Shallow Water Hydraulics

Oscar Castro-Orgaz
Willi H. Hager
Shallow
Water
Hydraulics
Shallow Water Hydraulics
Oscar Castro-Orgaz Willi H. Hager
•
Shallow Water Hydraulics
123
Oscar Castro-Orgaz Willi H. Hager
University of Córdoba VAW, ETH Zürich
Córdoba, Spain Zürich, Switzerland
ISBN 978-3-030-13072-5 ISBN 978-3-030-13073-2 (eBook)

https://doi.org/10.1007/978-3-030-13073-2
© Springer Nature Switzerland AG 2019
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To our Families
Preface
The purpose of Shallow Water Hydraulics is to present the theory and the com-
putation of open channel flows using analytical, numerical, and experimental
results. The book’s target audience includes graduate and undergraduate students as
well as practitioners of civil and environmental engineering. The book is concep-
tually divided into four parts: fundamental equations (Chap. 1), steady open channel
flows (Chaps. 2–4), unsteady open channel flows (Chaps. 5–9), and special topics
(Chaps. 10 and 11). As supporting learning material for students, a library of
numerical codes is available in Chap. 12. The complexities of theory and numerical
methods are progressively increased toward the end of the book, rendering pieces
of the material useful for courses of different levels.
It appears impossible to start the presentation of a new open channel flow book
without acknowledging the former works of Ven Te Chow (Open Channel
Hydraulics, 1959) and Francis M. Henderson (Open Channel Flow, 1966). Despite
Chow produced an encyclopedic treatise, it is Henderson’s outstanding book that
the authors followed. His presentation of concepts is simply unique, and although it
was published more than 50 years ago, it is still a main reference. Credit deserves
also the books of Subhash C. Jain (Open Channel Flow, 2001) and Hubert Chanson
(The Hydraulics of Open Channel Flow: An introduction, 2004), similar to
Henderson’s, pursuing a clear and brilliant presentation of open channel flow
concepts. These books offer a first open channel flow course, covering topics from
steady flow to fundamental questions relating to unsteady flow. However, they are
lacking a detailed exposition of numerical methods to solve the unsteady open
channel flow equations. The books of Sergio Montes (Hydraulics of Open Channel
Flow, 1998), M. Hanif Chaudhry (Open Channel Flow, 2008), and Eleuterio Toro
(Shock-Capturing Methods for Free Surface Shallow Flows, 2001) cover these
aspects in detail and can be used in an advanced open channel flow course. The
above-quoted books were used by the first author to offer a course on Operation of
Rivers and Reservoirs in the Environmental Hydraulics Master program held at the
University of Córdoba, Spain. The purpose was to give students a complete
overview on the computation of open channel flows, starting with the basic
equations, continuing with the solution of steady open channel flows, penetrating
vii
viii Preface
into the computation of unsteady open channel flows using modern numerical
methods, and finally introducing advanced topics including flows on movable beds
and sediment transport and non-hydrostatic flow modeling. The material of this
book originated from the lecture notes prepared for this course, and deep recog-
nition is therefore given to the books quoted above, which were used as a source of
knowledge and inspiration for years.
The authors are of the opinion that it is better to start from generalized equations
and then simplify the results using suitable approximations. From this perspective,
the fundamental equations of open channel flow are settled producing a rigorous
vertical integration of the 3D RANS equations (Chap. 1). The resulting mathe-
matical statements are then simplified for selected flows; in particular, the impli-
cations of assuming the hydrostatic pressure distribution are discussed in detail. The
emphasis of the book is directed to one-dimensional problems. The corresponding
steady flows are therefore described using the energy and momentum principles,
and a variety of numerical solutions to the algebraic and ordinary differential
equations governing these flows are presented (Chaps. 2 and 3). The transitions
across the critical depth, along with the formation of singular points in weir flow
and the hydraulic jump beyond a sluice gate, are detailed (Chap. 4). The funda-
mental equations of unsteady open channel flows along with their continuous and
discontinuous solutions are then presented, and modern shock-capturing
finite-difference and finite volume methods to solve them are extensively descri-
bed (Chaps. 5–9). A detailed description of dam break waves (Chap. 6) and sluice
gate maneuvers in open channels treating them as the solution of a Riemann
problem (Chap. 8) is provided. Special topics selected are the inclusion of sediment
transport and movable beds in shallow water models, as used to predict geomorphic
dam break waves (Chap. 10), and the computation of steady and unsteady
non-hydrostatic free surface flows (Chap. 11).
A collection of hand-solved exercises along the book is not presented. Rather,
the book offers to students and instructors a collection of source codes where each
type of problem discussed in the book is implemented step by step. The collection
of source codes is written in Visual Basic, and each code is inserted as a macro in
Microsoft Excel®. Teaching experience indicated that this approach permits stu-
dents an easy use of the material, and a productive interaction with the lecturer
during the classes is generated.
The authors hope that after studying the book, the reader will have a solid
background on the theory and computation of steady and unsteady open channel
flows, permitting to advance to the study of more complex problems relating to
two-dimensional numerical modeling of free surface flows.
Córdoba, Spain Oscar Castro-Orgaz

Zürich, Switzerland Willi H. Hager
January 2019
Acknowledgements
Oscar Castro-Orgaz would like to deeply thank Jerónimo Puertas Agudo,

University of Coruña, Spain, for kindly providing him with his open channel flow
lecture notes. This material was extensively used by the first author for teaching and
to prepare the original part of this book relating to steady flows. In turn, it has had a
profound influence on the first author’s background on open channel hydraulics.
Deep thanks are also for Prof. M. J. Polo, University of Córdoba, for her sug-
gestion to prepare this course.
ix
Contents
1 Fundamental Equations of Free Surface Flows . . . . . . . . . . . . . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Classification of Open Channel Flows . . . . . . . . . . . . . 1
1.1.2 Aims and Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 General Depth-Integrated Equations over a 3D Terrain . . . . . . . 5
1.2.1 Previous Open Channel Flow Developments . . . . . . . . 5
1.2.2 3D Flow Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.3 General Kinematic Boundary Conditions . . . . . . . . . . . 10
1.2.4 Continuity Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.5 Momentum Equations . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Saint-Venant Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.1 2D Flows in the Horizontal Plane . . . . . . . . . . . . . . . . 19
1.3.2 1D Cross-Sectional Averaged Equations . . . . . . . . . . . 24
1.4 Non-hydrostatic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.4.1 1D Shallow Water Non-hydrostatic Equations . . . . . . . 33
1.4.2 Uniform and Gradually Varied Flow on Steep
Slopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 37
1.5 Sediment Transport and Movable Beds . . . . . . . . . . . . . . . ... 41
1.5.1 1D Cross-Sectional Averaged Continuity Equation . ... 41
1.5.2 1D Cross-Sectional Averaged x-Momentum
Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 43
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... 47
2 Energy and Momentum Principles . . . . . . . . . . . .. . . . . . . . . . . . . 51
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . 51
2.2 Energy Principle . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . 54
2.2.1 Specific Energy Head . . . . . . . . . . . .. . . . . . . . . . . . . 54
2.2.2 Location of Critical Flow . . . . . . . . .. . . . . . . . . . . . . 59
2.2.3 Specific Energy Head for General Sections . . . . . . . . . 66
2.2.4 Critical Flow and Wave Motion . ........ . . . . . . . . 67
xi
xii Contents
2.2.5 Computation of Critical Flow in Arbitrary Sections . . . 73

2.2.6 Channel Transitions by Change of Bed Elevation . . . . . 78
2.2.7 Channel Transitions by Change of Channel Width . . . . 89
2.3 Momentum Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.3.1 Specific Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.3.2 Bélanger’s Equation for the Hydraulic Jump . . . . . . . . 98
2.3.3 Computation of Sequent Depths for General Cross
Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
2.3.4 Transitions in Supercritical Flow . . . . . . . . . . . . . . . . . 106
2.4 Control Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.4.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.4.2 Uniform Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2.4.3 Critical Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
2.4.4 Weir Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
2.4.5 Gate Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
2.5 Application: Numerical Solution of Energy–Momentum
Equations for Gate Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
3 Computation of Steady Gradually-Varied Flows . . . . . . . . . . . . . . . 137
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
3.2 Governing Equation of Non-uniform Flow . . . . . . . . . . . . . . . . 137
3.3 Uniform Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
3.3.2 Flow Resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
3.3.3 Uniform Flow Depth . . . . . . . . . . . . . . . . . . . . . . . . . 143
3.4 Flow Profiles in Prismatic Channels . . . . . . . . . . . . . . . . . . . . . 147
3.4.1 Qualitative Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 147
3.4.2 Sketching of Flow Profiles . . . . . . . . . . . . . . . . . . . . . 156
3.4.3 Flow Profile for Unknown Discharge . . . . . . . . . . . . . 157
3.5 Computation of Steady Flow Profiles . . . . . . . . . . . . . . . . . . . . 159
3.5.1 Governing Integral and Differential Equations . . . . . . . 159
3.5.2 Boundary Condition . . . . . . . . . . . . . . . . . . . . . . . . . . 161
3.5.3 Analytical Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
3.5.4 Implicit Integral Method . . . . . . . . . . . . . . . . . . . . . . . 166
3.5.5 Explicit Integral Method . . . . . . . . . . . . . . . . . . . . . . . 169
3.5.6 Euler’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
3.5.7 Fourth-Order Runge–Kutta Method . . . . . . . . . . . . . . . 170
3.5.8 Predictor–Corrector Method . . . . . . . . . . . . . . . . . . . . 172
3.6 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
3.6.1 Test of High-Resolution Method . . . . . . . . . . . . . . . . . 173
3.6.2 Comparative of Numerical Methods . . . . . . . . . . . . . . 173
Contents xiii
3.6.3 Flow on Mild Slope . . . . . . . . . . . . . . . . . . . . . . . . . . 175

3.6.4 Flow on Steep Slope . . . . . . . . . . . . . . . . . . . . . . . . . 175
3.7 Compound Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
4 Computation of Steady Transcritical Open Channel Flows . . . . . . . 183
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
4.2 Transition from Sub- to Supercritical Flow . . . . . . . . . . . . . . . . 183
4.2.1 Formation of Singular Points in Free Surface Flows . . . 183
4.2.2 Determination of Water Surface Slope at Critical
Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
4.3 Transition from Super- to Subcritical Flows . . . . . . . . . . . . . . . 189
4.3.1 Control Volume Equation for Hydraulic Jump . . . . . . . 189
4.3.2 Determination of Hydraulic Jump Position . . . . . . . . . . 193
4.4 Computational Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
4.4.1 Flow over Round-Crested Weirs . . . . . . . . . . . . . . . . . 196
4.4.2 Hydraulic Jump Beyond a Sluice Gate . . . . . . . . . . . . 198
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5 Unsteady Open Channel Flows: Basic Solutions . . . . . . . . . . . . . . . 201
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5.2 Shallow Water Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5.3 Discontinuous Solutions: Basic Equations of a Positive
Surge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.3.2 Reynolds Transport Theorem . . . . . . . . . . . . . . . . . . . 205
5.3.3 Positive Surges Moving Upstream in Supercritical
Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
5.3.4 Formulation Used in Finite Volume Methods . . . . . . . . 213
5.4 Methods of Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
5.4.1 Shock-Capturing Methods . . . . . . . . . . . . . . . . . . . . . . 214
5.4.2 Outline of Solution Methods . . . . . . . . . . . . . . . . . . . . 216
5.5 Method of Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
5.5.1 SWE in Characteristic Form . . . . . . . . . . . . . . . . . . . . 217
5.5.2 Initial and Boundary Conditions . . . . . . . . . . . . . . . . . 221
5.5.3 Wave Celerity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.6 Simple Wave Problem: Basic Equations of Rarefaction
Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.6.1 The Simple Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.6.2 Rarefaction Wave: A Negative Surge . . . . . . . . . . . . . 232
5.7 Simplified Models: The Kinematic Wave . . . . . . . . . . . . . . . . . 234
5.8 Use of Non-conservative Form of SWE . . . . . . . . . . . . . . . . . . 238
5.9 Limitations of SWE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
xiv Contents
5.10 Hydrologic Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

5.10.1 Reservoir Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
5.10.2 Muskingum Channel Routing . . . . . . . . . . . . . . . . . . . 248
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
6 Ideal Dam Break Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
6.2 Dam Break Wave Under Dry Tailwater Conditions . . . . . . . . . . 254
6.3 Dam Break Wave Under Wet Tailwater Conditions . . . . . . . . . 260
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
7 Finite Difference Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
7.2 Basic Numerical Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
7.2.1 Remark on Basic Numerical Concepts . . . . . . . . . . . . . 273
7.2.2 Finite-Difference Approximations . . . . . . . . . . . . . . . . 274
7.2.3 Shock-Capturing Schemes . . . . . . . . . . . . . . . . . . . . . . 278
7.2.4 Explicit and Implicit Schemes: Stability . . . . . . . . . . . . 280
7.3 Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . 285
7.3.1 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
7.3.2 Boundary Conditions for Continuous Flows . . . . . . . . . 286
7.3.3 Boundary Conditions for Discontinuous Flows . . . . . . . 288
7.4 Explicit Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
7.4.1 FTCS Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
7.4.2 Lax’ Diffusive Scheme . . . . . . . . . . . . . . . . . . . . . . . . 290
7.4.3 MacCormack Predictor–Corrector Scheme . . . . . . . . . . 291
7.4.4 MacCormack Scheme with Calibrated Artificial
Viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
7.4.5 TVD MacCormack Scheme . . . . . . . . . . . . . . . . . . . . 293
7.4.6 Upwind Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
7.5.1 Subcritical Dam Break Wave with Wet Tailwater . . . . . 300
7.5.2 Transcritical Dam Break Wave with Wet Tailwater . . . 303
7.5.3 Subcritical Surge . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
7.5.4 Supercritical Surge . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
7.5.5 Positive Surge with Friction . . . . . . . . . . . . . . . . . . . . 308
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
8 The Riemann Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
8.2 Wet-Bed Exact Riemann Solver . . . . . . . . . . . . . . . . . . . . . . . . 315
8.2.1 Wave Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
8.2.2 Solution of Star Region . . . . . . . . . . . . . . . . . . . . . . . 318
Contents xv
8.2.3 Complete Wave Profiles . . . . . . . . . . . . . . . . . . . . . . . 320

8.2.4 Possible Wave Patterns and Computation of Fluxes . . . 321
8.2.5 Computational Examples . . . . . . . . . . . . . . . . . . . . . . 323
8.3 Dry-Bed Exact Riemann Solver . . . . . . . . . . . . . . . . . . . . . . . . 327
8.3.1 Wet–Dry Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
8.3.2 Possible Wave Patterns . . . . . . . . . . . . . . . . . . . . . . . . 327
8.4 Application: Gate Maneuvers in Open Channels . . . . . . . . . . . . 332
8.4.1 Complete Gate Opening . . . . . . . . . . . . . . . . . . . . . . . 332
8.4.2 Partial Gate Opening . . . . . . . . . . . . . . . . . . . . . . . . . 333
8.4.3 Complete Gate Closure . . . . . . . . . . . . . . . . . . . . . . . . 337
8.4.4 Partial Gate Closure . . . . . . . . . . . . . . . . . . . . . . . . . . 338
8.4.5 Comparison with Experiments . . . . . . . . . . . . . . . . . . . 341
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
9 Finite Volume Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
9.2 Godunov-Type Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
9.2.1 Conservative Formula . . . . . . . . . . . . . . . . . . . . . . . . . 348
9.2.2 Conservative Property: Definition of Numerical
Discharge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
9.2.3 Godunov Upwind Numerical Flux . . . . . . . . . . . . . . . . 351
9.2.4 Stability Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
9.2.5 Computational Sequence for Godunov First-Order
Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
9.3 Approximate Riemann Solvers . . . . . . . . . . . . . . . . . . . . . . . . . 361
9.3.1 HLL Riemann Solver . . . . . . . . . . . . . . . . . . . . . . . . . 361
9.3.2 Lax Numerical Flux . . . . . . . . . . . . . . . . . . . . . . . . . . 365
9.3.3 Roe Riemann Solver . . . . . . . . . . . . . . . . . . . . . . . . . . 368
9.4 Dry-Bed Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
9.5 Source Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
9.5.1 Splitting Technique . . . . . . . . . . . . . . . . . . . . . . . . . . 370
9.5.2 ODE Solvers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
9.5.3 Well-Balanced Schemes . . . . . . . . . . . . . . . . . . . . . . . 373
9.5.4 Treatment of Friction . . . . . . . . . . . . . . . . . . . . . . . . . 375
9.6 One-Sided First-Order Upwind Scheme . . . . . . . . . . . . . . . . . . 376
9.7 MUSCL-Hancock Second-Order TVD Scheme . . . . . . . . . . . . . 378
9.7.1 MUSCL Reconstruction . . . . . . . . . . . . . . . . . . . . . . . 378
9.7.2 Slope Limiting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
9.7.3 Hancock Step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
9.7.4 Computational Sequence . . . . . . . . . . . . . . . . . . . . . . . 385
9.7.5 Surface Gradient Method . . . . . . . . . . . . . . . . . . . . . . 386
xvi Contents
9.8 Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . 388

9.8.1 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 388
9.8.2 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
9.9.1 Ideal Dam Break Waves . . . . . . . . . . . . . . . . . . . . . . . 392
9.9.2 Dam Break Waves with Friction . . . . . . . . . . . . . . . . . 393
9.9.3 Positive Surge with Friction . . . . . . . . . . . . . . . . . . . . 396
9.9.4 Hydraulic Jump Beyond a Sluice Gate . . . . . . . . . . . . 399
9.9.5 Flow over Round-Crested Weirs . . . . . . . . . . . . . . . . . 401
9.9.6 Dam Break Wave Evolution over a Bottom Sill . . . . . . 405
9.9.7 Solitary Wave Run-Up on a Plane Beach . . . . . . . . . . . 408
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
10 Sediment Transport and Movable Beds . . . . . . . . . . . . . . . . . . . . . 417
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
10.2 Flow Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
10.2.1 Shallow Water Equations . . . . . . . . . . . . . . . . . . . . . . 417
10.2.2 Sediment Transport Layers and Bed Deformation . . . . . 419
10.2.3 Modified Flow Equations . . . . . . . . . . . . . . . . . . . . . . 420
10.2.4 Total-Load Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 420
10.2.5 Sediment Transport Closure . . . . . . . . . . . . . . . . . . . . 421
10.3 Numerical Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
10.3.1 Conservation Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
10.3.2 First-Order Upwind Finite Volume Method . . . . . . . . . 424
10.3.3 Source Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
10.3.4 Stability Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
10.3.5 Computational Sequence . . . . . . . . . . . . . . . . . . . . . . . 426
10.4 Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
10.4.1 Dam Break Wave over Movable Bed . . . . . . . . . . . . . 426
10.4.2 Dike Erosion due to Overtopping . . . . . . . . . . . . . . . . 430
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
11 Numerical Modeling of Non-hydrostatic Free Surface Flows . . . . . 435
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
11.2 Two-Dimensional Steady Potential Flow . . . . . . . . . . . . . . . . . 436
11.2.1 Basic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
11.2.2 Picard Iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
11.2.3 Spillway Crest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
11.2.4 Steep Slope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
11.2.5 Flow over Round-Crested Weir: Numerical
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
11.2.6 Transition from Mild to Steep Slope . . . . . . . . . . . . . . 454
11.2.7 Flow over Trapezoidal Profiled Weir . . . . . . . . . . . . . . 456
Contents xvii
11.3 Unsteady Ideal Fluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 456

11.3.1 Two-Dimensional Serre Equations . . . . . . . . . . . . . . . . 456
11.3.2 One-Dimensional Serre Equations . . . . . . . . . . . . . . . . 461
11.3.3 Solitary Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
11.3.4 MUSCL-Hancock Scheme . . . . . . . . . . . . . . . . . . . . . 467
11.3.5 High-Order Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . 473
11.4 Unsteady Flow Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
11.4.1 Flow over Round-Crested Weir . . . . . . . . . . . . . . . . . . 475
11.4.2 Solitary Wave Propagation . . . . . . . . . . . . . . . . . . . . . 480
11.4.3 Dam Break Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
12 Numerical Library of Shallow Water Equations . . . . . . . . . . . . . . . 493
12.1 The Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
12.1.1 Introduction: Aims and Scope . . . . . . . . . . . . . . . . . . . 493
12.1.2 Summary of Available Codes . . . . . . . . . . . . . . . . . . . 494
12.2 Examples of Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
12.2.1 Gradually Varied Flow Computation Using
Newton–Raphson Method . . . . . . . . . . . . . . . . . . . . . . 494
12.2.2 Dam Break Wave Code . . . . . . . . . . . . . . . . . . . . . . . 501
12.2.3 Flooding over an obstacle code . . . . . . . . . . . . . . . . . . 513
12.3 Using the Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
12.3.1 Short Tutorial: Flooding of an Obstacle . . . . . . . . . . . . 532
12.3.2 Example of Homework 1: Computation of Gradually-
Varied Steady Flow Profiles . . . . . . . . . . . . . . . . . . . . 538
12.3.3 Example of Homework 2: Wetting and Drying over
Uneven Topography . . . . . . . . . . . . . . . . . . . . . . . . . . 542
12.3.4 Example of Homework 3: Study of Different
Discretizations of the Friction Source Term . . . . . . . . . 546
12.3.5 Example of Homework 4: Surge Reflection at a
Wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
12.3.6 Example of Homework 5: Analytical Solution for
Transcritical Weir Flow . . . . . . . . . . . . . . . . . . . . . . . 552
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557

