheduling problem in a system for which
the hydroelectric component is the primary source. Use the following notation to
formulate the hydrothermal coordination problem:
Lm is the load of the system in period m (m = 1, …, M).

Tm is the power generation by thermal units in period m.
Pm is the power generation by hydro units in period m.
Ttot is the total power generated by thermal units.
λ(Tm) is the cost of the thermal units, which is a function of power generation.
Ctot is the total cost in the planning horizon.
M is the total number of time periods in the planning horizon (m = 1, …, M).
im is the number of hours in each period.
MT is the total number of periods for which the thermal plant is running.
Solution: It is desired to supply the entire amount of load from the hydroplant
in such a way that the cost of the thermal plant is minimized. The objective function
is to minimize the total cost within the planning horizon, which can be written as:
MT
Minimize Ctot = ∑ λ (T ) ⋅ i
m =1
m m
MT M M
Subject to: ∑
m =1
Tmim − ∑
m =1
Lmim + ∑P i
m =1
m m =0 (10.31)
This constraint shows that the total power generation in the thermal and hydro plants
should supply the total load of the system. It should be noted that all physical
constraints related to power generation capacity of hydro and thermal plants should
also be incorporated and are not presented here.
Example 10.8
In Example 10.7, the cost of power generation in thermal plants can be estimated
using the following relation:
λ(Tm ) = aTm2 + bTm + c (10.32)
Use the Lagrange function to find the optimal rate of power generation in the thermal
units, assuming that the optimal rate of power generation in the thermal plant (T*)
is constant for the entire period it is operating (Wood and Wollenberg, 1996).
Solution: The Lagrange function can be written as:
M
⎛ MT M M
⎞
L= ∑
m =1
λ(Tm ) ⋅ im + α⎜ −
⎝ m=1
Tm im + ∑ ∑ m =1
Lm im − ∑ P i ⎟⎠
m =1
m m
Then,
∂L dλ(Tm )
= −α = 0 for m = 1, …, M
∂Tm dTm
or,
dλ(Tm )
=α for m = 1, …, M
dTm
This means that the thermal plant should be run at a constant incremental cost for
the entire period it is on. Considering the constant rate of power generation in thermal
plants, it can be written as:
MT MT
Ctot = ∑m =1
( )
λ T * ⋅ im = λ T * ( )∑ i
m =1
m ( )
= λ T * ⋅ MT
Considering the cost function for the thermal plant (Eq. (10.32)), it can be written
that:
(
Ctot = aT ∗2 + bT ∗ + c ⋅ MT )
Using Eq. (10.31), it can be written:
M M
T ∗ MT − ∑m =1
Lm im + ∑P i
m =1
m m =0
Therefore,
M M
∑ Lm im − ∑P i m m
Ctot = aT( ∗2
+ bT + c ⋅∗
) m =1
T∗
m =1
To minimize the total cost, it can be written that:
c ⎞ ⎛ ⎞
M M
∑ ∑ P i ⎟⎠ = 0
dCtot ⎛
= a − ⋅ ⎜ L i −
dT * ⎝ T ∗2 ⎠ ⎝ m=1 m m m =1
m m
Therefore,
T∗ = c / a
The optimal rate of power generation of the thermal plant is estimated to be c/a .
10.8 CONFLICT ISSUES IN THE OPERATION OF

HYDROPOWER SYSTEMS
Hydropower plants are usually an important element of multipurpose water resources
systems. The major conflict issues in operation of hydropower systems arise when
a reservoir that supplies water for different activities must also provide enough
discharge and head for energy supply. Monthly variations of power loads (energy
demand) and water demands usually do not follow the same pattern; therefore,
supplying one of these demands in a specific period of time might conflict with
supplying the other demands in the future. As mentioned in Chapter 1, various
organizations are responsible for water and energy supply, and the requirements and
constraints for these sectors and agencies should be incorporated when formulating
the conflict resolution problem. The major conflict issues in operation of hydropower
reservoirs can be summarized as follows:
• Keeping enough head in reservoirs for power generation with high effi-
ciency is in conflict with
• Supplying water demands, especially in dry seasons
• Keeping flood control storage
• Improving the water quality in the reservoir
• Water and energy demands do not follow the same pattern; therefore,
optimal scheduling of water release from the reservoir for these purposes
might vary.
The first step for formulating the conflict resolution problem should be recognizing
the main conflict issues for all the relevant organizations. For example, consider a
river–reservoir system that supplies the following demands:

• Part of energy demand of a city located downstream of the reservoir
1.2
1
Utility 0.8
0.6
0.4
0.2
0
Pa Pb Pc Pd
Power generation (MWh)
FIGURE 10.15 Utility function for a range of power generation.
Management of water supply and consumption in each of these sectors is usually

done by a specific agency. For example, even though a specific department (for our
purposes, the Department of Water Supply) is responsible for supplying water for a
variety of demands, the Department of Agriculture is responsible for managing
irrigation demands and maintaining an acceptable level of production efficiency. The
Department of Industries is focused on supplying enough water for industrial pro-
duction, and supplying energy demands is the main focus for the Department of
Energy, whose actions could have local, regional, or national impacts. Importing
energy from electricity distribution networks of other states or countries and export-
ing energy to these networks are other important issues for the operation of power
systems. Finally, coordination of hydrothermal units is a critical component of
conflict resolution in these systems.
In the second step, the priorities and favorable ranges of water and energy supply
for each of these agencies should be recognized. Each of the above-mentioned
departments has their own set of priorities for allocating water to different demands.
For example, the Department of Energy could define the utility function for a range
of energy supply, as shown in Figure 10.15. In this figure, the most favorable annual
volume of water allocation for this organization ranges from Pb to Pc MWh. This
range can be estimated based on the variability of energy demands and coordination
with other sources of power supply such as thermal units.
The increasing segment in the left-hand side of this figure shows the range of
least favorable to most favorable rate of power generation. The points on this line
are not totally rejected by this organization because the shortage can be supplied
from other sources, such as more expensive thermal generation. The decreasing limb
on the right-hand side of this graph also shows that even though the power is
generated in the hydropower plant these conditions are also less favorable because
of probable problems in coordination with other sources of electricity supply and
limitation in the power transmission network.
All other agencies selected in the first step should also provide a favorable range
of water supply for their different demands. Besides water demands and the priorities
of agencies in supplying those demands, each agency has a specific level of authority
in regard to changing the water allocation schemes and imposing a favorable range
of water supply within the existing political and institutional climate of a region.
For example, the Department of Agriculture has an interest in supplying a relatively
higher volume of water to the agricultural lands than to the industrial or municipal
sectors. Analysts can also define a set of relative weights based on the authority of
the various agencies to make decisions regarding water allocation.
The final step is to formulate the conflict resolution problem. For this purpose,
different methods can be used, as briefly explained in Chapter 2. The steps that
should be taken for conflict resolution in hydropower systems modeling can be
• The conflict issues should be recognized based on the objectives of the

hydropower system operation.
• The agencies engaged in water and energy resources allocation and con-
sumption should be recognized.
• These agencies should provide their own sets of favorable levels of water
supply for their demands.
• Analysts should define the relative authority of agencies in imposing the
water and energy allocated to each demand.
Example 10.9
Consider a reservoir that supplies water to the following demand points:
• A city located downstream of the reservoir

• Industries located downstream of the reservoir near the city
• Agricultural networks downstream of the reservoir near the city
A run-of-river power plant is also located on the river downstream of the water
diversion for the above demand points. The head available at the site of this power
plant is 10 m and plant efficiency is about 60%. The following agencies are affected
by decisions made to release water from the reservoir:

• Agency 4: Department of Energy
The main objective is to meet the water and energy demands of these agencies.
The Department of Water Supply has a twofold role — namely, to allocate water to
different purposes as well as supply water for domestic purposes (DD). The irrigation
demand (RD), industrial demand (ID), and energy demand (ED) should also be
accounted for. Decision makers in these agencies are asked to set their most favorable
ranges of water allocation and energy production by assigning either a 1 (most
favorable) or a 0 (least favorable); see Figures 10.16 to 10.19. Analysts set the
following weights on the relative importance of water demands within the political
climate of that region:
Agricultural: 0.19
Domestic: 0.38
Industrial: 0.19
Energy: 0.24
1.2
1
0.8
Utility
0.6
0.4
0.2
0
120 130 140 150 160 170 180 190 200
Volume (million cubic meters)
FIGURE 10.16 Utility function for water allocation for the domestic demand defined by
Agency 1.
0.8
0.6
Utility
0.4
0.2
0
600
700
800
900
1000
1100
1200
1300
1400
1500
1600
1700
Volume (million cubic meters)
FIGURE 10.17 Utility function for water allocation for the agricultural demand defined by
Agency 2.
0.8
0.6
Utility
0.4
0.2
0
50 70 90 110 130 150 170
Volume (million cubic metersV)
FIGURE 10.18 Utility function for water allocation for the industrial demand defined by
Agency 3.
FIGURE 10.19 Utility function for the power generation defined by Agency 4.
Average annual inflow to the reservoir is estimated to be 1400 million m3. Find the
most appropriate water allocation scheme for a normal year and a year with the
projected inflow of 1000 million m3.
Solution: The discharge required for generating the range of power defined by
Agency 4 (see Figure 10.19) can be estimated as follows:
⎧0 Q = 980, 000
(0.6 × 9.81 × 10 × 24 × 365) = 1.9 m sec = 60 million m / month
3 3
⎪
⎪
⎪1 Q = 1, 140, 000
3 3
⎪
Utility = ⎨
⎪1 Q = 1, 635, 000
3 3
⎪
⎪
Q = 2, 125, 000
3 3
⎪0
⎩
The nonsymmetric Nash solution as explained in Chapter 2 is the unique optimal

i =1
i i
wi
(10.33)
Subject to: di ≤ fi ≤ fi∗ ∀i (10.34)
where wi is the relative weight of objectives, fi is the utility function, di is the

disagreement point, and fi* is the ideal point of player (agency) i. In order to check
the effect of the relative authority of these agencies (MATA), the symmetric Nash
problem is also solved, and the optimal results are compared with the nonsymmetric
results. Table 10.8 shows the results of the symmetric and nonsymmetric Nash
problems. For this particular system, the annual water demand has been estimated
TABLE 10.8
Results of Symmetric and Nonsymmetric Nash Solutions for Example 10.9
Nonsymmetric Solution Symmetric Solution
(million m3) (million m3)
Inflow = Inflow = Inflow = Inflow =
1000 1400 1000 1400
Domestic 150 170 150 162

Agricultural 700 1030 700 1034
Industrial 80 130 80 130
Turbine Discharge (m3/sec) 2.21 2.21 2.21 2.32
Power Generation (MWh) 1140 1140 1140 1196
by an independent party to be at minimum about 110, 960, and 150 million m3 and
1830 kWh for domestic, agricultural, industrial, and energy demands, respectively.
In dry years (inflow = 1000 million m3), domestic and agricultural demands are not
supplied according to either solution; however, based on the symmetric solution, the
energy demands have been supplied. In normal years, all demands have been supplied
in both scenarios.
10.9 PROBLEMS
10.1 Monthly streamflow data for a gauging station are presented in the fol-
lowing table. Plot the flow–duration curve for this river.
1999 2000 2001

Month (m3/sec) (m3/sec) (m3/sec)
January 10 44 10
February 12 28 86
March 64 37 64
April 105 55 60
May 138 198 130
June 45 195 93
July 10 36 16
August 3 7 9
September 3 5 5
October 4 10 9
November 8 10 35
December 20 9 30
10.2 In Problem 10.1, a run-off river power plant is proposed at the site. The
head available at the site is 50 m and plant efficiency is about 75%.
Determine the expected firm energy at this plant.
10.3 In Problem 10.2, find the turbine design discharge if the turbines should
run full 40% of the time.
10.4 Assume that a reservoir with storage of 800 million m3 storage is built at
the site explained in Problem 10.2. Firm release from the reservoir is
considered to be 35 m3/sec, which is also the turbine discharge capacity.
The storage–water elevation relation at the reservoir site is given in the
following table:
Elevation Storage
(m) (million m3)
10 100
20 350
30 550
40 700
50 800
Use following relation for estimating the tailwater elevation:
Tailwater elevation (m) = Release (m3/sec) × 0.009
Assume that the reservoir storage at the beginning of the first month is
750 million m3. Use the SSR method to estimated the expected monthly
power generation in the reservoir.
10.5 In Problem 10.4, assume the reservoir should be operated based on the
following rule curve. Find the monthly power generation of the reservoir
and the firm energy.
Eta 25 30 40 50 50 40 30 40 40 40 40 40
a Et is the reservoir elevation at the beginning of month t (m).
10.6 For the system explained in Problem 10.4, consider that reservoir supplies
irrigation demands of agricultural lands downstream of the reservoir as
shown in the following table:
IDta 0 0 5 20 30 60 50 30 10 0 0 0
a IDt is the irrigation demand in month t (106 m3).
It also supplies part of the energy demands of the city, estimated to be

100,000 MWh per month. The extra power generated in this plant can be
exported to other states. Formulate an optimization model for finding the
optimal releases. Use the following cost functions for estimating the losses
associated with shortages in supplying water demands and energy loads:
(
⎧⎪1.56 × DW − R )
3
if DtW > Rt
loss = ⎨
1
t
t t
⎪⎩0 otherwise
losst2 = ⎨
(
⎧⎪7.35 × DtL − Pt ) if DtL > Pt
⎪⎩0 otherwise
where losst1 and losst2 are losses associated with water supply and power
generation objectives, respectively. DtW and DtL are water and energy
demands (in m3/sec and MWh, respectively). Rt and Pt are water release
and power generation, respectively, in month t.
10.7 In Problem 10.6, use the water and energy prices to formulate the opti-
mization problem. Discuss the proper form of the objective function.
References
American Society of Civil Engineers, Civil Engineering Guidelines for Planning and Design-
ing Hydroelectric Developments, Vol. 1, Planning, Design of Dams, and Related
Topics, and Environmental, American Society of Civil Engineers, New York, 1989.
Creager, W. P. and Justin, J. D., Hydropower Handbook, John Wiley & Sons, New York, 1950.
Georgakakos, A. P., Yao, H., and Yu, Y., Control Model for Hydroelectric Energy-Value
Optimization, ASCE J. Water Resources Planning Manage., 123(1), 30–38, 1997.
Gulliver, J. S. and Arndt, R. E. A., Hydropower Engineering Handbook, McGraw-Hill, New
York, 1991.
Karamouz, M., Energy and resource sustainability in developing countries, in Proc. of the
First Annual Conf. on Technological Advancement in Developing Countries, Colum-
bia University, New York, 1993.
Karamouz, M., Zahraie, B., Araghi-Nejad, Sh., A decision support system for monthly oper-
ating of hydropower reservoirs: A case study, ASCE J. Computing Civil Eng., 2003
(Submitted).
Linsley, R. K., Franzini, J. B., Freyberg, D. L., and Tchobanoglous, G., Water Resources
Engineering, McGraw-Hill, New York, 1992.
Rangarajan, S., Surminski, H., and Simonovic, S. P., Reliability-based loss functions applied
to Manitoba hydro energy generation system, ASCE J. Water Resources Planning
Manage., 125(1), 34–40, 1999.
Wong, K. P. and Wong, Y. W., Short-term hydrothermal scheduling. Part 1: Simulated anneal-
ing approach, IEE Proc. Gener. Transm. Distrib., 141(5), 497–501, 1994.
Wood, A. J. and Wollenberg, B. F., Power Generation, Operation, and Control, John-Wiley
& Sons, New York, 1996.
World Resources Institute, World Resources: A Guide to the Global Environment: Toward
Sustainable Development, a joint publication by the World Resources Institute, The
United Nations Environment Programme, The United Nations Development Pro-
gramme, and The World Bank, Oxford University Press, London, 1992.
World Resources Institute, World Resources 1998–99, a joint publication by the World
Resources Institute, The United Nations Environment Programme, The United
Nations Development Programme, and The World Bank, Oxford University Press,
London, 1998.
Yeh, W. W.-G. and Becker, L., Economic optimization of hydrothermal power systems oper-
ation, in UERG Energy Science and Technology Report, UER-114, University of
California, Berkeley, CA, 1983.
Yeh, W. W.-G., Becker, L., Hua, S.-Q., Wen, D.-P., and Liu, J.-M., Optimization of real-time
hydrothermal system operation, ASCE J. Water Resources Planning Manage., 118(6),
636–653, 1992.
Yi, J., Labadie, W., and Stitt, S., Optimal unit commitment scheduling in hydropower systems,
in Proc. of ASCE International Conference of WaterPower ’97, 1997.
11 Water Demand Analysis
and Management
11.1 INTRODUCTION
Water required for various purposes may come from groundwater or surface
resources such as lakes, reservoirs, and rivers. It is called untreated or raw water
and is usually transferred to a treatment plant. The degree of treatment depends on
the raw water quality and the purpose for which that water will be used. Different
water quality standards for domestic, industrial, and agricultural purposes are pre-
sented in Chapter 9. After treatment, the water usually enters a water distribution
network. Details of the facilities required for water distribution are presented later
in this chapter.
Providing an adequate water supply and sanitation to rapidly growing urban
populations and developing agricultural and industrial activities is becoming increas-
ingly difficult for different states throughout the world. The predominant approach
to supplying these increasing water demands has been supply augmentation schemes.
But, in many developed or developing countries, the costs of developing new sources
or expanding existing sources are growing and becoming more physically and
economically infeasible. The real costs of water per cubic meter in second- and
third-generation projects in some cities have doubled between the first and the second
project and then doubled again between the second and third (Bhatia and Falken-
mark, 1993).
In recent years, more consideration has been given to saving water rather than
developing new sources of water. In many countries, this approach has been accepted
from both an economic and environmental point of view as being the best solution
for meeting growing water demands. Therefore, water demand management is an
appropriate strategy for improving efficiency and the sustainable use of water
resources, taking into account economic, social, and environmental considerations
(Wegelin-Schuringa, 2002).
The ability to manage existing water resources and then to plan for developing
new water resources is directly tied to the ability to describe both present and future
water use. The main objective of water demand management is to improve the
efficiency and equity of water use and sanitation services. For this purpose, various
instruments have been developed that can be generally classified into the following
categories:
• Water conservation measures

• Economic measures
• Informational and educational measures
• Legal measures
483
The efficiency of each of these instruments depends greatly on local conditions. This
chapter presents various aspects of managing water demands for domestic, industrial,
and agricultural purposes.
11.2 WATER USE AND DEMAND: BASIC DEFINITIONS

Water demand is the scheduling of quantities that consumers use per unit of time
for particular prices of water and is an analytical concept (Mays and Tung, 1992).
Water use can be classified into two basic categories: consumptive and nonconsump-
tive. As shown in Table 11.1, domestic, agricultural, industrial, and mining water
uses are consumptive. In contrast, hydropower, transportation, and recreation are the
main nonconsumptive water uses, which are also instream uses. In this type of water
TABLE 11.1
Definitions of Water Use Terms
Term Definition
Consumptive use The part of water withdrawn that is evaporated, transpirated, incorporated
into products or crops, consumed by humans or livestock, or otherwise
removed from the immediate water environment
Conveyance loss The quantity of water that is lost in transit from a pipe, canal, conduit, or
ditch by leakage or evaporation
Delivery and release The amount of water delivered to the point of use and the amount released
after use
Instream use Water that is used but not withdrawn from a ground- or surface-water source
for such purposes as hydroelectric-power generation, navigation, water
quality improvement, fish propagation, and recreation
Offstream use Water withdrawn or derived from a ground- or surface-water source for public
water supply, industry, irrigation, livestock, thermoelectric-power
generation, and other uses
Public supply Water withdrawn by public or private water suppliers and delivered to users
Return flow The water that reaches a ground- or surface-water source after release from
the point of use and thus becomes available for further use
Reclaimed wastewater Wastewater treatment plant effluent that has been diverted for beneficial use
before it reaches a natural waterway or aquifer
Self-supplied water Water withdrawn from a surface or groundwater source by a user rather than
being obtained from a public supply
Withdrawal Water removed from the ground or delivered from a surface–water source for
offstream use
Source: Mays, L. W., Ed., Water Resources Handbook, , McGraw-Hill, New York, 1996. With permission.
Water Demand Analysis and Management 485
use, a use is made of a body of water without withdrawing water from it. Onsite
water use refers to maintaining swamps and wetlands for wildlife habitats and
ditching and ponding for soil management. The U.S. Geological Survey (USGS)
definitions for ten water use accounts are given in Table 11.1. In the water-use
circulars prepared by the USGS, the following water uses are identified within the
overall water accounting system (Solley et al., 1993):
• Water withdrawal for offstream purposes

• Water deliveries at point of use or quantities released after use
• Consumptive use
• Conveyance loss
• Reclaimed wastewater
• Return flow
• Instream flow
Water use can also be classified as offstream use and instream use. Different
water demands resulting from offstream uses can be classified as:
• Domestic or municipal water uses include residential (apartments and

houses), commercial (stores and businesses), institutional (hospitals and
schools), industrial, and other water uses (firefighting, swimming pools,
park watering).
• Industrial water uses include water required for industrial processes such
as cooling water for steam electric power generation units, refineries,
chemical and still manufacturing, textiles, food processing, pulp and paper
mills, and mining.
• Agricultural water uses include those for irrigating fields and for the
drinking and care of animals.
• Miscellaneous water uses include fisheries, recreation, and mining.
These uses require withdrawal of water from the surface or groundwater resource
system. Part of the water withdrawn may return to the system, perhaps at another
location and time with different quality. The percent of return flows is an important
factor in evaluating water use efficiency. It should also be considered in water
resources management schemes. Table 11.2 shows a more detailed classification of
water demands. As shown in this table, water use can be classified as municipal,
agricultural, industrial, environmental, or infrastructure (public work).
Water demand changes from year to year and month to month. Many physical,
economic, social, and political factors account for these variations. Significant cli-
mate changes have been observed in recent years in many parts of the world,
producing more intense floods, greater precipitation, and even unusual droughts in
some regions. Prediction of the demand variations in the future is the main objective
in water demand forecasting, which is discussed in the next section of this chapter.
TABLE 11.2
Detailed Classification of Water Uses
Use-Type Rate of
Water Demand Purpose Classification Classification Consumption
Drinking Municipal Withdrawal Low

Domestic (cooking, Municipal Withdrawal Low
washing, etc.)
Fish and wildlife Agricultural and Withdrawal, instream, Moderate
environmental onsite
Livestock Agricultural, municipal Withdrawal Moderate
Drainage Agricultural Withdrawal, onsite High
Irrigation Agricultural, municipal Withdrawal High
Wetland habitat Agricultural, Onsite Moderate
environmental
Soil moisture Agricultural Onsite High
conservation
Utilization of estuaries Agricultural Instream, onsite —
Recreation and water Municipal, infrastructure Instream —
sports
Aesthetic enjoyment Municipal, infrastructure Instream —
Navigation Infrastructure Instream —
Hydropower Infrastructure Instream —
Mining Industrial Withdrawal Moderate
Cooling Industrial and municipal Withdrawal High
Boiling Industrial and municipal Withdrawal High
Processing Industrial and municipal Withdrawal Moderate
Steam power Industrial and municipal Withdrawal High
Waste disposal Industrial, municipal, Instream —
agricultural
11.3 WATER DEMAND FORECASTING FOR REGIONAL

AND NATIONAL PLANNING
Water demand should be forecasted in time and space. Many water resource projects
have a relatively long useful life; therefore, in studies of regional water resources
development, the time horizon for water demand forecasting should be extended to
about 50 years. In medium-scale development plans, a lead time of 15 to 25 years
may be applied. It should also be kept in mind that the forecast is made for general
planning purposes for a relatively large region, rather than simply for construction
of a single project. These projections should be made for at least three levels, namely
normal, minimum, and maximum conditions.
The basic information required for long-term water demand forecasting includes
projections of population and economic activities. The forecasts should reflect tech-
nological changes in production processes, product outputs, wastewater treatment
methods, social preferences, and public policies with respect to water use and devel-
opment (Mays and Tung, 1992); otherwise, the forecasts would be of limited value
to decision makers. The growing number of conflicts among water users is another
important factor that should be incorporated in long-term water demand forecasting.
Because land, water, and other natural resources have region-specific character-
istics, the response of industrial and agricultural activities to development policies
would be expected to vary within and between regions. The following items indicate
some of the main differences encountered among regions corresponding to the
development of natural resources:
• Availability of natural resources

• Economic development
• Technological advancements in operation and utilization of natural
resources
• Costs of water withdrawal
• Availability of labor skills
• Costs of transportation
• Natural resources markets
• Assimilative capacities of the environment
Water demand forecasts should reflect these regional characteristics. High water
effluent standards and water prices may cause shifts in population. Within a long-
term planning horizon, they may shift water demands from one region to another.
In contrast, programs that support agricultural development (such as price-support
programs) may encourage populations to move to regions offering more job oppor-
tunities. Water demand predictions based on simple extrapolation of current trends
may lead to serious errors. The following data and information are necessary for
forecasting the regional water demand over a long-term lead time:
• Population projections based on demographic studies and studies done by

the agencies responsible for multiple-sector investment decisions
• Distribution of urban and rural population among subregions
• Gross national product (GNP)
• Projected outputs of agricultural, mining, electric power, and major man-
ufacturing sectors for determining the regional distribution of activities
based on the projected GNP
• Projected rates of per capita unit water use based on technological
advancements and relative share of instream and offstream uses
• Expected use of brackish or ocean water
The water demand projections can be made without incorporating the water supply,
at least at the first step. But, because these projections are likely to be revised if the
costs of water are taken into account, a projection of demand must be accompanied
by information about the supply curve of water for each planning region (see
Chapter 4 for more information about the supply curve). Some of the water supply
issues are presented in the following section.
11.4 WATER SUPPLY ISSUES

Four major characteristics of water supply are quantity, quality, time variation, and
price. If the quantity and time variation of water resources conform with the water
use in a region, then we do not need to store or regulate water. Otherwise, certain
facilities should be implemented and the costs of the initial investment, operation,
and maintenance should be incorporated in the supply curve. In the same way, if
water quality does not satisfy the standards set for particular water uses, treatment
plants should be implemented and their costs should be incorporated in water
resources development studies.
In the previous chapters, large-scale and conventional methods of surface and
groundwater resource development were discussed. Large-scale facilities such as
dams, water transfer structures, and well fields are not always applicable to rural
areas, small communities, and basins with limited surface and groundwater
resources. In these cases, some unconventional methods should be considered for
supplying water demands.
11.4.1 RAINWATER COLLECTION

Collecting rainfall is a primary or supplementary source at the household or small
community level, especially in places with relatively high rainfall. Roofs of buildings
are the most common collecting surfaces. Natural and artificial ground catchments
are also used. When designing rainfall collection systems, the following issues should
be considered (Falkland, 1991):
• Quantity issues: Rainwater collection systems often suffer from insuffi-

cient storage tank volumes or catchment areas. Leakage from tanks due
to poor design, selection of materials, construction, or a combination of
these factors is a major problem with rainwater collecting systems.
• Quality issues: The rainwater quality in many parts of the world is good,
but in rainwater collection systems quality problems occur. Physical,
chemical, and biological pollution of rainwater collection systems is found
where improper materials have been used or where maintenance of roofs
and other catchment surfaces, gutters, pipes, and tanks are lacking.
Figure 11.1 shows an example of a well-designed rainwater catchment system.
11.4.2 DESALINATION
Among the unconventional water supply methods, desalination is the most widely
used. Seawater has a salinity of about 35,000 mg/l which is primarily composed of
sodium chloride. Desalination of seawater in coastal areas or brackish water in inland
areas is technically feasible, but it is expensive. The cost of desalination of seawater
ranges from $.80 to $3/m3 (Bouwer, 1994). In addition to a high rate of energy
consumption, a major problem at inland desalination plants is the disposal of rejected
brine. Evaporation ponds, injection in deep wells, and transfer to the ocean are
common methods that depend on the volume of rejected brine, site location, and
(a) Non-toxic smooth

dense roofing
Even slope of roof Gutter of small

uniform slope
Downpipe
Screen on
Manhole cover Foul flush top of
diverter downpipe
Covered
tank
Outlet Overflow
Draining tank
(b) Inflow pipe
Foul flush diverter
Screened intake
Manhole cover
Drain
Storage cover
Seal
Bottom Screened
Drain sloped outlet
Screened overflow
Sand Delivery
Tap
Drainage
Drainage sump
ditch
FIGURE 11.1 Examples of rainwater catchment systems. (From Falkland, A., Ed., Hydrology
and Water Resources of Small Islands: A Practice Guide, United Nations Educational, Sci-
entific, and Cultural Organization (UNESCO), New York, 1991.)
geographical and climatic conditions. The following methods are used for removing
dissolved solids:
• Distillation: Water is heated to its boiling point to convert it into steam,

then the steam is condensed to yield salt-free water.
• Reverse osmosis (RO): Water is forced through a semipermeable mem-
brane under pressure, and the dissolved solids are held back.
• Electrodialysis: Ions are separated from the water by attraction through
selective ion-permeable membranes using an electrical potential.
Among these methods, the first two are the most common processes for desaliniza-
tion of seawater. Electrodialysis is usually preferred for brackish groundwater.
11.4.3 ARTIFICIAL RECHARGE

The volumes of groundwater naturally replaced in each year are relatively small
because of the slow rates of groundwater movement and the limited rate of infiltra-
tion. Artificial recharge can be used to reduce adverse groundwater conditions such
as progressive lowering of water levels or saline water intrusion. Methods that have
been used for artificial recharge include:
• Utilization of holding basins: Water is recharged by releasing it into basins

formed by construction of dikes or levees or by excavation. Besides
surface runoffs, local storms can also be diverted to these basins for
artificial recharge.
• Flooding: This method can be used where topography is relatively flat so
the flood can be spread over a large area. This method does not have land
preparation costs.
• Stream channel: In this method, channels are modified to enhance the
infiltration process. For this purpose, the time and area over which water
is recharged should be increased. Upstream reservoirs will also help in
regulating the streamflows based on the absorptive capacity of down-
stream channels.
• Recharge well: Recharge wells draw water from the surface to aquifers.
Well recharging is practical where deep confined aquifers must be
recharged or where economy of space, such as in urban areas, is an
important consideration (Todd, 1980).
• Furrow: Water is distributed to a series of furrows. The furrows are usually
shallow, flat bottomed, and closely spaced to provide maximum water
contact area.
11.4.4 GROUNDWATER DAMS

Groundwater dams are subsurface dams constructed in aquifers. Their advantages
over surface water storage dams are reduction in evaporation and improvement in
water quality. This method has been used in Africa, India, and Cape Verde Islands
(Hanson and Nilsson, 1986). Suitable geological condition must be present where
the water is stored. For this purpose, the rock where the dam is to be constructed
should be relatively impermeable. Also, a suitable storage formation consisting of
coarse sediments should be situated over the impermeable base. Further information
on groundwater dams can be found in Nilsson (1988).
11.4.5 WEATHER MODIFICATION

One of the most widely used methods of weather modification is precipitation
enhancement by cloud seeding, which requires an aircraft equipped with devices for
distributing chemicals such as silver and iodide into clouds. Radar is usually used
for detecting suitable clouds. Another method of weather modification is evaporation
reduction, which is mainly used for small communities. Covering water tanks and
deeper water storages (such as groundwater dams) can be used for evaporation
reduction.
11.5 FLOW MEASUREMENT EQUIPMENT

Flow measurement is an important issue in water demand management. The water
supply for various uses and demand points within a complex system, such as an
urban area, is usually estimated based on flow measurement within the system. Illegal
withdrawals and water losses can be detected by water measurement at different
points in the water transfer and distribution system. Water measurement equipment
includes those used in water networks (such as water meters) or those used in open
channels or rivers for measuring flow flux (such as Parshal flumes or weirs).
11.5.1 FLOW MEASUREMENT IN PIPES

Three common methods are used to measure water flow in pipes. In the first method,
differential pressure meters such as venturi tube, orifices, and nuzzles are used. In
venturi tube meters, water flows through the tube and velocity is increased in the
constricted portion, which temporarily lowers the static pressure in accordance with
the energy equation. The pressure difference between the inlet and throat is measured
and correlated to the rate of flow. Venturi meters are shaped to maintain streamline
flow for minimum head loss.
The second method is based on measuring a chamber of known volume con-
taining a disk that rotates as water passes through. The rotation resulting from the
filling and emptying of the chamber is transmitted to a recording register. The
advantages of this meter are simplicity of construction, high sensitivity and accuracy,
small loss of head, and low maintenance costs. This oscillating disk meter is usually
used for small customer services, such as individual households and apartments.
The third method used for measuring water flux in a pipeline utilizes a wheel
with blades that rotates at a speed in proportion to the quantity of water that passes
through the blades. A recording register is geared to the turbine wheel. The disad-
vantage of this method is poor accuracy at low flow rates when the water is not
moving at sufficient velocity to rotate the blades. Consequently, current meters are
used only in limited applications.
11.5.2 FLOW MEASUREMENT IN OPEN CHANNELS AND RIVERS

Open channels are used for two purposes: (1) conveying water from water reservoirs
for agricultural and other uses and (2) conveying wastewater. Wastewater contains
suspended and floating solids that prohibit the use of enclosed meters. A Parshal
flume is the device most commonly used to measure water flux in open channels.
A typical flume consists of a converging and dropping open channel section. Flow
moving freely through the unit can be calculated by measuring the upstream water
level. A stilling well is normally provided to hold a float bubble tube or other depth-
measuring device that is connected to a transmitter and flow recorder. The main
advantage of an open channel flume is low head loss. Weirs can also be used for
measuring flow of water in open channels. Water flowing over the sharp edge crest
must discharge to the atmosphere and air must be allowed to pass freely under the
jet. If these conditions are met, the rate of flow can be directly related to the height
of water measured behind the weir. The most common type of weir used for mea-
suring wastewater flows is the 90 V-notch weir. This kind of weir is usually installed
on a temporary basis to make flow measurements associated with industrial waste-
water surveys.
11.6 MUNICIPAL WATER DEMAND

As mentioned previously, municipal water uses include residential (apartments and
houses), commercial (stores and businesses), institutional (hospitals and schools),
industrial, and other water uses (firefighting, swimming pools, park watering).
Municipal water use is usually quantified as the volume of water used by each person
in each day, referred to as per-capita water use. It is usually measured in liters per
day or gallons per day per person or per metered service.
Globally, variation in the withdrawal of water for municipal purposes varies
enormously. Average municipal water use in the United States per metered service
is about 2270 l/day. Water consumption of residential areas in eastern and southern
states is about 790 l/day; in central states, about 1060 l/day; and in western states,
about 1740 l/day (Hammer and Hammer, 1996). Lawn sprinkling can have a signif-
icant influence on water demand in areas with large residential lots. In these areas,
50 to 75% of the total daily volume may be attributed to irrigation. Figure 11.2
shows per-capita water use in some of the big cities in the world.
Municipal water demand depends greatly on health standards in urban and rural
areas. The quality of the drinking water has a direct effect on health, and the amount
and quality of water available for bathing and cleaning have a significant indirect
impact on health. Per-capita water use in most residential areas has increased over
the past decades. Most new houses have more water fixtures, spacious lawns, modern
appliances, and other conveniences that consume larger volumes of water.
11.7 MUNICIPAL WATER DEMAND ESTIMATION AND

FORECASTING
Costs of water supply services and technological developments designed to lower
these costs have a major influence on the level of water demand in developing
FIGURE 11.2 Per-capita water use in large cities of the world. (Data tables from World Resources Institute, 1998.)
Per-Capita Municipal Water Use
(L/day)
100
200
300
400
500
600
700
800
900
0
Addis Ababa
Nairobi
AFRICA
Rabat
Lagos
Harare
Khartoum
Prague
Paris
EUROPE
Budapest
Amsterdam
Moscow
Stockolm
N. AMERICA C. AMERICA
Toronto
Atlanta
New York
Havana
San Salvador
Guatemala City
Rio de Janiro
S. AMERICA
Lima
Bogota
Brasillia
Shanghi
Bombay
Delhi
ASIA
Jakarta
Almaty
Lahore
Sana
OCEANIA
Melbourne
Auckland
493 Water Demand Analysis and Management

countries. In rural areas, the distance between standpipes or the number of persons
served by a single tap or well are decisive factors controlling the level of demand.
Precise estimation of municipal water use can be obtained by breaking down the
total delivery of water to urban areas into a number of classes of water use and
determining separate average rates of water use for each class. This method is
referred to as disaggregate estimation of water use. Previous studies have shown
that water use within some homogenous sectors is less variable compared to the
total water use; therefore, greater accuracy in estimation of water use can be obtained.
The components of urban water usage can be classified as follows:
• Domestic
• Washing and cooking
• Toilet
• Bath and/or shower
• Laundry
• House cleaning
• Yard irrigation
• Swimming pool
• Car washing
• Other personal uses (hobbies, etc.)
• Public services
• Public swimming pools
• Governmental agencies and private firms
• Educational services (such as schools, universities, and dormitories)
• Firefighting
• Irrigation of parks, golf courses, etc.
• Health services (such as hospitals)
• Public baths, public toilets, etc.
• Cultural public services (such as libraries and museums)
• Street cleaning and sewer washing
• Entertainment and sport complexes (such as cinemas and clubs)
• Food and beverage services
• Accommodation services
• Barber shops and beauty parlors
• Small industries (such as laundries or workshops)
• Construction and public works
• Water losses
• Leakage from pipes, valves, meters, etc.
• Evaporation in open reservoirs
• Overflow of reservoirs
• Disrepair of elements of a water distribution network, such as cracked
reservoirs, flow back through one-way valves and pumps, etc.
• Loss in production process (cooling, pumping, etc.)
• Transportation
• Taxies, buses, and other conveyances (stations, garages, etc.)
• Ports and airports
• Railways (stations and workshops)
In order to estimate the total municipal water use in a city, the study area should
first be divided into homogenous areas. These homogenous areas are usually selected
based on pressure districts or land-use units, and the water use rates can be assumed
to be constant for different users within each area. Temporal (e.g., annual, seasonal,
monthly) variation may also be considered when subdividing the water uses.
The most commonly used method to forecast water use is regression analysis.
A simple regression model can be formulated as follows:
Qt ,i, j = aXt ,i, j + b + ε t (11.1)
where:
Qt,i,j is the average rate of water use in time period t for disaggregated use i
in homogenous subarea j.
Xt,i,j is the independent variable in time period t for disaggregated use i in
homogenous subarea j.
εt is the error term in time period t.
a and b are regression coefficients.
Qt,i,j is the dependent variable of the regression. The independent variable, Xt,i,j,
should be selected based on the available data of different factors affecting the water
use and the relative importance of them in increasing or decreasing water uses. For
example, the most important factor in estimating water use in an urban area is the
population in each subarea. In Eq. (11.1), if the independent variable is set to be
equal to the number of users in each subarea, then the regression slope coefficient,
a, would be the per-capita water use. The water use can also be estimated based on
the number of connections to the water distribution system as follows:
Qt ,i, j = e ⋅ NCi, j ⋅ Wt ,i, j (11.2)
where:
NCi,j is the number of connection for disaggregated use i in homogenous

subarea j.
Wt,i,j is the water use per connection for disaggregated use i in homogenous
subarea j in time period t.
e is the efficiency of the water distribution system, which is a function of
leakage and other water losses in the system.
Multiple regression methods can also be used to incorporate other variables

correlated with water use in municipal areas to estimate and forecast water use in
the future. Population, price, income, temperature, and precipitation are some of the
variables that have been used. Based on the independent variables, water use models
can be classified as requirement models and demand models. In requirement models,
only the physical and ecological variables related to water use are incorporated, but
demand models are based on economic reasoning and include only variables that
are correlated significantly with water use and are expected to be causally related
to water use (Mays and Tung, 1992). Linear and logarithmic water use models have
been suggested by different investigators:
Q = a1 x1 + a2 x2 + ⋯ + an xn + ε (11.3)
Q = a1 ln x1 + a2 ln x2 + ⋯ + an ln xn + ε (11.4)
Besides regression models, time series analysis has also been used to forecast
future variations of water demands. For this purpose, time series of municipal water
use and related variables are used to model historical patterns of variations in water
demand. Long memory components, seasonal and nonseasonal variations, jumps,
and outlier data should be carefully identified and used to model water demand time
series. Details of time series analysis and statistical modeling are provided in
Chapter 5.
11.8 ELEMENTS OF URBAN WATER DISTRIBUTION

NETWORKS
Water distribution networks usually consist of the following elements:
• The reservoir provides water to the water distribution system. When

modeling water distribution systems, a reservoir is usually considered as
a boundary node with constant surface water elevation. In other words, it
is assumed that inflow and discharge from the reservoir do not affect its
surface water elevation. Important information about reservoirs includes
the surface water elevation and water quality.
• The tank stores excess water within the water distribution system and
releases it when the water use is high. Tanks are similar to reservoirs but
their storage capacity is smaller, and the hydraulic grade line in tanks
varies with respect to inflow and outflow variations.
• Pipes convey water from one point to the other. Pipes used in distributing
water under pressure include ductile iron, plastic, concrete, and steel.
Copper and plastic pipes with small diameters are usually used for house
connections. Pipes in water distribution networks should have enough
tensile and bending strength to withstand external loads due to trench
backfill and earth movement caused by freezing, thawing, or unstable soil
conditions. They should also have enough bursting strength to withstand
internal water pressures and be able to resist impact loads encountered in
transportation, handing, and installation. The inner side of the pipes should
be smooth and noncorrosive to create minimum resistance to water flow
(Hammer and Hammer, 1996). Two important characteristics of pipes are
length and diameter. When modeling water distribution networks, the
length assigned to a pipe should represent the total distance between two
nodes and not necessarily the straight-line distance. Pipes are usually
produced in lengths of around 6 m and are referred to by their nominal
diameter. The internal diameter of the pipes, which should be used in
modeling, is usually different from the nominal diameter. Internal diam-
eters may change over time due to tuberculation, corrosion, etc.
• A junction refers to any of the following nodes in a water distribution
network:
• Location where two or more pipes meet
• End of a single pipe (dead-end pipe)
• Location where water is withdrawn from the network
• Location where water is injected into the network
• Pumps are used for a number of purposes in water distribution networks;
however, they are usually used to boost the head at specific locations to
overcome piping head losses and physical elevation differences. Some of
the main types of the pumps can be classified as:
• Low-lift pumps elevate water from a source to a treatment plant.
• High-service pumps discharge water under pressure to a distribution
system.
• Booster pumps increase pressure in a water distribution system.
• Circulation pumps move water in a treatment plant.
• Well pumps lift water from wells for water supply.
• In water distribution networks, the most commonly used type of pump is
the centrifugal pump. A centrifugal pump has a motor that spins a piece
within the pump called an impeller. The mechanical energy of the rotating
impeller is imparted to the water, resulting in an increase in head.
• A valve is an element for controlling either the magnitude or direction of
water flow through a pipeline. Some of the main types of the valves can
be classified as:
• Shutoff valves, the most common types of the valves in water distri-
bution systems, are used to stop the flow of water through a pipeline.
The primary purpose of this type of valve is to turn off a portion of
the system, for example, for the time it takes to replace a broken pipe.
Different types of shutoff valves include gate, butterfly, globe, and plug
valves. Among all these types of shutoff valves, gate valves are the
most widely used valve.
• Check valves are used to ensure that water flows only in one direction
in the pipeline. Water flowing backward through the valve causes it to
close. They are usually installed in the discharge pipes of centrifugal
pumps to prevent backflow when the pump is off.
• Altitude valves are used to automatically control the flow into and out
of an elevated storage tank or standpipe to maintain desired water-level
elevation. When the tank level rises to a specific upper limit, the valve
closes to prevent any further flow from entering.
• Air release valves release trapped air during the operation of water
distribution systems. Air can enter a pipe network through leaking
joints, from pipes drawing air into the suction pipes, etc. These air
pockets increase resistance to the flow of water. Air release valves can
be used at these locations to discharge the trapped air.
• Control valves can be classified as flow control valve, throttle control
valve, and pressure-based control valves. Pressure-based control valves
are designed to maintain the hydraulic grade or the pressure at a specific
level.
11.9 MUNICIPAL WATER DEMAND MANAGEMENT

MEASURES
The primary measures of domestic water demand management can be classified as:
• Water loss reduction:

• Leakage detection and reduction
• In-house retrofitting
• Reduction of illegal connections
• Education and training:
• Public awareness
• In-school education
• Training and education of the staff in water related agencies
• Economic incentives and water pricing:
• Water metering
• Tariff structure
• Institutional measurements and effective legislation:
• Regulations for water demand management
• Regulations on resale of water
• Water reuse
A brief explanation is presented here about each of the above water management
measures.
11.9.1 WATER LOSS REDUCTION

Throughout the world, a large amount of water is lost through urban water supply
systems due to leakage in distribution networks and home appliances or through
illegal connections. These water losses are referred to as unaccounted-for water, a
measure that is often used to quantify the efficiency of a water supply system.
Unaccounted-for water can be defined as the difference between the amount of water
supplied from the water works, as measured through its meters, and the total amount
of accounted-for water. Accounted-for water includes water consumption as recorded
by customers’ meters, water stored in service reservoirs, and authorized free use
such as for flushing and sterilization of mains and routine cleaning of service
reservoirs. In many cities unaccounted-for water is estimated to be as high as 50%
of water supplied. A certain level of unaccounted-for water cannot be avoided, but
in some cities such as Singapore the level has been reduced to about 6% (United
Nations, 1998). Unaccounted-for water can be classified as:
• Apparent water loss, which includes errors due to meter inaccuracy and
improper accounting of water used in the commissioning and filling of
new mains, connections, and service reservoirs and for cleaning and
flushing of the water distribution system during maintenance
• Real water loss, which includes water loss due to leaks and illegal draw-
offs from the transmission and distribution system
The following items can be considered as control measures for reducing the
amount of unaccounted-for water:
• Leakage detection. A specific program is required to detect the leakage.

It can include visual inspection for leaks along transmission and distribu-
tion pipelines and also leak detection tests at night for distribution mains.
Nighttime is selected because the pressure is usually higher in the system,
making it is easier to detect the leaks and also minimizing inconvenience
to customers. To locate the leaks, mechanical, electronic, and computer-
ized acoustic instruments such as stethoscopes, geophones, electronic leak
detectors, and leak noise correlators can be used. The time frequency of
testing is important. It depends on the percentage of unaccounted-for water
and share of leakage and the size of the water distribution systems.
• Leakage control. Better quality pipes and fittings can be used for leakage
control. More durable and corrosion-resistant pipe materials such as duc-
tile iron pipes internally lined with cement mortar, copper, and stainless
steel can be used for this purpose. When designing the water distribution
network, minimizing the number of joints is important to reduce minor
leaks. Teflon packing for repairing valve glands and leaks due to wear
and tear and dezincification-resistant brass fittings can also reduce the
minor leakage in the water distribution network.
• Full and accurate metering policy. Meters with less accuracy such as
Venturi/Dall tubes can be replaced with more accurate meters such as
electromagnetic meters. A specific program for maintaining and replacing
meters regularly should also be considered. This can be done through bulk
changing programs.
• Proper accounting of water used. In most water distribution networks in
urban areas, significant quantities of water are used in the commissioning
and filling of new mains, for connection and service reservoirs, for clean-
ing and flushing the distribution system during maintenance, and for
firefighting. Water used for such purposes should be accurately reported
to ensure proper accounting.
• Strict legislation on illegal drawoff. The illegal drawoff should be detected
and anyone who is responsible for it should be prosecuted.
• Education and training. Public education and training are important issues
in achieving the water demand management goals. Various methods have
been tried in developed and developing countries such as:
• Training programs in the mass media
• Billboards along streets and on public transport vehicles
• Training brochures enclosed with water bills

Educating school children has been given top priority in many
countries. These programs present basic information about the hydro-
logic cycle, costs of water supply, maintenance of water distribution
networks, water pollution, and costs of wastewater treatment. Education
and training can also be of benefit to other water management strategies,
such as changing water tariffs and charging effluents. The staffs of
water-related agencies should be well trained. Training staff to be tech-
nically competent and customer service oriented should be considered
an important objective when designing staff training programs.
11.9.2 ECONOMIC INCENTIVES AND WATER PRICING

Pricing is a powerful tool for controlling urban water use. In many countries, water
and water-related services have been provided almost free or at a very low charge,
especially to domestic customers. Water charges in many urban areas do not even
cover the cost of operation and maintenance of a water supply system; therefore,
those services have been highly subsidized. When defining tariffs for urban areas,
groups with high, medium, and low levels of income and the areas in which they
live should be identified. Various structures of tariffs can then be applied to create
relative equity in the allocation of the costs of water services and to control the rate
of domestic water use. According to the United Nations (1998), increased tariff rates
can be of benefit if:
• They reflect the real costs of the services provided.

• They are linked to consumption levels.
• Differential charge increases are large enough to encourage water savings.
• Any changes in the tariff rates are accompanied by well-organized public
information campaigns.
11.9.3 INSTITUTIONAL MEASUREMENTS AND EFFECTIVE LEGISLATION

Fragmentation and overlapping of responsibilities among agencies involved in
water resources management are obstacles to efficient water use in urban areas. In
most countries, different organizations are responsible for the following tasks:
• Water resources development planning

• Water supply to urban water distribution network
• Wastewater collection
• Water pollution control
The levels of responsibility for these agencies may also differ at national or more
limited provincial or city levels. These tasks may be handled by a number of
organizations, and the decisions made in each of them can significantly affect other
responsible agencies. In addition, the urban sector is closely interrelated with other
economic sectors, which is why any proposed program for urban water conservation
will not be successful if it attempts to isolate the water supply and sanitation sector
from other urban sectors. Thus, integrated management policies are required to
achieve economical use of water in urban areas (United Nations, 1998); however,
such policies can only be put into practice with the commitment and close cooper-
ation of all the organizations and agencies involved. For this purpose, conflict
resolution issues should be incorporated in the development of planning and man-
agement policies.
The efficiency of implementing water resources development projects and utility
management can be improved by allowing the private sector to manage and handle
the associated risk, but the governmental agencies should maintain overall control
through legislation and regulation. The injection of private-sector efficiency and
productivity also results in reduced public-sector expenditure.
11.9.4 WATER REUSE

Details of water reuse for different purposes are presented in Section 11.16.
11.10 AGRICULTURAL WATER DEMAND

Water use in agriculture is continuing to increase as world food demand rises.
Agriculture already accounts for about 70% of water consumption worldwide. The
United Nations projects up to a 100% increase in irrigation water by the year 2025
(World Resources Institute, 1998) and forecasts that nations will develop new sources
of water. In Asia, Central America, and Africa, the share of agricultural water use
is about 85 to 90% of all water uses; therefore, water demand management issues
in the agricultural sector can play a significant role in reducing water demands.
Crop water demand is usually measured in terms of evapotranspiration. The rate
of evapotranspiration depends on meteorological factors such as relative humidity,
temperature, wind, radiation, number of daylight hours, precipitation, etc. and field
factors such as soil moisture and surface properties. Crop water requirements are
usually estimated based on the rate of potential evapotranspiration, ETo. Potential
evapotranspiration is defined as the maximum evapotranspiration rate that the atmo-
sphere is capable of extracting from a well-watered field. Methods for estimating
potential evapotranspiration have been developed by Blaney and Criddle (1945),
Thornthwaite and Mather (1955), Penman (1948), and Thomas (1981). Details of
the Thornthwaite method are presented in Chapter 12. The crop water requirement
can then be estimated as follows:
ETcrop = Kc ⋅ ETo (11.5)
where ETcrop is the water requirement for a specific crop, and Kc is the empirical
crop coefficient, which depends on the following factors:
• Climate
• Temperature
• Wind
• Humidity
• Crop
• Stage of growth
• Crop reflectivity
• Crop height and roughness
• Degree of ground cover
• Canopy resistance to transpiration
For most crops, the Kc value varies from 0.6 to 0.9 for the total growing season. In
order to estimate the agricultural water demands, the following items should be
incorporated:
• Crop water requirement is the amount of water that crops require to grow
and is considered to be equal to evapotranspiration.
• Irrigation requirement is the crop water requirement minus effective rain-
fall available for plant growth. In estimating the irrigation requirements,
water consumed for leaching and other miscellaneous requirements such
as for germination, frost protection, and plant cooling should be included,
and the soil moisture loss should be subtracted.
• Farm delivery requirement is estimated as the irrigation requirement plus
farm water losses due to evaporation, deep percolation, surface waste, and
nonproductive consumption.
• Gross water requirement is estimated based on farm delivery requirements
plus seepage losses in the canal system from the headworks to the farm
unit plus the waste of water due to poor operation.
Agricultural water demands are changed from year to year and month to month.
The parameters affecting agricultural water demands can be summarized as follows:
• Crop mix. Each crop has its own water demands; therefore, the types of
crops and the areas covered by each crop, which usually vary from year
to year, can dramatically change the agricultural water demands of a
specific farm.
• Irrigated and dry-land farming. Substitution of irrigated farming for dry-
land farming frequently occurs in areas with a sufficient rate of effective
rainfall.
• Period and sequence of cropping. The time of planting may also change
the water demands of crops. These variations are highly dependent on
climatic characteristics.
• Physical characteristics of the water transfer and irrigation systems. The
physical characteristics of irrigation systems are important in estimating
water losses through a transfer and irrigation system. The water supply
efficiency can be defined based on the project, field, and farm efficiency.
Project efficiency can be estimated based on the proportion of flow at the
source and water applied to crops. Field efficiency deals with the propor-
tion of water flowing into the field to the amount of water that is applied
to crops. Farm efficiency can be estimated based on the proportion of

water at the turnout to the amount of water that is applied to the crops.
In addition to water supply efficiency, irrigation efficiency can be defined
in terms of financial, physiological, and sociological aspects. Financial
efficiency can be defined as the financial return for a particular amount of
investment in water supply and can vary significantly from year to year
and from place to place. It is not always possible to quantify the long-
term sociological benefits of irrigation. In drought events, where the
indices of drought are not predictable, even a modest contribution to food
security arising from assured irrigation can have very profound conse-
quences on the quality of life of the people.
• Market prices. A change in the market prices of agricultural products is
another important issue in changing the agricultural water demands. Farm-
ers quickly respond to market changes due to economic incentives.
• Climate variation. Hot weather, a windy climate, and low humidity are
factors that increase crop water requirements. Also, the net effective rain-
fall supplies part of the crop water demands, so crop irrigation demands
will be less in years with greater rainfall.
• Policies related to pricing, importing, and exporting of agricultural prod-
ucts. National policies about importing and exporting agricultural products
can drastically change product prices. Agricultural price-support programs
have significant effects on increasing agricultural water requirements.
11.11 IRRIGATION METHODS