Subject Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561
Chapter 1
Fundamental Equations
of Free Surface Flows
1.1 Introduction
1.1.1 Classification of Open Channel Flows
Open channel flow is the study of the movement of liquids with a free surface,
which is by definition an interface in contact with the atmosphere. Beautiful
examples of these flows are seen at dam spillways (Fig. 1.1a) or along rivers
(Fig. 1.1b).
While these flows can be tackled with techniques of general fluid mechanics, this
book is devoted to the use of approximate methods widely employed in civil and
environmental engineering, namely the use of shallow water models. These refer to
depth- or cross-sectional averaged models where the variation of the field variables
in the vertical direction is only approximately accounted for, or even neglected
(Vreugdenhil 1994; Toro 2001). Typical open channel flow problems are solved by
assuming one-dimensional (1D) or two-dimensional (2D) flows in a horizontal
plane. Common open channel flow sections in practice are irregular as in a river
(Fig. 1.2a), trapezoidal for a hydropower canal (Fig. 1.2b), rectangular for a
spillway chute (Fig. 1.2c) and circular for a sanitary sewer (Fig. 1.3d), among
others.
To classify open channel flows, assume for the moment 1D flow. While the
water moves in a channel, there are variations in depth and velocity both in space
(coordinate x) and time t. Let the water depth be h and the velocity U without
rigorously defining these physical quantities at the current introductory stage. An
open channel flow is classified based on the following features:
Steady and unsteady flows
A flow is steady if the flow variables do not change with time at a given position. If
these change with time at a fixed position, the flow is referred to as unsteady. The
flow is said also to be variable as it changes with time.
© Springer Nature Switzerland AG 2019 1

O. Castro-Orgaz and W. H. Hager, Shallow Water Hydraulics,
https://doi.org/10.1007/978-3-030-13073-2_1
2 1 Fundamental Equations of Free Surface Flows
Fig. 1.1 Examples of free surface flows a spillway of Aldeadávila Dam; photo taken from https://
es.wikipedia.org/wiki/Presa_de_Aldeadávila, b Amazon river flowing through the Amazon rain-
forest (photo of public domain by NASA)
Uniform and non-uniform flows

A flow is uniform if the flow variables are constant in space at a given instant of time. If
the flow changes in space at an instant of time, the flow is non-uniform or varied.
Gradually varied and rapidly varied flows

Depending on the intensity of the change of flow variables with distance, the flow is
gradually varied if this variation is small and rapidly varied if the variation is large.
1.1 Introduction 3
Fig. 1.2 Typical channel flow sections a irregular, b trapezoidal, c rectangular, d circular
Hydrostatic and non-hydrostatic flows

A useful approximation for fluid pressure in open channel flows is the hydrostatic
law, implying that the pressure pb at the bottom equals the weight of the water
column above, that is, pb = qgh, where h is the water depth, q the fluid density, and
g the gravity acceleration. A non-hydrostatic channel flow does not obey this law,
among other reasons, if streamline curvature or the channel bottom slope is large
(Castro-Orgaz and Hager 2017).
Examples of open channel flows are presented in Fig. 1.3. In Fig. 1.3a, the flow
is steady and uniform, whereas it is steady but non-uniform in Fig. 1.3b. A weak
variation of the flow variables with distance is sketched in the latter gradually varied
flow. Figure 1.3c represents unsteady uniform flow, a case rarely found in nature.
The flow in Fig. 1.3d shows a smooth flood wave as an example of unsteady
gradually varied flow. Figure 1.3e represents the propagation of a surge, which is
basically a discontinuity in depth and velocity moving at a constant celerity. This
flow is therefore unsteady and rapidly varied, given the abrupt variation in depth
and velocity at the wavefront. In Chap. 5, this particular unsteady flow will be
reduced to a steady flow problem in a reference system moving with the surge.
Figure 1.3f represents the steady flow over a spillway crest, down the chute up to
a nearly horizontal tailwater channel ending on a free drop. At the spillway crest
(Fig. 1.4a), the curvature of streamlines is large, inducing a non-hydrostatic pres-
sure distribution (Castro-Orgaz and Hager 2017). This flow is therefore rapidly
varied and non-hydrostatic.
Once the water reaches the chute, the variation of depth with distance remains
small, and streamline curvature can be neglected, but the bed slope is so large that
the resulting bottom pressure is less that the water weight qgh. This flow is
therefore gradually varied and non-hydrostatic (Castro-Orgaz and Hager 2017). As
the water reaches the tailwater channel, the flow depth abruptly increases on a very
short distance (Fig. 1.4b), producing the so-called hydraulic jump. If air–water
Fig. 1.3 Open channel flow examples a steady uniform, b steady gradually varied, c unsteady
uniform, d unsteady gradually varied, e unsteady rapidly varied, f varied flow over a dam and in
tailwater channel
mixture effects are neglected, the pressure distribution within the hydraulic jump
can be assumed to be hydrostatic (Khan and Steffler 1996a; Castro-Orgaz and
Hager 2009). The flow is therefore rapidly varied and hydrostatic. Beyond the
jump, the water surface profile in the nearly horizontal tailwater channel has a small
variation of depth with distance, resulting in negligible streamline curvature. The
flow is therefore gradually varied and hydrostatic. In the vicinity of the free drop,
1.1 Introduction 5
Fig. 1.4 Rapidly varied flows at a spillway crest, b hydraulic jump
the variation of flow depth with distance is large, the streamline curvature is sig-
nificant and the pressure distribution non-hydrostatic. The flow is therefore rapidly
varied and non-hydrostatic.
1.1.2 Aims and Scope
The main purpose of this book is to present the theory and computation of open
channel flows using the hydrostatic pressure approach. This type of model can be
collectively named shallow water model. Both steady and unsteady flows will be
tackled, as well as sediment transport problems. Unsteady flow computations will
include both finite-difference and finite volume methods, focussing on 1D models.
Extension to non-hydrostatic flows is considered in Chap. 11.
1.2 General Depth-Integrated Equations

over a 3D Terrain
1.2.1 Previous Open Channel Flow Developments
In civil and environmental engineering applications, free surface flows are often
treated using the one-dimensional (1D) continuity, momentum, and energy equa-
tions, in which information on the field variables across the flow section are
averaged (Chow 1959; Henderson 1966; Liggett 1994; Montes 1998; Sturm 2001;
Jain 2001; Chaudhry 2008). de Saint-Venant (1871) and Boussinesq (1877) are the
fathers of depth-averaged open channel flow modeling, both proposing approximate
models (Castro-Orgaz and Hager 2017). The rational development of 1D flow
models starts with the 3D equations of continuity, momentum and energy, which
are integrated across a section normal to the channel bed (Keulegan 1942; Strelkoff
1969; Yen 1973). Prior to the averaging process, the full Navier–Stokes equations
are time-averaged for a turbulent flow, resulting in the Reynolds equations. This
development was first presented by Keulegan (1942) and Keulegan and Patterson
(1943). Further studies were conducted by Jaeger (1956), Chen and Chow (1968),
Yen (1975), Lai (1986), Liggett (1994), Montes (1998), Jain (2001) and Chaudhry
(2008). The contributions of Strelkoff (1969) and Yen (1973, 1975) are notable,
because they detail the fundamental differences between the 1D energy and
momentum equations, retaining the flow equations as a function of averaging
coefficients accounting for arbitrary distributions of velocity and pressure.
Garbis Hovannes Keulegan was born at Sebastia-Sivas, Armenia, in today’s

Turkey, on July 12, 1890, and passed away at age 99 on July 28, 1989, at
Vicksburg MI. He left in 1912 his home country for the USA and started as
an engineer at Ohio State University, graduating as a mathematician in 1915.
He joined the American Forces in 1918 as a translator, starting in 1921 as a
physicist at the National Bureau of Standards NBS. Until retirement in 1962,
he was there primarily engaged as an expert in soil mechanics. He submitted a
Ph.D. thesis in 1928 to the Johns Hopkins University, Baltimore MD.
Keulegan’s interest in hydrodynamics was initiated with the inauguration
of the National Hydraulic Laboratory established at NBS. He was one of the
three staff members and greatly contributed with by now classic papers on
turbulent flow in open channels, roll wave formation, water wave theory, and
flow in curved pipes. During World War II, Keulegan was mainly active for
the Beach Erosion Board in connection with the Allied Landing in France. He
thus developed the theory of tides, the water wave theory furnishing infor-
mation relating to the prediction of sea currents. His famous 1948 report on
the experimental study of submarine sand bars was one of the few that
remained unclassified. After the war, NBS was asked to furnish information
1.2 General Depth-Integrated Equations over a 3D Terrain 7
toward the basic laws of similitude involving density currents and the mixing
of salt with freshwaters. Keulegan investigated both questions thoroughly and
presented classic papers relating to the lock exchange experiments. His results
on wave propagation and density currents were published in the 1950 book
Engineering hydraulics of Rouse, and the 1966 book Estuary and coastline
hydrodynamics of Ippen, respectively. Keulegan was awarded a number of
prestigious decorations, including the National Medal of Science, Honorary
Membership of ASCE, and in 1979 election to the National Academy of
Engineering.
The coordinate system considered by Strelkoff (1969) and Yen (1973) to pro-
duce the cross-sectional averaged Reynolds equations was orthogonal and fitted to
the terrain. This system is widely used in geophysical flow research (Savage and
Hutter 1989, 1991; Pudasaini and Hutter 2007). Strelkoff (1969) and Yen (1973)
transformed the Navier–Stokes equations in Cartesian coordinates to the
terrain-fitted reference system stated by means of a rotation of axes to render two of
the coordinates tangent to the terrain, and the third normal to it. While it is feasible
to produce such a transformation from Cartesian to terrain-fitted coordinates, a
simple axes rotation is not the exact mathematical operator, so that the channel flow
equations developed by this procedure are not general. Dressler (1978) transformed
the plane Euler equations in Cartesian coordinates to a boundary-fitted system of
reference by using a Jacobian matrix. In his modified form of the Euler equations,
the channel bed curvature appears. This feature is also observable in the granular
mass flow equations by Savage and Hutter (1989, 1991). The discrepancy between
Dressler (1978) and Savage and Hutter (1989, 1991), and the form of the Navier–
Stokes equations to obtain 1D equations used by Strelkoff (1969) and Yen (1973),
indicate that the equations developed by the latter apply strictly to flows over
constant slope channels. The development by Strelkoff (1969) is also available from
Jain (2001). The key message is that the Navier–Stokes equations in the devel-
opments of Strelkoff (1969) and Yen (1973) should be expressed in the terrain-fitted
system of coordinates using a Jacobian matrix rather than a simple rotation of axes,
because only then the averaging process is to be conducted. If the developments of
Strelkoff (1969) and Yen (1973) are repeated with these considerations, additional
terms including the terrain curvature appear, as those contained in the
depth-averaged avalanche flow equations by Savage and Hutter (1989, 1991). The
transformation of the Reynolds equations to any curvilinear system (orthogonal or
not) is comprehensively reported by Rouse (1959) and Schlichting and Gersten
(2000). A disadvantage of the 1D channel flow equations of Strelkoff (1969) or Yen
(1973) is that undetermined averaging coefficients need to be mathematically closed
for computations. These are difficult to evaluate, like those of the 1D energy
equation for turbulent unsteady flow.
An alternative to terrain-fitted coordinates is the resort to Cartesian coordinates,
despite this option was less used in open channel flows. Notable contributions to be
considered below are the works by Steffler and Jin (1993), Khan and Steffler
(1996b, c), Denlinger and Iverson (2004), Iverson (2005), Denlinger and O’Connell
(2008), and Iverson and Ouyang (2015). Note that the generalized free surface flow
equations in Cartesian coordinates were given by Yen (1975), thereby removing the
previous problem by the use of terrain-fitted coordinates.
Due to the increase of mathematical complexities while doing an averaging
process in terrain-fitted coordinates, Cartesian coordinates will be generally used in
this book. Further, given the difficulties in using the energy equation for
depth-averaged turbulent unsteady flow computations over 3D terrain, the use of
this (important) principle is limited to steady flow problems, and therefore pre-
sented in Chap. 2. The general 3D flow equations of mass and momentum will be
depth-integrated to produce a general 2D averaged model in this chapter. This
general formulation applies in practice to produce 2D computational models
making suitable approximations to the field variables. For the case of 1D flow, the
2D-averaged flow equations will be laterally integrated, producing the corre-
sponding cross-sectional averaged model. Sediment transport and movable beds are
finally considered.
1.2.2 3D Flow Equations
Here the general depth-integrated equations for a mixture of fluid and sediments,
bounded by arbitrary non-material interfaces, are developed. Consider a river flow
represented by the movement of a continuum mixture of a fluid and solid with
density q(x, y, z, t), where (x, y, z) are the Cartesian coordinates and t is the time.
The flow is therefore bounded by two general interface surfaces representing the
free surface (subscript s) and the terrain surface (subscript b for bed), given by the
mathematical statements z = zs(x, y, t) and z = zb(x, y, t), respectively (Fig. 1.5).
The velocity components in the (x, y, z) directions are (u, v, w).
With the mixture velocity V = (u, w, v) as the barycentric velocity (Iverson
2005; Wu 2008), the mass conservation equation for the mixture is
@q @ ðquÞ @ ðqvÞ @ ðqwÞ

þ þ þ ¼ 0: ð1:1Þ
@t @x @y @z
Likewise, in the horizontal–vertical Cartesian system of coordinates used, Newton’s

second law gives for the mixture the momentum equations (Iverson 2005; Wu
2008).

@ @ 2 @ @ @sxx @sxy @sxz
ðquÞ þ qu þ ðquvÞ þ ðquwÞ ¼ þ þ ; ð1:2Þ
@t @x @y @z @x @y @z
Fig. 1.5 3D open channel flow in a river a longitudinal section, b transverse section, c plan view

@ @ @ 2 @ @syx @syy @syz
ðqvÞ þ ðquvÞ þ qv þ ðqvwÞ ¼ þ þ ; ð1:3Þ
@t @x @y @z @x @y @z

@ @ @ @ 2 @szx @szy @szz
ðqwÞ þ ðquwÞ þ ðqvwÞ þ qw ¼ qg þ þ :
@t @x @y @z @x @y @z
ð1:4Þ
Here g is the gravity acceleration and sij the stress tensor, with (i, j) = (x, y, z),
which is symmetric, i.e. sij = sji. Subscript i indicates the axis along which the
stress acts, and j is the axis normal to the plane containing the stress. Moreover, sij
is here introduced with the notation used in the environmental context, i.e., it is the
negative value of the common stress tensor definition used in civil engineering
(Iverson 2005). Equations (1.1)–(1.4) provide the starting point to generate a
depth-integrated model.
1.2.3 General Kinematic Boundary Conditions
Prior to integrating Eqs. (1.1)–(1.4) in the z-direction, the general kinematic

boundary condition at an interface will be presented (Iverson and Ouyang 2015;
Castro-Orgaz and Hager 2017). Consider for illustrative purposes a river bed,
whose general mathematical equation is written as
F ðx; y; z; tÞ ¼ z zb ðx; y; tÞ ¼ 0: ð1:5Þ
The material derivative of F following its motion is, therefore, given by
DF @F
¼ þ U rF ¼ 0; ð1:6Þ
Dt @t
where U is the velocity vector representing the deformation of the river bed. Note
that the velocity of displacement of an interface is not necessarily equal to that of
the particles that are momentarily lying upon it. In general, the velocity of the
displacement of a surface is given by the kinematic statement
U ¼ Vb Mb n: ð1:7Þ
Here Vb = (ub, vb, wb) is the fluid mixture velocity on the surface, Mb is the net
volume of fluid mixture crossing normal the interface “b” per unit area and time,
and n is the unit vector normal to the interface. Inserting Eq. (1.7) into Eq. (1.6)
produces
DF @F @F
¼ þ ðVb Mb nÞ rF ¼ þ Vb rF Mb jrF j
Dt @t @t
ð1:8Þ
@F @F @F @F
¼ þ ub þ vb þ wb Mb jrF j ¼ 0:
@t @x @y @z
The derivatives of F are
@F @zb @F @zb @F @zb @F

¼ ; ¼ ; ¼ ; ¼ 1: ð1:9Þ
@t @t @x @x @y @y @z
Inserting Eqs. (1.9) into Eq. (1.8), one finds
DF @zb @zb @zb

¼ ub vb þ wb Mb jrF j ¼ 0: ð1:10Þ
Dt @t @x @y
Noting that the modulus of F is given by

" 2 2 2 #1=2 " 2 2 #1=2
@F @F @F @zb @zb
jrF j ¼ þ þ ¼ þ þ1 ;
@x @y @z @x @y
ð1:11Þ
Equation (1.10) gives after re-arrangement

" 2 #1=2
@zb @zb @zb @zb 2 @zb
þ ub þ vb wb ¼ Mb þ þ1 ; ð1:12Þ
@t @x @y @x @y
or
@zb @zb @zb

þ ub þ vb wb ¼ Mb nb ; ð1:13Þ
@t @x @y
where
" 2 #1=2
@zb 2 @zb
nb ¼ þ þ1 : ð1:14Þ
@x @y
Equation (1.13) is the general kinematic boundary condition at the river bed. This
equation must be preserved in depth-integrated models, regardless of using a
non-slip or slip (ub = vb = wb = 0) assumption at the river bed. A similar equation
applies for the free surface, resulting in
@zs @zs @zs

þ us þ vs ws ¼ Ms ns ; ð1:15Þ
@t @x @y
where
" 2 #1=2
@zs 2 @zs
ns ¼ þ þ1 : ð1:16Þ
@x @y
1.2.4 Continuity Equation
Vertical integration of Eq. (1.1) yields
Zzs
@q @ ðquÞ @ ðqvÞ @ ðqwÞ
þ þ þ dz ¼ 0: ð1:17Þ
@t @x @y @z
zb
This integral relation is transformed by computing the following identities by

application of Leibniz’s rule:
Zzs Zzs
@q @ @zs @zb
dz ¼ qdz qs þ qb ; ð1:18Þ
@t @t @t @t
zb zb
Zzs Zzs
@ ðquÞ @ @zs @zb
dz ¼ qudz qs us þ qb ub ; ð1:19Þ
@x @x @x @x
zb zb
Zzs Zzs
@ ðqvÞ @ @zs @zb
dz ¼ qvdz qs vs þ qb vb ; ð1:20Þ
@y @y @y @y
zb zb
Zzs
@ ðqwÞ
dz ¼ qs ws qb wb : ð1:21Þ
@z
zb
Summing Eqs. (1.18)–(1.21) produces
Zzs Zzs Zzs

@ @ @ @zs @zb @zs
qdz þ qudz þ qvdz qs þ qb qs u s
@t @x @y @t @t @x
zb zb zb ð1:22Þ
@zb @zs @zb
þ qb ub qs v s þ qb vb þ qs ws qb wb ¼ 0;
@x @y @y
or,
Zzs Zzs Zzs

@ @ @
qdz þ qudz þ qvdz
@t @x @y
zb zb zb

@zs @zs @zs @zb @zb @zb
qs þ us þ vs ws þ qb þ ub þ vb wb ¼ 0:
@t @x @y @t @x @y
ð1:23Þ
Inserting the general kinematic boundary conditions Eqs. (1.13) and (1.15) into
Eq. (1.23) yields
Zzs Zzs Zzs

@ @ @
qdz þ qudz þ qvdz þ qs Ms ns qb Mb nb ¼ 0: ð1:24Þ
@t @x @y
zb zb zb
This is the general depth-integrated continuity equation.
1.2.5 Momentum Equations
Now, the x-momentum Eq. (1.2) is vertically integrated as
Zzs
@ @ 2 @ @
ðquÞ þ qu þ ðquvÞ þ ðquwÞ dz
@t @x @y @z
zb
Zzs
@sxx @sxy @sxz
¼ þ þ dz: ð1:25Þ
@x @y @z
zb
The following transformations are produced using again Leibniz’s rule:
Zzs Zzs
@ @ @zs @zb
ðquÞdz ¼ qudz qs us þ qb u b ; ð1:26Þ
@t @t @t @t
zb zb
Zzs Zzs
@ 2 @ @zs @zb
qu dz ¼ qu2 dz qs u2s þ qb u2b ; ð1:27Þ
@x @x @x @x
zb zb
Zzs Zzs
@ @ @zs @zb
ðquvÞdz ¼ quvdz qs us vs þ qb u b v b ; ð1:28Þ
@y @y @y @y
zb zb
Zzs
@
ðquwÞdz ¼ qs us ws qb ub wb ; ð1:29Þ
@z
zb
Zzs Zzs
@sxx @ @zs @zb
dz ¼ sxx dz ðsxx Þs þ ðsxx Þb ; ð1:30Þ
@x @x @x @x
zb zb
Zzs Zzs
@sxy @ @zs @zb
dz ¼ sxy dz sxy s þ sxy b ; ð1:31Þ
@y @y @y @y
zb zb
Zzs
@sxz
dz ¼ ðsxz Þs ðsxz Þb : ð1:32Þ
@z
zb
Equations (1.26)–(1.32) form a set of identities that are substituted into Eq. (1.25),
transforming it into
Zzs Zzs Zzs

@ @zs @zb @ @zs @zb @
qudz qs us þ qb ub þ qu2 dz qs u2s þ qb u2b þ quvdz
@t @t @t @x @x @x @y
zb zb zb
2
Zzs
@zs @zb @ @zs @zb
qs us vs þ qb ub vb þ qs us ws qb ub wb ¼ 4 sxx dz ðsxx Þs þ ðsxx Þb
@y @y @x @x @x
zb
3
Zzs
@ @zs @zb
þ sxy dz sxy s þ sxy b þ ðsxz Þs ðsxz Þb 5:
@y @y @y
zb
ð1:33Þ
Collecting terms, one finds
Zzs Zzs Zzs

@ @ @ @zs @zs @zs
qudz þ qu dz þ
2
quvdz qs us þ us þ vs ws
@t @x @y @t @x @y
zb zb zb
( Zzs Zzs
@zb @zb @zb @ @
þ qb ub þ ub þ vb wb þ sxx dz þ sxy dz
@t @x @y @x @y
zb zb
)
@zs @zs @zb @zb
ðsxx Þs þ sxy s ðsxz Þs þ ðsxx Þb þ sxy b ðsxz Þb ¼ 0:
@x @y @x @y
ð1:34Þ
Using Eqs. (1.13) and (1.15) produces
Zzs Zzs Zzs Zzs

@ @ @ @
qudz þ qu dz þ
2
quvdz þ qs us Ms ns qb ub Mb nb þ sxx dz
@t @x @y @x
zb zb zb zb
Zzs
@ @zs @zs @zb @zb
þ sxy dz ðsxx Þs þ sxy s ðsxz Þs þ ðsxx Þb þ sxy b ðsxz Þb ¼ 0:
@y @x @y @x @y
zb
ð1:35Þ
Equation (1.35) represents the general x-momentum depth-integrated equation for

non-hydrostatic mixture flows. Integrating Eq. (1.3) follows the same steps and no
additional insights are regained. The result is
Zzs Zzs Zzs Zzs