Increasing irrigation efficiency is the most important issue in agricultural water
demand management; therefore, it is important to recognize various methods of
irrigation and their advantages and disadvantages. The irrigation methods can be
classified as:
• Gravity or surface irrigation. Water is delivered to farms via canals and

then distributed to individual fields by gravity whenever possible. Pump-
ing, if required, increases operating costs proportionally. Surface irrigation
includes furrow, flood, and strip methods. Furrow irrigation is used on
crops grown in rows, with the water confined to running laterally between
the raised rows of crops. Flood irrigation is used on close-growing field
crops (such as rice) where the field or part of the field is surrounded by
shallow levees so that the entire field can be flooded. For the surface
irrigation method, the soil is the medium through which water is conveyed,
distributed, and infiltrated. This method is mechanically simple, has low
energy requirements, and can be easily adopted to small holdings. Its
serious deficiencies are low application frequency, high conveyance
losses, and wastage of water, which lead to adverse consequences such
as water logging and salinization. Figure 11.3 shows examples of flood
and furrow irrigation systems.
(a)
(b)
FIGURE 11.3 Examples of (a) flood and (b) furrow irrigation systems.
• Sprinkler irrigation systems: In these systems, sprayed water falls on

plants like rain. These systems are designed either to be portable and
move across a field with the water supply hose attached by cable (traveling
sprinklers) or to rotate continuously around a pivot point (center pivot
systems). Sprinkler systems have the advantage of providing a more con-
trolled application rate than surface irrigation systems. The conveyance
losses are minimal, and it allows soil aeration. The main disadvantages
of this method are high capital requirements and operation and mainte-
nance costs, as well as high sensitivity to wind. Figure 11.4 shows exam-
ples of sprinkler irrigation systems.
• Drip irrigation: In this method, water is applied to soils directly under
the surface; to the root zone. Water is delivered drop by drop through
perforation or emitters in the pipes. The trickling rate is maintained at a
level less than the rate of infiltrability of soil. This method can be used
for fields for which the upper subsoil is very permeable with uniform
topography and moderate slopes. The advantages of this method are low
labor costs and high crop yields. The capital costs of drip irrigation
systems are high. The method requires rigorous adherence to scheduling
and maintenance. A microprocessor-based drip irrigation system suitable
for use in developing countries is now commercially available. This device
uses low-cost ceramic sensors and operates on a solar-cell battery. The
device continuously monitors the soil moisture and controls the drip rate
so as to maintain the moisture within the desired limits. Figure 11.5 shows
examples of drip irrigation.
• Microsprayer irrigation systems. In these systems, the water is applied
only to a fraction of the ground surface. The advantages of these systems
are similar to those for drip irrigation systems, including high-frequency
irrigation and allowing injection of fertilizers, if desired. Microsprayer
irrigation systems also have the following advantages over drip irrigation
systems:
• As microsprayers have larger nozzle orifices, clogging of emitters is
not a serious problem, and it is not necessary to filter the irrigation
water.
• They are operated at a pressure on the order of 2 atmospheres, a level
that is much lower than for the drip systems.
• They can be scaled down for use in small farms.
The minor disadvantages of this method include:
• Because the area wetted is larger than that in drip systems, some evapo-
ration losses occur.
• Because the leaves get wet, brackish water cannot be used in irrigation.
• Low-head irrigation: In a closed conduit system, the conveyance losses
are avoided but the system requires energy to pressurize water for distri-
bution. Low-head systems have the advantages of a piped system but do
not require any pumps or nozzles. In these systems, water is simply
(a)
(b)
(c)
FIGURE 11.4 Different types of sprinkler irrigation: (a) center pivot system; (b) sprinkler
irrigation; (c) wheel line field irrigation.
FIGURE 11.5 Examples of drip irrigation.

allowed to bubble out from open, vertical standpipes that rise from buried
lateral irrigation tubes. These systems are often used for flow rates
typically on the order of 30 to 200 l/sec and are particularly suitable for
widely spaced crops such as fruits trees and grapevines. Low-volume,
high-frequency, partial-area irrigation is possible with this system. The
initial cost of the system is comparable to or even lower than other
systems, it incurs no energy costs, and the useful life of the system is
longer.
11.12 DRAINAGE AND SALINITY CONTROL

Drainage refers to artificial removal of excess water from agricultural lands. Drain-
age may be of two kinds: surface drainage and groundwater drainage. Surface
drainage refers to shaping the land so as to facilitate overland flow, which drains
excess water accumulation on the surface. Lowering the groundwater table or pre-
venting its rise (groundwater drainage) drains excess water within the soil or subsoil.
While soil saturation for short periods is not harmful by itself, prolonged satu-
ration (water logging) can adversely affect plant growth. Excess water in the soil
blocks the soil pores, impedes the movement of oxygen from the atmosphere, and
limits the respiration activity of the plants. A decrease in the solubility of oxygen
in water and an increase in the respiration rate of both plants and microorganisms
result in an increase in temperature. Thus, water logging has more severe conse-
quences in warm climates than in cold climates. The land becomes infertile due to
the steady rise of the water table, water logging, and salinization. Thus, all irrigated
lands must have a drainage system, and irrigation without drainage can have disas-
trous consequences.
Excess water from the soil or subsoil is drained through ditches, perforated
pipes, or machine-formed mole channels by gravity flow or pumping. The drained
water is transferred into a stream, lake, or an evaporation pond. The rate of flow
from the soil to the drains depends upon the following factors:
• Permeability of the soil

• Depth of the water table
• Depth of the drain
• Horizontal spacing between the drains
• Configuration of the drains (open ditches or tubes) and their diameter and
slope
• Nature of the drain-enveloping material (such as gravel) used to increase
the seepage surface
• Rate of recharge of groundwater
As the field conditions tend to be highly complex and variable, empirical equations
must be used to estimate the desirable depths and spacing of drain pipes under
particular soil and groundwater conditions.
11.13 AGRICULTURAL WATER DEMAND

MANAGEMENT
The agricultural water demand management measures can be classified as:
• Increasing irrigation efficiency

• Development of proper crop mix
• Soil development and management
• Wastewater reuse in irrigation
11.13.1 INCREASING IRRIGATION EFFICIENCY

The types of irrigation systems were explained in the previous section. When select-
ing the proper irrigation system for a specific district, the following considerations
are important:
• Physical site condition

• Social and institutional considerations
• Economic considerations
General site conditions as listed in Table 11.3 should be considered when

selecting the proper irrigation system. In addition to the general classification of
irrigation systems, the following items also affect the irrigation efficiency:
TABLE 11.3
Physical Site Conditions to Consider in Irrigation System Selection
Crops Land Climate Water Supply Energy
Crops grown Field shape Precipitation Source Availability

Crop rotation Obstructions Temperature Quantity Reliability
Crop height Topography Frost condition Quality Cost
Cultural practices Soil Humidity Salinity
Disease potential Texture Wind Sodicity
Pests Uniformity Sediments
Water requirements Depth Organics
Climate modification Intake rate Reliability
Water-holding capacity Delivery
Erodibility Schedule
Salinity Frequency
Drainability Rate
Bearing strength Duration
Flood hazard Water cost
Water table
Source: Mays, L. W., Ed., Water Resources Handbook, McGraw-Hill, New York, 1996. With permission.
• Appropriate on-farm facilities (measuring devices, regulatory structures,

ditches) ensure good managerial control over the delivery and distribution
system. For this purpose, training programs for the proper use of the
facilities are of paramount importance.
• Loss of water by direct evaporation from the soil surface is a significant
source of water loss and inefficient irrigation. Improved irrigation tech-
nologies should be used to reduce evaporation losses.
• A preference for using groundwater as a source of agricultural water
supply reduces losses at the source, particularly in regions with a hot
climate and high evaporation rate. If surface water is the main source of
water supply, reduction of evaporation and seepage losses from reservoirs
should be considered.
• Automated control and operation of irrigation systems greatly contribute
to increasing the efficiency of water use by preventing or reducing losses
due to managerial failures and over-irrigation.
• When farmers are confident water will be available when needed, they
are less likely to irrigate to excess. Farmers frequently over-irrigate to
ensure that enough water is applied to last until the next time water is
available. Delivery systems in which water is always available and noti-
fication is not required are referred to as on-demand delivery systems.
• Legal issues may also affect the efficiency of water use in irrigation. Water
rights are the most important issue related to the legal system. Transfer
restrictions on water use rights may prevent the utilization of water on
more productive lands or for more productive purposes. Quality criteria
are not usually specified in water use rights; however, they have a signif-
icant impact on the efficiency of water use. Water rights are sometimes
issued on a first-come, first served basis, which may tend to prevent or
restrict the efficient use of water in terms of larger areas and longer
periods.
• Pricing is another important issue in agricultural water demand manage-
ment. Some districts charge a fixed rate based on the land area served,
regardless of the amount of water delivered. Others charge for the volume
of water delivered, and the price for unit volume is usually defined based
on the price of agricultural products in each field. This encourages con-
servation but makes the financial management more difficult.
11.13.2 DEVELOPMENT OF THE PROPER CROP MIX

In many irrigation systems, the water is supplied from local rivers and streams with
no or insufficient regulation. In such cases, on-time use of available surface runoffs
may be an important objective in increasing productivity of agricultural lands. For
this purpose, an optimal crop mix can be developed. In order to develop an optimal
mixture of crops for specific agricultural land, an optimization model can be devel-
oped. In the following formulation for this model, on-time use of available water is
maximized considering monthly variations of demand and streamflows. The objec-
tive function of this model is as follows:
⎧ N 12
⎪⎪ ∑ ∑(I
Minimize z = ⎨ j =1 i =1
ij −Dij ) if I ij ≥ Dij
(11.5)
⎪
⎪⎩0 otherwise
where:
N is the number of crops.

i is the number of months (i = 1, …, 12).
Iij the volume of water allocated to crop j in month i.
Dij is the volume of water demand of crop j in month i.
This objective function will minimize losses for months in which the surface runoffs
are not allocated. To limit possible shortages when supplying irrigation demands,
constraints can be considered to limit the maximum shortage rates in different
months. Agencies responsible for irrigation water supply usually define tariffs based
on the price of agricultural products; therefore, water allocated to agricultural lands
may be priced differently. The difference in price may also be incorporated in finding
the optimal crop mix for each specific region.
11.13.3 SOIL DEVELOPMENT AND MANAGEMENT

Suitability of lands for irrigation is also an important issue in increasing irrigation
efficiency and productivity of agricultural lands. For this purpose, relocation of
agricultural activities in more cultivated lands, if possible, can be considered as an
option at the local and regional agricultural planning level.
11.14 INDUSTRIAL WATER DEMAND MANAGEMENT

In developing countries, industrial water demands generally have been rising, but
in other countries such as Japan the increasing rate of water demand has slowed
down or even reversed. Technological advancements in such industries as pulp and
paper production can significantly reduce the water demands. Development of recy-
cling techniques also has played a vital role in reducing industrial water demand in
recent years. Recycling water in cooling systems has been initiated in many coun-
tries, and water reuse has become an important issue in industrial water demand
management. Water quality standards for many industries can be achieved by con-
ventional treatment methods. Domestic and industrial wastewater can be treated and
reused in industrial processes. Water recycling and reuse are also appropriate options
for reducing the costs of wastewater treatment before discharge to water bodies.
Economic incentives can also be effectively used for controlling water-intensive
and water-polluting industries such as steel and petrochemicals and textiles. Water
tariffs, effluent charges, and tax concessions for the use of water-saving and waste-
water treatment equipment can encourage industries to invest in water-saving tech-
nologies and recycling and to reuse treated wastewater. If water-polluting industries
are charged according to the volume of wastewater they discharge to water bodies,
they have an incentive to recycle and reuse water.
11.15 ENVIRONMENTAL WATER DEMANDS

As mentioned previously, in-stream uses are those that utilize water in situ as it
resides in or moves through the water body itself. In many water bodies, minimum
in-stream requirements are specified that may have the highest (inalienable) priority
within the water rights system. An environmental water demand (EWD) is the water
regime required to sustain the ecological values of an aquatic ecosystem at a low
level of risk. If this water requirement is adopted, then it is likely that a water body
will:
• Be healthy.
• Look after the needs of animals and plants such as fish, macroinverte-
brates, and plants.
• Maintain its biodiversity.
Three steps to identify environmental water requirements in a basin are:
1. Prioritize sub-basins for assessment using information on current water

use, river and estuarine health indicators, and water management planning
priorities.
2. Consult basin stakeholders and relevant scientific experts to determine
important values for each sub-basin in the following broad categories:
a. Ecosystem
b. Recreation
c. Aesthetics
d. Physical landscape
e. Consumptive/nonconsumptive use values
3. Assess environmental water requirements using the most appropriate sci-
entific methodology on a catchment-by-catchment basis.
Estimation of environmental water requirements depends on identifying those com-

ponents of the environment that are to be protected. These components depend upon
the amount and timing of water use within the basin. For example, the majority of
water use in a catchment may occur in the summer, and this may have an effect on
species that have particular requirements in these months (e.g., spawning and fish
migration). Other basins may have high water use all year, or dams could be built
to alter the flow regime. The rivers may require different forms of assessment to
identify higher flows required for such processes as fish spawning and sustaining the
ecology of wetlands and estuaries. In heavily regulated rivers where flows can be
affected all year by dams or high water use, it is important to understand the timing
and duration of high flow events required to maintain the vitality of river systems.
Water chemistry, temperature, nutrient cycling, oxygen availability, and the
geomorphic processes that shape river channels and floodplains are often tightly
coupled to streamflow variation. Natural flow regimes are therefore intimately linked
to many different aspects of ecological integrity. In the river channel, low flows will
concentrate fish and other aquatic organisms, benefiting predators such as larger
fish. If low flows are too severe or last for too long due to human influences, large
numbers of individuals may perish and jeopardize the local populations of certain
species. Thus, rather than trying to prescribe a flow regime that benefits some species
all of the time, a better approach is to restore or sustain a flow regime that benefits
each species some of the time. The species that are found in each river have endured
many trials of adverse flow conditions, exploited many occasions of favorable flow,
and have managed to persist in their native rivers over long periods of time. Most
environmentalists believe that perpetuation of the natural flow regime is the best
approach for conserving the full richness of the biological diversity of a river.
11.16 WATER REUSE

Reused water can be added to the supply system as an additional source of water,
thus having a quantitative value, or certain portions of water resources can be
reutilized that might otherwise be classified as wastewater, thus having qualitative
value. Reused water has been utilized for agricultural and landscaping irrigation,
industrial processes, and cooling water while complying with environmental
instream flow requirements, groundwater recharge, and direct consumption. Appli-
cations of reused water have been increasing as a result of severe droughts and water
pollution control regulations (Lund et al., 1995).
Water reuse requires careful control of treatment processes and can be more
expensive than the use of freshwater. In evaluating the cost of reuse as a water supply
source, the costs of additional treatment, the redistribution system, and operation
and maintenance should be considered. The cost of treatment can make it impossible
for small communities to reuse water, but large communities may be able to increase
their water supply by 50% or more by reusing sewage. Water reuse can be classified
as direct and indirect reuse:
• Indirect reuse, in which water is taken from a river, lake, or underground

aquifer that contains sewage. The practice of discharging sewage to sur-
face waters and withdrawal for reuse allows the processes of natural
purification to take place.
• Direct reuse, which is the planned and deliberate use of treated wastewater
for some beneficial purpose. Direct reuse of reclaimed waters is practiced
for several applications without dilution in natural water resources.
Water can be reused for the following purposes:
• Agricultural irrigation. The largest user of reused water is agriculture.

The original form of sewage treatment was sewage farming, or the disposal
of sewage on farmlands. Most sewage farms were replaced 70 to 90 years
ago by biological treatment plants, which discharge to the nearest water-
course (Dean and Lund, 1981). Reused water can also be used for urban
irrigation. Irrigation with raw or partially treated sewage can conserve

water and fertilize crops economically by capturing nutrients that would
normally be wasted. This irrigation method is also an effective way to
prevent contamination of nearby waterways with disease organisms such
as coliform and other bacteria that sewage contains. The most serious
drawback of using sewage for irrigation is its role in transmitting infec-
tious diseases to agricultural workers and the general public. Infections
related to Ascaris and Trichuris worms are commonly associated with
wastewater irrigation. Eating uncooked vegetables that have been irrigated
with such water could transmit these worms. Wastewater irrigation has
also been linked to transmission of enteric diseases such as cholera and
typhoid, even in areas where these diseases are not endemic (World
Resources, 1998). The advantages of reclaiming wastewater and sewage
for irrigation can be summarized as follows:
• Increasing crop yield
• Reduction of pollution of freshwater resources by sewage discharge
• Decreased use of fertilizers
• Soil conservation
• Reduced cost of sewerage and plumbing
• Urban irrigation. Urban irrigation includes parks, golf courses, and land-
scape medians. Limited exposure risk is presented by areas such as golf
courses, cemeteries, and highway medians, where public access is
restricted and where water is applied only during night hours without
airborne drift or surface runoff into public areas. Other urban uses include
toilet flushing, fire protection, and construction.
• Intentional reuse. Reused water forms part of a potable water supply. The
term “intentional” indicates that at least part of the treatment has been
given because the wastewater will soon become part of the supply of
potable water.
• Groundwater recharge. Recharging underground aquifers with treated
sewage is one of the most generally accepted forms of water reuse. Water
reuse by groundwater recharge is widely believed to entail a lesser degree
of risk than other means of reuse in which the recycling connection is
more direct. The degree of treatment depends on the type of application
to the soil, soil formation and chemistry, depth to groundwater table,
dilution available, and residence time to the point of first extraction.
• Recreation. Wastewater is sometime used to fill lakes as part of a recre-
ational system. Biological treatment in a series of lagoons usually provides
the quality required for recreational purposes.
• Industrial use. The quantity of water used in many manufacturing pro-
cesses and power generation is very large. The treated wastewater is an
appropriate source of water for such industries, especially for regions
facing water shortages. The water necessary for cooling purposes does
not have to be very pure and in some industries can be supplied from
treated wastewater.
Various methods have been developed for wastewater treatment, and a brief expla-
nation about these methods and their efficiency in treating sewage from different
sources is presented next.
11.17 TREATMENT OPTIONS FOR WASTEWATER REUSE

The main purpose of municipal wastewater treatment is to prevent pollution of the
receiving water bodies. The degree of required treatment can be determined based
on the beneficial uses of the receiving water bodies. Stream pollution and lake
eutrophication resulting from municipal wastewater recharge are particularly trou-
blesome in water reuse projects. Conventional municipal wastewater treatment meth-
ods usually consist of the following phases:
• Preliminary treatment
• Primary treatment
• Secondary treatment
Table 11.4 shows a summary of treatment processes required to meet health criteria
for wastewater reuse.
11.17.1 PRELIMINARY PROCESS

The preliminary process generally includes pumping, screening, and grit removal.
The setup of different units of preliminary treatment depends upon raw-wastewater
characteristics, subsequent treatment processes, and the preliminary steps employed.
Pretreatment is also necessary when sewage is highly foul smelling because of the
presence of hydrogen sulfide, mercaptans, and amines. Chlorine, hydrogen peroxide,
chlorine dioxide, and ferric salts can be used to deodorize the sewage. Solids can
also be removed in this phase by screening. Flotation is used to remove grease and
fats and it can be performed in a pre-aeration tank also used for grit removal.
Flocculation may also be practiced on high-strength municipal wastes to provide
increased primary removal and prevent excessive loads on the secondary treatment
process. Possible arrangements of preliminary treatment units for municipal waste-
water treatment are shown in Figure 11.6.
11.17.2 PRIMARY TREATMENT

In primary treatment, sewage is kept in settling tanks for a specific detention time.
Completely mixed activated-sludge processes can be used to treat unsettled raw
wastewater; however, this method is restricted to small municipalities because of
the costs involved in sludge disposal and operation. Primary sedimentation can
remove 30 to 50% of the suspended solids in raw municipal wastewater. A good
percentage of biological oxygen demand (BOD) and suspended solids (SS) are
removed in this phase. In some cases, lime or ferric salts are added to increase the
pH to about 11 in order to precipitate calcium carbonate or calcium phosphate.
516
TABLE 11.4
Treatment Processes Needed to Meet the Health Criteria for Wastewater Reuse
Irrigation Recreational Use Municipal Use
Crops Not for Direct Crops Eaten Cooked; Crops Eaten Industrial
Human Consumption Fish Culture Raw No Contact Contact Use Nonpotable Potable
Health criteria A+F B+F/D+F D+F B D+G C/D C E

Primary treatment Essential Essential Essential Essential Essential Essential Essential Essential
Secondary Essential Essential Essential Essential Essential Essential Essential
treatment
Sand filtration or Sometimes necessary Sometimes Essential Sometimes Essential May also be
equivalent necessary necessary essential
polishing methods
Nitrification Sometimes Essential
necessary
Denitrification Sometimes May also be
necessary essential
Chemical Sometimes May also be
clarification necessary essential

Carbon adsorption May also be
essential
Ion exchange or Sometimes May also be
other means of necessary essential
removing ions
Disinfection Sometimes necessary Essential Sometimes Essential Sometimes Essential Essentiala
necessary necessary
Note: A = freedom from gross solids and significant removal of parasite eggs; B = same as A plus significant removal of bacteria; C = same as A plus more effective removal of bacteria
plus removal of viruses; D = not more than 100 coliform organisms per 100 ml in 80% of samples; E = no fecal coliform organisms in 100 ml plus no virus particles in 1000 ml plus
no toxic effects on humans plus other drinking water criteria; F = no chemicals that lead to undesirable residues in crops or fish; and G = no chemicals that lead to irritation of mucous
membranes or skin.
a Free chlorine in one hour.
Source: WHO, Reuse of Effluents: Methods of Wastewater Treatment and Health Safeguards, Technical Report 517, World Health Organization, Geneva, 1973.
Combined Sanitary
wastewater wastewater
Hand-cleaned Mechanically Communicator

coarse bar cleaned bar with bypass
rack screen bar screen
Mechanically Variable speed

cleaned bar Partial flume
pumps
screen
Grit-removal Communicator Lift pumps

unit
Aerated grit
Partial flume chamber
Partial
flume
Preaeration
and grit
Lift pumps removal
To primary
treatment
FIGURE 11.6 Three possible arrangements for preliminary treatment units in municipal
wastewater processes. (From Viessman, W., Jr. and Hammer, M. J., Water Supply and Pollution
Control, Harper & Row, New York, 1985.)
11.17.3 SECONDARY TREATMENT

Secondary treatment is also known as biological or chemical treatment. The follow-
ing processes and methods have been widely used for secondary treatment:
• Activated sludge processes

• Trickling filters
• Biological towers
In the activated sludge process, wastewater is continuously fed into an aerated tank,
where microorganisms synthesize the organics. The resulting microbial floc (acti-
vated sludge) is settled from the aerated mixed liquor under quiescent conditions in
a final clarifier and returned to the aeration tank. Advantages of liquid aeration are
high BOD removal, ability to treat high-strength wastewater, and adaptability for
future use in plant conversion to advanced treatment.
A high degree of operational control is needed in secondary treatment. Shock

loads may upset the stability of the biological process, and sustained hydraulic or
organic overloading results in process failure. The effluent is made to pass through
trickling filters and rotating biological contactors, which provide a media to support
the microbial films. Oxygen is supplied from air moving through voids in the media.
The residual solids settle out in the clarifier and are removed as stable sludge.
Trickling filters act as contact beds where settled sewage is spread by a rotary
distributor over relatively deep (about 2 m) circular beds containing stones 50 to
100 mm in diameter. A typical trickling filter plant consists of primary settling tanks,
a trickling filter, final settling tanks, and a recirculation system to maintain a mini-
mum flow through the filter in order to operate the distributor. Trickling filters are
more resistant to shock from high organic and hydraulic loads and less expensive
to operate than activated sludge plants, but they also cost more to build. Deeper beds
up to 12 m deep, biological towers, have high-void-ratio plastic media to reduce
bed plugging and improve treatment efficiency.
Current models of rotating biological contactors (RBCs) are large-diameter
polyethylene media with about 40% of the surface area submerged in the waste and
rotated slowly on a horizontal shaft in a concrete tank. A biomass film about 1 to 4
mm thick grows on the surface and shears off periodically into the tank below, to
be removed by final sedimentation. The process may have oxygen transfer limitations
with high-strength wastes; however, for most wastes this method has the lowest
power requirement of any competitive biological process (Henry and Heinke, 1989).
11.17.4 ADVANCED WASTEWATER TREATMENT

Advanced wastewater treatment (often called tertiary treatment) methods remove
more contaminants compared with the above-mentioned conventional methods.
Common advanced water and wastewater treatment methods remove:
• Heavy metals
• Organic chemicals
• Suspended solids
• Inorganic salts
• Phosphorus
• Pathogens
They also oxidize ammonia to nitrate. Lime precipitation is one of the suggested
methods for removing heavy metals. Applying lime to raw wastewater precipitates
phosphates and hardness cations along with organic matter. Suspended solids can
also be removed by granular-media filtration. Phosphorus, like the majority of
soluble compounds in wastewater, is removed only sparingly by plain sedimentation.
About 20 to 40% of phosphorus can be removed through conventional treatment
methods. Chemical precipitation using aluminum or iron coagulants is effective in
phosphate removal. Chemical/biological phosphorus removal processes also com-
bine chemical precipitation with biological removal of organic matter. Alum or iron
salts are added prior to the primary clarification, directly during the biological
process, or prior to final clarification.
Where the ammonia content of the effluent causes pollution of the receiving
watercourse, nitrification of wastewater is performed. The process does not remove
the nitrogen but converts it to the nitrate form. Nitrate can be reduced to nitrogen
gas by heterotrophic bacteria in an anoxic environment. An organic carbon source
is required to act as a hydrogen donor and to supply carbon biological synthesis. In
biological nitrification and denitrification, unoxidized organic matter can be used as
an oxygen acceptor for conversion of nitrate to nitrogen gas.
Pathogens include all categories of microorganisms such as viruses, bacteria,
protozoa, and helminths. Disinfection of biologically treated wastewater requires
coagulation and granular-media filtration followed by chlorination with an extended
contact time. Cysts of protozoa and helminth eggs are resistant to chlorine, and they
must be physically removed by effective chemical coagulation and granular-media
filtration.
11.18 GRAY WATER: DEFINITION AND REUSE ISSUES

Municipal wastewater has been classified as:
• Black water — toilet wastewater (in no-mix toilets, black water can also
be classified as brown water or yellow water).
• Gray water — household wastewater without black water
Urban sanitation is the first step for controlling water pollution in urban areas. For
this purpose, sewage systems were introduced in the middle of the 19th century.
They consisted of single-pipe drainage with water-flushed toilets for sanitation. In
large cities around the world, surface runoff is also accepted by the single-pipe
drainage system that transfers all urban water into a sewage treatment plant. Large
amounts of surface runoff containing a variety of pollutants cause inefficiency of
most sewage treatment plants.
In a sustainable society, strong coordination between urban and rural components
for nutrient and biomass recycling is required. After consumption, the question arises
as to how to deal with the resulting urine and feces and include them in the recycling
of nutrients and the biomass. One solution is diversion of urine from the conventional
sewage pipe drainage outlet. This process of ecological sanitation utilizes a dry
toilet to divert urine at the source (Winblad et al., 1998). The diverted urine can be
then transferred to the nutrient processing plants, where the nitrogen and phosphorus
are recovered as chemical fertilizers.
If the urine and feces are collected and used for agriculture, sewage character-
istics may change to so-called gray water, which is easier to treat by conventional
wastewater treatment processes. The collected gray water can be treated locally and
discharged to surface waters in the vicinity of the collection points (Maksimovic
and Tejada-Guibert, 2001). The nutrient content of gray water is comparable to
waters that by different standards are regarded as “clean” (Gunther, 1996).
A gray-water reuse system should be able to receive the effluent from one or
more households in all seasons of the year. Where saturation of garden soils of low
permeability occurs in winter rainfall, a facility to divert effluent to sewers or
alternative disposal should be present. The system must protect public health and
the environment, meet community expectations, and be cost effective (Anda et al.,
1996).
Efficient reuse of wastewater and nutrients requires new sanitation designs.
Conventional wastewater systems can often handle gray water and municipal waste-
water without urine in a reuse-efficient way. Otterpohl and Grottker (1996) suggested
the following objectives for designing sanitation systems:
• Sanitation systems must strictly avoid spreading of diseases.

• Nutrient contents of human wastes should be completely recycled to food
production (food and water cycles should be kept separate).
• Sanitation systems should have low water/energy consumption and should
not promote accumulation of nutrients in ecosystems.
• The use and operation of systems must comply with social and religious
rules.
• Costs of installation should be reasonable (an estimate of potential envi-
ronmental damages must be included).
• Operation should be simple and possible not only for highly specialized
experts.
Simple biological treatments can be applied to gray waters. Pretreatment is

necessary for separation of solids or grease from kitchen wastewater. Activated
sludge systems that rely on settling of sludge can cause problems with the lack of
nutrients; therefore, biofilm methods such as trickling filters, biofiltration, rotating
disks, or constructed wetlands (bio-sandfilters) are advantageous. Even though gray
water does not contain toilet wastewater, it may contain some fecal pathogens from
showers and washing machines which makes treatment methods based on sand
filtration and membranes preferable. Gray water contains a high amount of household
chemicals, thus producing high-quality effluents and reuse without restrictions
becomes simpler with the use of household chemicals that can be easily mineralized,
not simply degraded (Maksimovic and Tejada-Guibert, 2001).
11.19 CONFLICT ISSUES IN WATER DEMAND

MANAGEMENT
Water resources systems generally have multiple objectives, and many of those
objectives can conflict. Supplying demand is a main objective of most water
resources systems. A common conflict issue in water resources systems planning
and operation occurs when the system should supply water to a number of demand
points for different purposes with some constraints on water quality, flood control,
etc. In many large cities of the world, water demands are raised to a level that
resources are not capable of supplying. Reduction of water demands by water

demand management is appropriate in such cases.
In addition to diversity in the objectives of users of water resources, important
issues also arise from an institutional point of view. For example, a Department of
Agriculture might be interested in supplying irrigation demands and water needed
for livestock, while a Department of Industries might be interested in supplying
water required for manufacturing processes, cooling systems, etc. In municipalities,
responsible agencies are more concerned with increasing the reliability of supplying
urban and rural water demands. Thus, the options for water demand management
should be selected and applied by the various decision makers who usually come
from different agencies and might have different levels of authority to apply their
favorite policies. The following example shows how demand management alterna-
tives can be selected by the Nash theorem for resolution of conflict issues.
Example 11.1
Assume city A is located near river X and municipal water demands of this city are
supplied from this river. Industrial complex B is also located near this river. Water
required for the manufacturing processes and cooling systems of the industries
located in this complex is also supplied by a water diversion system implemented
downstream of the diversion point for city A. Remaining flow in the river supplies
the irrigation demands of agricultural lands located downstream of industrial com-
plex B. The river is capable of supplying total water demands in high flow seasons.
For simplicity, the average monthly streamflow in the river and the municipal,
agricultural, and industrial water demands are as shown in Table 11.5. As can be
seen in this table, about 33% of the water demands cannot be supplied by the river.
The following options are proposed for demand management by the responsible
agencies:
• Department of Water Supply

• Public education in the city and vicinities
• Reduction of unaccounted-for water by leakage control
• Urban irrigation by extracted water from groundwater aquifer
TABLE 11.5
Average Monthly Streamflow in River X and
Water Demands for Example 11.1
Volume
(million m3)
Streamflow in river X 40
Municipal water demand 20
Industrial water demand 15
Agricultural water demand 25
• Implementation of a system for recycling water for cooling systems in
the complex
• Modification of the manufacturing process
• Implementation of drip and pressure irrigation system (for increasing
irrigation efficiency)
• Education and training programs for farmers
Table 11.6 shows the estimated costs and expected percentage of reduction in water
demands for each sector. Each of the above-mentioned agencies has proposed a
favorite range of investment for demand management based on the costs of each of
these options and the benefits associated with supplying water demands with higher
reliability. The results of these proposals are summarized in Figures 11.7 to 11.9,
where the most favorable range of investment by each organization is indicated by
a utility equal to 1. Other investment levels are ranked by lower values of utility.
The highest possible level of investment is assigned by each organization for 0 utility.
Find the optimal set of demand management options to be able to supply the whole
demands of the system by the river.
TABLE 11.6
Estimated Costs and Expected Percentage of Reduction in the Water
Demands for Each Sector
Percentage of Reduction
in Water Demands Estimated Cost
Option for Demand Management of Each Sector (monetary units)
Public education in the city and vicinities 1.5 0.8

Reduction of unaccounted-for water by 15 20.50
leakage control
Urban irrigation by extracted water from 4 3.5
groundwater aquifer
Implementation of a system for recycling 45 12
water in cooling systems in the complex
Modification of manufacturing process 12 18
Implementation of drip and pressure 38 9.5
irrigation system (for increasing
Education and training programs for 8 0.6
farmers
Utility
1.0
0.5
0 5 10 15 20 22 Investment
(monetary unit)
FIGURE 11.7 Utility function of the Department of Water Supply for investing money in
municipal water demand management.
Utility
1.0
0.7
0 10 20 25 30 Investment
(monetary unit)
FIGURE 11.8 Utility function of the Department of Industries for investing money in indus-
trial water demand management.
Utility
1.0
0.8
0 5 8 15 20 Investment
(monetary unit)
FIGURE 11.9 Utility function of the Department of Agriculture for investing money in
agricultural water demand management.
Solution: The nonsymmetric Nash theorem has been used and the optimization
problem for finding the optimal set of demand management options is formulated
as follows:
3
i =1
i i
wi
Subject to di ≤ fi ≤ fi∗ ∀i
3 NOi
∑∑ Ri =1 j =1
i, j ≥ 20
where wi is the relative authority of each agency, fi is the utility function, di is the
disagreement point, and fi* is the ideal point of player (agency) i. In this example,
the relative authority of each agency is considered to be equal to one; therefore, the
problem is simplified to a symmetric problem. The disagreement point is also set
to zero. The above equations show that the total water demand reduction of the
system should be more than 20 million m3, which is the mean monthly deficit in
the current situation. In this equation, Ri,j is the volume of expected water demand
reduction for option j (j = 1, …, NOi), proposed for sector i. NOi is the total number
of water demand management options proposed for sector i. Table 11.7 shows the
TABLE 11.7
Optimal Set of Water Demand Management Options for Example 11.1
Total Total Water
Water Demand Reduction in Estimated Demand
Management Water Demand Cost for Each Reduction
Sector Option (million m3) Sector Utility (million m3)
Municipal Public education in the 0.3 21.3 0.041 3.3

city and vicinity
Reduction of 3
unaccounted-for water
by leakage control
Industrial Implementation of a 6.75 12 0.96 6.75
system for recycling
water in the cooling
systems in the complex
Agricultural Implementation of a drip 9.5 10.4 0.66 11.50
and pressure irrigation
system (for increasing
Education and training 2 — — —
programs for farmers
Total reduction in water demand 21.55
optimal set of water demand management options found by solving the above
formulation. As can be seen in this table, the total reduction in water demand is
21.55 million m3, which is higher than the deficit of 20 million m3 in the current
situation.
11.20 PROBLEMS
11.1 Assume that you want to model the water demand of a city in order to
forecast the municipal demands over the next several years. Historical
data for water use, population, price, and precipitation for a 10-year time
horizon are shown in the following table. Formulate a multiple regression
model for estimating the water use data and comment on selecting the
independent variables.
Annual
Price Precipitation Water Use
Year Population (monetary unit/m3) (mm) (m3/year)
1991 21,603 0.25 1015 2,540,513

1992 22,004 0.27 810 2,754,901
1993 23,017 0.29 838 2,722,911
1994 23,701 0.30 965 2,832,270
1995 24,430 0.31 838 2,904,727
1996 25,186 0.32 787 3,004,690
1997 26,001 0.33 720 3,309,927
1998 26,825 0.35 695 3,382,633
1999 27,781 0.31 820 3,375,392
2000 28,576 0.34 800 3,406,259
11.2 Agricultural lands near a river are irrigated by water diverted from the
river. The total area of irrigated lands is 400 ha (4,000,000 m2). Use the
following notation to formulate an optimization problem for finding the
optimal crop mix. Define the objective function so that as the shortages
in supplying irrigation demands are minimized:
Di,j is the irrigation demand of crop i in month j.
Qi,j is the water allocated to crop i in month j.
Ai is the area of the field allocated to crop i.
α is the water supply (irrigation) efficiency
Ij is the river discharge in month j.
11.3 Use the following data to solve Problem 11.2. Assume that the irrigation
efficiency is 65%.
Net Water Demand (mm)

Crop Sept. Oct. Nov. Dec. Jan. Feb. Mar. Apr. May June July Aug.
Wheat — 14 5 4 2 1.5 33 142 138 2 — —

Alfalfa — 14 5 5 1 2 33 141 97 — — —
Cotton 114 61 — — — — — 24 82 190 225 186
River Discharge (m3/sec)

Sept. Oct. Nov. Dec. Jan. Feb. Mar. Apr. May Jun. Jul. Aug.
0.2 0.35 1 1.5 1.7 0.8 0.75 0.6 0.36 0.25 0.2 0.15
11.4 In Problem 11.2, formulate the optimization problem for maximizing the
real-time use of river discharge.
11.5 Use the data for Problem 11.3 to solve Problem 11.4.
References
Anda, M., Ho, G., Mathew, K., and Monk, E., Greywater reuse options: areas for further
research in Australia, in Integrated Water Management in Urban Areas, Niemczynow-
icz, J., Ed., United Nations Educational, Scientific, and Cultural Organization
(UNESCO)-IHP, New York, 1996, pp. 347–356.
Bhatia, R. and Falkenmark, M., Water Resources Policies and the Urban Poor: Innovative
Approach and Policy Imperatives, Water and Sanitation Currents, United Nations
Development Program-World Bank Water and Sanitation Program, New York, 1993.
Blaney, H. F. and Criddle, W. D., Determining water requirements in irrigated areas from
climatological data, U.S. Department of Agriculture, Soil Conservation Service,
Washington, D.C., 1945.
Bouwer, H., Role of geopurification in future water management, in Soil and Water Science:
Key to Understanding Our Global Environment, Soil Society of America Special
Publication 41, Cincinnati, OH, 1994, p. 7381.
Dean, R. B. and Lund, E., Water Reuse: Problems and Solutions, Academic Press, San Diego,
CA, 1981.
Falkland, A., Ed., Hydrology and Water Resources of Small Islands: A Practical Guide, United
Nations Educational, Scientific, and Cultural Organization (UNESCO), New York,
1991.
Gunther, F., Simplifying wastewater treatment by source separation outline for a biologically
based greywater purification plant, in Integrated Water Management in Urban Areas,
Niemczynowicz, J., United Nations Educational, Scientific, and Cultural Organization
(UNESCO)-IHP, New York, 1996, pp. 301–314.
Hammer, M. J. and Hammer, M. J., Jr., Water and Wastewater Technology, Prentice-Hall,
Englewood Cliffs, NJ, 1996.
Hanson, G. and Nilsson, A., Groundwater dams for rural-water supplies in developing coun-
tries, Groundwater, 24(4), 497–506, 1986.
Henry, J. G. and Heinke, G. W., Environmental Science and Engineering, Prentice-Hall,

Englewood Cliffs, NJ, 1989.
Lund, J. R., Jenkins, M., and Wilchfort, O., Integrated Planning and Management for Urban
Water Supplies Considering Multiple Uncertainties, report prepared by the Depart-
ment of Civil and Environmental Engineering, University of California, Project Num-
ber UCLA-WRC-W-813, 1995.
Maksimovic, C. and Tejada-Guibert, J. A., Frontiers in Urban Water Management, IWA
Publishing, London, 2001.
Mays, L. W., Ed., Water Resources Handbook, McGraw-Hill, New York, 1996.
Mays, L. W. and Tung, Y.-K., Hydrosystems Engineering and Management, McGraw-Hill,
New York, 1992.
Nilsson, A., Groundwater Dams for Small Scale Water Supply, Intermediate Technology
Publications, London, 1988, 64 pp.
Otterpohl, R. and Grottker, M., Possible sustainable sanitation concepts for urban areas,
Integrated Water Management in Urban Areas, Niemczynowicz, J., Ed., United
Nations Educational, Scientific, and Cultural Organization (UNESCO)-IHP, New
York, 1996, pp. 269–278.
Penman, H. L., Natural evaporation from open water, bare soil, and grass, Proc. Royal Soc.
(London), A, 193, 120–145, 1948.
Solley, W. B., Pierce, R. R., and Perlman, H. A., Estimated Use of Water Resources in the
United States in 1990, U.S. Geological Survey Circular 1081, Washington, D.C.,
1993.
Thomas, H. A., Improved Methods for National Water Assessment, Contract WR 1524927,
Harvard Water Resources Group, Harvard University, Cambridge, MA, 1981.
Thornthwaite, C. W. and Mather, J. R., The water balance, Publication in Climatology VIII,
Drexel Institute of Technology, New Jersey, 1955.
Todd, D. K., Groundwater Hydrology, John Wiley & Sons, New York, 1980.
United Nations, The Demand for Water: Procedures and Methodologies for Projecting Water
Demands in the Context of Regional and National Planning, Department of Economic
and Social Affairs, United Nations, New York, 1976.
United Nations, Towards Efficient Water Use in Urban Areas in Asia and the Pacific, Economic
and Social Commission for Asia and the Pacific, United Nations, New York, 1998.
Viessman, W., Jr. and Hammer, M. J., Water Supply and Pollution Control, Harper & Row,
New York, 1985.
Wegelin-Schuringa, M., Water Demand Management and the Urban Poor, IRC International
Water and Sanitation Center, at WHO at Kobe Center, 2000.
Winblad, U., Ed., Ecological Sanitation, SIDA, Stockholm, 1998.
WHO, Reuse of Effluents: Methods of Wastewater Treatment and Health Safeguards, Technical
Report 517, World Health Organization, Geneva, 1973.
Programme, The United Nations Development Programme, and The World Bank,
Oxford University Press, London, 1998.
12 Drought Management
12.1 INTRODUCTION
Population increases and expansion of agricultural and industrial sectors in many
parts of the world have resulted in a growing demand for water. The impact of climate
change and widespread water contamination has resulted in many irregularities in
the supply and demand structure. Drought events have occurred with smaller recur-
rence intervals with higher peaks and levels of severity. Many limitations have been
imposed on water reuse due to quality and environmental restrictions (see Lettenmiare
et al., 1996; Aswathanarayana, 2001). All of these factors have led to a need for
integrated water resources management that takes into account all of the system
components, the interactions between them, and analyses of risk and uncertainty.
In recent years, much attention has been given to labeling droughts based on
precipitation data. Precipitation data alone, however, are not sufficient indicators for
detecting real droughts based on hydrologic conditions. In fact, the difference
between actual demand and user perceptions about water use is the key measure
used to evaluate the severity of a drought by hydrologists, meteorologists, water
resources engineers, agricultural researchers, economists, and other stakeholders.
Integrated water resources management in a region requires a detailed analysis
of normal conditions as well as dry and wet periods. In North America, drought has
not been a main focus of water resources studies due to relatively high per capita
resources. Recently, though, some attention has been given to droughts in the United
States due to water shortages in some areas, such as along the Mexican border.
Agreements between the United States and Mexico in regard to water allocation for
the Conchos Basin in Mexico and the Rio Grande basin in the United States have
been a key factor in promoting drought studies.
This chapter discusses types of drought events, including climatic, hydrologic,
and agricultural drought and demonstrates drought analysis and prediction as tools
for drought management studies.
12.2 DROUGHT DEFINITION

In general, droughts can be divided into three specific groups: climatic droughts,
hydrological droughts, and agricultural droughts. Drought that results from a rainfall
deficit is climatic drought. For industries and water consumers, drought can occur
when the water supply declines; hydrological drought is a result of deficit in effective
rainfall in a period compared to previous periods. For farmers, drought is a condition
that reduces their harvests; agricultural drought usually is defined as a reduction of
agricultural products and loss of plants. Continuation of rainfall deficits may cause
529
a permanent shortage in soil moisture in the growth season; therefore, the average
shortage in soil moisture over a period of several months for crops that are sensitive
to soil moisture and have high water demands can be used as an indicator of drought
severity.
A number of definitions of drought events have been suggested by researchers.
Yevjevich (1967) stated that if a widespread region is affected by a severe water
shortage for a long time, then it is possible to distinguish drought from other events
such as water shortage in that region. Dracup et al. (1980) stated that it is not possible
to define and analyze drought without paying attention to the nature of water
shortage, sequence of time for discretization of drought and truncation levels that
are used to separate drought from other phenomena, and regional aspects of the
drought.
Based on these definitions, droughts can be divided into two categories: point
(localized) or regional. In the point drought analysis, drought in station or small
area is investigated. Yevjevich (1967), Griffits (1990), and Beric et al. (1990) have
studied point drought. In point investigations, a time series of hydroclimatological
factors or a drought indicator is evaluated at a specific point. A time sequence and
then a truncation level are assigned to determine drought events in a time series in
order to perform the analytical evaluation.
According to Yevjevich (1967), the application of run theory is a fundamental
method of analyzing point drought. According to this theory, one could detect a
consecutive period of shortage or surplus after choosing a truncation level for existing
data by assuming that they are stationary. Then, from detected consecutive shortage
periods, probability distributions of drought parameters such as duration and the
amount of shortage are evaluated (Guerrero-Salazer and Yevjevich, 1975). Each of
these investigators chose a truncation level based on their specific goals. For example,
Sen used various probabilities of successful events (wet years) and failure events
(shortage). Griffits (1990) used average and Beric et al. (1990) used consecutive
time intervals with no rain (15 days) in the growth season as truncation levels.
In the study of regional drought, severe water shortage conditions in contiguous
and extended regions are investigated (e.g., Santos, 1983; Sen, 1980). Santos (1983)
suggested that, in addition to defining truncation level, a critical area should also be
defined. Based on this approach, regional drought occurs when the drought-affected
area is greater than the critical area. Many investigators such as Tase and Yevjevich
used standardized normal distribution values for drought events with different return
periods as a truncation level, assuming that all the specified rain gauges in the study
area follow the normal distribution. Santos (1983), who has done intensive studies
on the analysis of regional droughts, used 10 to 20% of average regional precipitation
as a truncation level.
12.3 DROUGHT PREPAREDNESS/MANAGEMENT

The flowchart of Figure 12.1 describes the types of droughts and how they can be
integrated into a drought management/preparedness algorithm (Karamouz et al.,
2002). The primary objective of this algorithm is to put climatic, hydrologic, and
agricultural droughts into some perspective to understand how they can be related
Drought Management
FIGURE 12.1 Algorithm for drought preparedness/management. (From Karamouz, M. et al., Proc. 2002 EWRI Conference on Water Resources Planning
and Management, Roanoke, VA, 2002.)
531
to real drought and how their results can be used for development of water resource
strategies in a region. The factors affecting hydrologic droughts are so complex and
interrelated that they cannot be modeled unless a precise water balance model is
developed. Such a water balance model can be used to make a realistic assessment
of rainfall, runoff, changes in temperature, and soil moisture conditions. It is impos-
sible to assess the occurrence and continuity of drought events with reasonable
certainty. The results of climatic drought, especially on a regional scale, as well as
assessment of effective rainfall, moving averages, and determination of drought-
affected areas can guide short- and mid-range water and agricultural resources
management. Topics such as drought prediction, forecasting, and monitoring can be
investigated once the types of droughts, their thresholds, and how they are related
are determined.
12.4 CLIMATIC DROUGHT

12.4.1 POINT EVALUATION
Climatic drought is defined based on rainfall deficit. In point evaluation of climatic
drought to detect dry events, the best probabilistic distributions should be chosen to
describe the behavior and variation of precipitation at each sub-basin represented
by a rain gauge. The concept of effective rainfall and excess deficit and surplus on
a monthly scale can be extended to determine the onset, duration, and the severity
of drought. The following steps could be taken in point evaluation of climatic
drought:
• Fitting statistical distribution

• Analysis of moving averages
• Point evaluation of drought events in monthly scale
12.4.1.1 Fitting Statistical Distribution
Assume that {Xi , i=1, …, N} is a discrete point series vector of a drought indicator
in N existing rain gauges of the study area. In analysis of climatic drought, the
drought indicator is precipitation and the time sequence is usually considered to be
annual. To choose the best statistical distribution for the time series of each station,
Xi, the following assumptions should be verified:
• Data are independent.

• Precipitation time series are stationary.
• Xi may have different probability distributions.
To choose the best statistical distribution, the classical approach can be used; how-
ever, it has many shortcomings. In this approach, first all of the existing data for
point (station) i (Xi) should be sorted in ascending order to form a new set (x(1), x(2),
x(n)) in such a way that x(1) = min(Xi) and x(n) = max(Xi). Then, if the data are sufficient
(at least 30 values, as recommended by Kottegoda and Rosso, 1998; Mendenhall
Drought Management 533
et al., 1990), the range between x(1) and x(n) is divided into K intervals and a histogram
for point i can be plotted. The best distribution can be chosen by considering the
shape of the histogram and comparing it to the shape of different statistical distri-
butions and by using the result of different goodness of fit tests such as the Kol-
mogorov–Smirnov and chi-square tests. Alternatively, to choose the best distribution
the observed data distribution plot can be compared with the expected values from
a statistical distribution plot. Quantile–quantile (Q-Q) and probability–probability
(P-P) graphs are two methods used for comparing observed values with values from
a given distribution. By these methods, the best distributions can be found with more
flexibility and accuracy as compared with the classical method.
In general, it is possible to obtain (n – 1) values in the observed sample in such
a way as to divide the cumulative frequency of the sample equally into n values.
Each of these values is a quintile. The objective is to compare the observed value
of events with their corresponding values from a statistical distribution function. To
prepare a Q-Q graph, the observed values should be ordered (x1 ≤ x2 ≤ … ≤ xn) and
then these values are plotted against the inversion of the cumulative probability
distribution function as:
Q = F −1 ⎜
(
⎛ i−r
adj ) ⎞⎟ i = 1, 2,..., n (12.1)
(
⎜ n+n
⎝ adj ) ⎟⎠
where nadj and radj are constants and should be ≤ 0.5, and F–1 is the inverse of the
cumulative distribution function (CDF) for a statistical distribution. This yields a
graph of observed values vs. standardized expected values from a specific distribu-
tion. The best statistical distribution is the one that falls into a linear regression
relationship of high correlation coefficients with the distribution of observed values.
nadj and radj are limiting the p-values for the CDF to fall between 0 and 1.
To prepare a P-P graph, the cumulative distribution function of the observed
values is plotted vs. the CDF of a statistical distribution. To plot P-P graphs, after
sorting the observed values in ascending order, one axis is the CDF of the observation
i vs. a cumulative distribution function of a statistical distribution for the observed
values (Xi). If the selected statistical distribution estimates the observed values
closely, then in a P-P graph all of the points should be on a 45° line.
To find a good fit for statistical distribution, parameters of a distribution should
be calibrated. One of the most sensitive parameters is the location parameter, which
is a measure of the truncation level (threshold) for a given distribution. For example,
in a Weibull distribution, the location parameter has been used in the following
density function:
*e [
− ( x −θ ) b ]
f ( x ) = c b *[( x − θ) b]
c
c −1
0≤x<∞ (12.2)
b > 0; c > 0; θ > 0
where:
b is the scale parameter.

c is the shape parameter.
θ is the location parameter (threshold).
The feasible range for this distribution is between 0 and infinity. The location
parameter (θ) should be less than the smallest observed value. In this case, this value
is subtracted from all of the existing data and the Weibull distribution is then fitted.
The scale parameter (b) could also be considered. Variation of this parameter
can change the size of the distribution. As can be seen in Eq. (12.2), as b increases,
the size of the distribution (the peak value) becomes smaller. In this case, to keep
the area under the curve equal to 1, the distribution will spread out more. By changing
these parameters, we can obtain a better fit of a Weibull distribution to the observed
data. The other types of distributions have similar parameters.
Beta distribution is another suitable type of distribution for predicting the behav-
ior of precipitation data but has not been given much attention in previous drought
studies. This distribution may change from a normal distribution to a very skewed
distribution and offers the flexibility necessary for drought studies.
The beta distribution has a probability density function of:
[
f ( x ) = Γ( υ + ω ) [Γ( υ)Γ(ω )] * ( x − θ)
υ−1
* (σ + θ − x )
ω −1
]σ υ +ω +1
(12.3)
0 < x <1 ; υ > 0 ; ω > 0
where:
Γ is the gamma function.