@ @ @ @
qvdz þ qv dz þ
2
quvdz þ qs vs Ms ns qb vb Mb nb þ syy dz
@t @y @x @y
zb zb zb zb
Zzs
@ @zs @zs @zb @zb
þ syx dz syy s þ syx s syz s þ syy b þ syx b syz b ¼ 0:
@x @y @x @y @x
zb
ð1:36Þ
Finally, the z-momentum Eq. (1.4) is vertically integrated, producing
Zzs
@ @ @ @ 2
ðqwÞ þ ðquwÞ þ ðqvwÞ þ qw dz
@t @x @y @z
zb
Zzs Zzs
@szx @szy @szz
¼ g qdz þ þ dz ð1:37Þ
@x @y @z
zb zb
Employing the transformations:
Zzs Zzs
@ @ @zs @zb
ðqwÞdz ¼ qwdz qs ws þ qb wb ; ð1:38Þ
@t @t @t @t
zb zb
Zzs Zzs
@ @ @zs @zb
ðquwÞdz ¼ quwdz qs us ws þ q b ub w b ; ð1:39Þ
@x @x @x @x
zb zb
Zzs Zzs
@ @ @zs @zb
ðqvwÞdz ¼ qvwdz qs vs ws þ q b vb w b ; ð1:40Þ
@y @y @y @y
zb zb
Zzs
@ 2
qw dz ¼ qs w2s qb w2b ; ð1:41Þ
@z
zb
Zzs Zzs
@szx @ @zs @zb
dz ¼ szx dz ðszx Þs þ ðszx Þb ; ð1:42Þ
@x @x @x @x
zb zb
Zzs Zzs
@szy @ @zs @zb
dz ¼ szy dz szy s þ szy b ; ð1:43Þ
@y @y @y @y
zb zb
Zzs
@szz
dz ¼ ðszz Þs ðszz Þb ; ð1:44Þ
@z
zb
applied to Eq. (1.37), and subsequently grouping terms, leads to
Zzs Zzs Zzs

@ @ @ @zs @zs @zs
qwdz þ quwdz þ qvwdz qs ws þ us þ vs ws
@t @x @y @t @x @y
zb zb zb
Zzs ( Zzs Zzs
@zb @zb @zb @ @
þ qb wb þ ub þ vb wb þ g qdz þ szx dz þ szy dz
@t @x @y @x @y
zb zb zb
)
@zs @zs @zb @zb
ðszx Þs þ szy s ðszz Þs þ ðszx Þb þ szy b ðszz Þb ¼ 0:
@x @y @x @y
ð1:45Þ
Using the kinematic boundary conditions results in
Zzs Zzs Zzs Zzs

@ @ @
qwdz þ quwdz þ qvwdz þ qs ws Ms ns qb wb Mb nb þ g qdz
@t @x @y
zb zb zb zb
Zzs Zzs
@ @ @zs @zs
þ szx dz þ szy dz ðszx Þs þ szy s ðszz Þs
@x @y @x @y
zb zb

@zb @zb
þ ðszx Þb þ szy b ðszz Þb ¼ 0: ð1:46Þ
@x @y
Equations (1.24), (1.35), (1.36), and (1.46) are the general non-hydrostatic
depth-integrated mass and momentum conservation equations for mixture flows.
This set of generalized equations accounts for density variations, movable beds,
flows across the free surface and non-hydrostatic stresses over a 3D terrain. The
flow can be turbulent or laminar, steady or unsteady, rotational or irrotational,
uniform or non-uniform, and gradually varied or rapidly varied. The bed may be
rigid or erodible, pervious or impervious, with steep or small slopes, and the free
surface a material surface or receive flows. The fluid may be compressible or
incompressible, with sediments in suspension or clear water (Yen 1973, 1975).
With the general stress tensor considered, the model also applies to avalanche
dynamics using the Mohr–Coulomb yield criterion (Iverson 2005). Open channel
flows are turbulent; Reynolds decomposition is employed in these flows, splitting
the flow variables into mean (denoted by bars) and fluctuating (denoted by primes)
components, e.g., u ¼ u þ u0 ; v ¼ v þ v0 ; w ¼ w þ w0 ; p ¼ p þ p0 ; q ¼ q þ q0 . By
definition, the time-averaged value of a fluctuating component is zero. Employing
these transformations, doing a time-averaging of the 3D equations, and neglecting
viscous contributions, the symmetric stress tensor is (Rodi 1980; White 1991, 2009)
sxx ¼ p rxx ; syy ¼ p ryy ; szz ¼ p rzz ;

ð1:47Þ
sxz ¼ rxz ; sxy ¼ rxy ; syz ¼ ryz :
Here r denotes the turbulent Reynolds stress due to time averaging of the Navier–
Stokes equations for fluid flow. The velocity field corresponding to this stress tensor
is a time-averaged mathematical approach representing the mean turbulent motion.
The Reynolds stresses are in reality terms originating from the acceleration, and,
thus, are “apparent stresses.”
With this set of generalized equations, including variable density, turbulence,
and non-hydrostatic pressures, simplified versions will be presented in the ensuing
developments.
Ben Chie Yen was born on April 14, 1935, at Guangzhou, China, and passed
away aged 66 years on December 23, 2001, at St. Louis IL, USA. He
obtained his BS degree from the National Taiwan University in 1956, the
civil engineer degree from University of Iowa in 1959, and earned his Ph.D.
title there in 1965. After a short stay at Princeton University, he was from
1966 to 1976 Assistant and Associate Professor at the University of Illinois,
Urbana, and from 1976 there Professor of civil engineering. Yen visited later
a number of universities, including University of Karlsruhe in 1974, EPFL
Lausanne in 1982, and National Taiwan University in 1983. He passed away
following cardiovascular aneurysm.
Yen’s Ph.D. thesis was in open channel flow, a topic that he pursued until
the 1970s. He was able to distinguish between the energy and the momentum
equations, resulting in two different formulations with the appropriate aver-
aging coefficients and corresponding loss slopes. He then turned to the
stormwater technology thereby investigating a number of its elements. He, for
instance, pointed to the significant energy losses induced by junction man-
holes, a fact that was often neglected. Yen chaired the second Conference on
Urban Drainage in 1982. He recognized the need to merge water quality and
quantity experts by forming a new international group, the Joint Committee
on Urban Storm Drainage JCUSD, which he chaired also from 1982. He
further investigated unsteady sewer flows and eventually developed into an
expert in this field. He was awarded the Hunter Rouse Hydraulic Engineering
Lecture from the American Society of Civil Engineers ASCE in 1999 for “his
fundamental work on open channel and flow resistance.” He was further
promoted during the 29th IAHR Congress held in Beijing to Honorary
Member of IAHR.
1.3 Saint-Venant Theory 19
1.3 Saint-Venant Theory
1.3.1 2D Flows in the Horizontal Plane
In this section, the shallow water equations (SWE) for 2D flows of clear water
(q = const.) over a horizontal plane are derived. If the flow depth h(x, y, t) is smaller
than the characteristic length in the (x, y)-plane, a scaling analysis reveals that,
except for the near-bed boundary layer, the velocity components u and v can be
assumed constant across h and equal to their depth-averaged values U and
V (Liggett 1994). Therefore,
Zzs
1
uðx; y; z; tÞ U ðx; y; tÞ ¼ udz; ð1:48Þ
h
zb
Zzs
1
vðx; y; z; tÞ V ðx; y; tÞ ¼ vdz: ð1:49Þ
h
zb
The depth-independent horizontal velocity components imply a slip velocity at the

bed, thereby neglecting the high-velocity gradient confined to the thin bed boundary
layer (Steffler and Jin 1993). Inserting this approximation into Eq. (1.23), noting
that q = const., assuming that there is no flow across the free surface and that the
bottom is rigid and impervious, the depth-averaged continuity equation reduces to
@h @ @
þ ðUhÞ þ ðVhÞ ¼ 0: ð1:50Þ
@t @x @y
Using the approximations stated above in Eq. (1.35), and neglecting stresses at the
free surface, yields the depth-averaged x-momentum as
@ @ 2 @
ðUhÞ þ U h þ ðUVhÞ
@t @x @y
2 3
Zzs Zzs ð1:51Þ
14@ @ @zb @zb
¼ sxx dz þ sxy dz þ ðsxx Þb þ sxy b ðsxz Þb 5:
q @x @y @x @y
zb zb
This needs further considerations about the stress tensor. Consider first the impli-
cations of using Eqs. (1.48)–(1.49) on the x-momentum balance. Basically, it
amounts to assume that the Boussinesq velocity correction coefficients (Liggett
1994; Katopodes 2019), e.g.,
Rh Rh Rh
u2 dg v2 dg uvdg
bx ¼ 0
; by ¼ 0
; bxy ¼ 0
; ð1:52Þ
U2h V 2h UVh
are equal to unity, with η as the vertical distance above the channel bed. In uniform
and one-dimensional gradually varied turbulent open channel flows, the velocity
distribution is typically approximated by the logarithmic law of the wall (Fig. 1.6a),
which for a rough bed is (White 1991, Montes 1998)

u 1 g

¼ ln þ B: ð1:53Þ
u j ks
Here u* = (sb/q)1/2 is the shear velocity, j (=0.41) the von Kármán constant, ks the
equivalent roughness height and B (8.5) a constant of integration. Other velocity
distributions as the power law and the wall-wake law are also used in open channel
flow problems (Montes 1998; Chanson 2004).
The Boussinesq velocity correction coefficient is of the order of bx 1.05 for
these flows (Chow 1959; Henderson 1966; Chaudhry 2008). Thus, given the small
contribution of the differential advection originating from the non-uniformity of
u and v with depth, the Boussinesq velocity correction coefficients in the x- and
y-directions are assumed to be unity. Likewise, Eq. (1.39) then reads for the
depth-averaged y-momentum balance
Fig. 1.6 Examples of velocity distributions in open channel flows a log-law velocity distribution
in rough-bed uniform open channel flow, b velocity distribution in hydraulic jumps, with a jet flow
and a roller flow above
@ @ 2 @
ðVhÞ þ V h þ ðVUhÞ
@t @y @x
2 3
Zzs Zzs ð1:54Þ
14@ @ @zb @zb
¼ syy dz þ sxy dz þ syy b þ sxy b syz b 5:
q @y @x @y @x
zb zb
Caution is claimed here by pointing out that taking the velocity correction coeffi-
cients equal to unity is a legitimate approximation for gradually varied flows
(setting aside compound channel flows), but not so in rapidly varied flows.
A prominent example is the hydraulic jump (Fig. 1.4b). Within the region of high
turbulence and shear, the u-velocity profiles are highly non-uniform in the vertical
direction. The correction coefficients are easily of the order of bx = 2.5 (Fig. 1.6b),
encompassing a jet-like flow which spreads below a recirculation vortex or roller
(Khan and Steffler 1996a; Castro-Orgaz and Hager 2009). Obviously, these coef-
ficients cannot be neglected. As will be shown in Chap. 5, the SWE permit to catch
hydraulic jumps as local discontinuities, excluding the detailed variation of the field
variables across it, including the free surface and the velocity profiles. This fact has
double implications: (1) the SWE can detect a hydraulic jump, and identify it as a
point-like discontinuity, but cannot resolve the flow within it; (2) there is no need to
introduce any correction coefficient b given that the equations do not resolve the
internal flow phenomena of a hydraulic jump. This feature must be fully accounted
for only if a higher-order depth-averaged model is considered for resolving the
detailed free surface profile of the hydraulic jump (Khan and Steffler 1996a;
Castro-Orgaz and Hager 2009).
Using Eqs. (1.47) for the stress tensor sij, and neglecting stresses of
depth-averaging in Eqs. (1.51) and (1.54), produces
@ @ 2 @
@t @x @y
2 3
Zzs ð1:55Þ
14@ @zb @zb
¼ pdz þ pb rxy b þ ðrxz Þb 5;
q @x @x @y
zb
@ @ 2 @
@t @y @x
2 3
Zzs ð1:56Þ
14@ @zb @zb 5
¼ pdz þ pb rxy b þ ryz b :
q @y @y @x
zb
A typical assumption in the derivation of the gradually varied SWE is that the
pressure distribution is hydrostatic. From Eq. (1.4), the vertical momentum balance
is thus for a hydrostatic flow
dp
¼ qg: ð1:57Þ
dz
Integrating in the z-direction, taking zero pressure as reference at the free surface,
yields
g g
p ¼ qgh 1 ¼ pb 1 : ð1:58Þ
h h
This is a linear distribution of fluid pressure (Fig. 1.7), with the bed pressure equal
to the weight of the water column above, that is,
pb ¼ qgh: ð1:59Þ
To highlight the conditions under which this approximation is valid, consider the
general Eq. (1.46) for the z-momentum balance. This is an equation for the bottom
pressure, which yields, by using Eq. (1.47) for the stress tensor, after assuming that
(u, v, w) are approximated by the corresponding depth-averaged values (U, V, W),
material surfaces at the free surface and the bed (Ms = Mb = 0), depth-averaged
density, zero free surface stresses, and neglecting turbulent stresses due to
depth-averaging,
@ @ @ @zb
pb ¼ qgh þ ðqWhÞ þ ðqWUhÞ þ ðqWVhÞ þ ðrzz Þb ðrxz Þb
@t @x @y @x
@zb
ryz b : ð1:60Þ
@y
In addition to the simplifications introduced above ab initio, the following

approximations are now required to reduce Eq. (1.60) to Eq. (1.59):
– Temporal rate of variation of W must be small. This is equivalent to neglect the

local vertical acceleration.
– Velocity W must be negligible in comparison to U and V. The ratios W/U and
W/V representing streamline inclinations are small, or, more precisely, their
Fig. 1.7 Hydrostatic

pressure distribution in open
channel flow
variations in x- and y-directions, which represents the flow curvatures. This is

equivalent to neglect the convective vertical acceleration.
– Turbulent normal stress rzz at the base must be negligible. This is only possible
if the channel slope is small (see Sect. 1.4.1).
– Bed slopes in the x- and y-directions must be small.
Therefore, a hydrostatic flow resembles those occurring in a small slope channel
with almost horizontal streamlines, a low turbulence level, and uniform velocities in
the vertical direction. For non-hydrostatic problems, all or part of these effects must
be accounted for in the momentum equations (Serre 1953; Stansby and Zhou 1998;
Castro-Orgaz and Hager 2017). Introducing the hydrostatic pressure approximation
in Eqs. (1.55)–(1.56) produces
@ @ 2 @
@t @x @y
ð1:61Þ
@h @zb 1 @zb 1
¼ gh gh þ rxy b ðrxz Þb ;
@x @x q @y q
@ @ 2 @
@t @y @x
ð1:62Þ
@h @zb 1 @zb 1
¼ gh gh þ rxy b ryz b :
@y @y q @x q
To be coherent with Eq. (1.59), the bed slopes in the x- and y-directions are
assumed to be small, eliminating the corresponding stress contributions due to
slopes, resulting in the depth-averaged SWE or Saint-Venant equations
(Vreugdenhil 1994; Toro 2001)
@ @ 2 @ @zs 1
ðUhÞ þ U h þ ðUVhÞ ¼ gh ðrxz Þb ; ð1:63Þ
@t @x @y @x q
@ @ 2 @ @zs 1
ðVhÞ þ V h þ ðVUhÞ ¼ gh ryz b ; ð1:64Þ
@t @y @x @y q
where the gradients of zs = zb + h, the free surface elevation, are used. Finally, with
Cf as a bed friction coefficient, closure for the bed turbulent stresses is given by
1=2 1=2
ðrxz Þb ¼ qCf U U 2 þ V 2 ; ryz b ¼ qCf V U 2 þ V 2 : ð1:65Þ
1.3.2 1D Cross-Sectional Averaged Equations
Consider an open channel flow with a predominant movement in the x-direction, so

that the flow geometry can be characterized taking vertical sections of area
A (Fig. 1.8) and an axis connecting these sections. In general, the free surfaces
within the cross sections will be inclined, but if these variations are gentle, a
horizontal free surface on each flow area A can be assumed. Further, if the river is
curved in plan, it is feasible to approximate x as the curvilinear distance following
the axis if the channel curvatures are weak and can thus be neglected.
Consider a generic cross section with a horizontal free surface, as sketched in
Fig. 1.9. The depth-averaged continuity Eq. (1.50) will be laterally integrated from
the left to the right channel banks (points 1 and 2 in Fig. 1.9), resulting in
Zyr
@h @ @
þ ðUhÞ þ ðVhÞ dy ¼ 0: ð1:66Þ
@t @x @y
yl
Fig. 1.8 Open channel flow

with a predominant
movement in the longitudinal
direction
Fig. 1.9 Generic cross

section with horizontal free
surface
Note that within the flow section both h and zb are variable, but zs is a constant.
Now, the following identity is developed by application of Leibniz’s rule:
Zyr Zyr
@h @ @yr @yl @A @yr @yl
dy ¼ hðx; y; tÞdy hr þ hl ¼ hr þ hl ; ð1:67Þ
@t @t @t @t @t @t @t
yl yl
where the flow area is defined by
Zyr Zzs Zyr

Aðx; tÞ ¼ dydz ¼ hðx; y; tÞdy: ð1:68Þ
yl zb yl
The lateral integration of the second term in Eq. (1.66) yields
Zyr Zyr
@ ðUhÞ @ @yr @yl @Q @yr @yl
dy ¼ Uhdy ðUhÞr þ ðUhÞl ¼ ðUhÞr þ ðUhÞl ;
@x @x @x @x @x @x @x
yl yl
ð1:69Þ
where the discharge Q is given by
Zyr Zzs Zyr

Qðx; tÞ ¼ udydz ¼ Uhdy: ð1:70Þ
yl zb yl
Integration of the remaining term related to V produces
Zyr
@ ðVhÞ
dy ¼ ðVhÞr ðVhÞl : ð1:71Þ
@y
yl
Summing Eqs. (1.67), (1.69) and (1.71) yields

@A @Q @yl @yl @yr @yr
þ þ hl þ Ul Vl hr þ Ur Vr ¼ 0: ð1:72Þ
@t @x @t @x @t @x
To further simplify Eq. (1.72), Chen and Chow (1968) assumed that the
depth-averaged velocity field (U, V) must satisfy a kinematic boundary condition at
the banks, whereas Jain (2001) considered the usual case of a cross section of zero
water depth at the banks, e.g., zs = zb (Fig. 1.9). One may state, therefore,

@y
@yl @yl þ Ul @y
@x Vl ¼ 0 ) Option 1
l l
hl þ Ul Vl ¼ 0 ) @t ð1:73Þ
@t @x hl ¼ 0 ) Option 2

@y
@yr @yr þ Ur @y
@x Vr ¼ 0 ) Option 1
r r
hr þ Ur Vr ¼ 0 ) @t ð1:74Þ
@t @x hr ¼ 0 ) Option 2
By using either of the two arguments, Eq. (1.72) simplifies to the section-averaged
continuity equation
@A @Q
þ ¼ 0: ð1:75Þ
@t @x
Further, the x-momentum equation [Eq. (1.51)] will be laterally integrated. For
convenience in ensuing developments, the identity
Zzs Zzs
@sxy @sxz @ @zb
þ dz ¼ sxy dz þ sxy b ðsxz Þb ; ð1:76Þ
@y @z @y @y
zb zb
is used to revert a step of the depth-integration process, resulting in

2 3
Zzs Zzs
@ @ 2 @ 14@ @zb @sxy @sxz
ðUhÞ þ U h þ ðUVhÞ ¼ sxx dz þ ðsxx Þb þ þ dz5:
@t @x @y q @x @x @y @z
zb zb
ð1:77Þ
Considering the stress tensor given by Eqs. (1.47), (1.77) transforms to

2
Zzs
@ @ 2 @ 14@ @zb
ðUhÞ þ U h þ ðUVhÞ ¼ ðp rxx Þdz þ ðp rxx Þb
@t @x @y q @x @x
zb
3
Zzs
@rxy @rxz
þ dz5:
@y @z
zb
ð1:78Þ
Assuming hydrostatic pressure distribution and neglecting rxx, one may write
Eq. (1.78) as
2 3
Zzs
@ @ 2 @ 14@ 1 @zb @rxy @rxz
ðUhÞ þ U h þ ðUVhÞ ¼ qgh þ qgh
2
þ dz5;
@t @x @y q @x 2 @x @y @z
zb
ð1:79Þ
or,
2 z 3
Zs
@ @ 2 @ @zs 14 @rxy @rxz
ðUhÞ þ U h þ ðUVhÞ ¼ gh þ þ dz5:
@t @x @y @x q @y @z
zb
ð1:80Þ
Equation (1.80) is now laterally integrated as

2 3
Zyr Zyr Zzs
@ @ 2 @ @z 1 @r @r
ðUVhÞ dy ¼ 4gh dz5dy:
s xy xz
ðUhÞ þ Uh þ þ þ
@t @x @y @x q @y @z
yl yl zb
ð1:81Þ
The following identities are generated to integrate the acceleration terms [left-hand
side of Eq. (1.81)]
Zyr Zyr
@ @ @yr @yl
ðUhÞdy ¼ Uhdy ðUhÞr þ ðUhÞl ; ð1:82Þ
@t @t @t @t
yl yl
Zyr Zyr
@ ð U 2 hÞ @ @yr 2 @yl
dy ¼ U 2 hdy U 2 h r þ U h l ; ð1:83Þ
@x @x @x @x
yl yl
Zyr
@ ðUVhÞ
dy ¼ ðUVhÞr ðUVhÞl : ð1:84Þ
@y
yl
Summing Eqs. (1.82)–(1.84) yields
Zyr
@ @ 2 @ @Q @ Q2
ðUhÞ þ U h þ ðUVhÞ dy ¼ þ
@t @x @y @t @x A
yl ð1:85Þ

@yl @yl @yr @yr
þ ðUhÞl þ Ul Vl ðUhÞr þ Ur Vr ;
@t @x @t @x
and, using Eqs. (1.73)–(1.74),
Zyr
@ @ 2 @ @Q @ Q2
ðUhÞ þ U h þ ðUVhÞ dy ¼ þ : ð1:86Þ
@t @x @y @t @x A
yl
In this process, the variation of U in the lateral direction was neglected. If the
variation of the velocity component u within the cross section is accounted for, the
momentum flux term Q2/A should be multiplied by the Boussinesq correction
coefficient for the section (Liggett 1994; Montes 1998), given by
R
u2 dA
b¼ A
: ð1:87Þ
ðQ2 =AÞ
For a horizontal water surface within the cross section, the free surface slope term
reads
Zyr Zyr
@zs @zs @zs
g hðx; y; tÞdy ¼ g hðx; y; tÞdy ¼ g A: ð1:88Þ
@x @x @x
yl yl
Using Green’s theorem, the following identity is generated (Keulegan and Patterson
1943)
Zyr Zzs Z I
@rxy @rxz @rxy @rxz
þ dzdy ¼ þ dA ¼ rxy dz rxz dy ;
@y @z @y @z
yl zb A C
ð1:89Þ
where the line integral extends along a contour C as the sum of the wetted perimeter
P (Fig. 1.10) and the free surface width B. The integrand of the line integral is the x-
component of the fluid force exerted on the channel solid contour and free surface.
This shear force is denoted by sb, from which a mean shear stress so is defined by
Fig. 1.10 Definition of

wetted perimeter
I I

rxy dz rxz dy ¼ sb dP ¼ so P; ð1:90Þ
C C
where stresses acting on the free surface are neglected.

Inserting Eqs. (1.86), (1.88), and (1.90) into Eq. (1.81) yields the cross-sectional
averaged x-momentum balance as

@Q @ Q2 @zs 1
þ ¼ gA so P: ð1:91Þ
@t @x A @x q
This equation is valid for non-prismatic channels and is a form widely used in river
hydraulics computations (Cunge et al. 1980; Wu 2008). Defining the friction slope
Sf as
so P so
Sf ¼ ¼ ; ð1:92Þ
qg A qgRh
where Rh = A/P is the hydraulic radius, the following “slope form” results (Cunge
et al. 1980; Chow et al. 1988)

@Q @ Q2 @h
þ ¼ gA So Sf ; ð1:93Þ
@t @x A @x
where So ¼ @zb =@x is the bottom slope. For practical computations, consider a
flow depth and bed elevation taking as reference the thalweg; thus h = h(x, t) and
zb = zb(x, t) (Fig. 1.11). Consider further a transformation of Eq. (1.93), which is
widely used in practice. The hydrostatic pressure force on an arbitrary cross section
is
Z Z ZhðxÞ
Fp ¼ pdA ¼ qg ðh gÞdA ¼ qg ðh gÞbðx; gÞdg; ð1:94Þ
A A 0
Fig. 1.11 Definition of flow

depth and bed elevation in
non-prismatic channel section
where the width b at elevation η is a function of x in non-prismatic channels

(Fig. 1.11). The centroid of the flow section is defined by its depth below the free
surface h. Thus, from Eq. (1.94), one can write
Z ZhðxÞ
Fp
¼ Ah ¼ ðh gÞdA ¼ ðh gÞbðx; gÞdg: ð1:95Þ
qg
A 0
The product Ah is obviously the fundamental quantity determining the pressure

forces. Now, its x-gradient is computed as follows (Cunge et al. 1980)
ZhðxÞ ZhðxÞ ZhðxÞ ZhðxÞ

@ @ @ @h @b
Ah ¼ ðh gÞbðx; gÞdg ¼ ½ðh gÞbdg ¼ bdg þ ðh gÞ dg
@x @x @x @x @x
0 0 0 0
ZhðxÞ ZhðxÞ ZhðxÞ

@h @b @h @b
¼ bdg þ ðh gÞ dg ¼ Aþ ðh gÞ dg:
@x @x @x @x
0 0 0
ð1:96Þ
This relation expresses the spatial variation of hydrostatic pressure forces per unit
weight in a non-prismatic channel.
Inserting Eq. (1.96) into Eq. (1.93) produces the alternative form for
non-prismatic channels
ZhðxÞ
@Q @ Q2 @b
þ þ gAh ¼ gA So Sf þ g ðh gÞ dg: ð1:97Þ
@t @x A @x
0
In compound channels, the equivalent roughness height of the main channel is

lower than in the flood plains, resulting in a significant lateral variation of the
depth-averaged velocity U (Fig. 1.12). The result is that b becomes significant and
must be accounted for in the momentum balance [Eq. (1.97)] as (Montes 1998;
Sturm 2001; Jain 2001)
2 ZhðxÞ
@Q @ Q @b
þ b þ gAh ¼ gA So Sf þ g ðh gÞ dg: ð1:98Þ
@t @x A @x
0
If the pressure is assumed to be hydrostatic and the velocity uniform, the specific
momentum M in free surface flows is defined as (Jaeger 1956; Montes 1998)
Fig. 1.12 Lateral distribution of depth-averaged velocity component in x-direction for uniform
flow in compound channel cross section, assuming that the velocity is constant in main channel
and flood plains, each linked to a different value of the equivalent roughness ks
Z 2
u p Q2 Q2
M¼ þ dA b þ Ah þ Ah: ð1:99Þ
g qg gA gA
A
The definition of M arises thus naturally from the application of the x-momentum
balance, as evidenced on inspecting Eq. (1.97). For a prismatic channel, Eq. (1.98)
reduces to

@Q @ Q2
þ þ gAh ¼ gA So Sf : ð1:100Þ
@t @x A
This is a form widely employed to compute 1D unsteady free surface flows in

prismatic channels (Chaudhry 2008) (Fig. 1.13).
A practical result is directly regained from Eq. (1.100). For steady flow in a
hydraulic jump on a horizontal channel, neglecting friction forces, Eq. (1.100) may
be integrated between the boundary sections upstream and downstream of the jump,
where the flow is gradually varied, resulting in

Q2 Q2
þ gAh ¼ þ gAh : ð1:101Þ
A upstream A dowstream
Fig. 1.13 1D unsteady free

surface flow in prismatic
channel
This states conservation of M between the boundaries of a control volume con-

taining inside the rapidly varied flow produced within a hydraulic jump (Fig. 1.6b).
This is the fundamental equation of hydraulic jumps, to be used in Chap. 4.
For steady flows, Eq. (1.93) yields

d Q2 dh dzb
¼ gA gA gASf ; ð1:102Þ
dx A dx dx
or,
Q2 dA dh dzb
¼ gA gA gASf : ð1:103Þ
A2 dx dx dx
For a non-prismatic channel, the x-derivative of the flow area A can be expressed as

dA @A dh @A
¼ þ : ð1:104Þ
dx @h dx @x h¼const:
These partial derivatives are trivially simple to evaluate in man-made channels; for
example, the result for a rectangular channel of variable width is

@A db @A
Aðx; hÞ ¼ bð xÞh ) ¼h ; ¼ b: ð1:105Þ
@x h¼const: dx @h
Inserting Eq. (1.104) into Eq. (1.103) yields

Q2 @A dh Q2 @A
gA 1 3 ¼ gA So Sf þ 2 ; ð1:106Þ
gA @h dx A @x h¼const:
from which the ordinary differential equation describing the flow surface profile is
(Le Méhauté 1976; Katopodes 2019)

Q2 @A 2
dh So Sf þ gA3 @x h¼const: So Sf þ FB @A @x h¼const:
¼ ¼ : ð1:107Þ
dx Q2
1 gA 3 B 1 F2
Here the Froude number F is defined as
Q=A
F¼ ; ð1:108Þ
ðgA=BÞ1=2
and the free surface width B is determined from
Zh
@A @
¼ bðx; gÞdg ¼ B: ð1:109Þ
@h @h
0
The Froude number is a fundamental dimensionless number governing free surface

flows. Its physical meaning will be explained in detail in Chap. 2. For a prismatic
channel @A=@x ¼ 0; and Eq. (1.107) simplifies to the steady gradually varied flow
equation for prismatic channels
dh So Sf
¼ : ð1:110Þ
dx 1F2
Equations (1.107) and (1.110) are widely used in open channel hydraulics, but it
should be borne in mind that they are unreliable in highly curvilinear flows, as for
flows over a dam spillway crest operating at a high head (Montes 1998).
1.4 Non-hydrostatic Theory
1.4.1 1D Shallow Water Non-hydrostatic Equations
Steffler and Jin (1993) and Khan and Steffler (1996b, c) developed a depth-averaged
non-hydrostatic 1D momentum model for flows in the vertical plane (x, z), to be
described here. Other approximations are presented in Chap. 11. Consider a 2D
steady flow over a wavy bed (Fig. 1.14a). The velocity component u in the x-
direction is approximated by its depth-averaged value U, like in the Saint-Venant
theory, but the vertical velocity w and the non-hydrostatic fluid pressure p are
assumed to vary linearly in the vertical direction (Fig. 1.14b). Assuming 1D flow in
the x-direction, using Eq. (1.47) for the stress tensor and neglecting depth-averaged
turbulent stresses, Eq. (1.51) takes the form
0 1
Zzs
@ @ @ 2 p A 1 @zb @zb
ðUhÞ þ U hþ dz ¼ pb ðrxz Þb þ rxx ; ð1:111Þ
@t @x q q @x @x
zb
whereas Eq. (1.60) yields for 1D flow
@ @ @zb
pb ¼ qgh þ ðqWhÞ þ ðqWUhÞ þ ðrzz Þb ðrxz Þb : ð1:112Þ
@t @x @x
Fig. 1.14 1D
non-hydrostatic flow over
uneven terrain a longitudinal
profile, b assumed horizontal
and vertical velocity, and
pressure profiles
Simple shear along the bed slope with tangential stress sb is considered. For this
choice, the bed Reynolds stresses are given by (Steffler and Jin 1993; Castro-Orgaz
and Hager 2017)
ðrzz Þb ¼ 2sb cos h sin h;

ðrxx Þb ¼ 2sb cos h sin h; ð1:113Þ

ðrxz Þb ¼ sb cos2 h sin2 h ;
where h is the bed slope angle. It is therefore obvious that the bed Reynolds stresses
are not negligible on steep slopes. For h ! 0 the channel is of a small slope,
resulting in
ðrzz Þb 0; ðrxx Þb 0; ðrxz Þb sb : ð1:114Þ
This is an approximation used in the Saint-Venant theory, but not here. The
Reynolds stresses contribution in the x-momentum equation is thus

@zb sin h
ðrxx Þb þ ðrxz Þb ¼ 2sb cos h sin h þ sb cos2 h sin2 h
@x cos h ð1:115Þ

¼ sb cos2 h þ sin2 h ¼ sb :
1.4 Non-hydrostatic Theory 35
The Reynolds stresses contribution to the z-momentum balance is

@zb sin h
ðrzz Þb ðrzx Þb ¼ 2sb cos h sin h sb cos2 h sin2 h
@x cos h
2 ð1:116Þ
sin h þ cos h
2
sin h @zb
¼ sb sin h ¼ sb ¼ sb :
cos h cos h @x
Inserting Eqs. (1.115)–(1.116) for a pure bed shear into Eqs. (1.111) and (1.112)
gives
0 1
Zzs
@ @ @ 2 p A pb @zb sb
ðUhÞ þ U hþ dz ¼ ; ð1:117Þ
@t @x q q @x q
zb
@ @ @zb
pb ¼ qgh þ q ðWhÞ þ q ðWUhÞ þ sb : ð1:118Þ
|{z} @t @x @xffl}
hydrostatic term
|fflfflfflfflffl{zfflfflfflfflffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} |fflffl{zffl
local acceleration convective acceleration Reynolds stresses
The vertical velocity w is assumed to vary linearly with depth (Fig. 1.14b). Its
depth-averaged value W is thus given by
ws þ wb
W¼ ; ð1:119Þ
2
where the free surface and bed kinematic boundary conditions are, respectively,
@ @
ws ¼ ðh þ zb Þ þ U ðh þ zb Þ; ð1:120Þ
@t @x
@zb @zb
wb ¼ þU : ð1:121Þ
@t @x
A distribution p = p(z) must be introduced into Eq. (1.117) for model closure.
A linear distribution is assumed following Khan and Steffler (1996b) and Denlinger
and Iverson (2004)
g g
p ¼ pb 1 ¼ ðp1 þ qghÞ 1 ; ð1:122Þ
h h
where p1 is the difference between the actual pb and the hydrostatic bottom pressure
(qgh). The non-hydrostatic pressure force is thus
Zzs Zh
h2 hp1
pdz ¼ pdg ¼ qg þ : ð1:123Þ
2 2
zb 0
Inserted into Eq. (1.117) yields

@ @ h2 hp1 pb @zb sb
ðUhÞ þ U hþg þ
2
¼ ; ð1:124Þ
@t @x 2 2q q @x q
or

@ @ h pb pb @zb sb
ðUhÞ þ U hþ
2
¼ : ð1:125Þ
@t @x 2q q @x q
This equation is the non-hydrostatic Boussinesq-type x-momentum equation

developed by Khan and Steffler (1996b, c), with closure for pb given by
Eq. (1.118). For steady flow, the system of equations to solve is
dM pb dzb sb
¼ ; ð1:126Þ
dx q dx q
h pb
M ¼ U2h þ ; ð1:127Þ
2q
d dzb
pb ¼ qgh þ q ðWUhÞ þ sb : ð1:128Þ
dx dx
ws þ wb
W¼ : ð1:129Þ
2
d
ws ¼ U ðh þ zb Þ; ð1:130Þ
dx
dzb
wb ¼ U : ð1:131Þ
dx
Inserting Eqs. (1.129)–(1.131) in Eq. (1.128) produces after some manipulations

" 2 #
U 2 d2 h dh dh dzb d2 z b dzb
pb ¼ qgh þ q h 2 2 þ 2h 2 þ sb : ð1:132Þ
2 dx dx dx dx dx dx
Using the result for pb, the specific momentum M is thus

" #
h2 1 d2 h 1 dh 2 1 d2 zb 1 dh dzb h dzb sb
M ¼ g þ hU 1 þ h 2
2
þ h 2 þ :
2 4 dx 4 dx 2 dx 2 dx dx 2 dx q
ð1:133Þ
Inserting Eqs. (1.132)–(1.133) into Eq. (1.126), a third-order ODE describing

h = h(x) results. This is a Boussinesq-type equation; the solution of this type of
equation is presented in Chap. 11. This equation can be used to describe the rapidly
varied non-hydrostatic flow over a spillway crest (Figs. 1.3f and 1.4a). For details
see Castro-Orgaz and Hager (2017). Note that important non-hydrostatic effects
accounted for in Eqs. (1.132)–(1.133) originate from the curvature terms
@ 2 h=@x2 and @ 2 zb =@x2 ; but non-negligible contributions are also introduced by the
slope terms @h=@x and @zb =@x: An alternative but similar equation suitable for
potential flows is used in Chap. 11.
1.4.2 Uniform and Gradually Varied Flow on Steep Slopes
Consider uniform flow on a steep slope (Fig. 1.15), for which the flow depth is a
constant and there are not spatial variations of any flow variable. Equations (1.126)
and (1.128) simplify then to
sb ¼ pb tan h; ð1:134Þ
pb ¼ qgh þ sb tan h: ð1:135Þ
Combining Eqs. (1.134) and (1.135) results in the non-hydrostatic bottom pressure
on a steep slope as (Chaudhry 2008; Castro-Orgaz et al. 2015; Castro-Orgaz and
Hager 2017)
Fig. 1.15 1D
non-hydrostatic uniform flow
on a steep slope
pb h
¼ ¼ hcos2 h: ð1:136Þ
qg 1 þ tan2 h
Equation (1.136) indicates that the bottom pressure on a steep slope is

non-hydrostatic. This is a classic result (e.g. Chow 1959; Henderson 1966) already
highlighted by Lauffer (1935). Consider gradually varied flows on a steep slope
(Fig. 1.16), for which the variation of h with x is small. Castro-Orgaz and Hager
(2017) demonstrate that Eq. (1.136) can be used in gradually varied non-hydrostatic
flows on steep slopes. Splitting the bottom pressure in hydrostatic and
non-hydrostatic (p1) components reads
pb h p1
¼ ¼ þ h; ð1:137Þ
qg 1 þ tan2 h qg
which yields for p1

2
dzb
p1 dx
¼ 2 h: ð1:138Þ
qg dzb
1þ
dx
Fig. 1.16 1D
non-hydrostatic gradually
varied flow on a steep slope
a vertical pressure profile,
b free surface and bottom
pressure head line
The vertical pressure distribution is therefore linear with this value of p1 for the
non-hydrostatic correction (Fig. 1.16a). The specific momentum M is thus
2
dzb
2 2
h hp1 h dx h2 0h
2
M ¼ U2h þ g þ ¼ U2h þ g g 2 ¼ U h þ g ; ð1:139Þ
2
2 2q 2 dzb 2 2
1þ
dx
where g0 is the enhanced gravity to account for slope effects (Denlinger and Iverson
2004; Denlinger and O’Connell 2008), given by
1
g0 ¼ g 2 : ð1:140Þ
dzb
1þ
dx
The bottom pressure head is thus rewritten as

2 3

dzb 2
6 7
6 dx 7 0
pb ¼ qgh þ p1 ¼ qgh61 2 7 ¼ qg h: ð1:141Þ
4 dzb 5
1þ
dx
Inserting Eqs. (1.139) and (1.141) into Eq. (1.126) produces the gradually varied
flow equation on a steep slope as

d h2 dzb sb
U 2 h þ g0 ¼ g0 h : ð1:142Þ
dx 2 dx q
With U = q/h as the depth-averaged horizontal velocity, Eq. (1.142) transforms into

dh q2 dzb sb
1 0 3 ¼ : ð1:143Þ
dx gh dx qg0 h
The stress sb acts tangentially to the sloping plane (Steffler and Jin 1993)
(Fig. 1.16b). Therefore, it is parametrized as (Khan and Steffler 1996b)
sb ¼ qCf Uo2 ; ð1:144Þ
with Cf as the bed friction coefficient and Uo as the velocity component parallel to
the sloping plane. The velocity components of the Boussinesq-type model [given
by Eqs. (1.126–1.131)] in the (x, z) directions are (U, W). The following approx-
imation is now used (Khan and Steffler (1996b)
1=2
Uo U 2 þ W 2 : ð1:145Þ
For uniform flow on a steep slope,

dzb
W ¼U : ð1:146Þ
dx
Using this relation for gradually varied flows, the coupling of Eqs. (1.145)–(1.146)
generates
" 2 #1=2
dzb
Uo ¼ U 1 þ : ð1:147Þ
dx
Thus, combining Eqs. (1.144) and (1.147) yields the bottom shear stress as
" 2 #
dzb
sb ¼ qCf U 1 þ
2
; ð1:148Þ
dx
from which the friction slope on a steep bottom slope is

" 2 #
sb dzb
Sf ¼ ¼ Cf F 1 þ
2
: ð1:149Þ
qgh dx
1=2
Equation (1.143) can be rewritten with F ¼ q=ðgh3 Þ as
" 2 #
dzb dzb
Sf 1 þ
dh dx dx
¼ " # : ð1:150Þ
dx dzb 2
1 F 1þ
2
dx
Inserting Eq. (1.149) then yields

" 2 #2
dzb dzb
C f F2 1 þ
dh dx dx
¼ " # : ð1:151Þ
dx dzb 2
1 F 1þ
2
dx
This is the gradually varied flow equation for non-hydrostatic flow on steep rect-
angular channels. It will be considered in Chap. 3.
1.5 Sediment Transport and Movable Beds 41
1.5 Sediment Transport and Movable Beds
1.5.1 1D Cross-Sectional Averaged Continuity Equation
Consider in this section the development of the cross-sectional averaged continuity

equation for flows of variable density q(x, t) flowing over a movable bed
(Fig. 1.17). Important flows of this kind are dam break waves over erodible beds
(Wu 2008; Cantero-Chinchilla et al. 2016). We start with the general
depth-integrated continuity equation [Eq. (1.23)]
Zzs Zzs Zzs

@ @ @
qdz þ qudz þ qvdz
@t @x @y
zb zb zb

@zs @zs @zs @zb @zb @zb
qs þ us þ vs ws þ qb þ ub þ vb wb ¼ 0:
@t @x @y @t @x @y
ð1:152Þ
The free surface is taken as a material surface, and the bed is considered erodible,
but a non-slip kinematic boundary conditions is implemented, that is, u = v =
w = 0 at the bed (Wu 2008). Equation (1.152) reduces then to
Zzs Zzs Zzs

@ @ @ @zb
qdz þ qudz þ qvdz þ qb ¼ 0: ð1:153Þ
@t @x @y @t
zb zb zb
Taking average values of u, v, namely (U, V), and q, denoted by the same symbol to
simplify notation, Eq. (1.153) simplifies to
@ @ @ @zb
ðqhÞ þ ðqUhÞ þ ðqVhÞ þ qb ¼ 0: ð1:154Þ
@t @x @y @t
Note that the movable bed due to sediment transport is accounted for by inclusion
of the time derivative @zb =@t in the continuity equation. An average value of
density is used, but a Reynolds decomposition of this variable into mean and
Fig. 1.17 One-dimensional flow with movable bed and sediment transport
fluctuating parts is not done here for the sake of simplicity. Lateral integration of
Eq. (1.154) from the left to the right channel bank yields
Zyr
@ @ @ @zb
ðqhÞ þ ðqUhÞ þ ðqVhÞ þ qb dy ¼ 0: ð1:155Þ
@t @x @y @t
yl
The following identities are generated by application of Leibniz’s rule:
Zyr Zyr
@ @ @yr @yl
ðqhÞdy ¼ qhdy ðqhÞr þ ðqhÞl
@t @t @t @t
yl yl
@ @yr @yl
¼ ðqAÞ ðqhÞr þ ðqhÞl ; ð1:156Þ
@t @t @t
Zyr Zyr
@ ðqUhÞ @ @yr @yl
dy ¼ qUhdy ðqUhÞr þ ðqUhÞl
@x @x @x @x
yl yl ð1:157Þ
@ @yr @yl
¼ ðqQÞ ðqUhÞr þ ðqUhÞl ;
@x @x @x
Zyr
@ ðqVhÞ
dy ¼ ðqVhÞr ðqVhÞl ; ð1:158Þ
@y
yl
Zyr Zyr
@zb @ @yr @yl
qb dy ¼ qb þ qb ðzb Þl
zb dy qb ðzb Þr
@t @t @t @t
yl yl ð1:159Þ