υ and ω are the shape parameters.
θ is the threshold (location) parameter.
σ is the scale parameter.
Note that the shape of the beta distribution depends on the values of the parameters
ω and υ. The standardized beta distribution has a valid range from 0 to 1. Smaller
thresholds can be considered by standardizing the variable as (x – θ)/σ. Note that
the location parameter must be less than the smallest observed value, and the (σ +
θ) must be greater than the largest observed value. In general, if the points in the
Q-Q plot form a straight line, then the respective family of distributions (a beta
distribution with υ and ω parameters, in this case) provides a good fit to the data;
thus, the intercept and slope of the fitted line can be interpreted as graphical estimates
of the threshold (θ) and scale (σ) parameters, respectively.
12.4.1.2 Analysis of Moving Averages

In the previous section, the goal was to find the best distribution and the best criteria
to separate dry events from the others. In the moving average method, the cumulative
TABLE 12.1
Average Precipitation Data for 10 Years for Two Regions
(Example 12.1)
Average Precipitation Average Precipitation
Year of Region A (mm) of Region B (mm)
1 640 230
2 456 127
3 236 261
4 456 301
5 235 261
6 601 322
7 532 400
8 731 223
9 256 128
Current year 541 112
impact of precipitation deficits and surpluses in previous years are considered. By

using this method, the drought investigation is diverted from climatic drought to
hydrological (real) drought. The n-year moving average, X ′j in the jth year, can be
calculated from the following expression:
n −1
X ′j =
1
n ∑X
i=0
j +i n = 3,5,9,... (12.4)
Usually, 3-, 5-, and 10-year moving averages are investigated to recognize drought
spells and to predict their trend.
Example 12.1
Use the 3-year moving averages to compare the two series of annual average
precipitation for two different regions presented in Table 12.1. Which of these regions
is facing a severe drought in the current year? If the long-term average precipitation
in regions A and B is estimated to be 400 and 250 mm, respectively, discuss the
drought periods and their durations.
Solution: Table 12.2 shows the estimated 3-year moving averages for regions
A and B. Figures 12.2 and 12.3 show the 3-year moving average and long-term
average of precipitation for the two regions. Region A has experienced one drought
event of a 2-year duration. The current year in that region is experiencing a wet
spell. Region B was in the middle of a drought period for the first years of the
record, and the current year seems to be the first year of a drought event.
TABLE 12.2
3-Year Moving Average for Two Regions (Example 12.1)
3-Year Moving Average 3-Year Moving Average
Year for Region A (mm) for Region B (mm)
1 — —
2 — —
3 444.00 206.00
4 382.67 229.67
5 309.00 274.33
6 430.67 294.67
7 456.00 327.67
8 621.33 315.00
9 506.33 250.33
Current year 509.33 154.33
700.00
600.00
Precipitation (mm)
500.00
400.00
300.00
200.00 3-year moving average
100.00 long-term average
0.00
3 4 5 6 7 8 9 10
Year
FIGURE 12.2 Long-term average annual precipitation and 3-year moving average for
region A in Example 12.1.
350.00
300.00
Precipitation (mm)
250.00
200.00
150.00
100.00 3-year moving average
50.00 long-term average
0.00
3 4 5 6 7 8 9 10
Year
FIGURE 12.3 Long-term average annual precipitation and 3-year moving average for
region B in Example 12.1.
12.4.1.3 Point Evaluation of Drought Events on a Monthly Scale

In this method, which was initially proposed by Herbst et al. (1966), it is assumed
that farming in any region is adapted to the prevailing climatic pattern so that
maximum advantage is taken of the months of high average rainfall (Chang, 1990).
Due to variations in mean monthly rainfall, a seasonal drought of a certain severity
and duration is often a normal feature of that particular climate and as such should
not be included in the assessment of droughts with severe consequences. In the
following sections, it is assumed that the benefit to vegetation continues for some
time after excessive rain due to the storage of moisture in the soil and, conversely,
that the recovery of vegetation after a drought is not immediate and the impact
persists for some time.
12.4.1.3.1 Effective Rainfall and Rainfall Difference Series
In the literature, effective or excess rainfall refers to that portion of precipitation
that is converted to runoff, but in drought studies the definition of effective rainfall
is different and can be calculated as follows:
(
ERi,n = Mi,n + Wi ERi −1,n − Mi −1 ) (n = 1, …, NY ) , (i = 1,…,12)
(12.5)
ER1,1 = M1
⎛ ⎞
⎜
Wi = C⎜1 +
Mi ⎟ (12.6)
1 ⎟
⎜ × MAR ⎟
⎝ 12 ⎠
where:
MAR is the mean annual rainfall.

C is a constant.
ERi,n is the effective rainfall in month i of year n.
Mi,n is the rainfall in month i of year n.
Mi is the long-term monthly mean for month i (i = 1, …, 12).
Wi is the weighting factor for month i.
NY is the number of years of the record.
C is a constant that can be obtained by trial and error in order to better simulate the
observed wet and dry periods. In the previous studies a value of about 0.1 is selected
for C. The weighting factors obtained from this equation show the effect of rainfall
at each month on the drought situation of the next months. An average deficit ( DF i )
for month i is defined as:
NY
∑ DELTA i ,n
RDi = n =1
(i = 1,…,12) (12.7)
NY
Mi − ERi,n ERi,n < Mi

DELTAi,n = (n = 1, …, NY ) (12.8)
0 ERi,n ≥ Mi
The carryover from month to month is defined as the effect of rainfall deficit or
surplus from the previous month:
⎧⎪ ERi,n − Mi if ERi,n ≥ Mi
ERSi,n = ⎨ (12.9)
⎪⎩0 otherwise
ERDi,n = Mi − ERi,n − RDi (12.10)
[
RDi,n = Min ( Mi − ERi,n ), RDi ] (n = 1, …, NY ) (12.11)
where:
ERSi,n is the excess rainfall surplus in month i of year n.

ERDi,n is the excess rainfall deficit in month i of year n.
RDi,n is the rainfall deficit in month i of year n.
RDi is the long-term monthly mean for rainfall deficit in month i.
The carryover from month to month is determined by subtracting the mean rainfall
for a particular month from the effective rainfall for the same month so that a deficit
or surplus rainfall for that month is obtained. The deficit and surplus rainfall is
multiplied by a weighting factor for the next month and the product, whether negative
or positive, is added algebraically to the rainfall figures of that month. This sum is
the effective rainfall. The weighting factor is used to calculate the carryover effect
of a particular month and shows the relative effect of the deviation from the mean
in a month as it contributes to the soil moisture of the following month.
For these equations, the carryover is assumed to be 0 for the first month of the
record. The process is continued to obtain the effective monthly rainfall for the entire
period of the record. Excessive rainfall deficit (ERD) is of particular importance to
drought studies. This measure indicates how large the deficit of a system could
tolerate beyond the normal amount of deficits of the system naturally and over the
coarse of experienced historical data. This parameter is the main feature of calcu-
lating drought severity.
Example 12.2
The monthly rainfall data for a period of 24 months and the long-term monthly
average rainfall are presented in Table 12.3. The mean annual rainfall (MAR) is
estimated as 535.2 mm from long-term data. Assume C (the constant in Eq. (12.6))
TABLE 12.3
Rainfall Data for Example 12.2
Rainfall Mean Monthly Rainfall
Month (mm) (mm)
1 0.99 1.25
2 3.99 1.13
3 1.59 0.67
4 19.59 10.43
5 2.45 57.51
6 21.06 77.7
7 72.31 76.75
8 75.21 80.91
9 116.42 106.08
10 34.81 75.22
11 14.62 43.37
12 0.00 4.17
13 1.02 1.25
14 0.86 1.13
15 0 0.67
16 0.7 10.43
17 67.4 57.51
18 51.7 77.7
19 45.4 76.75
20 81.2 80.91
21 38.1 106.08
22 57.1 75.22
23 2.3 43.37
24 0.1 4.17
is 0.1. Calculate the effective rainfall, rainfall deficit, and excess rainfall deficit for
the region.
Solution: Equations (12.5) and (12.6) are used for estimating the effective
rainfall. Based on Eq. (12.5), the effective rainfall in the first month is considered
to be equal to the average rainfall in that month, which is 1.25 mm. For the second
month, the weighting factor can be estimated as follows:
⎛ ⎞
⎜ 1.13 ⎟
W2 = 0.1 × ⎜1 + ⎟ = 0.103
1
⎜ × 535.2 ⎟
⎝ 12 ⎠
Then, the effective rainfall for this month can be estimated using Eq. (12.5):
ER2,1 = 3.99 + 0.103 × (0.99 − 1.25) = 3.96 mm

Equations (12.7) and (12.8) are used for estimating the rainfall deficit. For example,
for the first month, it can be written:
ER1,1 = M1 < M 1 ⇒ DELTA1,1 = 1.25 − 0.99 = 0.26 mm
ER1,2 > M 1 ⇒ DELTA1,2 = 1.25 − 0.18 = 1.07 mm
Therefore,
∑ DELTA i ,n
0.26 + 1.07
RD1 = n =1
= = 0.665 mm
2 2
Using Eq. (12.11), it can be written that:
RD1,1 = Min[(0.99 − 0.99), 0.665] = 0.0 mm
The excess rainfall deficit for the first month is estimated using Eq. (12.10) as
follows:
ERD1,1 = 1.25 − 0.99 − 0.665 = 0.405 mm
Table 12.4 shows the values of effective rainfall, rainfall deficit, and excess rainfall
deficit estimated for the 15 months of data presented in Table 12.3.
12.4.1.3.2 Onset and Termination of Drought
In a test to signify the onset of a drought, it can be assumed that a single month
could constitute a drought only if no rain occurred in the month of highest mean
rainfall. The test is based on a comparison of the sum of the negative differences
(deficits) from the point in time the test begins, with a sliding scale of 12 values
calculated by linear interpolation between the maximum value of the mean monthly
rainfall (MMMR) and the mean annual deficit (MAD). A monthly increment x is thus
obtained from the following equation:
MAD − MMMR
x= (12.12)
11
where MMMR = Max[Mi ], and MAD is equal to the ratio of total deficits in the
historical period to the number of the years in which deficits have occurred.
The sliding scale is a line connecting the MAD to the MMMR. The first value
on the sliding scale is equal to the MMMR and is the maximum deficit that can
occur in a single month. The second value on the sliding scale is obtained by adding
x to the MMMR; the third, by adding 2x; and so on up to MMMR + 11x, which is
equivalent to the MAD. To start the test, it can be assumed that no drought prevailed
prior to the start of the available rainfall record. If the first month of available data
TABLE 12.4
Effective Rainfall, Rainfall Deficit, and Excess Rainfall Deficit Estimates
for Example 12.2
Effective Rainfall —
Effective Excess Rainfall Average Monthly
Rainfall Rainfall Deficit Deficit Rainfall
Month (mm) (mm) (mm) (mm)
1 0.99 0.00 0.13 –0.26

2 3.96 0.03 –2.18 2.83
3 1.88 –0.29 –0.85 1.21
4 19.74 –0.15 –8.13 9.31
5 4.58 –2.13 52.87 –52.93
6 6.55 6.21 69.10 –71.15
7 52.95 12.93 14.08 –23.80
8 68.51 6.70 3.72 –12.40
9 112.23 3.84 –9.82 6.15
10 36.46 –1.65 29.98 –38.76
11 6.98 7.50 28.96 –36.39
12 –3.98 3.98 3.51 –8.15
13 0.18 0.27 0.80 –1.07
14 0.75 –1.34 1.71 –0.38
15 –0.04 –0.43 1.14 –0.71
16 0.61 –2.20 12.02 –9.82
17 65.15 2.25 –17.05 7.64
18 53.80 –2.10 –11.81 –23.90
19 38.90 6.50 10.54 –37.85
20 70.55 10.65 –0.51 –10.36
21 34.60 3.50 56.69 –71.48
22 37.90 19.20 –0.36 –37.32
23 –5.06 7.36 9.03 –48.43
24 –5.20 5.30 0.91 –9.37
Sum. — 85.93 — —
with negative difference (Mi – M i ) could be considered as the start of a potential

drought, then a comparison is made between the absolute value of this deficit and
the first value of the sliding scale. If the absolute value of (Mi – M i ) exceeds the
sliding scale, then a drought is deemed to have started. Likewise, if the difference
of the next month is checked and if it is negative, then it is added to the negative
difference of the first month and compared with the second value on the sliding
scale. If this criterion is exceeded by the absolute value of the two deficits combined,
a drought is deemed to have started from the first month and the process continues
until the sliding scale reaches the MAD.
In the test of drought termination, the test is applied to the period following the
first month with a positive difference (surplus) occurring after the start of a drought.
The test for termination of a drought consists of either a continuous period of slightly
above-average rainfall or a few successive months of abnormally high rainfall,
indicating drought termination. The duration of the drought is the period between
the onset and termination of the drought. One way to determine drought termination
is to look at successive surplus periods to see how they compare with the maximum
of 3-, 4-, …, 12-month successive averages. If the successive surpluses exceed the
maximum successive averages, then the drought could be considered terminated.
After determining a drought period, its severity is obtained from the following
equation (for more details, see Herbst et al., 1966):
Sd =
∑ ( ERD ) * D
j
j d
(12.13)
∑ RD d
d
where:
Sd is the severity of drought d.

d is the index of drought.
j is all the months of excess rainfall deficit in drought d.
D is the drought duration (month).
ERD j is the excess rainfall deficit observed in month j.
RDd is the average deficit in each month of period d.
Example 12.3
Determine the onset, duration, and severity of drought for the data presented in
Example 12.2. Consider that the mean annual deficit is 150 mm and the maximum
values of the n-month average rainfall are shown in Table 12.5.
Solution: As shown in Table 12.3, the maximum value of the mean monthly
rainfall (MMMR) is 106.08 mm. The sliding scale is then estimated using Eq. (12.12)
as follows:
150 − 106.08
x= = 3.99
11
TABLE 12.5
Maximum n-Month Average of the Rainfall
Series (mm) for Example 12.3
Maximum 2-month average 175.0 mm
TABLE 12.6
Estimated Values for the Sliding
Scale for Example 12.3
Month Sliding Scale Value
1 106.08
2 110.07
3 114.06
4 118.05
5 122.04
6 126.03
7 130.02
8 134.01
9 138.00
10 141.98
11 145.97
12 150.00
TABLE 12.7
Cumulative Differences and Sliding Scales
for Example 12.3
Month Sliding Scales Cumulative Difference
1 106.08 –0.26
2 110.07 2.57
3 114.06 3.78
4 118.05 13.09
5 122.04 –39.84
6 126.03 –110.99
7 130.02 –134.79
8 134.01 –147.19
9 138 –141.04
10 141.98 –179.80
11 145.97 –216.19
12 150.00 –224.34
Table 12.6 shows the values estimated for the sliding scale for different months. To
test the onset of droughts, the differences between effective rainfall and the long-
term average of rainfall in each month are estimated, which is shown in Table 12.4.
As it can be seen in Table 12.4, the first negative difference occurs in month 1;
therefore, it could be considered as the onset of a drought event. The cumulative
summations of the differences after the first negative value are compared with the
sliding scales (x) in Table 12.7. As shown in this table, the absolute values of the
difference series become greater than the sliding scale series after month 7. Now it
TABLE 12.8
Results of Drought Termination Analysis for Example 12.3
Differences = Effective Cumulative Rainfall Maximum
Rainfall – Monthly After the First n-Month
Month Rainfall Mean Rainfall Positive Difference Averages
(1) (2) (3) (4) (5)
1 0.99 –0.26 — —
2 3.99 2.83 — —
3 1.59 1.21 — —
4 19.59 9.31 — —
5 2.45 –52.93 — —
6 21.06 –71.15 — —
7 72.31 –23.80 — —
8 75.21 –12.40 — —
9 116.42 6.15 116.42 —
10 34.81 –38.76 151.23 175.0
11 14.62 –36.39 165.85 243.1
12 0.00 –8.15 165.85 265.3
13 1.02 –1.07 166.87 275.3
14 0.86 –0.38 167.73 283.9
15 0 –0.71 167.73 298.4
16 0.7 –9.82 168.43 310.5
17 67.4 7.64 235.83 345.6
18 51.7 –23.90 287.53 366.3
19 45.4 –37.85 332.93 380.5
20 81.2 –10.36 414.13 410.1
21 38.1 –71.48 — —
22 57.1 –37.32 — —
23 2.3 –48.43 — —
24 0.1 –9.37 — —
can be said that a drought has started from month 1. If this test is not passed after
considering 11 sliding scales, the test is performed again after observing another
negative difference.
The test of termination of a drought starts after the first positive difference
between effective rainfall and monthly mean rainfall is observed during a drought.
In this example, the application of the test starting from first positive value shows
that the drought has not been terminated yet. In the second iteration, the cumulative
summation of rainfall (column 4) is compared with the maximum n-month averages
(column 5), which shows that the values in column 4 become greater than the values
in column 5 after the 20th row. Therefore, it can be said that the drought has been
terminated and 20 months is considered as the duration of drought.
To calculate the severity of this drought, Eq. (12.13) is used as follows:
∑ (ERD ) = 212.31
j
j
Dd = 20.0
∑ RD
d
d = 233.7
212.31 × 20
Sd = = 18.1
233.7
12.4.2 REGIONAL ANALYSIS OF CLIMATIC DROUGHT

The main objective for regional analysis is to find a suitable criterion for defining
dry events on a regional scale. Three methods can be used for this analysis: weight-
ing, median of ratios, and combined methods.
12.4.2.1 Weighting Method
In this method, the weights of each station are calculated by applying the Thiessen
polygon network and the long-term average regional precipitation dataset is obtained
for the entire region. Then, the best statistical distribution is chosen for the long-
term average precipitation dataset ( P ) and based on that, the ratios of average
regional precipitation to long-term average precipitation for various probabilities
(different return periods) are calculated. These ratios are applied as regional criteria
to designate the precipitation datasets of each rain gauge of the region as wet or dry
events.
Example 12.4
The historical annual precipitation data for a basin are given in Table 12.9. Table
12.10 shows the annual rainfall in drought events with different return periods. If a
10-year climatic drought occurs in the region, determine which sub-basins face a
drought with a 10-year or more return period.
Solution: In order to find the annual rainfall associated with a 10-year drought
event, the data given in Table 12.9 are sorted in ascending order (see column 3 of
Table 12.9). Then, the probability for each data is estimated using the following
formula:
m
p=
n +1
where m is the rank of the sorted data and n is the total number of data.
TABLE 12.9
Annual Average Precipitation for a Basin (Example 12.4)
Precipitation Sorted Precipitation Return Period
No. of Years (mm) (mm) Probability (yr)
(1) (2) (3) (4) (5)
1 618.23 101.47 0.05 19.00

2 563.54 284.2 0.11 9.50
3 546.6 307.44 0.16 6.33
4 501.51 345.81 0.21 4.75
5 480.41 350.1 0.26 3.80
6 451.88 368.51 0.32 3.17
7 418.34 369.99 0.37 2.71
8 403.34 403.34 0.42 2.38
9 369.99 418.34 0.47 2.11
10 368.51 451.88 0.53 1.90
11 350.1 480.41 0.58 1.73
12 345.81 501.51 0.63 1.58
13 307.44 546.6 0.68 1.46
14 284.2 563.54 0.74 1.36
15 776.02 617.04 0.79 1.27
16 1001.96 618.23 0.84 1.19
17 101.47 776.02 0.89 1.12
18 617.04 1001.96 0.95 1.06
TABLE 12.10
Estimated Regional Precipitation Data for Drought Events (Example 12.4)
Drought Event Precipitation (mm)
Sub-Basin Area (%) 2-Year 5-Year 10-Year 25-Year 50-Year
A 20 380 280 180 100 70

B 30 600 540 510 490 350
C 15 550 520 480 320 250
D 15 280 250 200 180 70
E 20 400 350 310 280 190
TABLE 12.11
Recorded Precipitation Data for Stations in Example 12.5
Stations Representing Sub-Basins
A B C D E
Annual precipitation (mm) 380 520 320 250 410

As can be seen in Table 12.9, annual precipitation of 284 mm has a 10-year

return period. Sub-basins A and D are facing droughts with 10-year and more return
periods.
Example 12.5
In a specific year, the data shown in Table 12.11 have been recorded in the region
presented in Example 12.4. What is the appropriate return period of the regional
drought?
Solution: The average precipitation over the region can be estimated using the
Thiessen polygon network as follows:
Average rainfall =
(380 × 20) + (520 × 30) + (320 × 15) + (250 × 15) + (410 × 20)
= 399.5
100
Considering the data presented in Table 12.9, the regional drought is estimated to
have a 2.4-year (between 2 and 3 years) return period.
12.4.2.2 Median of Ratios Method
The objective is to obtain a regional truncation level; therefore, the medians of the
calculated P/Pmean ratios are determined for various return periods for point evalua-
tion of the different stations in the region:
⎛ P ⎞ ⎧⎪⎛ P ⎞ T ⎫⎪
⎜P ⎟ = Median ⎨ ⎜ ⎟ ; i = 1,..., N ⎬; T = 2, 5,10,... (12.14)
⎝ mean ⎠ T ⎪⎩⎝ Pmean ⎠ i ⎪⎭
where:
(P/Pmean)T is the regional criterion of precipitation ratio to long-term average

precipitation for return period T.
(P/Pmean) iT is the ratio of precipitation to long-term average precipitation in
rain gauge i for return period T.
N is the number of total rain gauges in the study region.
12.4.2.3 Combined Method
This method is a combination of the previous two methods. In this method, the
estimated values for (P/Pmean)T in the point evaluations for each station will be
weighted by using their relative area in the Theissen network. Therefore, the third
regional criterion can be calculated as follows:
T
⎛ P ⎞ ⎛ P ⎞
N
⎜P ⎟ =
⎝ mean ⎠ T ∑i =1
αi × ⎜ ⎟
⎝ Pmean ⎠ i
; T = 2, 5,10,... (12.15)
In addition to truncation level, other important factors should be considered in

regional drought analysis. Santos (1983) suggested that, in addition to defining
critical and truncation levels, a critical area should also be defined. Regional drought
is said to occur when the drought-affected area is larger than the critical area. The
critical area can be considered to be equal to a percentage of the total area, depending
upon the severity of the decisions imposed by using this threshold. For example, if
the threshold is selected as 50% of the area of a region, then the corresponding
return period that results in drought for an affected area equal to or greater than
50% is considered to be the specific return period for that year.
The other factor that should be considered in drought regionalization is the range
of variation for point averages in the entire region. For this purpose, dimensionless
precipitation has been used, which is a ratio of precipitation to average precipitation.
Rossi et al. (1992) suggested the use of dimensionless precipitation. The drought-
affected area in the current period is defined and analyzed by applying the observed
data, as follows:
• The total observed data for all of the rain gauges are changed to be
dimensionless.
• A regional isohyetal map is plotted over the entire region.
• Based on regional criteria definitions, drought-affected areas are estimated
by summation of all of the drought-affected subareas over different time
periods as follows:
ADT (t ) = ∑ β ⋅ a (t )
i i
(12.16)
⎪⎧1 if Pi (t ) ≤ PiT
βi = ⎨
⎪⎩0 otherwise
where:
ADT (t ) is the drought-affected area with return period T during time t.

ai(t) is the subarea represented by station i during time t.
Pi(t) is the precipitation in station i in a given time period t.
PiT is the precipitation at station i with a return period T obtained from the
best-fitted distribution for that station.
Figure 12.4 shows an isohyetal map of P/Pmean ratios for a 25-year return period
(drawn for a river basin in the central part of Iran) (Karamouz et al., 2002). This
map can be plotted using the spatial kriging method. Similar maps can be generated
for different return periods. According to these maps, if the region is affected by a
drought with return period T, one can identify parts of the region that are in the most
critical situation.
FIGURE 12.4 Isohyetal map for P/Pmean ratios for 25-year return period for a river basin in
the central part of Iran (the hatched sections show the drought affected areas in the region).
(From Karamouz, M. et al., Proc. 2002 EWRI Conference on Water Resources Planning and
Management, Roanoke, VA, 2002. With permission.)
12.5 PREDICTION OF DROUGHT TRENDS

One of the most important considerations for decision makers, operators, and man-
agers in drought condition is to predict the drought trends and how long a drought
will last based on previous climatic data. Among the existing methods, the Poisson
statistical method and kriging geostatistical method have been widely used. The time
between termination of drought and the start of the next drought event, known as
the expected waiting time, follows a Poisson distribution. The kriging geostatistical
method is used to predict inflow to a reservoir during critical water-shortage periods
in drought-affected regions. If the predicted inflow is not enough to meet the esti-
mated demand, then a drought has begun. If the time between the existing drought
and the next expected drought event (start time forecast for drought) is shorter than
the remaining water shortage period, then the operation strategies should be changed
from normal to water rationing. This method has been suggested by Rouhani and
Cargile (1989).
FIGURE 12.5 Three-year moving average of precipitation the Zayandeh-rud River basin in
the central part of Iran and polynomials fitted to the dry and wet spells. (From Karamouz,
M. et al., Proc. 2002 EWRI Conference on Water Resources Planning and Management,
Roanoke, VA, 2002. With permission.)
A third method is based on long-term regional average precipitation data and

its moving average has also been investigated by Karamouz et al. (2002). In this
method, the spells can be approximated by a polynomial function as follows:
Yki = a0 k + a1k × tik + a2 k × tik 2 + … + ank × tikn ; k = 1, 2,..., K ; i = 1, 2,..., Tk′ (12.17)
where:
Yki is the moving average value in the kth wet and dry spells during sequence i.
ajk is the jth polynomial coefficient in the kth wet and dry spell.
tik is the time sequence for the kth wet and dry spell.
k is the number of existing wet and dry spells in the observed dataset.
Tk′ is the wet and dry spell time period.
If a region is in the middle of a dry spell, a polynomial can be fitted to predict the
duration of the drought. Figure 12.5 shows the 3-year moving average and fitted
polynomial for the dry and wet spells for a river basin in the central part of Iran
(Karamouz et al., 2002).
12.6 HYDROLOGIC DROUGHT: DEFINITION AND

BACKGROUND
Hydrologic drought investigation consists of the analysis of water supply and demand
as well as socioeconomic factors. In general, hydrologic drought occurs when the
water resources in a region cannot supply water demands. To investigate the available
and potential water supply of a system, surface reservoirs, groundwater, and inflows
to the system must be considered. As far as water demands, different demands of
the system and their trends and variations, especially during drought periods, should
be studied. Natural and uncontrolled events include streamflows; others are induced
by human intervention, such as the operation of reservoirs. Because of the natural
uncertainty in streamflow and its relation with other parameters such as precipitation,
the most important part of a hydrologic drought study is streamflow investigation.
For hydrologic drought, most of the literature has concentrated on analysis of
historical and generated streamflow data, including estimation of the pdf (probability
density function) of the drought characteristics. These efforts can be classified into
three categories:
• Experimental investigation of data attempts to describe drought charac-

teristics based on observed series such as streamflows and demands. Based
on this method, Chang (1990) defined the expected time between drought
events using a Poisson distribution, and then he investigated the variation
in streamflow behavior during a drought using daily streamflow data.
Dracup et al. (1980) studied stationary and stochastic characteristics of
data and the correlation between various drought characteristics in the
historical data. They suggested that the results based on observed data
may be insufficient because of a short range of recorded data. Ben-Zvi
(1987) considered the volume of streamflows to be an indicator of drought.
He defined average data as the truncation level of water shortages during
small droughts and standard deviation as the truncation for severe droughts
and then studied droughts of different severity. Yevjevich (1967) proposed
that the difference between water supply and demand could be used as
an indicator of hydrologic drought. He defined sequential periods (runs)
as drought spells in series and emphasized that run theory is the best
method of studying hydrologic droughts.
• Investigation of generated data applies the previous methodology to gen-
erated data. Different methods may be used to generate data. Zucchini and
Adamson (1988) used the bootstrap method to evaluate the risk of shortage
of annual inflow to a reservoir. Yevjevich and Obeysekra (1985) studied the
results of different drought characteristics and the correlations among them
using generated data with various assumptions, such as the number of data
generated, and using different lag 1 correlation coefficients. Guerrero-
Salazer and Yevjevich (1975) analyzed the probability distribution function
of duration and severity after identifying the wet and dry spells in a year.
• Estimation of the pdf of drought characteristics is a common method in
hydrologic drought investigation. Drought studies were done by Sen
(1980), who tried to use sequential periods (runs) as analytical tools to
solve drought problems, such as the probability density function of
drought characteristics. Yevjevich and Obeysekera (1985) and Yevjevich
(1967) used generated data to determine the pdf of drought characteristics.
Griffits (1990) determined the pdf of each drought characteristic and the
multivariate distributions.
12.6.1 RUN THEORY

According to Yevjevich (1967), the basic method for local analysis of drought data
is the application of run theory to determine the time series of hydrologic variables
such as duration and severity of droughts. In this method, the following steps can
be taken:
1. Annual streamflow data is separated into dry and wet years.