@Ab @yr @yl @Ab @B
¼ qb qb z s ¼ qb qb zs ;
@t @t @t @t @t
where Ab is the bed area above the reference datum (Fig. 1.17). Summing
Eqs. (1.156)–(1.159) generates

@ @ @yl @yl @yr @yr
ðqAÞ þ ðqQÞ þ hl þ Ul Vl qhr þ Ur Vr
@t @x @t @x @t @x
@Ab @B
þ qb q b zs ¼ 0:
@t @t
ð1:160Þ
Using Eqs. (1.73)–(1.74), and dropping the term related to @B=@t; which is not
considered in river flow models for fast geomorphic flows (Wu 2008), Eq. (1.160)
yields finally
@ @ @Ab
ðqAÞ þ ðqQÞ þ qb ¼ 0: ð1:161Þ
@t @x @t
1.5.2 1D Cross-Sectional Averaged x-Momentum Equation
We start with the general depth-integrated x-momentum equation [Eq. (1.34)]
Zzs Zzs Zzs

@ @ @ @zs @zs @zs
qudz þ qu dz þ
2
quvdz qs us þ us þ vs ws
@t @x @y @t @x @y
zb zb zb
( Zzs Zzs
@zb @zb @zb @ @
þ qb ub þ ub þ vb wb þ sxx dz þ sxy dz
@t @x @y @x @y
zb zb
)
@zs @zs @zb @zb
@x @y @x @y
ð1:162Þ
This simplifies upon assuming that the free surface is a material surface and using a
non-slip velocity at the bed, to
Zzs Zzs Zzs ( Zzs Zzs

@ @ @ @ @
qudz þ qu dz þ
2
quvdz þ sxx dz þ sxy dz
@t @x @y @x @y
zb zb zb zb zb
)
@zs @zs @zb @zb
@x @y @x @y
ð1:163Þ
Neglecting stresses at the free surface, it further simplifies to
Zzs Zzs Zzs Zzs

@ @ @ @
qudz þ qu dz þ
2
quvdz þ sxx dz
@t @x @y @x
zb zb zb zb
ð1:164Þ
Zzs
@ @zb @zb
þ sxy dz þ ðsxx Þb þ sxy b ðsxz Þb ¼ 0:
@y @x @y
zb
Now, depth-averaged values of u, v, and q are taken, resulting in
@ @ 2 @
ðqUhÞ þ qU h þ ðqUVhÞ
@t @x @y
2 3
Zzs Zzs ð1:165Þ
@ @ @z @zb
¼ 4 ðsxz Þb 5:
b
sxx dz þ sxy dz þ ðsxx Þb þ sxy b
@x @y @x @y
zb zb
Using the identity [Eq. (1.76)]
Zzs Zzs
@sxy @sxz @ @zb
þ dz ¼ sxy dz þ sxy b ðsxz Þb ; ð1:166Þ
@y @z @y @y
zb zb
the following alternative form of Eq. (1.165) is obtained

2 3
Zzs Zzs
@ @ 2 @ @ @z @s @s
ðqUVhÞ ¼ 4 dz5:
b xy xz
ðqUhÞ þ qU h þ sxx dz þ ðsxx Þb þ þ
@t @x @y @x @x @y @z
zb zb
ð1:167Þ
The stress tensor given by Eqs. (1.47) is used, and again noted that even though the
velocity field (u, v, w) represents the time-averaged velocity field when this stress
tensor is used, we do not conduct here a Reynolds decomposition of density for
simplicity’s sake. Once the stress tensor is introduced into Eq. (1.167), neglecting
normal turbulent stresses and assuming a hydrostatic pressure distribution, the
resulting equations is
Zzs
@ @ 2 @ @
ðqUhÞ þ qU h þ ðqUVhÞ ¼ qgðzs zÞdz
@t @x @y @x
zb
Zzs
@zb @rxy @rxz
qgh þ þ dz;
@x @y @z
zb
ð1:168Þ
or,

@ @ 2 @ @ 1
ðqUhÞ þ qU h þ ðqUVhÞ ¼ qgh2
@t @x @y @x 2
Zzs
@zb @rxy @rxz
qgh þ þ dz:
@x @y @z
zb
ð1:169Þ
Expanding the pressure gradient terms yields
@ @ 2 @ @h 1 @q
ðqUhÞ þ qU h þ ðqUVhÞ ¼ qgh gh2
@t @x @y @x 2 @x
Zzs
@zb @rxy @rxz
qgh þ þ dz;
@x @y @z
zb
ð1:170Þ
and using the gradient of the free surface, Eq. (1.170) transforms into
@ @ 2 @ @zs
ðqUhÞ þ qU h þ ðqUVhÞ ¼ qgh
@t @x @y @x
Zzs
1 @q @rxy @rxz
gh2 þ þ dz:
2 @x @y @z
zb
ð1:171Þ
This is a suitable form of the depth-integrated x-momentum equation for flows with
sediment transport over movable beds. Lateral integration of Eq. (1.171) from the
left to the right channel bank yields
Zyr
@ @ 2 @
ðqUhÞ þ qU h þ ðqUVhÞ dy
@t @x @y
yl
2 3
Zyr Zzs
@z 1 @q @r @r
¼ 4qgh dz5dy:
s xy xz
gh2 þ þ ð1:172Þ
@x 2 @x @y @z
yl zb
The following identities are generated by application of Leibniz’s rule:
Zyr Zyr
@ @ @yr @yl
ðqUhÞdy ¼ qUhdy ðqUhÞr þ ðqUhÞl ; ð1:173Þ
@t @t @t @t
yl yl
Zyr Zyr
@ 2 @ @yr 2 @yl
qU h dy ¼ qU 2 hdy qU 2 h r þ qU h l ; ð1:174Þ
@x @x @x @x
yl yl
Zyr
@
ðqUVhÞdy ¼ ðqUVhÞr ðqUVhÞl : ð1:175Þ
@y
yl
Summing Eq. (1.173)–(1.175) yields, using Eqs. (1.73–1.74),
Zyr 2
@ @ 2 @ @ @ Q
ðqUhÞ þ qU h þ ðqUVhÞ dy ¼ ðqQÞ þ q : ð1:176Þ
@t @x @y @t @x A
yl
For a horizontal water surface across the cross section
Zyr Zyr
@zs @zs @zs
g hdy ¼ g hdy ¼ g A: ð1:177Þ
@x @x @x
yl yl
Using Green’s theorem, the average shear stress is given by
Zyr Zzs
@rxy @rxz
þ dzdy ¼ so P: ð1:178Þ
@y @z
yl zb
Lateral integration of the term containing the density gradient yields
Zyr Zyr Zyr Z

1 2 @q 1 @q 1 @q 1 @q 1 @q
gh dy ¼ g h dy ¼ g
2
hðhdyÞ ¼ g hdA ¼ g hp A:
2 @x 2 @x 2 @x 2 @x 2 @x
yl yl yl A
ð1:179Þ
Using Eqs. (1.176)–(1.179), the cross-sectional averaged x-momentum equation is

(Wu 2008)
2
@ @ Q @zs 1 @q
ðqQÞ þ q ¼ qgA so P g hp A; ð1:180Þ
@t @x A @x 2 @x
or
2
@ @ Q @zs 1 @q
ðqQÞ þ q ¼ qgA qgASf g hp A: ð1:181Þ
@t @x A @x 2 @x
This is a widely used momentum model for 1D river flows with sediment transport
(Wu 2008). In this equation, the variable fluid density in space and time due to
sediment transport is fully accounted for.
References
Boussinesq, J. (1877). Essai sur la théorie des eaux courantes [Essay on the theory of water flow].
Mémoires présentés par divers savants à l’Académie des Sciences, Paris 23, pp. 1–680
(in French).
Cantero-Chinchilla, F., Castro-Orgaz, O., Dey, S., & Ayuso, J. L. (2016). Nonhydrostatic dam
break flows II: One-dimensional depth-averaged modelling for movable bed flows. Journal of
Hydraulic Engineering, 142(12), 04016069.
Castro-Orgaz, O., & Hager, W. H. (2009). Classical hydraulic jump: Basic flow features. Journal
of Hydraulic Research, 47(6), 744–754.
Castro-Orgaz, O., & Hager, W.H. (2017). Non-hydrostatic free surface flows. In Advances in
geophysical and environmental mechanics and mathematics (696 pages), Berlin: Springer,
https://doi.org/10.1007/978-3-319-47971-2.
Castro-Orgaz, O., Hutter, K., Giráldez, J. V., & Hager, W. H. (2015). Non-hydrostatic granular
flow over 3D terrain: New Boussinesq-type gravity waves? Journal of Geophysical Research:
Earth Surface, 120(1), 1–28.
Chanson, H. (2004). The hydraulics of open channel flows: An introduction. Oxford, UK:
Butterworth-Heinemann.
Chaudhry, M. H. (2008). Open-channel flow (2nd ed.). Berlin: Springer.
Chen, C. L., & Chow, V. T. (1968). Hydrodynamics of mathematically simulated surface runoff.
Hydraulic Engineering Series No. 18, Civil Engineering Studies. Urbana-Champaign:
University of Illinois.
Chow, V. T. (1959). Open channel hydraulics. New York: McGraw-Hill.
Chow, V. T., Maidment, D. R., & Mays, L. W. (1988). Applied hydrology. New York:
McGraw-Hill.
Cunge, J. A., Holly, F. M., & Verwey, A. (1980). Practical aspects of computational river
hydraulics. London: Pitman.
de Saint-Venant, A.B. (1871). Théorie du mouvement non permanent des eaux, avec application
aux crues des rivières et à l’introduction des marrées dans leur lit [Theory of unsteady water
movement, applied to floods in rivers and the effect of tidal flows]. Comptes Rendus de
l’Académie des Sciences, 73, 147–154; 73, 237–240 (in French).
Denlinger, R. P., & Iverson, R. M. (2004). Granular avalanches across irregular three-dimensional
terrain: 1. Theory and computation. Journal of Geophysical Research: Earth Surface, 109(F1),
F01014. https://doi.org/10.1029/2003JF000085.
Denlinger, R. P., & O’Connell, D. R. H. (2008). Computing nonhydrostatic shallow-water flow

over steep terrain. Journal of Hydraulic Engineering, 134(11), 1590–1602.
Dressler, R. F. (1978). New nonlinear shallow flow equations with curvature. Journal of Hydraulic
Research, 16(3), 205–222.
Henderson, F. M. (1966). Open channel flow. New York: MacMillan.
Iverson, R. M. (2005). Debris-flow mechanics. In M. Jakob & O. Hungr (Eds.), Debris flow
hazards and related phenomena (pp. 105–134). Heidelberg: Springer-Praxis.
Iverson, R. M., & Ouyang, C. (2015). Entrainment of bed material by Earth-surface mass flows:
Review and reformulation of depth-integrated theory. Reviews of Geophysics, 53(1), 27–58.
https://doi.org/10.1002/2013rg000447.
Jaeger, C. (1956). Engineering fluid mechanics. Edinburgh: Blackie and Son.
Jain, S. C. (2001). Open channel flow. New York: Wiley & Sons.
Katopodes, N. D. (2019). Free surface flow: Shallow-water dynamics. Oxford, UK:
Keulegan, G. H. (1942). Equation of motion for the steady mean flow of water in open channels.
Journal of Research, US National Bureau of Standards, 29, 97–111.
Keulegan, G. H., & Patterson, G. W. (1943). Effect of turbulence and channel slope in translation
waves. Journal of Research, US National Bureau of Standards, 30, 461–512.
Khan, A. A., & Steffler, P. M. (1996a). Physically-based hydraulic jump model for depth-averaged
computations. Journal of Hydraulic Engineering, 122(10), 540–548.
Khan, A. A., & Steffler, P. M. (1996b). Modelling overfalls using vertically averaged and moment
equations. Journal of Hydraulic Engineering, 122(7), 397–402.
Khan, A. A., & Steffler, P. M. (1996c). Vertically averaged and moment equations model for flow
over curved beds. Journal of Hydraulic Engineering, 122(1), 3–9.
Lai, C. (1986). Numerical modelling of unsteady open-channel flow. In Advances in hydroscience,
14, 161–333.
Lauffer, H. (1935). Druck, Energie und Fliesszustand in Gerinnen mit grossem Gefälle [Pressure,
energy and condition of flow in channels of high bottom slope]. Wasserkraft und
Wasserwirtschaft, 30(7), 78–82 (in German).
Le Méhauté, B. (1976). An introduction to hydrodynamics and water waves. New York: Springer.
Liggett, J. A. (1994). Fluid mechanics. New York: McGraw-Hill.
Montes, J. S. (1998). Hydraulics of open channel flow. Reston VA: ASCE Press.
Pudasaini, S. P., & Hutter, K. (2007). Avalanche dynamics. Berlin: Springer.
Rodi, W. (1980). Turbulence models and their application in hydraulics: A state-of-the-art review.
Dordrecht: IAHR.
Rouse, H. (1959). Advanced mechanics of fluids. New York: Wiley and Sons.
Savage, S. B., & Hutter, K. (1989). The motion of a finite mass of granular material down a rough
incline. Journal of Fluid Mechanics, 199, 177–215.
Savage, S. B., & Hutter, K. (1991). The dynamics of avalanches of granular materials from
initiation to runout, I. Analysis. Acta Mechanica, 86(1–4), 201–223.
Schlichting, H., & Gersten, K. (2000). Boundary layer theory. Berlin, Germany: Springer.
Serre, F. (1953). Contribution à l’étude des écoulements permanents et variables dans les canaux
[Contribution to the study of steady and unsteady channel flows]. La Houille Blanche, 8(6–7),
374–388; 8(12), 830–887 (in French).
Stansby, P. K., & Zhou, J. G. (1998). Shallow-water flow solver with non-hydrostatic pressure: 2D
vertical plane problems. International Journal for Numerical Methods in Fluids, 28(3),
541–563.
Steffler, P. M., & Jin, Y. C. (1993). Depth-averaged and moment equations for moderately shallow
free surface flow. Journal of Hydraulic Research, 31(1), 5–17.
Strelkoff, T. (1969). One dimensional equations of open channel flow. Journal of the Hydraulics
Division, ASCE, 95(3), 861–876.
Sturm, T. W. (2001). Open channel hydraulics. New York: McGraw-Hill.
Toro, E. F. (2001). Shock-capturing methods for free-surface shallow flows. New York: Wiley.
Vreugdenhil, C. B. (1994). Numerical methods for shallow water flow. Dordrecht NL: Kluwer.
References 49
White, F. M. (1991). Viscous fluid flow. New York: McGraw-Hill.

White, F. M. (2009). Fluid mechanics. New York: McGraw-Hill.
Wu, W. (2008). Computational river dynamics. London, U.K.: Taylor and Francis.
Yen, B. C. (1973). Open-channel flow equations revisited. Journal of the Engineering Mechanics
Division, ASCE, 99(EM5), 979–1009.
Yen, B. C. (1975). Further study on open-channel flow equations. Sonder-Forschungsbereich, 80/
T/49, Universität Karlsruhe, Germany.
Chapter 2
Energy and Momentum Principles
2.1 Introduction
From Chap. 1, the cross-sectional averaged equations based on the three-

dimensional (3D) conservation laws of mass and momentum are (Montes 1998;
Jain 2001)
@A @Q
þ ¼ 0; ð2:1Þ
@t @x
1 @Q @M
þ ¼ A So Sf : ð2:2Þ
g @t @x
Here, Q is the discharge, A cross-sectional area, g the gravity acceleration, So the

bottom slope, Sf the friction slope, x the streamwise horizontal coordinate, t the
time, and M the momentum function. For a flow of uniform velocity and hydrostatic
pressure distributions, M reduces to (Jaeger 1956; Montes 1998)
Z 2
u p Q2
M¼ þ dA þ A
h: ð2:3Þ
g qg gA
A
Here, u is the velocity component in the x-direction, p fluid pressure, and h the
depth below the free surface of the centroid of area A. Equation (2.2) is only valid
for prismatic channels. Equations (2.1)–(2.2) are differential equations even for the
simple case of steady gradually varied flow. The conservation laws can be
expressed in integral form if an integration in the flow direction x is accomplished
by considering a control volume of finite length. However, not all hydraulic
problems are solved by the isolated consideration of mass and momentum con-
servations. Energy conservation in integral form needs to be added. Its use is

https://doi.org/10.1007/978-3-030-13073-2_2
52 2 Energy and Momentum Principles
alternative, or complementary, to the application of the momentum principle. In this

chapter, the one-dimensional (1D) integral forms of the mass, momentum and
energy conservation principles are presented and used to solve fundamental steady
open channel flow problems.
For steady flow, Eqs. (2.1)–(2.2) transform to
dQ
¼ 0; ð2:4Þ
dx
dM
¼ A So Sf : ð2:5Þ
dx
The differential form of the energy equation must be considered now. The total
energy head H is defined based on the total energy flow across section A as (Jaeger
1956)
Z
1 u2 þ v 2 þ w 2 p
H¼ þ þ z udA: ð2:6Þ
UA 2g qg
A
Herein, the kinetic energy due to the velocity components (u, v, w), the potential
energy due to elevation z, and the work done by fluid pressure p are accounted for
(Liggett 1994; Montes 1998). Consider gradually varied flow in a channel of small
bottom slope, and assume that the velocity components v = w = 0. Further, assume
that u is uniform across cross section A and equal to the mean value U = Q/A, and
that the vertical pressure distribution follows the hydrostatic law. Under these
simplifications, Eq. (2.6) yields with zb = zb(x) as the bottom elevation relative to
an arbitrary datum (Jaeger 1956; Montes 1998)
Z 2
1 u þ v2 þ w2 p U2 Q2
H¼ þ þ z udA zb þ h þ ¼ zb þ h þ :
UA 2g qg 2g 2gA2
A
ð2:7Þ
This relation is a true energy equation and shall not be mistakenly linked to
Bernoulli’s equation resulting from the integration of the momentum equation
along the streamlines in inviscid and incompressible steady fluid flow (Rouse 1938;
Liggett 1994; Montes 1998). For steady flow with constant discharge Q, the dif-
ferential energy equation obtained from the first law of thermodynamics states that
the variation of H equals the rate of conversion of energy into heat (Liggett 1994;
Montes 1998) (Fig. 2.1)
2.1 Introduction 53
Fig. 2.1 Definition sketch for differential form of energy principle under steady flow
dH
¼ Se ; ð2:8Þ
dx
where Se is the gradient of dissipated energy or the energy head slope. Note that
conceptually Se is different from Sf; the former is a measure of the internal energy
dissipation within a volume of fluid, whereas Sf represents the external shear forces
acting on the boundaries of the control volume.
Equations (2.4), (2.5), and (2.8) are the differential forms of the mass,
momentum, and energy balances for hydrostatic pressure and uniform velocity
distributions (Yen 1973); their integration along a control volume of length (x1–x2)
(Fig. 2.2) yields (Jain 2001)
Q1 Q2 ¼ 0; ð2:9Þ
Zx1

M1 M2 ¼ A So Sf dx; ð2:10Þ
x2
Zx1
H1 H2 ¼ Se dx: ð2:11Þ
x2
Fig. 2.2 Control volume for steady flow in a prismatic channel of arbitrary cross section
These equations are statements to be satisfied for any flow within a control volume.
Non-hydrostatic pressure and non-uniform velocity distributions could be
accounted for in H and M (Jaeger 1956; Montes 1998; Jain 2001). The boundary
sections are usually positioned in a gradually varied flow zone, allowing for a
simple evaluation of H and M as functions of the local flow depth. Given the
absence of mass sources/sinks (e.g., rain/infiltration), Eq. (2.9) states constant
discharge across the control volume. However, neither M nor H is conserved within
the control volume given the terms related to Sf and Se, acting as sinks of
momentum and energy. Even though this is the general situation, there are many
cases where these terms can be neglected. For example, in a hydraulic jump on a
horizontal bottom, the shear forces are smaller than the pressure forces plus
momentum flux, resulting in M = const. from Eq. (2.10). The energy dissipation is
significant and cannot be overlooked, however. This is an example on how energy
and momentum laws are used as complementary principles: The momentum
equation is used to compute the flow depth h2 at the downstream section of the
jump for known values of Q and h1, whereas Eq. (2.11) is applied to evaluate the
energy dissipation produced by the hydraulic jump.
2.2 Energy Principle
2.2.1 Specific Energy Head
The concept of specific energy was introduced by Bakhmeteff (1912, 1932); it is

one of the most powerful steady open channel flow tools. Consider a fluid flow
without energy dissipation (Se = 0). Thus, Eq. (2.11) yields
Q2
H 1 ¼ H 2 ¼ H ¼ zb þ h þ ¼ const: ð2:12Þ
2gA2
The specific energy head E is defined taking the channel bottom as reference. Thus,
Q2
E ¼ hþ : ð2:13Þ
2gA2
It represents therefore the total energy head above the channel bottom at any
section. Inserting Eq. (2.13) into Eq. (2.12) produces
zb ð xÞ þ Eð xÞ ¼ const: ð2:14Þ
This is the equation governing ideal fluid flow across channel transitions, to be used
below. It is a suitable approximation in gradual transitions involving a smooth
2.2 Energy Principle 55
variation of bottom elevation and channel shape, without zones of flow separation
where energy losses have to be accounted for (Rouse 1938; Montes 1998). What
are the properties of Eq. (2.13)? Consider a rectangular channel of width b, with
A = bh, and unit discharge q = Q/b. Equation (2.13) is rewritten as
U2 q2
E ¼ hþ ¼ hþ : ð2:15Þ
2g 2gh2
Consider the variation of E with h for q = const. Equation (2.15) is re-arranged

as
q2
ðE hÞh2 ¼ ¼ const: ð2:16Þ
2g
This is a cubic in h, plotted in Fig. 2.3. Of the three roots, one is negative, lacking
of physical meaning and therefore discarded. The curve has the asymptotes
(E–h) = 0 and h = 0. The first asymptote is a straight line inclined 45° with respect
to the E-axis. The second asymptote is the E-axis. The three roots of Eq. (2.16) are
determined analytically for a given value of E (Chanson 2004; Jeppson 2011).
Consider a given value of E. If a vertical line is plotted in Fig. 2.3, the inter-
section with the curve E = E(h) yields two points 1 and 2. For a given discharge, a
certain specific energy head is possible for two different values of the flow depth,
namely h1 and h2. These are called alternate depths (Henderson 1966; Jain 2001).
Note from Fig. 2.3 that at point 1, flow depth h corresponds to the dominant part of
E, whereas the velocity head U2/(2g) is small. At point 2, where the value of E is
identical, the trend is reversed: U2/(2g) is the dominant term, whereas h is small.
This observation suggests that each of the alternate depths represents a different
regime, one of small velocity and large flow depths (hydrostatic forces) and the
other representing a flow of small depth at high velocity. The slow velocity regime
is referred to as “subcritical flow,” whereas the high-speed flow regime involves
“supercritical flow.” The limit between these two states is marked by point c in
Fig. 2.3, referred to as the “critical flow condition,” representing the point for which
the value of E is the minimum possible for a given value q. It is impossible to
Fig. 2.3 Specific energy

head curve for constant unit
discharge
transport the flow rate q with a value of E below Emin, therefore. The flow depth at
critical condition is called critical (subscript c) flow depth hc. Critical flow is an
important tool in open channel hydraulics and includes important properties, the
first introduced in Fig. 2.3, namely minimum specific energy. From Eq. (2.15), a
quadratic relation between the alternate depths exists.
Boris Aleksandrovich Bakhmeteff (written Bakhmetev in Russian) was

born on May 14, 1880, at Tbilisi, and passed away aged 71 years on July 21,
1951, at Brookfield, CT, USA. He graduated in 1903 as a civil engineer from
the Institute of Engineers, St. Petersburg, receiving the Ph.D. degree in 1911.
He then was appointed as Professor of hydraulics and mechanics at the
Polytechnic Institute and in parallel focused on water power as Consulting
Engineer. Successes in his career led in 1917 to his appointment as the first
Russian Ambassador to the USA following the czarist regime, a position held
until 1922. He never returned to his native country but established a practice
as a hydraulic consultant in New York, NY. He became US citizen in 1935.
He was Professor of civil engineering since 1931 at Columbia University,
New York.
Bakhmeteff was an innovator of advanced theories of fluid mechanics and
an authority on open channel hydraulics. His textbooks Hydraulics of Open
Channels, and Mechanics of Turbulent Flow were outstanding and marked
the dawn of hydraulic research in the USA. He specifically was able to
introduce the specific energy and momentum functions and the distinction
between sub- and supercritical flows in open channels based on these con-
cepts. His research on hydraulic jumps was notable. He was awarded the
1937 James Laurie ASCE Prize, the 1944 J. C. Stevens Award, and the 1947
Norman Medal. He was a quiet but generous philanthropist and helped dis-
placed persons to establish themselves in the USA. He was elected Honorary
Member of ASCE in 1945.
Let us quantify critical flow. Differentiation of Eq. (2.15) with respect to

h produces
dE q2
¼ 1 3: ð2:17Þ
dh gh
Setting this expression equal to zero, the extreme of the E-h curve yields the critical
flow condition, namely
q2
1 ¼ 0: ð2:18Þ
gh3c
The value of the critical depth in the rectangular channel is thus

2 1=3
q
hc ¼ : ð2:19Þ
g
Inserting Eq. (2.18) into Eq. (2.15), the minimum specific energy head is

q2 1 q2 3
Emin ¼ hc þ ¼ h c 1 þ ¼ hc : ð2:20Þ
2gh2c 2 gh3c 2
The critical depth equals two-thirds of the minimum specific energy, therefore,
2
hc ¼ Emin : ð2:21Þ
3
The velocity head at the critical flow condition is thus
Uc2 2 U2 Uc2 1 1
Emin ¼ hc þ ¼ Emin þ c ) ¼ Emin ¼ hc : ð2:22Þ
2g 3 2g 2g 3 2
In addition, Eq. (2.18) can be rewritten as
q2 Uc2
¼ ¼ 1; ð2:23Þ
gh3c ghc
or
Uc
F¼ ¼ 1: ð2:24Þ
ðghc Þ1=2
The Froude number F is unity at the critical flow condition, therefore. From
Fig. 2.3, subcritical flow (e.g., point 1) is characterized by U < Uc and thus F\1.
Likewise, for supercritical flow (e.g., point 2) U > Uc and F [ 1. Using the critical
depth hc as scaling length, Eq. (2.15) is rewritten as

E h 1 h 2
¼ þ : ð2:25Þ
hc hc 2 hc
This is the dimensionless specific energy head curve, plotted in Fig. 2.4. It is a universal
curve for rectangular channel flow, involving the critical flow condition E/hc = 3/2.
There is another interesting feature of Eq. (2.15). Let us rewrite it as
q2 ¼ 2gh2 ðE hÞ: ð2:26Þ
Consider the variation of q with h under constant specific energy head E (Fig. 2.5). Note
that q = 0 either if the flow depth is zero or if E = h. The maximum of the curve follows
from differentiation of Eq. (2.26) with respect to h for E = const., yielding
dq
2q ¼ 4ghðE hÞ 2gh2 ¼ 4ghE 6gh2 : ð2:27Þ
dh
For dq/dh = 0 in Eq. (2.27) results critical flow, expressed as
2
hc ¼ E: ð2:28Þ
3
Inserting Eq. (2.28) into Eq. (2.26) produces

2 2 3
2 2 2 21 2
q2max ¼ 2g E E E ¼2 gE E ¼ gE 3 : ð2:29Þ
3 3 3 3 3
Fig. 2.4 Dimensionless

specific energy head curve for
rectangular channel
Fig. 2.5 Discharge curve for

constant specific energy
The maximum discharge is thus

3=2
2 3 1=2
qmax ¼ gE : ð2:30Þ
3
Under the critical flow condition, the discharge is a maximum for the available
value of the specific energy head. This is an important result of great relevance to
establish the head–discharge curve of control structures based on the critical flow
theory (Jaeger 1956; Bos 1976; Ackers et al. 1978; Montes 1998), as shown below.
Another way of expressing Eq. (2.30) is
1=2
qmax ¼ Cd gE 3 ; ð2:31Þ
where Cd is the so-called discharge coefficient

3=2
2
Cd ¼ : ð2:32Þ
3
2.2.2 Location of Critical Flow
Consider the position where critical flow is established. Figure 2.6 shows a sketch for
shallow ideal fluid flow over a dam crest. This flow is discussed in Chap. 4, but here
introduced to illustrate an important case of critical flow, following Henderson (1966).
Fig. 2.6 Critical flow over a

dam spillway
Given that the total energy head H is a constant, Eq. (2.14) is applied. The flow
over the spillway crest changes from upstream subcritical to downstream super-
critical. The detailed computation of the free surface profile is described in Chap. 4.
Differentiation of Eq. (2.14) with respect to x produces
dE dzb dE dh dzb
þ ¼ þ ¼ 0; ð2:33Þ
dx dx dh dx dx
or, using Eq. (2.17),

q2 dh dzb dh dzb
1 3 þ ¼ 1 F2 þ ¼ 0: ð2:34Þ
gh dx dx dx dx
This represents a mathematical identity to be satisfied at any point of the flow

profile. Just upstream of the weir crest dzb/dx > 0, the flow is subcritical ðF\1Þ, so
that dh/dx < 0. Downstream of the crest dzb/dx < 0, the flow is supercritical
ðF [ 1Þ, and dh/dx is again negative, based on Eq. (2.34). At the weir crest, there is
continuity in the free surface slope, which must be therefore negative. Note that at
the weir crest dzb/dx = 0. Thus, Eq. (2.34) is satisfied for the case dh/dx = 0, which
implies a nonzero value of dE/dh, but also for the case F ¼ 1 with dh/dx 6¼ 0
(Henderson 1966). This last option corresponds to the minimum value of the
specific energy head at the weir crest, so that the critical depth is established there.
The flow over the spillway is represented in the E-h plane in Fig. 2.7.
The discharge over the spillway crest is then the maximum possible for the
available energy head, with Cd = (2/3)3/2. In this analysis, it is assumed that energy
dissipation and non-hydrostatic effects due to curvilinear streamlines can be
neglected. Consideration of these flow phenomena would modify the value of Cd
from its ideal value in practice. Energy dissipation would tend to decrease Cd,
whereas the effects of curvilinear motion are to increase it (Castro-Orgaz and Hager
2017).
Critical flow can be demonstrated to occur at the maximum of the bottom profile,
but not at a minimum. Differentiation of Eq. (2.34) with respect to x produces
Fig. 2.7 Critical flow over a

dam spillway represented in
the E-h plane

q2 d2 h 3q2 dh 2 d2 zb
1 3 þ þ 2 ¼ 0: ð2:35Þ
gh dx2 gh4 dx dx
1=2
Inserting F ¼ q=ðgh3 Þ gives

d2 h 3 2 dh 2 d2 zb
1F 2
þ F þ 2 ¼ 0: ð2:36Þ
dx2 h dx dx
Let F ¼ 1 for critical flow, the free surface slope at the critical section is (Hager
2010)
1=2
dh hc d2 z b
¼ : ð2:37Þ
dx c 3 dx2 c
Real solutions to Eq. (2.37) result for a bed curvature at the critical section
d2zb/dx2 < 0; e.g., the bed profile must be convex, as is characteristic for a spillway.
For a concave bottom profile imaginary numbers result, excluding a real solution at
a minimum elevation of the bottom profile.
Consider a rectangular horizontal channel with a gradual variation of channel
width. Examples of this structure are flumes used for discharge measurement
(Montes 1998) (Fig. 2.8).
To apply the 1D approach pursued here, the contraction–expansion of the sec-
tions must be gradual, to avoid flow separation and 3D effects. For an ideal fluid
flow, the energy head remains constant, given by
Q2
E ¼ hþ ¼ const: ð2:38Þ
2gb2 h2
Differentiation of Eq. (2.38) with respect to x yields
dE dh Q2 dh Q2 db
¼ 2 3 3 2 ¼ 0; ð2:39Þ
dx dx gb h dx gb h dx
Fig. 2.8 Critical flow in a

flume with gradual channel
contraction–expansion
or
dh h db
1 F2 F2 ¼ 0: ð2:40Þ
dx b dx
Critical flow F ¼ 1 is therefore formed where db/dx = 0 from Eq. (2.40). An

important complementary property is that critical flow is established at the section
of minimum width in a channel contraction, but not at a maximum width section in
an expansion. To demonstrate this feature, let us differentiate Eq. (2.39) with
respect to x, producing

d2 h Q2 dh Q 2 d2 b 3Q2 db 2
1 þ
dx2 gb2 h3 dx gb3 h2 dx2 gb4 h2 dx
ð2:41Þ
4Q2 dh db 3Q2 dh 2
þ 3 3 þ 2 4 ¼ 0;
gb h dx dx gb h dx
or, written as function of F,

d2 h
2 dh h 2 d2 b 3h 2 db 2 4 2 dh db 3 2 dh 2
1 F F þ F þ F þ F ¼ 0:
dx2 dx b dx2 b2 dx b dx dx h dx
ð2:42Þ
Noting that F ¼ 1 is formed where db/dx = 0, Eq. (2.42) simplifies to

hc d2 b 3 dh 2
þ ¼ 0; ð2:43Þ
b dx2 h dx
from which the water surface slope at the critical section is (Hager 2010)
1=2
dh h2c d2 b
¼ : ð2:44Þ
dx c 3b dx2 c
To obtain a physically relevant solution, the channel sidewall curvature at the

critical section, d2b/dx2, must be positive. Critical flow is thus only established at
the flume throat. The typical transcritical flow profile in the Venturi flume is shown
in Fig. 2.9.
The specific energy head curve along the Venturi flume is not unique, as for
spillway flow, due to the streamwise variation of b = b(x). The function is in this
case (Fig. 2.10)
Fig. 2.9 Transcritical flow in

Venturi flume: a free surface
profile h(x) along channel axis
and b plan view with b = b(x)
Fig. 2.10 Critical flow in a

Venturi flume represented in
the E-h plane
Q2 qð x Þ 2
E ðx; hÞ ¼ h þ ¼ hþ : ð2:45Þ
2gh2 bð xÞ2 2gh2
The effect induced by the variation of b with x is better revealed by considering the
unit discharge q = q(x) = Q/b(x). An increase of q produces an increase of the
critical depth and thus a displacement of the E-h function (Fig. 2.10). For a constant
value of the specific energy head Eo across the Venturi flume, critical flow takes
place at the section of minimum width, whereas points 1 and 2 in the curve
corresponding to width B are the up- and downstream sub- and supercritical flow
depths, respectively (Fig. 2.10).
The two basic mechanisms to produce critical flow, namely the (1) contraction of
the channel section and (2) elevation of the bottom profile, can be combined
simultaneously in a transition structure. An example is the inlet to a tunnel diversion
(Fig. 2.11). The inlet structure is used to covey the flow of a river across the
diversion tunnel, while a dam is under construction (Vischer and Hager 1998). The
transition between the river and the circular tunnel can be designed using a rect-
angular channel of gradual decrease in the width and increase in the bottom slope,
to accelerate the river flow from sub- to supercritical tunnel flow, thereby avoiding
backwater effects. Another example is the non-prismatic spillway channel used in
small earth dams (Chow 1959; Castro-Orgaz et al. 2008). The variation of bottom
slope combined with channel width moves the critical flow section away from the
Fig. 2.11 Channel transition

involving variations of width
and elevation: sketch of inlet
to diversion tunnel (Vischer
and Hager 1998)
spillway crest to an undetermined position within the channel structure. To deter-

mine the position of critical flow in these transition structures, the total energy head,
e.g.,
Q2
H ¼ z b ð xÞ þ h þ ¼ const: ð2:46Þ
2gbð xÞ2 h2
is differentiated with respect to x, resulting in
dH dzb dh Q2 dh Q2 db
¼ þ 2 3 3 2 ¼ 0: ð2:47Þ
dx dx dx gb h dx gb h dx
1=2
Rewriting it as function of the Froude number F ¼ Q=ðgb2 h3 Þ produces
dzb dh h db
þ 1 F2 F2 ¼ 0: ð2:48Þ
dx dx b dx
For critical flow F ¼ 1, the location of the critical section must satisfy the relation
dzb hc db
¼ 0: ð2:49Þ
dx b dx
The definition of critical depth is from Eq. (2.24)

1=3
Q2
hc ¼ : ð2:50Þ
gb2
Inserting into Eq. (2.49) yields
dzb Q2=3 db
1=3 5=3 ¼ 0: ð2:51Þ
dx g b dx
Accounting for the functions zb = zb(x) and b = b(x) in Eq. (2.51) gives a mathe-
matical identity that allows for computing the position of the critical section xc in a
rectangular transition structure under ideal fluid flow.
To illustrate the application of Eq. (2.51), consider an example of Jain (2001)
(Fig. 2.12). A rectangular transition structure ends at a free overfall, involving a
linear width reduction and a parabolic weir profile, given by the equations (di-
mensions in m)
x x x
zb ¼ 2 ; b¼6 : ð2:52Þ
15 15 10
The discharge is Q = 30 m3/s and the length of the structure 30 m.

Computing the derivatives
dzb 2 x db 1
¼ 1 ; ¼ ; ð2:53Þ
dx 15 15 dx 10
and inserting into Eq. (2.51) yield the implicit equation
2 xc 1 Q2=3 xc 5=3
f ð xc Þ ¼ 1 þ 6 ¼ 0: ð2:54Þ
15 15 10 g1=3 10
Solving it by trial-and-error yields xc 20 m, whereas the position of maximum

elevation is x = 15 m. Therefore, the channel contraction effect shifts the critical
flow section downstream of the weir crest. A transition structure would keep the
Fig. 2.12 Transition

structure with linear reduction
of channel width and
parabolic bottom profile
(adapted from Jain 2001)
critical flow section located at the weir crest (dzb/dx = 0) only if the throat (db/
dx = 0) is located there too. This is a practical design approach used for weirs and
flumes for discharge measurement (Montes 1998). In practice, energy dissipation
needs to be accounted for (Castro-Orgaz et al. 2008), as well as the inclusion of
arbitrary cross sections.
2.2.3 Specific Energy Head for General Sections
For any cross section, the specific energy head is
Q2
E ¼ hþ : ð2:55Þ
2gA2
Its representation in the E-h plane yields a curve similar to that of Fig. 2.3. The
critical flow condition is in this case
dE Q2 dA
¼1 3 ¼ 0; ð2:56Þ
dh gA dh
or, with dA/dh = B as the free surface width,
Q2
B ¼ 1: ð2:57Þ
gA3
In general, Eq. (2.57) must be solved numerically to compute the critical depth hc
for a given discharge Q and cross-sectional geometry. Inserting Eq. (2.57) into
Eq. (2.55), the minimum (subscript min) specific energy head is
1 Ac
Emin ¼ hc þ : ð2:58Þ
2 Bc
Equations (2.55)–(2.58) are applied to non-rectangular cross sections with gradu-

ally varied flow. However, there is a particular case where these are inaccurate, due
to the assumption of uniform velocity across the cross section, namely for com-
pound channel flow. The specific energy head then is (Sturm 2001)
Q2
E ¼ hþa ; ð2:59Þ
2gA2
where a is the Coriolis velocity correction coefficient, given for 1D gradually varied
flow by the expression (Jaeger 1956)
R R 3
ðu2 þ v2 þ w2 ÞudA u dA
a¼ A
3
A 3 : ð2:60Þ
U A U A
For gradually varied flows, a is of the order of 1.1–1.15 (Chow 1959; Chaudhry
2008), and its consideration is seldom a critical issue (Liggett 1994). However, in
compound channel flows, it can be large, and more importantly, it forces a notable
modification of the behavior of the E = E(h) curve, with two points of minimum
specific energy head (Sturm 2001).
2.2.4 Critical Flow and Wave Motion
Consider a static mass of water of depth h on which a perturbation is produced, say,

by the impact of a rigid body, i.e., a stone (Fig. 2.13a). It produces a series of
circular waves that propagate in the radial direction with celerity c (Fig. 2.13b, c).
Assume that the stone acts as a continuous source of perturbations, not decaying in
time.
The circular wave pattern generated at point “O” after time t has advanced a
distance ct in the radial direction (Fig. 2.13c). If the wave amplitude is small, the
celerity c is deduced by applying the energy conservation principle, because the
energy loss linked to the traveling wave front can be neglected for a small
amplitude wave. For finite amplitude waves, the momentum balance shall be used,
given that the energy loss may be significant, as detailed in Chap. 5. Consider a
small amplitude 1D wave traveling over still water (Fig. 2.14a). This unsteady
motion is reduced to a steady flow by considering an observer traveling with wave
speed (Chow 1959). The steady flow observed in this moving frame is depicted in
Fig. 2.14b. It is obtained by superimposing the wave celerity (–c) to the unsteady
flow. The continuity equation then simply yields
q ¼ ch: ð2:61Þ
Energy conservation for the equivalent steady flow produces (Chow 1959;
Henderson 1966)
c 2 h2 c2
H ¼ h þ Dh þ 2
¼ hþ : ð2:62Þ
2gðh þ DhÞ 2g
Solving Eq. (2.62) for c and assuming that Dh h, given the small wave height,
yields
Fig. 2.13 Perturbations

generated on still water:
a stone falling, b wave
generation, c circular wave
pattern
Fig. 2.14 Small gravity

perturbation traveling on still
water: a unsteady flow motion
and b equivalent steady flow
seen by observer traveling
with the wave
" #1=2
2gðh þ DhÞ2
c¼ ðghÞ1=2 : ð2:63Þ
2h þ Dh
The quantity c = (gh)1/2 is the propagation celerity of small gravity waves over still
water. Consider now that the fluid is moving at a constant velocity U (Fig. 2.15a).
This fluid flow produces a displacement of the circular wave pattern generated on
still water. In this case, c is still the wave celerity, i.e., its velocity of displacement
relative to fluid flow. Consider first the case U < c. The absolute displacement
velocity of each wave front is
dx
¼ U c: ð2:64Þ
dt
Given that the wave spreads to both the left and the right, and that c > U, there is
one wave front traveling to the left at speed c–U and another to the right at
U + c (Fig. 2.15b). In still water, the wave patterns are concentric circular fronts
(Fig. 2.16a), but for U < c, there is a displacement of circular waves to the right,
with a wave front propagating to the left (Fig. 2.16b).
Manipulation of Eq. (2.64) yields

dx U
¼Uc¼c 1 ¼ cðF 1Þ; ð2:65Þ
dt c
Fig. 2.15 Perturbations

generated on a water flow
slower than celerity c of the
gravity waves: a stone falling
and b wave generation
Fig. 2.16 Wave patterns in a still water, b subcritical flow, c critical flow, d supercritical flow;
adapted from Chow (1959)
where F is the Froude number, defined as
U U Fluid velocity
F¼ ¼ ¼ : ð2:66Þ
c ðghÞ1=2 Celerity of small gravity wave
Francis (Frank) Martin Henderson was born on December 28, 1921, at