2. All sequential years in which the annual runoff is below long-term average
are considered as dry events.
3. All sequential years in which the annual runoff is above long-term average
are considered as wet events.
The above procedure is initialized on an annual time scale. However, if the time
scale is less than one year and includes periodic parameters, a seasonal or monthly
time scale can be used. Analysis of sequential periods helps the evaluation of the
probability distribution of drought duration and severity. It should be noted that the
time series usually are not long enough for this analysis; therefore, sufficient data
must be generated to obtain a long-term time series. The bootstrap, tree ring, and
autoregressive moving average (ARMA) methods have been used for data generation
in various investigations.
12.6.2 PROBABILITY DISTRIBUTION FUNCTION OF DROUGHT

CHARACTERISTICS
The final objective of analyzing a natural phenomenon is to quantify its behavior.
This can usually be done by using the probability distribution function of the data.
For this purpose, the pdf curves are obtained by frequency analysis of generated
data after determining the duration and severity of droughts.
It is common to use the concept of return period for frequency analysis, but it
is not suitable here as drought is not an annual event (there is not one specific
observation per year). A long-term series (e.g., 500 years) is generated to estimate
the pdf for various durations. Other questions that must be answered include:
• What is the largest drought duration and what is its severity?

• What is the most severe drought between the probable drought events?
Answering these basic questions requires a large number of statistical data series
for which only long-term generated data are not sufficient; therefore, a planning
horizon should be considered that is usually the same as the available historical
data period or an existing planning horizon. The number of samples that must be
generated is also determined. Because each sample represents only one value for
each predefined objective (for example, each sample has only one drought with
maximum intensity), the number of generated series should be large enough for
frequency analysis.
12.6.3 RESERVOIR OPERATION AND DEMAND ALLOCATION

One of the most critical aspects of hydrologic drought is consideration of water
demand. Reservoirs are usually used as the water supply system in large basins. The
operating policies of existing reservoirs are an important tool for dealing with
hydrologic droughts. Some researchers have studied the effects of reservoirs during
drought (Chang et al., 1995). Optimal operating policy of a reservoir tends to allocate
water to meet demands during droughts as much as possible.
12.7 AGRICULTURAL DROUGHT: DEFINITION

AND BACKGROUND
In the study of agricultural droughts, many investigators believe that soil moisture
deficit is the best and most practical drought indicator. Soil moisture shortage is the
result of moisture loss during dry periods when evapotranspiration exceeds precip-
itation. Such losses depend on the type of plants. Once the soil moisture drops to a
lower level than the plant roots, the actual evapotranspiration falls to a very low rate
compared to the potential evapotranspiration. It is necessary to use water balance
models to determine soil moisture shortage in different time periods, especially in
regions with limited data availability.
Water balance models are based on theoretical methods to simulate runoff and
soil moisture variations in a river basin. Water balance models were first developed
by Thornthwaite (1948) and later modified by Thornthwaite and Mather (1955).
These models include methods for estimating the water balance between the river
basin inflow (from precipitation and snowmelt) and outflow (evapotranspiration and
groundwater supply). Since development of the first water balance model by Thorn-
thwaite and its subsequent modifications, these models have been used widely for
irrigation management, prediction of runoff, and reservoir and groundwater supply
(Haan, 1972; Howard and Lloyd, 1979; Mather, 1981; Van Bavel, 1953). Revelle
and Waggoner (1983), Gleick (1987), and Schaake (1990) have also used water
balance to estimate the effect of general climatic changes on hydrologic character-
istics of specific regional models.
Water balance models are used to obtain a drought indicator. The Palmer drought
severity index (PDSI) (Palmer, 1965) uses a water balance model between moisture
supplied and demand for a two-layer soil. The PDSI is a tool for determining spatial
and temporal variation in soil moisture stresses. Its calculation includes determination
of Z index, the monthly soil moisture indicator that reflects the deviation of actual
precipitation, and soil moisture supply from the expected values for normal conditions
known to be climatically appropriate for existing conditions (CAFEC).
12.7.1 THORNTHWAITE WATER BALANCE MODEL

The model initially developed by Thornthwaite (1948) is an old model and may be
considered outdated, but it is still being used by many investigators (e.g., Alley,
1984), as is true for the PDSI. The Thornthwaite model is not data intensive and
works in regions with data deficiencies. Furthermore, it can be calibrated at different

stages so that it is less susceptible to error accumulation. This model is based on
factors such as existing soil moisture, soil moisture deficit, and total amount of
rainfall and provides a general view of soil moisture variation resulting from vari-
ations in precipitation over the historical record. In other words, results of this model
can be used for a general evaluation of current conditions caused by the impacts of
regional climatic parameters on hydrologic and agricultural conditions. This water
balance model is applicable to daily, monthly, or even annual time sequences.
The input data for the model are precipitation, monthly average temperature,
and potential soil moisture. The model output consists of actual soil moisture,
infiltration, excess rainfall, and actual and potential evapotranspiration. The Thorn-
thwaite equation can be used to calculate the monthly potential evapotranspiration
(PE) as follows:
PE = 16(10θ / I ) a * f (λ ) (12.18)
a = 6.75 * 10 −7 I 3 − 7.71 * 10 −5 I 2 + 1.79 * 10 −2 I + 0.49239 (12.19)
( )
1.514
I = θ5 (12.20)
where:
PE is the potential evapotranspiration (mm).

θ is the average temperature for a specific time period (e.g., month) in ºC.
I is the annual temperature indicator, which is the summation of temperature
indicator i over the 12 months of a year.
f(λ) is the regional correction factor (a function of latitude and month).
The procedure for actual evapotranspiration and soil moisture calculation is as

follows: When precipitation for a specific month is greater than the potential evapo-
transpiration, the actual evpotranspiration is equal to the potential evapotranspiration,
and the remaining rainfall is compared with both the soil moisture capacity and the
initial soil moisture of the previous month. The remaining rainfall supplies the soil
moisture until it reaches its potential value. No runoff occurs until the soil approaches
its moisture capacity. Obviously, if the summation of rainfall and soil moisture is
less than the potential evapotranspiration, the actual evapotranspiration is less than
the potential evapotranspiration.
12.7.2 AGRICULTURAL DROUGHT INDICATORS

For trend studies, climatic factors such as temperature and precipitation are consid-
ered in addition to hydrological factors such as runoff (Thornthwaite, 1948). The
following steps are taken to calculate the moisture anomaly index (Z), which is
necessary for calculation of the PDSI.
12.7.2.1 Potential Evapotranspiration

As explained previously, the Thornthwaite equation is applied to calculate potential
evapotranspiration. The advantage of this method is that temperature, the only
required input parameter, is measured for most regions at acceptable time intervals.
Other methods, such as the T model by Mather (1981) and P model by Maan (1972),
require a number of climatic parameters that are seldom measured for one region
at the same time intervals.
12.7.2.2 Soil Moisture Condition
The objective of this step is to identify the pattern of soil moisture variation as a
function of temperature and precipitation. First, the precipitation at a specific month
is compared with the potential evapotranspiration for that month. If the precipitation
is sufficient, then the actual and potential evapotranspiration are assumed to be equal.
After subtracting the potential evapotranspiration, the remaining amount will supply
the soil moisture until it reaches its capacity (Φ). The rain that remains produces
runoff and results in groundwater recharge. If the precipitation is less than potential
evaporation, then the actual evapotranspiration could be provided from soil moisture
from the previous month. The following equations express these situations:
⎧ ETi = PEi
⎪
{
⎪Si = Min ( Pi − PEi ) + Si −1,Φ
⎪
}
⎪
If Pi ≥ PEi then: ⎨ Ri = Si − Si −1 (12.21)
⎪
⎪ Li = 0
⎪
⎪⎩ EPi = (Pi − PEi + Si −1) − Φ
⎧ ETi = Min{PEi ,( Pi + Si −1 )}
⎪
⎪Si = ( Pi + Si −1 ) − ETi
⎪⎪
otherwise: ⎨ Ri = 0 (12.22)
⎪
⎪ Li = Si −1 − Si
⎪
⎪⎩ EPi = 0
where:
Pi is the precipitation in month i.

PEi is the potential evapotranspiration in month i.
ETi is the actual evapotranspiration in month i.
Si is the soil moisture content in month i.
Ri is the amount of water that the soil gained in month i.
Li is the soil moisture loss in month i.

Φ is the soil moisture capacity.
EPi is the excess rainfall in month i.
Then, the soil moisture deficit (Di) is calculated as follows:
Di = Φ − Si (12.23)
In the first condition, the excess rainfall can be greater than 0, which means that the
soil moisture deficit is equal to 0 and excess water is available for runoff.
Example 12.6
The monthly water balance parameters of a region are provided in Table 12.12.
Assume that the soil moisture content does not change before and after rainfall.
What is the evapotranspiration in that month (in mm)?
Solution: Direct runoff and groundwater recharge can be estimated as:
Direct runoff = 60 × 0.3 = 18 mm
Groundwater recharge = 100 × 0.06 = 6 mm
The actual evapotranspiration can then be estimated as the difference between rainfall
depth and depth of direct runoff and groundwater recharge:
Evapotranspiration = 60 – 18 – 6 = 36 mm
12.7.2.3 Runoff Calculation
Runoff is generated as a result of excess rainfall. The simplified water balance model
is applied to calculate excess rainfall. Separation of runoff and groundwater recharge
is difficult. Runoff can be calculated as follows:
ROi = F*EPi (12.24)
TABLE 12.12
Parameters of Water Balance Model for Example 12.6
Changes Storage Coefficient
Precipitation (mm) Runoff Coefficient in Water Table (cm) of Aquifer
60 0.3 +10 6%
where:
ROi is the calculated runoff from excess rainfall in month i.

F is the lag coefficient for converting excess rainfall to runoff.
EPi is excess rainfall in month i.
As it can be seen in this equation, a lag coefficient is necessary to calculate the

runoff, which depends on the geophysical condition of a region. This factor is less
than 1 and is calibrated by using long-term runoff data measured at gauging stations
and long-term excess rainfall data calculated in the previous step. The method of
least mean square error is used to calibrate the model and select the best lag factor
for the region.
12.7.2.4 Potential Climatic Values
Besides the potential evapotranspiration, other potential values should be calculated

as follows:
• Potential recharge (PR) is the moisture required by the soil to reach its
moisture capacity. It is the difference between soil moisture capacity and
the actual soil moisture which is the same as soil moisture deficit, Di.
• Potential moisture loss (PL) is the amount of moisture lost from the soil
to supply the evapotranspiration demand (PEi – Si–1). The moisture loss
approaches its potential when there is no precipitation.
PLi = Min{PEi , Si −1 } (12.25)
• Potential runoff (PRO) is the maximum runoff that can occur. It is a

function of precipitation less the soil moisture gain. In the most critical
conditions, it can be assumed that when the amount of runoff is high, the
amount of potential recharge (PR) is low, and when runoff is low, the
potential recharge is high. When S reaches its maximum value, the poten-
tial runoff is also maximized. The nature of potential runoff is much more
complicated than other potential values. Different investigators have
assumed values as high as 3 times the amount of precipitation and as
ambiguous as the difference between soil moisture capacity and potential
recharge. In the absence of evidence that could relate PRO to other
hydrologic parameters, the value of PRO for each month is assumed to
be the highest runoff experienced during the historical record for that
month.
12.7.2.4.1 Coefficients of Water Balance Parameters
• Evapotranspiration coefficient (α): In dry climates, actual evapotranspi-
ration is less than potential evapotranspiration. The amount of precipita-
tion required for near-normal conditions is dependent on the average
climate and on the prevailing meteorological conditions during the period

in question. α is used to estimate the expected evapotranspiration that is
climatically appropriate for existing conditions (CAFEC) from the poten-
tial evapotranspiration. This terminology has been used frequently in
drought studies. It simply implies the expected normal condition:
ETi
α= (12.26)
PEi
where ETi and PEi are the averages of the historical ET and PE, respec-
tively, in month i. The expected ET obtained by applying this coefficient
could be compared with its actual amount in a specific month to obtain
the deviation from normal conditions for that month.
• Recharge coefficient (β): In many regions, soil moisture recharge is a
seasonal property of the soil. The proportion of average recharge for month
i to average potential recharge is β:
Ri
β= (12.27)
PRi
This coefficient could be used to compare the expected recharge in a

specific climatic condition with its actual amount.
• Runoff coefficient (γ): Coefficient γ is calculated as the ratio of average
runoff for month i to the potential runoff:
ROi
γ= (12.28)
PROi
By applying this coefficient, the expected runoff in a specific climatic

condition is calculated.
• Moisture loss coefficient (δ): Coefficient δ is the ratio of average moisture
losses in month i to potential loss:
Li
δ= (12.29)
PLi
The amount of soil moisture losses calculated by this coefficient could be

subtracted from the sum of the other three factors to obtain the amount
of precipitation for a region in normal condition.
12.7.2.4.2 Climatically Appropriate Precipitation for the

Existing Condition (CAFEC) (P) ˆ
Using the above equations, the amounts of evapotranspiration, recharge, runoff,
moisture loss, and expected precipitation for each month under normal conditions
can be calculated as follows:
∧
ET i = α.PEi (12.30)
Rˆ i = β.PRi (12.31)
∧
ROi = γ.PROi (12.32)
Lî = δ.PLi (12.33)
∧ ∧
Pî = ETi + Rˆ i + ROi − Lî (12.34)
Each component of CAFEC precipitation has an average ∧ equal to the average

historical observations. The expected evapotranspiration, ETi , is equal to:
∧
∑ (ET )
i =1
i
ETi = α ⋅ PEi = n PEi , (12.35)

∑ (PE )
i =1
i
By taking the summation of both sides:
n ∑ (ET ) i n
∑ ∑ (PE )
∧
⎛ ET ⎞= i =1
(12.36)
⎝ i⎠ n i
i =1
∑ (PE )
i =1
i
i =1
Therefore:
n n
∑ ∑ (ET )
∧
⎛ ET ⎞= (12.37)
⎝ i⎠ i
i =1 i =1
and the average value of CAFEC evapotranspiration is equal to the average evapo-
transpiration from historical hydrologic accounting. The same reasoning holds for
the averages of other components of CAFEC precipitation; therefore, the estimated
value of CAFEC precipitation is unbiased. In spite of having the same averages, the
CAFEC values and the actual/observed values for each period are seldom in complete
agreement; however, in large period studies and observations they tend to converge.
12.7.2.4.3 Anomaly Moisture Index (Z)
After calculating CAFEC precipitation and comparing it with observed precipitation,
the difference (d) can be calculated as:
di = Pi − Pî (12.38)
These departures from normal series show the deviation of weather from its normal
condition. This new series provides a good description of the moisture or lack of
moisture in the climate.
12.7.2.4.4 Climatic Character (K)
Average moisture supply is not always dependent on the precipitation in a period.
Sometimes, when precipitation is insufficient, storage of moisture in the previous
period is used. Using the average of moisture supplied, a climatic character (ki) for
each month i is calculated as follows (Palmer, 1965):
12
∑ D ×K ′
_
K j = 17.67 K ′j / i j j = 1, …12 (12.39)
i =1
where
⎛ T + 2.8 ⎞
K ′j = 1.5 log10 ⎜ j _ ⎟ + 0.5 (12.40)
⎝ Dj ⎠
and
Tj = ( PE j + R j + ROj ) / ( Pj + L j ) (12.41)
Tj is a measure of the ratio of moisture demand to moisture supply for a month, and
(K) is a regional climatic character. To estimate the Z index for a specific region,
the coefficients of the above equations must be calculated based on the climatic
character of the region. The monthly constants (K) are used as weighting factors of
monthly deviations during dry spells. Then, a moisture anomaly index, Z, for month
i is defined as follows:
Zi = K × di (i = 1, 2,⋯, N ) (12.42)
12.7.2.4.5 Converting the Z Index to a Drought Severity Index

To convert the Z index to a drought severity index, some regional studies are
required. The Palmer studies were carried out in western Kansas, central Iowa, and
the northwestern part of North Dakota. The PDSI introduced by Palmer for these
regions is:
PDSIi = 0.897 × PDSIi −1 + (1 / 3) Zi i = 1, 2, ⋯, N

(12.43)
PDSI 1= (1 / 3) Z1
The coefficients of the above equation depend upon the region and will change in
various zones and regions with different climatic regimes. To obtain the above
equation for a specific region, all sub-basins of the region must be investigated. The
methodology of calculating the PDSI coefficients is described as follows. First, the
Z index series, which will serve as the basis of the agricultural index, is calculated
for all sub-basins. Based on the calculated Z index series for all sub-basins of the
region, the following steps convert the Z index to a drought severity index:
1. For all negative values in Z index series, cumulative Zs are calculated.

2. Periods representing the maximum rates of the accumulated negative
values are selected and plotted.
3. A straight line is drawn that indicates the approximate maximum rates
observed during extremely dry periods of various lengths.
4. The ordinate from normal to extreme Z is divided into four equal lengths
and numbered –1 to –4 to represent the severity of drought. The categories
for drought severity and the related degrees of severity for these categories
can be the same values used by Palmer (1965).
5. An equation between X (drought severity index), Z index, and t (duration)
can then be obtained.
Karamouz et al. (2002) obtained the following equation for a specific region in
the central part of Iran:
Xi = ∑Z
t =1
t (5.14t + 10.88) (12.44)
Extending the methodology proposed by Palmer (1965), Karamouz et al. (2002)

derived the following equation for a specific region in the central part of Iran.
PDSIi = 0.679 × PDSIi −1 + (1 / 16) Zi i = 1, 2, ⋯, N (12.45)

TABLE 12.13
Amount of Abnormal Dryness Required To Maintain a Given Drought
Severity
i −1 i
T Xi–1
∑t =1
Zi
Δ Xi Xi
∑Z
t =1
i
Zi
2 –1 –16.02 0 –1 –21.16 –5.14

10 –1 –57.14 0 –1 –62.28 –5.14
2 –3 –48.06 0 –3 –63.48 –15.42
10 –3 –171.42 0 –3 –186.84 –15.42
12.7.2.4.6 An Example of How to Derive the Drought Severity

Index for a Region
Equation (12.44) is only an approximation of the drought severity index. The prob-
lem must be changed to be able to evaluate the severity of drought in monthly
sequences. Such changes will provide the duration indirectly as a consequence of
the accumulation of successive monthly contributions to drought severity. In order
to evaluate the contribution of each month, we set i = 1 and t = 1 in Eq. (12.44) and
we have:
X1 = Z1 /16 (12.46)
Because this is an initial month,
X1 – X0 = ΔX1 = Z1/16 (12.47)
A certain amount of dryness (Z < 0) is required to maintain the severity of the existing
dry spell. Now we must determine this dryness for ΔX = 0 (Table 12.13) (Palmer,
1965). The rate of increasing Z, in order to maintain a constant value of X (ΔX = 0),
depends on the value of X that is to be maintained (Palmer 1965). This reasoning
adds additional term to Eq. (12.47), and the equation is changed to the form:
ΔXi = (Zi/16) + c Xi–1 (12.48)
where:
ΔXi = Xi – Xi–1
Now the problem is to determine the c coefficient. Using Eq. (12.44), the value
of zi will maintain a given value of X from month to month. These values for two
arbitrary values of Xi–1= Xi and two arbitrary values are shown in Table 12.13. If we
bring the values of Zi, Xi–1, and ΔX from Table 12.13 into Eq. (12.48), we obtain:
ΔX = 0 = (–5.14/16) – 1.0 c (12.49)
and
ΔX = 0 = (–15.42/16) – 3.0 c (12.50)
Therefore, c is –0.321, and the final equation for the monthly contribution to drought
severity is:
ΔXi = (Zi /16) – 0.321 Xi –1
In the other form, we have:
Xi = Xi –1 + (Zi /16) – 0.321* Xi –1 (12.51)
Then, Eq. (12.45) is obtained.

12.7.2.4.6 Palmer Drought Severity Index
The PDSI is an index for evaluating the severity of a drought. After determining the
Z index, the PDSI formulation is expressed as follows:
PDSIi = 0.897 PDSIi–1 + (1/3) Zi i = 1,2,…,N

PDSI = (1/3) Z1 (12.52)
where N is the number of time periods. Dry and wet spells are categorized by the
PDSI; see Table 12.14 for the drought classifications based on PDSI values.
TABLE 12.14
Drought Categories Based on PDSI
PDSI Drought Category
≤–4 Most severe drought

–4 to –3 Severe drought
–3 to –2 Medium drought
–2 to –1 Nearly drought
–1 to 1 Normal
1 to 2 Nearly wet
2 to 3 Medium wet
3 to 4 Severe wet
≥4 Most severe wet
Source: Palmer, W. C., Meteorological Drought, U.S. Weather Bureau

Research Paper No. 45, 1964.
To obtain drought and wet spells using the PDSI, one of the three different
indices ( Xik ) must be used as follows:
Xik = 0.897 * Xik−1 + (1 / 3) × Zi k = 1, 2, 3 i = 1, 2,…, N (12.53)
where:
Xi1 is the severity index for a wet spell in month i.

Xi2 is the severity index for a dry spell in month i.
Xi3 is the severity index for a spell that cannot be classified as dry or wet in
month i.
To determine the drought severity index, the following steps should be taken: If
Xi2 is greater than or equal to –1, then month i is in a dry spell. If Xi1 is greater
than or equal to 1, then month i is in a wet spell. During a drought period, Xi3 = Xi2 ,
and during a wet period Xi3 = Xi1 . In this situation, when the indicators ( Xi1 or Xi2 )
are between –0.5 and 0.5, then X 3i is equal to 0, which represents the termination
time of a dry or wet spell, respectively. Often only one indicator is not equal to 0,
and this indicator is the PDSI. The advantage of this method is that once a dry or
wet period is observed by either Xi2 or Xi1 , then the other resets to 0. Therefore, the
value of the PDSI does not grow without any limits, which could be the case if only
Eq. (12.52) is used (Alley, 1984).
The termination of an established drought is assumed to occur when Zi ≥ Zei,
where Zei is the moisture required to reduce the severity of a drought to –0.5.
Similarly, the termination of a wet spell is assumed to occur when Zi ≤ Zei. Zei can
be derived by solving Eq. (12.52) for Zi and substituting –0.5 and 0.5 for Xi, for dry
and wet spells, respectively. The parameter introduced by Palmer (1965), Pe, is used
to determine the correct value for the PDSI under various conditions. Pe is the
probability that an established wet or drought spell has ended and is defined as:
j*
100 ∑U
j =0
(i− j )
Pei = j*
(12.54)
Zei + ∑U
j =1
(i− j ) − Ui
where
Ui = Zi + 0.15 for an established drought spell (12.55)
and
Ui = Zi − 0.15 for an established wet spell (12.56)

Based on the study of Palmer (1965), in the case of an established drought a value
of Z = 0.15 will maintain an index of –0.5 from month to month. Therefore, any
value of Z greater than or equal to –0.15 will tend to end a drought. A similar
assumption could be considered for the termination of a wet spell, and Palmer
proposed the Ui as defined here.
The parameter j* corresponds to the number of successive values of Ui computed
prior to the current month. When the PDSI series shows a drought or wetness period,
Pe must be at its extreme values (0 or 100) to change the current condition. When
0 < Pe < 50, then the PDSI = X3; when 50 < Pe < 100, then the PDSI = X2 or X1,
depending on which term is the opposite of the sign of X3. Readers are referred to
Palmer (1965) and Alley (1984) for more details.
As mentioned in the methodology for calculating the PDSI, the coefficients of
the Palmer drought severity index are dependent on the regional climate and may
change with the climatic characteristics of various regions. So, it is necessary to
derive new coefficients for the PDSI equation for different study areas.
Example 12.7
An example of the estimated PDSI for a specific region is provided in Figure 12.6,
which shows the time series of the PDSI and standardized effective rainfall for the
Zayandeh-rud River basin in the central part of Iran for a period of 34 years as
monthly data from 1967 to 2002. Also, drought periods identified by the PDSI
through the method described in Section 12.7 and effective rainfall calculations
described in Section 12.4 are shown in the figure. The PDSI series shows seven
drought periods, and the seventh period has not yet terminated. Effective rainfall
series also show seven drought periods, and the seventh period has not yet terminated.
Comparison of the two series reveals an overlap between the dry periods in both
series. Similarities between the onset and termination times of both series show the
relative consistency of both methods in identifying drought periods. Figure 12.6
shows the variations in standardized effective rainfall and PDSI time series and the
drought spells identified by these methods.
12.8 GEOSTATISTIC TOOLS IN DROUGHT STUDIES

Geostatistics refers to the study of phenomena that fluctuate in space and/or in time.
Geostatistics offers a collection of deterministic and statistical tools for modeling
spatial and/or temporal variability. The basic concept of geostatistics is to determine
any unknown value z as a random variable Z, with a probability distribution that
models the uncertainty of z. The random variable is a variable that can take a variety
of values according to a probability distribution. One of the most well-known
geostatistic methods is kriging, which is widely used in the field of environment,
mining, surveying, and water resources engineering studies. Geostatistics has been
used in recent years for regional drought analysis. Because of their ability to char-
acterize spatiotemporal events, these methods are efficient tools for drought studies,
especially in the field of climatic drought assessment.
FIGURE 12.6 Dry and wet spells identification using PDSI (䡬) and effective rainfall
method (䡺).
12.8.1 KRIGING
Kriging is a collection of generalized linear regression techniques used to minimize
estimation variance which is defined from a prior model for a covariance. It is
important to have area-wide drought information before applying drought-control
measures to an entire basin. Kriging can be used in its spatial form to determine the
drought-affected areas in a region or for future installation of new gauging stations
in ungauged areas. Kriging methods may be used to study the temporal variation of
a process. In this case, a temporal domain, time variogram, and covariance function
could be used to predict variables for unobserved data. Regional data resulting from
dynamic processes in space and time use variogram and covariance functions that
allow prediction of that process in different points in space and with different lead
times.
Kriging in the simplest form applies first-order analysis based on the following
intrinsic hypotheses (Bogardi et al., 1985):
• The expectation of Z(x), a single realization of the parameter z, exists and

is independent of point x:
E[ Z ( x )] = m (12.57)
Or if there is a known function m(x) in which
E[ Z ( x )] = m( x ) (12.58)
a variable such as Y(x) could be used as:
Y ( x ) = Z ( x ) − m( x ) (12.59)
And
E[Y ( x )] = 0 for all x (12.60)
• For all vectors h, the increment [Z(x + h) – Z(x)] has a finite variance
independent of x:
Var[ Z ( x + h) − Z (h)] = 2γ(h) (12.61)
The γ function is the variogram. Variograms are obtained from the fol-
lowing three parameters: (1) nugget effect, which can partly be explained
by the measurement error (C0); (2) sill, which is the total variance of the
data beyond a certain distance or range (C); and (3) range or zone of
influence (R) (see Figure 12.7).
FIGURE 12.7 An example of empirical and theoretical variogram.
The first step in kriging analysis is to estimate a semivariogram. In all applica-

tions, the purpose is to analyze the correlation between neighboring measurements
of a property and thereby derive the spatial variability of such a property. The
functional description of this spatial structure, if it exists, is known as the semivar-
iogram in geostatistics. The computational form of a semivariogram is:
∑ [(z( x ) − z( x + h)) ]
1 2
γ(h) = i i (12.62)
2n
where γ(h) is the semivariogram value for distance h, z(xi) is the measurement of
location xi, and z(xi + h) is the measurement at location xi + h, where n is the number
of measurement pairs separated by vectorial distance h.
The empirical variogram could be replaced by a theoretical variogram model to
be used in a kriging problem. A theoretical semivariogram model consists of an
isotropic nugget effect and any positive linear combination of the following standard
semivariogram models:
• Spherical model (Sph(.)):
⎧ ⎡ h ⎛ h⎞ ⎤
3
⎪ . ⎢1 .5 − 0 . 5
⎝ a ⎠ ⎥⎦, if (h ≤ a)
c
γ(h) = c.Sph⎛ ⎞ = ⎨ ⎣ a
h
⎝ a⎠ (12.63)
⎪
⎩c, if (h ≤ a)
where a is an actual range and c is the positive variance contribution or sill.