Christchurch, NZ, and passed away on August 25, 2006, at Port Stephens,
NSW, AU. He graduated in 1943 as a civil engineer from the University of
Canterbury, NZ. After war service, he gained the M.Sc. degree from Victoria
University, Wellington, NZ. He then joined in 1952 its staff at the School of
Engineering, where he was particularly involved in the development of the
first axial flow jet boat. After two sabbaticals at the University of Michigan in
1956, and University of Cambridge in 1964, he was appointed as Professor of
hydraulics and Deputy Head of his alma mater. In 1968, he took over the
position of Head of the Civil Engineering Department, University of
Newcastle, NSW. During the following 15 years until retirement, he served
as Dean of Engineering and spent sabbaticals at the University of London in
1974 and at the University of Alberta, Edmonton, CA, in 1977. From 1983,
he was a consultant, with engineering projects in Australia, Southeast Asia,
and New Zealand.
The National Water Committee of the Institution of Engineers, Australia,
decided in 1998 to mark the lifelong achievements of Emeritus Professor
Henderson: During the Henderson Oration, his educational, professional, and
scientific achievements were awarded. The highlights of his professional
career are therein described, including model tests resulting in a shock wave
reduction, vortex generation at intakes, and computer works to investigate
fluid transients. Victor Streeter (1909–2015) has had a great influence on
Henderson’s professional knowledge during his visit to University of
Canterbury in 1952 and during Henderson’s stay at the University of
Michigan. The latter stay marked the initiation of two papers on flow over ski
jumps published in the French journal La Houille Blanche in the early 1960s.
Later, Henderson was involved in the preparation of his famous book Open
Channel Flow, which made him known among hydraulicians and may be
considered his legacy in hydraulics.
Physically, c is the velocity of information transmission relating to the variation of

the equilibrium conditions. The transmission of information by small gravity waves
is the mechanism used by free surface flow to adjust from non-equilibrium to a new
equilibrium state (steady flow) (Puertas and Sánchez 2001).
U < c implies that F\1. This flow is called subcritical, and physically, small
gravity perturbations travel both in the up- and in the downstream directions. The
case U = c results in F ¼ 1, that is, the flow is critical, and the wave pattern as
plotted in Fig. 2.16c. An additional feature of critical flow regained here is that the
wave celerity traveling against the fluid flow is zero. At the point of perturbation
origin (point “O” in Fig. 2.16c), the waves moving upstream are blocked and
cannot propagate to the left. Waves can only propagate in the downstream direction
with the absolute velocity dx/dt = 2c.
Fig. 2.17 Angle of shock

front in supercritical flow
U > c implies F [ 1. This flow is called supercritical; small gravity perturba-

tions travel only in the downstream direction. Remember that c is the celerity of
shallow water waves only if the wave amplitude is small as compared with the
water depth. Therefore, a small gravity wave cannot propagate against flow
direction in supercritical flow. However, as will be shown in Chap. 5, a finite
amplitude gravity wave can propagate in the upstream direction of supercritical
flow if it is large enough. The circular wave patterns are displaced along the
longitudinal axis a quantity Ut, and there is an envelope of all circular fronts, which
is itself a wave front or a supercritical shock (Fig. 2.16d). Its angle with the fluid
flow direction is from simple geometry (Fig. 2.17)
ct 1
sin b ¼ ¼ : ð2:67Þ
Ut F
The physical difference between sub- and supercritical flows is illustrated using
an example of Puertas and Sánchez (2001). Consider an obstacle (stone) placed in a
steady and uniform stream. If it is inserted into an initially uniform subcritical
stream (U < c), a gradual water-level variation is produced (Fig. 2.18a) (unsteady
and rapidly varied flow effects are overlooked for the sake of simplicity). The
gradual elevation of the water level is transmitted in the upstream direction as a
smooth wave front. The water level will increase in response to the upstream
discharge received, until the flow starts to spill over the stone, evolving in the long
term to a new steady state (equilibrium condition). Note from this example that the
obstacle influences the conditions of the upstream flow. The obstacle presence is
“informed to the approach flow by small gravity waves.” This information trans-
mission by gravity waves “informs” the upstream flow that there is an obstacle
downstream, so that the flow is able to adapt its condition to that situation.
If the stone is placed into an initially uniform supercritical stream
(U > c) (Fig. 2.18b), there is no possibility of transmitting the information in the
upstream direction, so that the approach flow is not informed about the obstacle
presence. The water then impacts the obstacle and spills over it.1 The present
analysis is qualitative, given that it is in reality an unsteady flow, but highlights the
1
In this introductory example, the possibility of having a hydraulic jump is overlooked. The
hydraulic jump is the way water changes from super- to subcritical flow, to be detailed in the next
section. It is basically a finite amplitude perturbation in steady flow.
Fig. 2.18 Transmission of

information by small gravity
waves in a subcritical flow
and b supercritical flow;
adapted from Puertas and
Sánchez (2001)
basic difference between sub- and supercritical flows: Subcritical flows are affected
by perturbations produced downstream of a section, whereas supercritical flows are
not affected. Any variation of flow depth or discharge in a channel produces small
gravity wave fronts transmitting information by these perturbations, producing a
transition to a new equilibrium condition.
2.2.5 Computation of Critical Flow in Arbitrary Sections

General
The critical depth for an arbitrary cross-sectional shape follows by solving the
equation
Q2
BðhÞ ¼ 1; ð2:68Þ
gAðhÞ3
or
AðhÞ3
f ð hÞ ¼ Q 2 g 0: ð2:69Þ
BðhÞ
The problem is, for a given channel shape and discharge Q, to find the value of the
flow depth by which f(h) = 0 in Eq. (2.69). This computation must be generally
conducted using numerical methods. Of the various root-finding methods available
in the literature, the Newton–Raphson method is applied (Hoffman 2001; Jeppson
2011). Consider a generic function f = f(h) sketched in Fig. 2.19. To conduct
iterations of Eq. (2.69), a previous trial value of h is required. Consider the known
value of h at iteration “k”, for which f(hk) is nonzero. The tangent to f(h) at k is
Fig. 2.19 Illustration of

Newton–Raphson method
k
df fk
f 0 hk ¼ ¼ : ð2:70Þ
dh hk hk þ 1
A better approximation to the root of Eq. (2.69) is therefore given by the inter-
section of the tangent line with the h-axis, e.g.,
fk
hk þ 1 ¼ hk : ð2:71Þ
ðdf =dhÞk
Equation (2.71) is known as Newton–Raphson iterative formula, allowing for an

ordered convergence to the root. Limitations are that the function f must admit to
form the derivative df/dh and that the initial value of h to start iterations must be a
suitable estimate of the root, to ensure convergence (see Hoffman 2001 for details).
The derivative needed in Eq. (2.71) is

df A3 dB
¼ g 3A2 2 : ð2:72Þ
dh B dh
A convergence criterion to the iteration algorithm based on Eq. (2.71) is

kþ1
f f k

fk e; ð2:73Þ
that is, if the new iteration value yields an error below a specific tolerance e, usually
taken as 10–6, the actual numerical value is of acceptable quality and therefore taken
as the numerical solution for the critical depth.
Trapezoidal cross section
For a trapezoidal section (Fig. 2.20), the flow area A and free surface width B are
Fig. 2.20 Relative critical

depth zhc/b and minimum
specific energy head
zEmin/b versus relative
discharge [z3/(gb5)]1/2Q in
trapezoidal channel
A ¼ bh þ zh2 ;
ð2:74Þ
B ¼ b þ 2zh:
A suitable initial value to start iterations is the critical depth of the rectangular
section of the trapezoidal bottom width b
1=3
Q2
hc ¼ : ð2:75Þ
gb2
A code with this scheme implemented is on the side file “criticaldepth_NR.xls”, in

Chap. 12. For a channel of base width b = 1 m, side slope z = 2 and Q = 5 m3/s,
the program yields hc = 0.8346 m after six iterations. The critical depth and min-
imum specific energy of the trapezoidal channel are shown in dimensionless form in
Fig. 2.20 (Henderson 1966). These functions are obtained as follows. Inserting
Eq. (2.74) into Eq. (2.69) yields
Q2
3 ðb þ 2zhc Þ ¼ 1; ð2:76Þ
g bhc þ zh2c
or, as plotted in Fig. 2.20

3
z3 Q2 1 þ zhbc zhbc
¼ : ð2:77Þ
gb5 1 þ 2 zhbc
The minimum specific energy head is thus
Q2
Emin ¼ hc þ 2 ; ð2:78Þ
2g bhc þ zh2c
or
z3 Q2
zEmin zhc gb5
¼ þ
2 : ð2:79Þ
b b 2 1 þ zhbc zhbc
To determine the head–discharge curve Q = Q(Emin), computations are explicit: A

value of hc is selected, from which the discharge Q is obtained from Eq. (2.76) and
the specific energy head from Eq. (2.78). An important application of the critical
flow theory is to determine the head–discharge relation of control structures gen-
erating critical flow, as the Venturi flume. Assuming that energy losses are negli-
gible, the total head H at the inlet is equal to Emin, with the critical depth established
at the throat, where b = bmin. The theory is compared with the data of Blau (1960)
in Fig. 2.21, for a Venturi flume of base width b = 0.0665 m and z = 0.299,
indicating a fair agreement. Deviations between the theory and experiments with
increasing discharge are due to non-hydrostatic effects at the critical flow section
(Montes 1998; Hager 2010).
Circular cross section
The basic flow variable needed to compute critical flow in a circular cross section of
diameter D is the central angle h
Fig. 2.21 Head–discharge

relation of Venturi flume:
a plan view and b relation
Q(H)

1 hc
h ¼ 2cos 12 : ð2:80Þ
D
The free surface width B and the flow area A are then given by (Jain 2001; Sturm
2001)

h
B ¼ D sin ;
2
ð2:81Þ
D2
A¼ ðh sin hÞ:
8
Q2 1 ðh sin hÞ3
¼ : ð2:82Þ
gD5 512 sinðh=2Þ
The minimum specific energy head follows by substituting Eq. (2.81) into
Eq. (2.58), resulting in
Emin hc 1 ðh sin hÞ
¼ þ : ð2:83Þ
D D 16 sinðh=2Þ
The computation of the head–discharge relation is simple. First, a value of hc is

selected, from which h is computed from Eq. (2.80). The value of Q2/(gD5) is then
given by Eq. (2.82) and the value of Emin/D by Eq. (2.83). These relations are
plotted in Fig. 2.22.
The theory is applied to compute the head–discharge relation of a circular culvert
with critical depth at its inlet (Henderson 1966). Without energy losses at the pipe
inlet, the specific energy head Emin equals the upstream water depth H, as long as
the approach flow velocity head is small. Computing the [Q2/(gD5); H/D] data pairs
along the interval of the dimensionless discharge investigated by Toman et al.
(2014) results in a fair agreement between the critical flow theory and the experi-
mental data (Fig. 2.23).
To compute the critical depth of a circular section of diameter D for a discharge
Q, however, numerical computations are required to solve Eq. (2.68) iteratively.
A code based on the Newton–Raphson scheme is implemented in Chap. 12 in the
file “criticaldepth_NR_pipe.xls”. For a channel of D = 1 m and Q = 1 m3/s, the
program yields hc = 0.573 m after three iterations.
Fig. 2.22 Critical depth

hc/D and minimum specific
energy head Emin/D versus
Q/(gD5)1/2 in circular pipe
Fig. 2.23 Head–discharge

relation in a circular culvert
with critical flow at the inlet
2.2.6 Channel Transitions by Change of Bed Elevation
Sub-subcritical flow
Consider the free surface profile in a channel transition consisting of a smooth
hump, to avoid separation losses, located in a prismatic rectangular channel in
which an ideal uniform fluid flow of flow depths up- and downstream of the hump
is established. First, the profiles will be discussed qualitatively using the specific
energy head diagram, and then computations are detailed. Consider ideal fluid flow
toward the hump under subcritical flow with q as the unit discharge, H the total
energy head, and Z the maximum hump elevation (Fig. 2.24). From Eq. (2.14) for
ideal and steady flow,
Fig. 2.24 Subcritical flow over a hump: a geometry and b specific energy head diagram
H ¼ E1 ¼ E2 þ Z ¼ E3 : ð2:84Þ
Therefore, the specific energy head at the hump crest (point 2) is
E2 ¼ H Z: ð2:85Þ
In this example, it was assumed, as observed on the specific energy head diagram,
that E2 > Emin. The flow depth at point 2 is therefore above hc, and the water
surface forms a depression as it crosses the hump (Figs. 2.24 and 2.25).
Fig. 2.25 Experiment with subcritical flow over a hump [taken from movie Fluid Motion in a
Gravitational Field, by Rouse (1961), IIHR-Hydroscience & Engineering, the University of Iowa]
Super-supercritical flow
If supercritical flow approaches the obstacle with identical values of H and q as for
F\1, a representation of the flow conditions in the specific energy head diagram
shows that the flow depth over the obstacle increases (Figs. 2.26 and 2.27), again
for E2 > Emin.
The value of Z resulting from equaling E2 and Emin is the maximum possible to
pass the flow rate q with the specific energy head H at the inlet of the hump. It is,
therefore,
Zmax ¼ H Emin : ð2:86Þ
Fig. 2.26 Supercritical free surface profile over a hump: a geometry and b representation on the
specific energy head diagram
Fig. 2.27 Experiment with supercritical flow over a hump [taken from movie Fluid Motion in a
Fig. 2.28 Sub- and

supercritical free surface
profiles over a hump with
critical flow at the crest:
a geometry and b specific
energy head diagram
For this value of Z, the crest flow depth equals the critical depth both for the sub-
and for the supercritical profiles (Fig. 2.28). The computation of the flow profiles is
now detailed.
As the water flows over the obstacle, each section is subject to a different value
of specific energy head imposed by the upstream condition, namely E(x) = H–zb(x).
The problem is to compute the alternate depth corresponding to the flow regime
selected (sub- or supercritical) for the value of E at section x. Equation (2.15) is
written as
3
h E h 2 1
þ ¼ 0: ð2:87Þ
hc hc hc 2
Equation (2.87) can be rewritten as (Selby 1973)
v3 þ av þ b ¼ 0; ð2:88Þ
where
" #
h 1E 1 E 2 1 E 3 27
v¼ ; a¼ ; b¼ 2 þ : ð2:89Þ
hc 3 hc 3 hc 27 hc 2
The two positive solutions to Eq. (2.88) are (Selby 1973; Chanson 2004; Jeppson
2011)

h E 1 2 C
¼ þ cos ; ð2:90Þ
hc hc 3 3 3
for subcritical flow, and

h E 1 2 C 4p
¼ þ cos þ ; ð2:91Þ
hc hc 3 3 3 3
for supercritical flow, where

27 E 3
cos C ¼ 1 : ð2:92Þ
4 hc
The free surface profile can thus be analytically computed for given zb = zb(x), q,
and H.
For non-rectangular sections, the alternate depths must be determined numeri-
cally. A numerical computation method is presented that is applied to these cases.
The algorithm for the rectangular section is presented, and the numerical solution is
compared with the analytical result. Let the specific energy head at section x be
Eð xÞ ¼ H zb ð xÞ: ð2:93Þ
To determine the flow depth, a root search procedure is applied to the function
q2
f ð xÞ ¼ h þ E ð xÞ: ð2:94Þ
2gh2
The Newton–Raphson iterative formula reads
f ð xÞ k
hð x Þ k þ 1 ¼ h ð x Þ k ; ð2:95Þ
½df ð xÞ=dhk
where the derivative of f is
df ð xÞ q2
¼ 1 3: ð2:96Þ
dh gh
To start iterations, a suitable initial value of h(x) is needed, corresponding to the

same regime of the root. For subcritical flow, the starting value is obtained by
assuming that the kinetic energy head can be neglected in the specific energy
equation, whereas the specific energy head equals the kinetic energy head for
supercritical flow, e.g.,
8
< E ð xÞ;
2
if 1 gh
q
3 \1
hð xÞk¼0 ¼ h q2 i1=2 ð2:97Þ
: ; if
2
1 gh
q
3 [ 1:
2gEð xÞ
Consider as test case a Gaussian hump of profile zb (m) = 0.2exp[–0.5{x (m)/

0.24}2]. A subcritical flow profile for q = 0.2 m2/s and H = 0.45 m was computed
analytically and numerically, with the results compared in Fig. 2.29. The codes
used for the simulations are available in the files “Analytical_hump_subsuper.xls”
and “Hump_NR.xls” in Chap. 12, respectively. Note that the accuracy of the
numerical method is excellent, supporting its extension to other cross-sectional
channel shapes.
A supercritical flow profile was generated using the same data as for F\1, and
the analytical results are compared with the numerical solution in Fig. 2.30. As
previously, the agreement between both solutions is excellent.
Fig. 2.29 Computation of

subcritical flow profile over a
Gaussian hump
Fig. 2.30 Computation of

supercritical flow profile over
a Gaussian hump
Sub-supercritical flow
An additional possible flow profile involves a change from upstream subcritical
flow to downstream supercritical flow (Fig. 2.31). This profile is obtained by setting
critical flow at the crest and joining the upstream subcritical profile obtained with
Eq. (2.90) with a downstream supercritical profile determined from Eq. (2.91).
Obviously, for this situation to occur, the flow depths up- and downstream from the
obstacle might not be equal. Rather, the two flow depths correspond to the alternate
depths for the specific energy head H = Zmax + Emin (Fig. 2.31).
It is pertinent to mention that for a given upstream subcritical alternate depth h1,
the value of h3 is determined analytically by the alternate depth equation (Fig. 2.31)
q2 q2
H ¼ h1 þ ¼ h 3 þ ; ð2:98Þ
2gh21 2gh23
so that

q2 1 1
h1 h3 ¼ ; ð2:99Þ
2g h23 h21
Fig. 2.31 Transcritical flow profile over a hump: a geometry and b specific energy head diagram
or

q2 h21 h23 q2 ð h1 h3 Þ ð h1 þ h3 Þ
h1 h3 ¼ ¼ : ð2:100Þ
2g h23 h21 2g h23 h21
After simplification, the result is (Subramanya 1986)
q2
h23 h21 ðh1 þ h3 Þ ¼ 0: ð2:101Þ
2g
For given q and h1, Eq. (2.101) is a quadratic equation in h3, which is easily solved.
Consider now how a transcritical flow profile as that of Fig. 2.31 is established
in a channel for given water depths up- and downstream of the obstacle. Consider
upstream subcritical flow (q, H) and a maximum hump elevation Z satisfying
E2 ¼ H Z\Emin : ð2:102Þ
As plotted in the specific energy head diagram (Fig. 2.32), the vertical line labeled
X-X′ does not cut the specific energy head diagram for discharge q. The line
E2 = const. is outside the range of possible physical solutions (Rouse 1938, 1950), so
that the flow rate q cannot be transported over the obstacle with the available specific
energy H E1. The upstream specific energy head must be raised until the specific
energy head on the crest corresponds to the minimum possible Emin for the flow to pass
the obstacle. This is in fact what the water flow does: As the water flow q approaches
the obstacle with the upstream head H, an unsteady flow motion is induced, given the
impossibility to pass the obstacle. The obstacle acts as a perturbation, and this situation
is gradually transmitted to the approach flow by small gravity waves propagating into
the upstream direction at celerity (gh)1/2 relative to the flow. In response to this
feedback, the upstream flow receives the message and begins to increase its upstream
specific energy, establishing a new equilibrium condition. Once the water flows over
the crest at critical flow condition, the upstream energy head is fixed by the new point
1′ as E1′ = Z + Emin > E1. The new head over the obstacle is then E1′. The upstream
specific energy increase from E1 to E1′ is accomplished by a backwater curve of
necessary length to produce an adaptation of flow depths from h1 to h1′ producing the
exact increase needed in the specific energy. This type of curve2 will be described in
Chap. 3, given that frictional effects are relevant.
2
This type of gradually varied flow profile is called M1-type curve. The role of friction is
important, as well as that of the upstream bottom slope So. As will be demonstrated in Chap. 3, a
form equivalent to Eq. (2.8) is dE/dx = So‒Se, with So > Se for the M1 curve, so that E is increased
in the flow direction. A certain length is necessary to increase the specific energy by the quantity
DE needed to pass the obstacle. This specific energy gain is locally lost beyond the obstacle in a
hydraulic jump. The relation describing the energy loss in a hydraulic jump will be presented in the
next section.
Fig. 2.32 Transcritical ideal flow profile over a hump as result of chocked upstream subcritical
flow: a geometry and b specific energy head diagram
Super-subcritical flow
From the crest, the water flows down the tailwater face of the hump under the
supercritical regime without any energy losses, reaching point 3′ downstream of the
hump. The specific energy head at this point is E1′, which is larger than the value of
the specific energy head E1 far downstream of the obstacle. To return to this flow
condition, the water must lose the specific energy head gained by approaching the
hump, namely E1 − E1′, and change from the super- to the subcritical regime. This
is accomplished by a hydraulic jump, the physical phenomenon producing such
flow transition, accompanied by significant energy losses equal to ΔE = E1′ − E1.
The flow profile over the obstacle is 1-1′-c-3′-3, therefore.
After having described the entire free surface profile, it is observed that the
obstacle “forces” the upstream flow to increase its water depths from h1 to h1′. As
this occurs, namely that the upstream flow conditions are affected, the flow is said to
be “chocked.”
Consider an upstream supercritical flow obeying Eq. (2.102). Again, the
upstream flow must increase its specific energy to pass the obstacle. However, from
gradually varied flow computations to be described in Chap. 3, it is not possible to
gain this specific energy in supercritical flow. To increase the specific energy head
and passing the obstacle, the upstream flow must change from super- to subcritical
flow. Obviously, this transition is accompanied by an energy loss to be supplied by
a backwater curve under subcritical flow. The flow over the obstacle is depicted in
Fig. 2.33. The upstream supercritical flow at section 1 changes to subcritical flow at
section 1′, with the energy loss ΔEj across the hydraulic jump. The energy head at
the crest must be the minimum possible to pass the flow, Emin. Thus, the required
Fig. 2.33 Transcritical flow profile over a hump resulting in chocked upstream supercritical flow:
a geometry and b specific energy head diagram
energy just upstream of the obstacle is E1′′ = Z + Emin. Therefore, the flow con-
ditions must change from section 1′ to section 1′′ by a gradually varied flow profile
(curve type S1, Chap. 3) of the length necessary to supply the increase in the
specific energy head needed, ΔE, namely (Fig. 2.32)
DE ¼ Emin þ Z þ DEj E1 : ð2:103Þ
The upstream supercritical flow is therefore chocked. Beyond the crest, the flow
reaches the toe of the hump under supercritical condition up to section 3′, without any
loss of energy. A gradually varied flow profile in the tailwater portion permits the
adaptation from 3′ to 3. The flow profile over the obstacle is therefore 1-1′-1′′-c-3′.
This type of flow profile is produced if a supercritical flow over an obstacle,
shown in Fig. 2.26, suffers a reduction of discharge; the effect is highlighted in
Fig. 2.34 using the specific energy head diagram, with q2 < q1. The upstream flow
depth h1 is kept constant with a gate. Then, the discharge is reduced, say, by
maneuvers with a pump. The conditions at the upstream section change from point
1 to 1′, which obviously has less specific energy head than the original upstream
flow. This specific energy head is insufficient to pass the obstacle, so it shall be
increased. The only way to increase the specific energy head of a supercritical flow
approaching an obstacle is by jumping to subcritical flow and producing a back-
water effect (Fig. 2.33). Rouse (1961) studied this phenomenon by starting with the
Fig. 2.34 Discharge reduction in supercritical flow over a hump under constant upstream flow
depth
Fig. 2.35 Experiment with supercritical flow over a hump chocked by discharge reduction [taken
from movie Fluid Motion in a Gravitational Field, by Rouse (1961), IIHR-Hydroscience &
Engineering, the University of Iowa]
Fig. 2.36 Experiment with transcritical flow over a hump [taken from movie Fluid Motion in a
supercritical profile of Fig. 2.27, then reducing the discharge, and inducing the
propagation of a surge in the upstream direction (Fig. 2.35), which, once reaching
steady state, transforms into a hydraulic jump located in the upstream channel,
leading to a transcritical flow profile over the hump (Fig. 2.36).
2.2.7 Channel Transitions by Change of Channel Width
Consider now the transition along a gradual contraction–expansion in a rectangular