• Exponential model (Exp(.)):
⎡ 3h ⎤
γ(h) = c.Exp⎛ ⎞ = c.⎢1 − exp⎛ − ⎞ ⎥
h
(12.64)
⎝ a⎠ ⎣ ⎝ a ⎠⎦
• Gaussian model:
⎡ ⎛ (3h)2 ⎞ ⎤
γ(h) = c.⎢1 − exp⎜ − 2 ⎟ ⎥ (12.65)
⎢⎣ ⎝ a ⎠ ⎥⎦
• Power model:
γ (h) = c.h ω (12.66)
where ω is a constant between 0 and 2.
• Hole effect model:
⎡ ⎤
γ (h) = c.⎢1.0 − cos⎛ .π⎞ ⎥
h
(12.67)
⎣ ⎝ a ⎠⎦
Figure 12.7 shows a schematic of empirical and theoretical variograms. Consider

the estimate of an unknown (unsampled) value z(u) from neighboring data values
z(ua) (a = 1, …, n). The random function model Z(u) is stationary with mean m and
covariance C(h). In its simplest form, also known as simple kriging, the algorithm
considers the following linear estimator:
∑ ∑ λ (u)) )m
n
*
Z SK (u ) = λ α (u) Z (u ) + ( (1 − α (12.68)
α =1 α
*
where Z SK is a simple kriging estimator, and λα(u) are the weights to minimize the
error variance, also referred to as the estimation variance. Minimization of the error
variance in a set of normal equations results:
∑
n
λ β (u)C(uβ − uα ) = C(u − uα )
β =1 (12.69)
∀α = 1,..., n
where C is the stationary covariance and C(0) is the stationary variance of Z(u).
The corresponding minimum estimated variance, or kriging variance, is:
∑
n
σ 2SK (u) = C(0) − λ α (u)C(u − uα ) ≥ 0 (12.70)
α =1
Ordinary kriging is the most commonly used method of kriging, whereby the sum
of weights
∑
n
λ α (u )
α=1
is limited to 1. This allows building an estimator Z *OK (u) that does not require
prior knowledge of the stationary mean m and remains unbiased in the sense that:
E{Z *OK (u)} = E{Z (u)} (12.71)
The relationship between weights λ and the semivariograms introduced earlier can
be represented in a matrix form as follows:
⎛ 0 γ 12 γ 13 ⋯ γ 1n 1⎞ ⎛ λ10 ⎞ ⎛ γ 10 ⎞
⎜ γ 21 0 γ 23 ⋯ γ 2n 1⎟ ⎜ λ2 ⎟ ⎜ γ ⎟
⎜ ⎟ ⎜ 0 ⎟ ⎜ 20 ⎟
⎜ ⋮ ⋮ ⋮ ⋱ ⋮ ⋮⎟ ⎜ ⋮ ⎟ =⎜ ⋮ ⎟ (12.72)
⎜γ γ n2 γ n3 … 0 1⎟ ⎜γn⎟ ⎜γ ⎟
⎜ n1 ⎟ ⎜⎜ 0 ⎟⎟ ⎜ n 0 ⎟
⎝ 1 1 1 ⋯ 1 0⎠ ⎝ µ⎠ ⎝ 1 ⎠
where λi0 represents the weight of station i, and 0 represents the origin station.
12.8.2 SPATIAL KRIGING IN DROUGHT STUDIES

An important factor that should be considered in drought regionalization is the
drought-affected area of the entire region. Kriging can be used in its spatial form to
determine the drought-affected areas in a region. Also, spatial kriging errors could
be used as guidance for future installation of new stations in an ungauged area. To
determine the drought-affected areas in a region, these steps should be followed:
• A spatial network with specific grids is considered over the entire region.
• The observed data of all rain gauges are considered as known measured
data at their nearest nodes of the network.
• The spatial variogram of precipitation is calculated and is fitted by a

theoretical variogram.
• The unknown precipitation at each node of the network is obtained using
the characteristics of the above variogram by a kriging method.
• As defined by regional criteria, drought-affected areas are estimated by
the summation of all drought-affected subareas that have experienced
precipitation less than the defined criteria.
An example of spatial kriging in regional drought assessment has been presented

in Section 12.4 of this chapter.
12.8.3 SPATIOTEMPORAL KRIGING

The behavior of environmental variables such as seasonal rainfall patterns can
usually be explained by time-dependent patterns and require different water supply
policies. These policies will be more important during drought events. A spatiotem-
poral study of precipitation over a region makes it possible to determine space and
time variations of precipitation in a region. Also, simulation methods in geostatistics
can be used to simulate realizations of precipitation time series at each location of
the region. The realizations are useful for calculating the probable characteristics of
a climatic drought (severity and/or duration) at each zone of a region at a given
significant level.
The spatiotemporal time series could be studied through the assessment of a
trend model and the residuals of precipitation time series at each station of a region.
A time series at each station is divided by deterministic and random components as
follows:
Z (u, t ) = M (u, t ) + R(u, t ), ∀u ∈ D, ∀t ∈ T (12.73)
Z(u, t) is a set of rainfall time series during period T at each location u, which is a
two-dimensional vector; D represents the two-dimensional domain of the study area;
M(u, t) is a deterministic spatiotemporal component that models the average variation
of rainfall; and R(u, t) is a random variable component, independent from M(u, t),
that models the fluctuations with higher frequency around the average spatiotemporal
variations. Deterministic models of rainfall variations, M(u, t), are developed inde-
pendently at each station u. The stochastic model of spaciotemporal variation is then
obtained by regionalizing the parameters of the deterministic models. The first goal
at each station is to determine the periodic behavior of the rainfall data, such as
inter-annual variations, which can be obtained through time series models such as
spectral analysis.
Deterministic variations at each station uα can be modeled as the sum of k + 1
unknown temporal basic functions fk(t) as follows:
M ( u α , ti ) = ∑ b (u ) f (t ),
k =0
k α k i i = 1, ⋯, Tα (12.74)
where bk(uα) is the coefficient of the kth function. These functions usually consist
of a linear function to show the long-term trend and the seasonal or annual periodic
components. The periodic nature of the basic functions is determined as a component
of M(u,t) by a time series analysis method such as Fourier analysis. The coefficients
of the model are regionalized by the kriging method. This makes it possible to predict
the rainfall trend at each location of the region. Also simulation of the regionalized
trend is possible using geostatistical simulation methods (Deutsch and Journel,
1999).
The residual values R(u,t) are obtained by subtracting the average variations
M(u,t) from rainfall values, Z(u,t) at each station. The model of the spatio-temporal
residual, R(u,t), is developed using spatially correlated residual time series.
Regionalizing and simulating data using various kriging and simulation methods
could be applied to both trends and residuals of time-series modeling. Most geosta-
tistics methods require time series to be in a Gaussian distribution environment;
therefore, transformation of the data to normal values will be necessary in most cases.
The most straightforward algorithm for generating the realizations of a multi-
variate Gaussian field is provided by the sequential Gaussian simulation (Deutsch
and Journel, 1998). Simulation through LU (lower and upper triangular matrices)
decomposition of the covariance matrix is another method in geostatistical simulation.
The final real precipitation time series of each realization is obtained by com-
bining the results of the simulated trend and residual models for all points (nodes)
of the region.
Example 12.8
Consider a region with 36 rain gauges (Figure 12.8) which have recorded 50 years
of precipitation data. The rain gauges are shown as points in a 30 by 30-unit network.
Set up an algorithm for monthly space–time regionalization of the precipitation time
series:
Solution: The algorithm is expressed by following steps:
1. At first, the real values of precipitation are transformed to normal values.

The trend model for each station is obtained by fitting time series models
such as spectral model. If we use a model with n parameter for all stations,
36 × n parameters will be produced at this step.
2. Using the coefficients of trend model and sequential time steps, the pre-
cipitation trend series are obtained at each station, which are equal to 36
vectors of 600 (50 × 12) monthly precipitation trend values.
3. The residual time series at each station is obtained by subtracting the trend
model from the real values of precipitation. This will produce 36 vectors
of 600 precipitation residual values.
4. The parameters of trend model are regionalized using kriging methods.
The results are presented as 900 (30 × 30) n values.
5. If m realizations are needed, the simulation of regionalized parameters
should be done m times using simulation methods.
FIGURE 12.8 The location of rain gages in the region.
6. Using the coefficients of trend model and sequential time steps, the pre-
cipitation trend series are obtained at each station, which are equal to
900 × m vectors of 600 values.
7. The precipitation residual time series at each station are regionalized using
geostatistical methods. This will provide 900 residual time series of 600
values.
8. To obtain m realizations, m simulation of residuals should be done. Then
the output will contain 900 × m vectors of 600 values.
9. Compose the trend values and residuals at 900 nodes of the network to
provide 900 × m vectors of 600 values of the normalized precipitation
time series.
10. Use inverse transformations to convert normalize data to real values.
11. Finally the result consists of m realizations of 50-year (600-month) time
series at each 900 nodes of the network.
12.9 PROBLEMS
12.1 Use the 3-year moving averages to compare the two series of annual average
precipitation for two different regions presented in the following table:
Precipitation of Region A Precipitation of Region B

No. of Years (mm) (mm)
1 50 71
2 3 40
3 8 118
4 59 99
5 58 48
6 141 134
7 30 107
8 22 92
9 41 28
10 47 32
11 103 45
12 23 53
13 64 66
14 46 142
15 129 153
16 82 59
17 128 46
18 89 121
19 17 11
20 15 191
21 28 101
Current year 15 45
(a)
Which of these regions is facing the most severe drought in the
current year?
(b) If the long-term average precipitation in regions A and B is esti-
mated to be 350 and 250 mm, respectively, discuss the drought
periods and their durations.
(c) Use Eq. (12.13) to estimate the drought severity for each of the
drought periods.
12.2 The monthly rainfall data for 22 years is presented in the following table.
Assume C (the constant in Eq. (12.6)) is 0.1 and calculate the following
items for the region:
(a) Effective rainfall
(b) Rainfall deficit
(c) Excess rainfall deficit.
12.3 Determine the onset, duration, and severity of drought for the first year
of data presented in Problem 12.2.
12.4 The historical annual precipitation in a basin and the annual rainfall in
drought events with different return periods are shown in the following
No. of
Years Sept. Oct. Nov. Dec. Jan. Feb. Mar. Apr. May June July Aug.
1 0.2 49.6 71.0 40.9 77.1 109.2 153.3 40.9 9.8 0.7 1.6 0.4
2 0.4 2.9 40.3 106.8 73.7 90.8 83.5 5.0 0.0 15.9 0.8 0.0
3 10.4 8.3 117.7 88.4 67.2 88.7 59.8 12.8 0.0 1.4 0.0 0.0
4 0.0 59.2 98.6 205.1 179.7 76.5 37.4 73.1 13.6 0.0 0.6 0.0
5 8.8 57.9 47.5 124.3 28.1 77.7 124.5 27.8 1.4 1.4 6.7 0.2
6 38.0 140.7 133.9 54.5 55.6 128.8 82.4 45.5 14.2 0.3 2.0 0.0
7 2.5 29.5 106.6 48.1 43.3 31.7 15.4 31.5 0.3 0.3 0.0 0.0
8 4.7 22.0 91.8 61.2 164.8 159.9 83.7 97.4 8.7 0.3 1.3 1.4
9 1.9 40.5 27.5 122.9 74.9 73.2 93.9 23.1 0.0 0.4 0.0 0.0
10 15.3 46.8 31.7 62.6 80.1 121.3 125.4 27.5 0.1 0.4 0.0 0.0
11 50.2 102.7 44.6 93.1 58.3 107.0 65.1 28.0 9.6 0.0 0.0 0.0
12 0.0 22.6 53.2 63.0 40.8 117.3 39.2 55.3 2.8 0.0 0.4 0.1
13 17.4 64.4 66.2 59.5 64.3 55.5 92.3 49.9 0.2 0.0 0.0 0.0
14 0.5 46.4 142.0 24.2 67.2 90.2 41.8 29.5 0.9 0.0 0.1 1.1
15 0.5 128.8 153.2 73.5 69.6 175.7 73.3 95.6 2.9 0.0 0.6 0.5
16 33.4 82.2 58.5 151.7 99.0 145.9 57.1 38.1 0.2 10.6 14.9 0.0
17 16.3 127.5 45.7 67.7 56.6 82.2 24.9 34.5 0.3 0.9 0.2 0.2
18 0.0 88.7 120.8 49.2 74.0 115.2 41.4 51.6 0.2 0.0 0.0 0.0
19 2.1 17.2 10.7 77.8 65.4 116.4 47.0 4.8 0.0 0.9 0.0 0.0
20 47.7 14.8 190.6 82.5 48.0 110.6 97.8 1.9 0.4 1.0 0.0 0.3
21 2.3 28.2 101.3 139.0 131.2 252.9 59.4 96.9 5.5 0.7 0.0 0.5
22 0.7 77.2 45.3 74.7 78.2 102.9 59.1 135.8 0.0 0.0 0.3 1.1
tables. If a 25-year climatic drought occurs in the region, determine which

sub-basins face droughts with 10-year or more return periods.
Regional Precipitation Data for Drought Events with Different Return

Periods (Problem 12.4)
Drought Precipitation (mm)
Sub-Basin Area (%) 2-Year 5-Year 10-Year 25-Year 50-Year
A 20 380 280 180 100 70

B 30 600 540 510 490 350
C 15 550 520 480 320 250
D 15 280 250 200 180 70
E 20 400 350 310 280 190
Annual Average Precipitation in a Basin (Problem 12.4)

Precipitation Sorted Precipitation Return Period
No. of Years (mm) (mm) Probability (yr)
1 618.23 101.47 0.05 19.00

2 563.54 284.2 0.11 9.50
3 546.6 307.44 0.16 6.33
4 501.51 345.81 0.21 4.75
5 480.41 350.1 0.26 3.80
6 451.88 368.51 0.32 3.17
7 418.34 369.99 0.37 2.71
8 403.34 403.34 0.42 2.38
9 369.99 418.34 0.47 2.11
10 368.51 451.88 0.53 1.90
11 350.1 480.41 0.58 1.73
12 345.81 501.51 0.63 1.58
13 307.44 546.6 0.68 1.46
14 284.2 563.54 0.74 1.36
15 776.02 617.04 0.79 1.27
16 1001.96 618.23 0.84 1.19
17 101.47 776.02 0.89 1.12
18 617.04 1001.96 0.95 1.06
12.5 The data presented in the following table have been recorded in the region
presented in Problem 12.4 for one specific year. What is the appropriate
return period of regional drought?
Station
A B C D E
Annual precipitation (mm) 310 490 350 275 390
12.6 Consider a basin with 12 rain gauges, the locations of which are shown
in the following figure. The recorded rainfall at each station for a year
are presented in the following table. Assume a threshold of 25-mm pre-
cipitation for the return period of a 10-year drought in the region. Specify
the 10-year drought-affected areas in the region.
Rain Gauges
1 2 3 4 5 6 7 8 9 10 11 12
Precipitation 320 315 205 105 120 240 230 231 242 250 317 318
References
Alley, W. M., The Palmer drought severity index: Limitation and assumptions” J. Climate
Applied Meteorol., 23, 1100–1109, 1984.
Aswathanarayana. U., Water Resources Management and the Environment, Balkema Publish-
ers, New York, 2001.
Beric, M., Zelenhasic, E., and Srdjevic, B., Extreme dry weather intervals of the growing
season in Backa, Yugoslavia, Water Resources Manage., 4, 37–73, 1990.
Ben–Zvi, A., Indices of hydrological drought in Israel, J. Hydrol., 22, 179–191, 1987.
Bogardi, I., Bardossy, A., and Duckstein, L., Multicriterion network design using geostatistics,
Water Resources Res., 21(2), 199–208, 1985.
Chang, G., Effects of droughts on the streamflow characteristics, ASCE J. Irrigation Drainage
Eng., 116(3), 332–341, 1990.
Chang, T. J., Xenia, A. and Choo B. T., Use of Flood-Control Reservoirs for Drought
Management” ASCE J. Irrigation Drainage Eng., 121(1), 34–42, 1995.
Deutsch C. V., Journel, A. G., GSLIB: Geostatistical Software Library and User’s Guide,
Oxford University Press, New York, 1998.
Dracup, J. A., Lee, K. S., and Paulson, E. G., On the definition of droughts, Water Resources
Res., 16(2), 297–302, 1980.
Gleick, P. H., Regional hydrologic consequences of increases in atmospheric CO2 and other
trace gasses, Climatic Change, 10(2), 137–161, 1987.
Griffits, G. A., Rainfall deficits: distribution of monthly runs, J. Hydrol., 115, 219–229, 1990.
Guerrero–Salazer, P. L. A. and Yevjevich, V., Analysis of Drought Characteristics by the
Theory of Run, Hydrology Paper 80, Colorado State University, Fort Collins, CO,
1975.
Haan, C. T., A water yield model for small watersheds, Water Resources Res., 8(1), 58–69, 1972.
Herbst, P. H., Bredenkamp, D. B., and Barker, H. M. G., A technique for the evaluation of
drought from rainfall data. J. Hydrol., 4, 264–272, 1966.
Howard, K. W. F. and Lloyd, J. W., The sensitivity of parameters in the Penman evaporation
equations and direct recharge balance, J. Hydrol., 41, 329–344, 1979.
Karamouz, M., Drought Management in Zayandeh-rud River Basin, final report to Regional
Water Authority of Isfahan, Iran, 2003.
Karamouz, M., Torabi, S., Araghi-Nejad, Sh., and D. Asgharzadeh, Analysis of droughts:
application to isfahan region in Iran, Proc. 2002 EWRI Conference on Water
Resources Planning and Management, Roanoke, VA, 2002.
Kottegoda, N. T. and Rosso, R., Statistics, Probability, and Reliability for Civil and Environ-
mental Engineers, McGraw-Hill, New York, 1998.
Lettenmiar, D. P., McCabe. G., and Stakhiv. E. Z, Global climate change: Effects on hydrologic
cycle, in Water Resources Handbook, L. W. Mays, Ed., McGraw-Hill, New York,
1996.
Maan, C. T., A water yield model for small watersheds, Water Resources Res., 8(2), 58–69,
1972.
Mather, J. R., Using computed stream flow in watershed analysis, Water Resources Bull.,
17(3), 474–482, 1981.
Mendenhall, W., Wackerly D. D., and Scheaffer, R. L., Mathematical Statistics with Appli-
cations, PWS-KENT, Boston Publishing Company, Boston, MA, 1990.
Palmer, W. C., Meteorological Drought, U.S. Weather Bureau Research Paper No. 45, 1965.
Revelle, R. R. and Waggoner, P. E., Effects of a carbon dioxide-induced climatic change on
water supplies in the western United States, Changing Climate, National Academy
Press, Washington D.C., 1983.
Rossi, G., Characteristics of drought over a region and shortage control strategies, Proc. of
the 18th IAHR Congress, Cagliari, Vol. 2, 1979.
Rossi, G., Benedini, M., Tsakiris, G., and Giakoumakis, S., On regional drought estimation
and analysis, Water Resources Manage., 6, 249–277, 1992.
Rouhani, S. and Cargile, K. A., A geostatistical tool for drought management, J. Hydrol., 108,
255–266, 1989.
Santos, M. A., Regional droughts: a stochastic characterizations, J. Hydrol., 66, 183–211,
1983.
Schaake, J. C., From climate to flow, in Climate Change and U.S. Water Resources, Waggoner,
P. E., Ed., Wiley Series in Climate and the Biosphere: Report of the American
Association for the Advancement of Science Panel on Climatic Variability, Climate
Change and the Planning and Management of U.S. Water Resources, John Wiley &
Sons, New York, 1990, pp. 177–207.
Sen, Z., Regional drought and flood frequency analysis: theoretical consideration, J. Hydrol.,
46, 1980.
Thornthwaite, C. W., An approach toward a rational classification of climate, Geogr. Rev.,
38(1), 55–94, 1948.
Thornthwaite, C. W. and Mather, J. R., The water balance, Publ. in Climatol. Lab. Climatol.,
8, 1–86, 1955.
Van Bavel, C. H. M., A drought criterion and its application in evaluating drought indices
and hazard, Agron. J., 45(4), 167–172, 1953.
Yevjevich, V., An Objective Approach to Definitions and Investigations of Continental Hydro-
logic Drought, Hydrology Paper No. 23, Colorado State University, Fort Collins, CO,
1967, 18 pp.
Yevjevich, V. and Obeysekera, J. T. B., Correlation between sample first autocorrelation
coefficient and extreme hydrologic runs, J. Hydrol., 79, 171–186, 1985.
Zucchini, W. and Adamson, P. T., On the application of the bootstrap to assess the risk of
deficient annual inflows to a reservoir, Water Resources Manage., 2, 245–254, 1988.
Index
ε-Constraint Method, 36 Basic solution, 28
Basic variable, 27
Basin
A Homogenous, 207
Bayesian decision theory, 103
Activated sludge process, 515, 517 Bellman optimality principal, 323
Activity coefficient, 389, 391 Benefit
Adequacy, 448 External, 119
Advection envelopes, 375–377 Marginal, 141
Aesthetic value, 136 Benefit–cost ratio, 132–135, 142, 309
Akaike criterion, 166 Biochemical oxygen demand (BOD)
Allele, 60 Carbonaceous, 386, 388
Alternating direction implicit method, 256 Nitrogenous, 388
Alternative, 21 Removal, 389
Ambient trend monitoring, 375 Biological tower, 517–518
Ancillary facilities, 299 Bioxidation
Anderson test, 162 Parameters, 393
Annual worth, 129, 132–133, 142 Rate constant, 389, 400–401
Aquatic life, 366–367, 387, 419 Bivariable function, 80
Aquiclude, 239 Black water, 519
Aquitard, 239, 241 BOD, See Biochemical oxygen demand
Area-volume-elevation curve, 300, 308, 445 Borda count, 49
ARIMA Box-Cox, 151
Forecasting, 173–175 Breakdown probability, 90
Modeling, 175 Brown water, 519
ARMA BSDP model, 328, 333
Forecasting, 172–173
Generation, 173
Modeling, 173 C
Artificial recharge, 490
Furrow, 490 Calibration, 212–213, 229
Flooding, 490 Capacity expansion, 273–274, 304, 310
Recharge well, 490 Capillary forces, 196
Stream-channel, 490 Capital recovery factor, 121–122
Assimilative capacity, 487 Catchment basin, 191, 193–194
Atmospheric water, 4 Chance constraint programming, 101, 114,
Attribute, 22 319–322
Autoregressive Channel modification, 206, 232
Coefficient, 167–170 Characteristic flow, 324–327
Modeling, 167 Chi-square, 533
Autocorrelation, 148, 161–163, 169–170, Test, 155–158, 182, 185
172–173, 182–183 Degree of freedom, 156–157
Chromosomes, 60
Circularity ratio, 194
B Climatic
Character, 560, 565
Backward shift operator, 150 Drought, See Drought
Bacteria, 364, 437 Cloud seeding, 491
579
Conceptual rules, 340–341 D

Confidence interval, 379–382, 412, 429
Confined aquifer, 239, 241, 243, 256, 258, 264, Dam
271, 286, 290 Arch, 298–299
Conflict Buttress, 298–299
Degree of, 14–15 Embankment, 298–299
Issues, 1, 12, 229–230, 284–286, 346–348 Gravity, 298–299
Qualitative aspects, 12, 15, 474, 476, 520–521 Types, 298
Quantitative aspects, 12, 14 Dantzig-Wolf method, 28
Resolution, 51–63, 285–286, 289, 347–348, Darcy’s Law, 211, 240, 245, 247, 250, 425
354, 419, 421, 428, 474–476, 501 Data
Conformal mapping, 251 Availability, 13
Conjunctive use, 277 Preparation, 147
Constraint, 22–29, 101, 111 Verification, 202
Equality, 24 Database management, 16, 17
Inequality, 24 DDSP model, 328, 333, 339
Consumptive use, 297 Decision
Contaminant transport, 422, 425 Maker, 21
Continuity equation, 243, 245, 248 Variable, 21–23, 25, 27, 35–36, 43–44, 54, 56,
Control limit theorem, 88 58, 60, 320, 322, 333, 338, 355
Control points, 201–219 Vector, 101
Conveyance loss, 297 Decision space, 21, 27–28, 30–33, 38, 52, 57,
Convolution equation, 210 60–61, 63–64
Copy system, 85 Continues, 21
Correlogoram, 161–162, 170 Discrete, 21
Cost function, 276, 278, 310, 333, 340, 342–343 Feasible, 21
Costs Decision support systems, 13, 16, 17
Environmental, 118 Adaptive, 16
Exclusion, 136 Intelligent, 16
External, 118 Spatial, 16
Fixed, 117–118 Decomposition techniques, 28
Incremental, 118 Degree of acceptance, 349, 355
Marginal, 136, 141 Degree of fulfillment, 340
Operation and maintenance, 127, 132–134 Degree-day method, 215
Private, 118 Density function, 102, 104–107, 112–113
Social, 118 Deoxygenation rate, 388, 389, 391, 409, 411, 430
Variable, 117–118 Desalination, 488
Co-terminated assumption, 128 Design problems, 8
Covariance, 176, 179–181 Detection problems, 8
Criterion-space, 22, 31, 34, 36–37, 40–42, 46, 48, Deterministic equivalent, 321–322, 358, 410
52, 64, Differencing, 150, 173, 182
Critical depth, 201, 218 Direction-based method, 41–44
Critical points, 201 Disagreement point, 421
Crop Discount gradient factor, 123–124
Coefficient, 501 Discount rate, 119, 127–130, 142
Height, 502, 509 Discrete stochastic dynamic programming, 102
Mix, 502, 509–511, 525 Disintegration phase, 6
Reflectivity, 502 Disaggregation, 147, 175–176, 179–180, 182
Roughness, 502 Dissolved oxygen, 364, 386–387, 388, 393, 395,
Water demand, 501–503 398–99, 406–407, 414, 419
Cross-correlation, 175–177 Dissolved pollutant concentration, 368, 423
Crossover, 60–61 Distance-based method, 44–48
Cumulative distribution function, 533 Distillation, 490
Cumulative periodogram test, 164 Distribution
Curse of dimensionality, 57, 102 Conjugate, 106
Index 581
Marginal, 107 Forward, 323, 358

Posterior, 106 Recursion relation, 102, 318, 322–323, 407
Prior, 105–106 State incremental, 57
Distribution function, 69–73, 87, 90, 95–96, 113 Stationary solution, 323, 328
Bernoulli, 68, 78, 86, 106 Stochastic, 322–323, 328–329, 333, 338–339,
Beta, 106, 534 359
Binomial, 68, 78, 86, 106
Chi-square, 74–75, 78
Degree of freedom, 74–75 E
Exponential, 72–74, 78, 87–88, 90–92
F, 75, 78 Ecological sanitation, 519
Gamma, 72, 78, 151 Economic life, 127
Geometric, 68–69, 78 Economic models, 136
Hypergeometrical, 69, 78
Effective porosity, 241
Log-normal, 150
Effective rainfall, 529, 532, 537–541, 543–544,
Negative binomial, 68, 78
565–566, 574
Normal, 73–75, 78, 87–89, 106–108, 114
Efficiency, 138, 302–303, 343
of Drought characteristic, 534
Poisson, 69, 78, 87, 95–96, 106, 112, 114 Hydraulic, 442
Standard-normal, 73–75, 88–89 Effluent standards, 365
t, 75, 78, 74–75 Eigenvalue, 85–86, 113
Uniform, 71, 74, 78, 86–88, 113, 115 Electrodialysis, 490
Weibul, 72–73, 78, 87–88, 90–92, 533–534 Electronic leak detector, 499
DO, See Dissolved oxygen Energy
DO Sag, See Oxygen-Sag model Commercial
Dominated, 34–35, 43 Firm, 443, 446, 451, 453, 455, 456, 459, 462,
DPR model, 316, 338 479, 480
Drainage Off-peak, 444
Area, 486, 508 On-peak, 444
Drawdown, 341 Primary, 442, 443
Dredging, 310
Production, 441–446, 450, 455–456, 464, 476
Drought
Energy balance method, 215
Affected area, 530, 532, 548, 549, 567,
570–571, 576 Engineering economy, 117
Agricultural, 529–530, 553–554, 561 Environment
Climatic, 529–530, 532, 535, 45, 565, 571, Cultural, 4, 6
575 Physical, 1, 6
Duration, 542, 552 Water right, 301, 303, 347
Hydrologic, 529–530, 532, 535, 550–551, 553 Environmental economics , 117
Management, 16, 17, 529 Epilimnion, 413–414
Onset, 532, 540–543, 565, 574 Equipotential lines, 250–252
Prediction, 532, 549–550 Equity, 138–139
Preparedness, 530–531 Eutrophication, 414
Severity, 530, 538, 553, 561–565, 574 Evaporation, 10–11, 193, 196
Termination, 540–541, 544, 549, 553,
Rate, 197, 200, 235, 307–308
564–565
Evapotranspiration, 193, 210–211, 222, 553–560
Dry-land farming, 502
Excess rainfall, 537, 554, 556–557
Dynamic programming, 55–60, 322–323, 328,
358–359 Deficit, 538–542, 574
Backward, 316, 338 Surplus, 538
Deterministic, 316, 322, 358, 407–408 Excess water, 207
Differential, 57 Expectation, 75, 77–78, 81–82, 89, 96, 101–102,
Discrete, 57–58 104, 112
Discrete differential, 57 Expectation optimization, 101
F Genetic algorithm, 60–63

Geologic formation, 239–240
Failure, 309–310, 343–345, 357 Heterogeneity, 240, 263
Feedback structure, 84–85 Homogenous, 240
Finite-difference, 253–256, 259, 270–272, 290 Isotropy, 240
Finite-element Geophone, 499
Grid, 263–264 Geostatistical methods, See Kriging methods
Weighted residual, 253–254, 264, 269 Goodness of fit, Tests, 147, 155–158, 167, 533
Weighting function, 264 Goods
Fitness, 60 Public, 135–136, 140–141
Flood Private, 135–136, 140–141
Control, 191, 200–201, 204–207, 213, Graphic groundwater, 271
218–220, 295, 300–303, 313–314,
Green-Ampt method, 211, 215
317, 340–341, 346
Grey water, 519–520
Damage, 204–207, 213, 218, 232–234
Grid design, 269
Diversion, 206
Gross national product, 487
Forecast, 207
Groundwater modeling, 239, 250, 271
Hydrograph, 212, 301, 314
Plain, 204–206, 212, 216, 219, 222 Calibration, 250, 269–270
Proofing, 207 Optimization, 272–279, 289
Regulation, 206 Simulation, 250, 269–271, 290
Walls, 205 Verification, 270
Warning system, 199, 207 Groundwater
Flood control Flow equation, 239, 243, 245, 248–250,
Structural, 204, 206 255–256, 272
Nonstructural, 219, 205–206 Contamination, 422, 424
Flow measurement Dams, 490–491
Equipment, 491 Management, 239, 250, 272
In open channels, 492 Motion equation, 247
In pipes, 491 Quality, 363, 369, 375, 422–423, 426–428
Flow net, 250–251 Sampling location, 375
Flow-stage relationship, 303 Group decision-making, 48–52
Forgetfulness property, 72, 90
Frequency analysis, 216, 232, 234, 378, 384
Freshwater, 4 H
Future value, 119, 121, 124, 129, 130
Future worth, 126, 128–129, 132 Hare system, 51, 65
Fuzzy Hazard rate, 90–92, 114
Membership function, 340 Head
Region, 340 Gross, 441–442, 444
Rules, 340 High, 444, 459
Fuzzy sets, 108–111 Hydraulic, 239, 240, 242, 249, 256, 445
Algebraic product, 108 Losses, 444, 446
Algebraic sum, 109 Low, 444, 449, 459
Complement, 109 Medium, 444
Concave, 109 Net, 441–442, 444, 46, 450, 457–460, 463,
Convex, 109 467
Intersection, 109 Heat, 364–365, 413
Support of a fuzzy set, 109 HEC-1, 198, 212–216, 229
Union, 109 HEC-2, 212, 216–218, 221
HEC-5, 212, 218–220
HEC-5Q, 414, 416, 419
G
HEC-6, 212, 219–221
Galerkin method, 264, 266–267, 290 Horizontal distribution, 139–140
Gamma function, 534 Horton’s equation, 215
Index 583
Hydraulic conductivity, 212, 234, 240–241, 245, J

249–251, 271, 290, 292, 376
Hydraulic radius, 374, 415 Joint distribution
Hydrograph Conditional distribution, 80
Dimensionless, 207, 209–210, 213, 215 Conditional expectation, 81
Instantaneous unit hydrograph, 209 Continuous, 81
Unit, 209 Correlation, 81
Hydrologic Covariance, 81
Expectation, 81
Cycle, 9–11, 196, 212, 296
Regression curves, 81–82
Seasons, 295
Variance, 82
Variables, 296, 319
Jump, 146
Hydropower
Facilities, 300
Plant efficiency, 302 K
Hydrothermal coordination, 467–470, 472
Hypolimnion, 413–414 Kinematics wave method, 215
Kolmogrov’s theorem, 84
Kolmogrov-Smirnov
I Statistic, 157–158, 187
Test, 155, 157, 164–165, 533
Infeasibility, 101 Kriging methods, 548–549, 565, 567–570
Infiltration, 10 Exponential model, 569
Estimation, 196, 211 Gaussian model, 569
Rate, 211, 215 Hole effect model, 569
Inflow discretization, 324, 327 Power model, 569
Installed capacity, 442–443, 446 Spatial, 548, 570–571
Instream flow, 297, 304, 312, 336, 347, 349 Spatiotemporal, 572
Intangible, 119
Interbasin water transfer, 304–307, 357
L
Interception, 196, 215
Interest Lag time, 199–200, 232
Compound, 120 Lagrange function, 472–473
Discrete compounding, 123–124 Lambda curve, 469–470
Period, 119–120 Land use, 204, 206, 213
Rate, 119–120, 125–126, 130, 134, 141 Laplace equation, 425
Simple, 119, 141 Leak
Intrinsic permeability, 240, 247 Control, 499
Invertibility condition, 171–172 Detector, 499
IRAS, 224–225 Noise correlator, 499
Irrigation efficiency, 503, 509, 511, 522, 524–526 Least square method, 148, 153–154
Levees, 205, 207, 219, 221
Farm, 503
Lifetime, 72, 78, 89–90
Field, 502
Lime precipitation, 518
Project, 502
Linear programming, 23–27
Irrigation methods, 503
Stochastic, 320
Drip, 503 Linearization, 29, 39
Gravity, 503 Linsley’s method, 199, 215, 232
Low-head, 505 Location parameter, 533–534
Microsprayer, 505 Longbein-Duram formula, 390, 400
Sprinkler, 505 Long-term probability vector, 99–100, 114
Surface, 503 Loss function, 311, 316, 329, 334
Isohyetal maps, 548–549 Low flow determination, 383
M Hidden layers, 83–84

Hidden nodes, 84, 113
Macro-location, 371–373 Hidden variables, 83–84
Management information systems, 13 Training process, 83
Markov, 167 Non-basic variable, 27
Markov chain, 96, 99–100, 114 Non-dominated, 34–35, 53
Maximum likelihood method, 147, 152, 154–155, Nonexclusive, 136
166 Nonlinear programming, 28–30, 46, 55
Maxmin rule, 103 Non-point sources, 363–364, 383
Median, 75–77, 112 Nonrival, 136
Median of ratios method, 547 Normalizing procedure, 38
Membership function, 67, 108–111 Numerical method, 250, 252–253
Memory, 7 Nutrients, 364, 386, 414, 512, 514, 519–520
Finite, 7
Infinite, 7
Zero, 7 O
Metering, 498–499
Method of moments, 147, 152–153 Object oriented, 212, 224–226
Micro-location, 371, 373 Objective function, 23–29, 31, 34–35, 43, 46–48,
MIKE-11, 212, 221–224 52, 54–56, 60–64, 101–104, 111,
MIS, See Management information systems 303, 305, 308, 311–312, 316,
Mixing zone, 373–374 319–320, 355
MODFLOW, 271–272 Deterministic, 102
Mode, 75–77, 112 Observers, 82, 84–86, 113
Moments generating function, 77, 112 On-demand delivery system, 510
Money-time relationship, 117, 119 Online measurement, 85
Monitoring, 2–3, 16–17 Open access resources, 119
Network, 202, 369–371, 373, 375–376, 378, Operating rules, 318–319, 337–340
381 Optimization
Wells, 202, 375–378 Model, 227–228
Monitoring programs, 369 Multi-criteria, 30–35
Monte Carlo simulation, 386, 409, 411–412, 431 Single-criterion, 22–23, 31
Moving average, 532, 534–538, 550, 574 Organic compounds, 365, 425
Model, 171–172 Outlying observation, 147
Process, 171 Oxygen
Multi-criterion-decision-analysis, 2, 15, 17, 19 Deficit, 392–393, 397, 407
Multiple criteria problem, 370, 419 Depletion, 365,387, 414, 419
Multiple-criterion-decision-making, 92 Sag model, 387
Multivariate, 147, 175–179, 182 Oxygen-demanding material, 364
Muskingum-Cung method, 219
Mutation, 61–62
Mutually exclusive, 126–127, 134, 142 P
Pair-wise comparison, 37, 50, 51, 65
N Palmer drought severity index (PDSI), 553–554,
561, 563–566
Nash theorem, 52–54, 65, 521, 524 Parameter, 7
Bargaining solution, 52, 55 Parshal flume, 491–492
Bargaining theory, 13, 287, 419, 428 Partial autocorrelation, 169–170, 172–173, 182
Nonsymmetric, 55, 65, 292 Partition coefficient, 368
Product, 54 Pathogens, 518–520
Solution, 230, 420, 478–479 PDSI, See Palmer drought severity index
Navigation, 298, 302 Penstock, 300–301, 307
Requirement, 301, 303, 337 Performance criteria, 343
Neural networks, 82–84, 89, 113, 339–340 Periodicity, 146–147, 164–165
Index 585
Removing, 161, 147 Probability

Periodogram, 164 Distribution, See Distribution function
Piecewise approximation, 308, 396 Vector, 97–100, 114
Planning models Probability density function, 310, 384, 534, 551
Long-term, 463 Probability-probability graphs, 533
Mid-term, 465 Process
Short-term, 465 Nonstationary, 146
Plant factor, 443, 446 Stationary, 146
Plurality voting, 48, 50–51, 65 Stochastic, 145–146
Point elevation, 532, 537, 547 Production
Point sources, 33–364, 373, 383, 395, 429 Function, 136
Poisson process, 95–96, 114 Possibility curve, 136–137
Policy decision, 319, 322 Progressive programs, 139
Pollutants, 363–365 Project lifetime, 125–126, 135
Conservative, 368 Pumping, 250, 256, 258–259, 273, 278–280, 290
Growing, 368 Pumps, 497, 505
Nonconservative, 368 Booster, 497
Ponding, 212 Centrifugal, 497
Population moments, 152–153 Circulation, 497
Population projection, 487 High-service, 497
Porte Manteau test, 162 Impeller, 497
Degree of freedom, 163 Lift, 517
Posterior probability, 329 Low-life, 497
Power Variable speed, 517
Firm, See Energy Well, 497
Generation, 302, 312–313, 343, 356, 359–360
Plant, 300–302, 356, 359
Primary, See Energy Q
Secondary, 443
QUAL2E, 386, 430–437
Power load, 446
QUAL2E-UNCAS, 386
Base, 443
Quantile-quantile graphs, 533
Factor, 443
Intermediate, 443
Peak, 443
R
Weekly, 447–448
Power load estimation, 446 Rainwater collecting, 488
Economic analysis, 446 Rainfall deficit, 529, 532, 538–542, 574
End-use analysis, 446 Rainfall-runoff modeling, 198, 210, 213, 216,
Trend analysis, 446 222, 229
Power plant, 444 Random component, 146, 148–150, 167
Efficiency, 450 Random number generator, 86, 89
Macrohydro, 444 Random variable, 67–71, 75, 77–78, 81, 86–87,
Microhydro, 444 89–90, 96, 101, 105–106, 112,
Pumped storage, 444 145–146, 152, 171, 174, 176
Run-of-river, 444 Continuous, 70–71, 77, 87
Storage, 444 Discrete, 67, 69–70, 75, 86, 112, 115
Power potential estimation, 449 Randomness, 146, 165–166, 181
Flow-duration curve, 449–451 Rate of return
Sequential stream flow routing, 449, 453 External, 126, 129–130
Precipitation, 10–11 Internal, 126, 130, 142
Present value, 121–122, 129–130, 305, 307, 343 Minimum attractive, 127, 129
Present worth, 121–122, 124–135, 142 Reaeration rate, 389, 390, 400
Primal form, 24, 26, 63 Real-time operation, 16–17, 462, 466
Probabilistic methods, 67, 82, 108 Recharge coefficient, 558
Recreation, 298, 341, 357 Runoff

Recursive equation, See Dynamic programming Calculation, 556
Refill, 304–341 Coefficient, 556, 558
Regional analysis, 545 Potential, 557–558
Regionalization, 216, 548, 570, 572
Regressive programs, 139, 140, 143
Real-time operation, 296, 337, 340 S
Reliability, 89, 91, 320–321, 338, 343–344, 446,
Safe range, 317, 342–343
448, 453, 466
Safe yield, 272
Function, 90–94
Salts, 364–365
Theory, 95
Salvage value, 131, 133, 141
Reliability-based design, 309–310, 357
Sample moments, 152–153
Repeatability assumption, 127–128, 131, 142
Sampling
Requirement models, 495
Frequency, 202–203, 375, 378–382, 429
Reservoir
Network, 202
Cascade, 334–335
SAR, 366–367, 371, 387, 434
Leakage, 307, 335
Satellite
Parallel, 334–36, 340–341
Information, 204, 207
Sizing, 309, 313 Remote sensing, 205
Reservoir operation Resolution, 205, 207, 235
Stochastic modeling, 201, 206, 219–220, 224, Scale parameter, 534
295, 320 Security, 448, 466–467
Reservoir storage Sedimentation, 310
Active, 299, 302–303, 310, 313 Semivariogram, 568, 570
Dead, 302 Sequential optimization, 35–36, 64
Discretization, 323, 325 Seven-day minimum average flow, 383–386,
Flood control, 32, 313–314, 317, 346 399–400
Zones, 302 Severity index, 345–346
Resiliency, 343–345 Shape factor, 194, 213
Resource-supply function, 137 Shape function, 264–265, 267
Response equation, 256, 264, 266, 273, 275, Shape parameter, 534
277–278, 285 Shear velocity, 374
Return flows, 201, 203, 229, 297, 335 Simplex method, 27–28, 48
Return period Simulation model, 82, 218, 226
309–310, 314–315, 357 Single criterion problem, 22–23, 31
Revenue, 117–118, 126, 142 Sinking fund factor, 121, 124
Reverse osmosis, 490 Skewness test, 158–160
Reynolds transport theorem, 367, 423 Slack variable, 28
Risk, 67, 103–104 Sliding scale, 540–544
Rival, 136 Sluice way, 299–300
River Snow
Channel form, 198 Cover, 205
Channel roughness, 197–198, 221 Line, 205
Cross section, 197–198, 218, 220–221 Water , 205
Length, 197, 200 Snow melt, 207, 215
Quality management, 228 Snyder’s method, 199, 215, 232
Slope, 193, 198 Social choice, 48–52, 65, 136
River basin modeling, 191 Sodium absorption ratio, See SAR
River-reservoir system, 295 Soil
Rotating biological contactors, 518 Characteristics, 195
Routing Development,
Reservoir , 213–215, 219 Hydraulics properties, 239–241
Stream, 219 Management
Rule curve, 337–339 Structure, 195
Run theory, 530, 551–552 Texture, 195
Index 587
Soil moisture, 195–196, 205, 207, 215, 530, 538, Parallel, 93, 114
554 Properties, 6
Anomaly index, 554 Quantized, 7
Capacity, 554, 556–557 Series, 93
Condition, 555 Simple, 7
Deficit, 553–554, 556–557
Simulation, 8
Loss coefficient, 558
Stable, 8
Sorbed pollutant concentration, 368
Source monitoring, 375 Stochastic, 7
Specific storage, 243, 249 Theory, 85, 98
Spinning reserve, 468
State of nature, 7
Stationarity, 172–173, 181 T
Stationary condition, 168
Stationary solution, 323, 328 Tactic, 462
Statistical distribution, 532–533, 545 Tail water, 441, 444–445, 456–460, 480
Steady state flow, 216 Tariff, 498, 00, 511
Stethoscope, 499 Termination, 162
Stochastic Test of independence in time, 162
Dynamic programming, 82, 86 Test of normality, 147
Matrix, 95 Thermal plant, 468–474
Process, 97 Thermocline, 413
Simulation, 102
Thiessen polygon network, 545, 547
Storage coefficient, 242, 248–249, 271, 287–288,
Thomas Fiering, 167
290, 532
Strategy, 462 Thornthwaite method, 210–211, 501, 553–555
Stratification, 374, 413–414 Time of concentration, 199, 206, 215
Stream flow Time series
Estimation , 207–209 Modeling, 11, 145
Hydrograph, 198, 210 Toxic metals , 364–365
Streeter and Phelps equation, 388, 393–394, 400, Transformation, 147, 150–152, 160
407, 409 Transition
Subsurface flow, 10, 250 Matrix, 96, 326
Superfluous variable, 102 Probability, 97–98, 114, 326, 328, 333, 358
Supply curve, 487–488
Transmission loss, 464, 468, 471
Suspended soils, 364, 368
Transmissivity, 242
Synthetic problems, 8, 319
Synthetic stream flows, 319 Transpiration, 197
Synthetic time series, 167, 176 Trash-rack, 446
System Treatment
Adaptive, 8 Cost, 395, 397–399, 405, 407, 411, 419, 426,
Causal, 8 429–431
Complex, 7 Efficiency, 395–396, 398–399, 403, 405, 407,
Continuous, 7 409
Damped, 8 Trend, 146
Definition, 6 Linear function, 146
Degree of integration, 6 Polynomial function, 146
Deterministic, 7
Power function, 146
Devised, 7
Removal, 146–150
Discrete, 7
Failure, 93 Trickling filters, 517–518, 520
Hydrologic, 8 Turbine, 442–446, 450, 453, 456, 459, 479–480
Identification problem, 8 Turnover, 413
Linear, 85, 98, 100 Turning point test, 165–166, 181
Natural, 7 Tylor’s polynomial, 29, 59
U Supply, 302, 304, 311–313, 320, 334,

340–341, 347–349, 354, 359–360
Unaccounted-for water, 498–499, 521–522, 524 Vapor, 196–197
Unconventional water supply methods, 488 Withdrawal, 1, 5
Unit commitment, 462, 46–467 Water demand, 296–298, 304–309, 311–312, 333,
Upgradiant zone, 376–378 335, 337–338, 344, 346–349,
Utility function, 38–39, 286–288, 347, 349–350, 353–354, 356–358, 483
352, 356, 360 Agricultural, 272, 276, 279, 282, 289, 297,
312, 337, 347, 349, 351, 354, 356,
484–487, 492, 501–503, 509
V Analysis, 483
Domestic, 285–289, 297, 347, 349–350, 354,
Valve 356, 359–360
Air release, 497 Environmental, 486, 512
Altitude, 497 Forecasting, 486–487
Check, 497 Function, 136–138
Control, 498 Industrial, 272, 277, 285–290, 327, 301,
Shutoff, 497 347–349, 352, 354, 356–357,
Variable 484–487, 511–514, 516
Characteristic, 146 Management, 16, 17, 483
Continuous, 11 Models, 496
Discrete, 11 Municipal, 485–486, 492–496, 498, 515–516
Distributed, 7 Points, 200–201, 229
Hydrologic, 145–146 Water distribution networks, 496
Lumped, 7 Junction, 497
Qualitative, 11 Pipe, 496
Quantitative, 11 Pumps, 497
Variance, 75, 78, 81, 88, 89 Reservoir, 496
Variogram, 567–571 Tank, 496
Venturi, 491, 499 Valve, 497
Vertical distribution, 139 Water loss
Virus, 364, 432 Apparent, 499
Vulnerability, 2, 343, 345–346 Real, 499
Reduction, 498
Water Quality
W Management, 363
Management models, 395, 42, 426
WASP5, 386, 436–437 Monitoring, 369–373, 381, 416
Wastewater reuse, 203, 501, 509, 511, 513–516, Primary standards, 366
519 Requirement, 201
Wastewater treatment, 395, 396, 409, 426, 484, Secondary standards, 366
486, 500, 511, 515, 519 Standards, 201, 229, 363, 365–367, 394, 399,
Advanced, 518 407, 426
Preliminary, 515, 517 Stations, 371–375, 381–383, 395, 422, 426,
Primary, 515–517 402
Secondary, 515–518 Water resources
Water Planning process, 2
Allocation, 200, 226–227, 230–231, 235, 273, Systems, 4–6
276–279, 285–287, 289, 292 Water reuse, 498, 501, 509, 513, 515
Availability, 4 Direct, 513
Budget, 10 Indirect, 513
Charge, 500 Intentional, 514
Pricing, 498, 500 Watershed, 10–11
Scarcity, 1 Area, 198–199, 207–208
Storage, 224 Boundary, 191
Index 589
Definition, 191 Onsite, 484

Geomorphology, 191 Per-capita, 492–493, 495
Improvement, 206 Weather modification, 491
Length, 193–194, 197 Weighting method, 36–37, 46, 64, 545
Perimeter, 194, 213 Willingness to pay, 117, 137
Physical characteristics, 191–144 Marginal, 117, 137
Shape, 193, 197, 199, 232 Worst-case analysis, 103
Slope, 193, 197
Storage coefficient, 199, 215, 232
Water use, 484–487 Y
Agricultural, 484
Yellow water, 519
Consumptive, 484 Yule-Walker, 168
Disaggregation, 494–495
Domestic, 484
Instream, 484 Z
Municipal, 484
Off-stream, 484, 485 Z-Index, See Drought severity index

Water Resources Systems Analysis

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Water Resources Systems Analysis

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Water Resources Systems Analysis

Uploaded by

Copyright:

Available Formats

WATER RESOURCES

Visit the CRC Press Web site at www.crcpress.com

© 2003 by CRC Press LLC

No claim to original U.S. Government works

Ferenc Szidarovszky is a graduate of the Eotvos University of Sciences, Budapest,

Banafsheh Zahraie is an assistant professor at the Department of Civil Engineering

Chapter 2 Decision Making: Optimization and Conflict Resolution.................21

Chapter 3 Decision Making under Uncertainty..................................................67

Chapter 4 Water Resources Economics............................................................117

Chapter 5 Time Series Analysis........................................................................145

Chapter 6 River Basin Modeling ......................................................................191

Chapter 7 Groundwater Management...............................................................239

Chapter 8 River–Reservoir Systems Modeling ................................................295

Chapter 9 Water Quality Management .............................................................363

Chapter 10 Hydroelectric Systems Analysis ....................................................441

Chapter 12 Drought Management ....................................................................529

1. Problem definition and data collection and processing

• Climatic, hydrologic, hydraulic, and environmental characteristics of the

FIGURE 1.2 Process of water resources planning.

economic (nontechnical) issues related to institutional framework and distribution

World ocean 1,338,000 96.5 —

1.2 AVAILABILITY OF WATER ON EARTH AND WATER USE

1.3 WATER RESOURCES SYSTEMS: BASIC CONCEPTS

Annual Internal Annual Withdrawal Sector Withdrawal

Africa 3.996 1995 0.15 4 7 5 88

The second phase in systems analysis is to integrate different components of the

1.4 SYSTEM DEFINITION AND PROPERTIES

• All systems have some structure and organization.

• Lumped, which do not change in space

Memory, which is another important characteristic of a system, is the length of time

Systems are generally classified as deterministic or stochastic. In deterministic

• Continuous systems — The output is produced continuously.

• Adaptive systems — These systems learn from their past history to

Hydrologists primarily deal with design and synthetic problems.

Time Time Time

a) Continuous systems b) Discrete systems c) Quantized systems

FIGURE 1.3 Outputs of continuous, discrete, and quantized systems.

Input operator Output

1.5 HYDROLOGIC SYSTEMS AND MODELING

• Theoretical approach, which focuses on modeling the physics of the

• Empirical vs. theoretical

1.5.1 HYDROLOGIC CYCLE

FIGURE 1.5 Schematic diagram of the hydrologic cycle.

ΔSs = Pl + Rin − Rout + Rg − Ts − I (1.1)

ΔSg = Gin − Gout − Rg − Eg + I (1.2)

1.5.2 HYDROLOGIC VARIABLES AND PARAMETERS

1.6 CONFLICT ISSUES IN WATER RESOURCES

• Several demand points exist and groundwater that is supplied to one of

• Use water resources for producing specific goods and products.

on organizational setups in some developing countries. The qualitative aspects of

1.7 DECISION TOOLS IN WATER RESOURCES

Water Resources Systems Analysis

User Interface (Q+A Sessions):

• Decisions under certainty

• Utilizing genetic algorithms and other optimization techniques to search

alternative by determining whether or not it satisfies all restrictions. If the alternatives

2.2 SINGLE-CRITERION OPTIMIZATION

where c1, c2, …, cn are real numbers, c = (c1, …, cn), and

it becomes nonnegative. If a decision variable can be both positive and negative,