and horizontal channel (Fig. 2.37). The upstream channel width is B, its minimum
width at the throat bc, and the upstream specific energy head Eo. The analytical
solution given by Eqs. (2.90)–(2.92) permits the computation of the flow profiles,
to be described using the specific energy head diagram. If the up- and downstream
flow conditions are subcritical, the flow passage across the contraction–expansion
produces a depression–elevation of the water surface. If Eo > Emin, the critical
depth is not reached at any point along the structure (Fig. 2.37). Note that the
critical depth in a channel of variable width is different at each section, given by
" #1=3
Q2
hc ð x Þ ¼ : ð2:104Þ
gbð xÞ2
However, as previously demonstrated, critical flow can only be established at the

throat. The value of Emin is therefore the minimum specific energy at the throat. Its
Fig. 2.37 Sub- and

supercritical profiles across a
contraction–expansion:
a plan, b geometry, c specific
energy head diagram
value is reached at the throat for the available value of Eo if the throat width is bmin
as
1=3
3 Q2
Eo Emin ¼ ; ð2:105Þ
2 gb2min
or
3=2
2 3 1=2
bmin ¼ Q gEo : ð2:106Þ
3
For the two cases plotted in Fig. 2.37, bc > bmin. If the up- and downstream flow
conditions are supercritical, the flow passage across the contraction–expansion
produces an elevation–depression of the water surface (Fig. 2.37). If the value of Eo
is kept constant and the value of bc is reduced to bmin, the sub- and supercritical
profiles reach critical flow at the throat (Fig. 2.38).
Fig. 2.38 Sub- and

supercritical profiles across a
contraction–expansion of
minimum throat width:
a plan, b geometry, c specific
energy head diagram
Fig. 2.39 Transcritical

profile across a contraction–
expansion of minimum throat
width: a plan, b geometry,
c specific energy head
diagram
An additional possible profile involves the transition from upstream subcritical

to downstream supercritical flow (Fig. 2.39). A further reduction of bc below bmin
produces a transcritical flow profile and chocking of the upstream flow, with a
hydraulic jump at the flume exit, for upstream subcritical flow, and an upstream
hydraulic jump if it were supercritical.
Transcritical flow in a Venturi flume tested by Khafagi (1942) is considered in
Fig. 2.40. The inlet is a converging circular arc transition of radius R = 0.545 m,
1=2
bð xÞ ¼ bc þ 2 R R2 ðx 0:3Þ2 : ð2:107Þ
The origin of the x-coordinate is at contraction start. The contraction is followed by

a constant width throat 0.05 m long and a 1:8 linear expansion to recover the
original flume width, given with bc = 0.12 m by
Fig. 2.40 Transcritical profile in a Venturi flume: a plan and b comparison of test data with theory

1
bð xÞ ¼ bc þ 2 ðx 0:35Þ : ð2:108Þ
8
Consider the measured free surface profile for Q = 0.01 m3/s. The analytical
solution of Eqs. (2.90)–(2.92) was used to generate a theoretical profile, noting that
E = Eo = Emin at all sections, e.g.,
1=3
3 Q2
E ¼ Eo Emin ¼ : ð2:109Þ
2 gb2c
The value of the critical depth at each section is different, however [see
Eq. (2.104)]. That is, upstream from the flume throat, the subcritical flow profile at
section x of the circular arc transition is

hð x Þ Eo 1 2 C
¼ þ cos ; ð2:110Þ
hc ð x Þ hc ð x Þ 3 3 3
where

27 Eo 3
cos C ¼ 1 : ð2:111Þ
4 hc ð x Þ
Along the 0.05-m-long throat, the water depth is constant and equal to the critical
depth for width bc = 0.12 m. Along the linear expansion, the supercritical flow
profile is

hð x Þ Eo 1 2 C 4p
¼ þ cos þ : ð2:112Þ
h c ð x Þ hc ð x Þ 3 3 3 3
A comparison of the analytical solution and the experimental data reveals fair
agreement, although the water depth along the throat is clearly not a constant.
Deviations of experiments and predictions are mainly due to non-hydrostatic effects
(Hager 2010). Note further the presence of a hydraulic jump in the tailwater, which
is not predicted by the present computations based on constant energy head.
The analytical solution of Eqs. (2.90)–(2.92) was used to compute the free
surface profile of the transition structure shown in Fig. 2.12. A subcritical flow
profile was computed upstream of the critical section using Eq. (2.90), whereas the
supercritical profile downstream from the critical section was determined using
Eq. (2.91). It is thereby necessary to evaluate the specific energy head E(x) and the
critical depth hc(x) at each section. The specific energy head along this transition is
different at each section and given by E(x) = H–zb(x). The total energy head is a
constant to be evaluated using the critical flow section. From Eq. (2.54), it is
located at xc 20.026 m using the Newton–Raphson method. Channel width and
bottom elevation at the critical section are thus bc = 3.997 m and zb = 0.887 m
from Eq. (2.52). The total energy head is then
1=3
3 Q2
H ¼ zb ðxc Þ þ Emin ¼ zb ðxc Þ þ
2 gb2c
1=3 ð2:113Þ
3 302
¼ 0:887 þ 3:573 m:
2 9:81 3:997 2
With the known value of the total head, the application of Eqs. (2.90)–(2.91) is
immediate, using Eq. (2.52) to evaluate b(x) and zb(x) (Fig. 2.41).
2.3 Momentum Principle
2.3.1 Specific Momentum
The use of the specific energy concept is useful for flows where the energy losses
can be neglected. This amounts to ensure that there are no flow separation zones. If
however, the energy loss is unknown in advance, an alternative method of analysis
must be devised. This method is based on the use of the momentum principle
(Liggett 1994; Montes 1998). Consider for illustrative purposes the hydraulic jump
Fig. 2.41 Transcritical profile in a flume of parabolic bottom profile with a linear width
contraction (see Fig. 2.12)
(Fig. 2.42a, b). Its internal flow is highly turbulent, with a markedly non-uniform
turbulent velocity profile. The energy loss is unknown, so the energy principle is
unsuited here. Consider steady flow in a horizontal channel where the discharge is
Q and the upstream water depth h1; the problem is to find the tailwater depth h2 of
the hydraulic jump. A control volume to apply the momentum principle is shown in
Fig. 2.42c. Its boundary sections are located outside the zone of rapidly varied flow,
where the pressure is hydrostatic and the velocity distribution close to uniform.
Neglecting the shear forces due to turbulent flow acting on the boundaries of the
control volume, Eq. (2.10) yields
M1 ¼ M2 ; ð2:114Þ
that is, the momentum function M is conserved in the hydraulic jump, with
Q2
M ¼ Ah þ ¼ const: ð2:115Þ
gA
For a rectangular channel of width b, Eq. (2.115) simplifies to
h Q2 h2 q2
M ¼ ðbhÞ þ ¼ b þb ; ð2:116Þ
2 gbh 2 gh
from which arises the definition of specific momentum S (Jaeger 1956; Henderson
1966; Montes 1998)
2.3 Momentum Principle 95
Fig. 2.42 Hydraulic jump: a side view of laboratory test, b top view showing details of
turbulence within the roller, c definition sketch to apply the control volume approach
M h2 q2
S¼ ¼ þ : ð2:117Þ
b 2 gh
The specific momentum S is only a function of flow depth h and unit discharge q, as
the specific energy [see Eq. (2.15)]; it is a quantity conserved in hydraulic jumps.
Equation (2.117) is rewritten as

h2 q2
h S ¼ ¼ const: ð2:118Þ
2 g
This is a cubic in h, plotted in Fig. 2.43. The curve has one asymptote for h = 0,
e.g., the S-axis, and tends to the hydrostatic thrust (1/2)h2 for h ! ∞. As observed,
there is a point of minimum S in Fig. 2.43. This point is determined from the
derivative

dS d h2 q2 q2 q2
¼ þ ¼h 2 ¼h 1 3 ; ð2:119Þ
dh dh 2 gh gh gh
rewritten as
dS dE
¼h ¼ h 1 F2 : ð2:120Þ
dh dh
Using Eq. (2.120), the minimum of the S = S(h) curve is therefore

2 1=3
dS q
¼0 ) F ¼ 1; hc ¼ : ð2:121Þ
dh g
This result indicates another property of critical flow (F ¼ 1), namely minimum
specific momentum. The minimum value of S is thus
h2c q2 3
Smin ¼ þ ¼ h2 : ð2:122Þ
2 ghc 2 c
Consider a vertical line plotted in Fig. 2.43, for which the intersection with the
S = S(h) curve yields points 1 and 2. The flow depth h1 < hc and, therefore, the
regime for this point involves supercritical flow. Likewise, for point 2 results
h2 > hc, implying subcritical flow. Points 1 and 2 are determined based on
S = const. and, thus, represent the boundary flow depths of a hydraulic jump. In
general, the two depths of the specific momentum curve obtained by setting
S = const. are called sequent depths or conjugate depths, and shall not be confused
with the alternate depths originating from the specific energy principle.
Equation (2.117) is written in dimensionless form as
1
S 1 h 2 h
¼ þ : ð2:123Þ
h2c 2 hc hc
This universal function is plotted in Fig. 2.44 with Eq. (2.25) for the dimensionless
specific energy head curve. The two curves together reveal that for a hydraulic
jump, which implies S = const., the value of E is not conserved, resulting in an
energy loss. This issue is of relevance to be explored below. For rectangular
channel flow, similar to what was done for the alternate depths as function of E, the
analytical solution for the sequent depths as function of S is from Eq. (2.118), the
solution of the cubic in h/hc
Fig. 2.43 Specific

momentum curve for constant
unit discharge

specific momentum and
energy head curves in
rectangular channel flow
3
h S h
2 2 þ 2 ¼ 0: ð2:124Þ
hc hc hc
By using a trigonometric transformation, the two positive solutions of Eq. (2.124)

are (Selby 1973; Jeppson 2011)

h 2 S 1=2 K
¼2 cos ; ð2:125Þ
hc 3 h2c 3

h 2 S 1=2 K 4p
¼2 cos þ ; ð2:126Þ
hc 3 h2c 3 3

2 S 3=2
cos K ¼ : ð2:127Þ
3 h2c
The solution for the sequent depths is thus trivially simple for rectangular channel
flow.
The specific energy and momentum concepts are two of the most notable tools
for steady open channel flow analyses. It is widely recognized in the literature that
the specific energy head curve was first developed by Bakhmeteff (1912). The first
diagram is reprinted in Fig. 2.45, corresponding to q = 1 m2/s. Inspecting
Fig. 2.45, it appears evident that the function S = S(h) is plotted superimposed with
the E = E(h) curve. Bakhmeteff denoted S by the letter “h” in his 1912 book.
Fig. 2.45 Original figure of

Bakhmeteff (1912) who
published the first
E-h diagram (pp. 39 and 40).
Note that the momentum
function h(h) is presented
simultaneously
Although it appears trivial, it is not, because the two curves of Fig. 2.45 are the
E(h) and S(h) functions, both having been first developed by Bakhmeteff (1912).
This confirms that the origin of the specific momentum concept is simultaneous to
that of the specific energy. Both should be credited to Bakhmeteff (Castro-Orgaz and
Sturm 2018).
2.3.2 Bélanger’s Equation for the Hydraulic Jump
Consider now Eq. (2.117) applied between the boundary sections of a hydraulic
jump (Fig. 2.42c)
h21 q2 h2 q2
S¼ þ ¼ 2þ : ð2:128Þ
2 gh1 2 gh2
This equation allows for a relation linking h1 with the unknown value of h2.

1 2 q2 1 1
h h2 þ
2
¼ 0; ð2:129Þ
2 1 g h1 h2
or

1 q2 h2 h 1
ðh2 h1 Þðh2 þ h1 Þ ¼ 0: ð2:130Þ
2 g h1 h2
Simplification yields
1 q2
h1 h2 ðh2 þ h1 Þ ¼ 0; ð2:131Þ
2 g
which is a quadratic equation in h2, e.g.,
1 1 q2
h1 h22 þ h21 h2 ¼ 0: ð2:132Þ
2 2 g
Its solution reads
h2 1 h 1=2 i q2
¼ 1 þ 1 þ 8F21 ; F21 ¼ : ð2:133Þ
h1 2 gh31
This relation is Bélanger’s equation for the sequent depth ratio of a hydraulic jump
in a horizontal, rectangular, and frictionless channel (Bélanger 1849; Montes 1998).
The important point on how momentum and energy principles act comple-
mentary is: The momentum balance applied to a hydraulic jump is a valuable tool
that permits, as shown, to compute the tailwater depth h2 of the hydraulic jump
regardless of the internal flow complexities. Once the momentum principle is
applied, the energy principle can be used to complete the analysis and determine the
hydraulic energy loss, unknown in advance. It is given by

q2 q2 q2 1 1
DE ¼ h1 þ h2 þ ¼ ð h1 h2 Þ þ
2gh21 2gh22 2g h21 h22
ð2:134Þ
q2 h22 h21
¼ ð h1 h2 Þ þ :
2g h21 h22
From Eq. (2.131), one gets
q2 1
¼ h1 h2 ðh2 þ h1 Þ; ð2:135Þ
g 2
which, inserted into Eq. (2.134), yields
ð h2 h1 Þ 3
DE ¼ : ð2:136Þ
4h2 h1
Using the specific energy E1 of the upstream supercritical flow, it is written in

dimensionless form as
3
DE
h2
h1 1
¼ : ð2:137Þ
E1 4ðh2 =h1 Þ 1 þ 12 F21
The computational process is illustrated graphically with the aid of Fig. 2.46, where
the E(h) and S(h) curves are plotted for the known unit discharge q. With the known
value of h1, the corresponding S value is determined, giving point “a” in Fig. 2.46b.
The vertical line plotted with this value of S yields a cut with S(h) at point “b”,
which is the subcritical tailwater depth h2 of the hydraulic jump. Horizontal lines
plotted from “a” and “b” cut with the E(h) curve, yielding the corresponding
specific energies at both points (points “c” and “d”, respectively). Their difference is
the energy head dissipated by the hydraulic jump. This sequence is conducted
analytically using Eqs. (2.133) and (2.137). Their accuracy is verified in Fig. 2.47,
where the experimental data of Bretz (1988) were used for validation. Note the
small deviation of the latter with predictions due to the neglected boundary shear
effect.
Fig. 2.46 Sequent depth h2 and energy loss DE of a hydraulic jump in a rectangular, horizontal,
and frictionless channel: a hydraulic jump profile and b specific energy and momentum diagrams
Fig. 2.47 Experimental

verification of the sequent
depth ratio h2 =h1 ðF1 Þ and
energy head loss DE=E1 ðF1 Þ
for the hydraulic jump in a
rectangular, horizontal, and
frictionless channel
2.3.3 Computation of Sequent Depths for General Cross

Sections
General
For an arbitrary cross section, the momentum function depends on the discharge
Q and the flow depth h as
Q2
M ¼ Ah þ : ð2:138Þ
gA
The functional relation with h is determined by the particular cross-sectional

geometry. For an arbitrary cross section, a plot M = M(h) has a similar shape to the
S = S(h) curve plotted in Fig. 2.43, with a point of minimum M, determined from

dM d Q2 d Q2 dA Q2 Q2
¼ Ah þ ¼ ðAhÞ 2 ¼ A 2B ¼ A 1 3B ;
dh dh gA dh gA dh gA gA
ð2:139Þ
rewritten as
dM dE
¼A ¼ A 1 F2 : ð2:140Þ
dh dh
Thus, the minimum value of M is reached for critical flow (F ¼ 1) regardless of

channel shape.
Symmetrical trapezoid
For a trapezoidal section of base width b and side slope 1:z, the flow area A is
A ¼ bh þ zh2 ; ð2:141Þ
and the Ah term is, upon decomposing the trapezoid into a rectangle and two
triangles,
h h h2 h3
Ah ¼ bh þ zh2 ¼ b þ z : ð2:142Þ
2 3 2 3
The momentum function is thus
h2 Q2
M¼ ð2zh þ 3bÞ þ : ð2:143Þ
6 g bh þ 12 zh2
To compute the sequent depths for a prescribed value of the momentum function
Mo, a numerical solution is necessary (Jeppson 2011). The Newton–Raphson
formula
fk
hk þ 1 ¼ hk ; ð2:144Þ
ðdf =dhÞk
with
df dM
f ¼ M Mo ; ¼ ; ð2:145Þ
dh dh
yields
ðM Mo Þk
hk þ 1 ¼ hk
k : ð2:146Þ
A 1 F2
Fig. 2.48 Example of sequent depth computation in a trapezoidal section
A code of this scheme is available on the file “conjugatedepth_trapezoidal.xls”,

Chap. 12. Figure 2.48 illustrates the computational process. For a supercritical flow
with Q = 1 m3/s and h1 = 0.1 m in a trapezoidal channel of b = 1 m and z = 2, the
problem is to determine the sequent depth of a hydraulic jump. Figure 2.48 shows
the M-h curve and a vertical line corresponding to the value of the momentum
function of the supercritical flow, Mo = 0.8551 m3. The program yields the con-
jugate depth h2 = 0.8678 m. Both sequent depths are plotted on the M-h diagram.
Circular
For a circular section of diameter D (Fig. 2.49), the A

h term of the momentum
function is determined by integrating the corresponding elementary relations,
developed as follows.
Consider the flow geometry as sketched in Fig. 2.49; the following identities
apply
1
b ¼ 2ðsin xÞ D;
2
1 1
g ¼ D ðcos xÞ D;
2 2
1
dg ¼ ðsin xÞ Ddx; ð2:147Þ
2
1 1
h g ¼ D cos x þ D cosðp bÞ
2 2
1
¼ Dðcos x cos bÞ:
2
Fig. 2.49 Determination of

the flow area and profundity
of the centroid in a circular
section
A first immediate result is the computation of the flow area as
Zh Zb
D3 2 D2
A¼ bdg ¼ sin xdx ¼ ðb sin b cos bÞ; ð2:148Þ
2 4
0 0
by using the primitive function
Zb
1 1
sin2 xdx ¼ b sin b cos b: ð2:149Þ
2 2
0
Now, the Ah term is defined by
Zh
Ah ¼ ðh gÞdA; ð2:150Þ
0
which, by use of Eq. (2.147), transforms to
Zb
D3
Ah ¼ ðcos x cos bÞsin2 x dx: ð2:151Þ
4
0
With the aid of Eq. (2.149) and the auxiliary primitive function
Zb
1
cos x sin2 xdx ¼ sin3 b; ð2:152Þ
3
0
integration of Eq. (2.151) produces

D3 1 3 1 1
Ah ¼ sin b b cos b þ sin b cos b
2
4 3 2 2
ð2:153Þ
3
D
¼ 3 sin b sin3 b 3b cos b :
24
Using Eqs. (2.149) and (2.153), the momentum function of the circular section is
finally (Henderson 1966)
Q2 D3
M ¼ Ah þ ¼ 3 sin b sin3 b 3b cos b
gA 24
ð2:154Þ
4Q2
þ :
gD ðb sin b cos bÞ
2
This function is used to compute the sequent depths of hydraulic jumps as previ-
ously explained for the trapezoidal section if the pipe flow is not pressurized; i.e.,
the subcritical sequent depth does not touch the pipe top.
If the upstream supercritical flow changes to pressurized flow, the hydraulic
jump is said to be “incomplete” (Montes 1998; Jeppson 2011) (Fig. 2.50). For a
transition from upstream supercritical to downstream pressurized flow, the
momentum balance applied to the hydraulic jump control volume produces (Montes
1998; Jeppson 2011)
Q2 D pt Q2
Ah1 þ ¼ At þ At þ : ð2:155Þ
gA1 2 qg gAt
Here, At = pD2/4 is the area of the pipe flowing full and pt the pressure at the pipe
top. Denoting by Mt the momentum function for a gravity pipe flow flowing full
D Q2
Mt ¼ At þ ; ð2:156Þ
2 gAt
Fig. 2.50 Incomplete hydraulic jump in a circular pipe

the pressure head on the pipe top is
pt M1 Mt
¼ ; ð2:157Þ
qg At
where M1 is the momentum function of the supercritical flow, to be evaluated using

Eq. (2.154). A simple method to check if an incomplete hydraulic jump appears is
to compare M1 with Mt: If M1 > Mt, the downstream flow is pressurized, whereas
for M1 < Mt the hydraulic jump is computed on the basis of Eq. (2.154).
2.3.4 Transitions in Supercritical Flow

General
Non-uniform sub- and supercritical flows in channel transitions are computed by
using the 1D approach and the specific energy diagram. This method is generally
valid for subcritical flows, but may not be accurate for supercritical flows, because it
is applied only to supercritical flows in a prismatic, rectangular channel. If changes
in boundary shape occur, 2D oblique standing waves are generated, and an alter-
native method of attack is required. Consider the flow in a straight rectilinear
channel with parallel walls (Fig. 2.51). Two steady sources of perturbations are
therefore generated at the inlet. Given that U > (gh)1/2, circular waves originating at
the sides do not include the source point itself, and the tangents to the circular
waves form two wave fronts intersecting and being reflected to the opposite wall
and further transmitted downstream. The phenomenon produces a diamond-type
wave pattern (Fig. 2.51). For supercritical flow, these standing wave patterns appear
due to flow perturbations, as a wall deflection. Their computation is detailed below.
Abrupt wall deflection
Given that the flow is 2D in the horizontal plane, the 1D control volume theory
previously development must be generalized. Reynolds’ transport theorem states
the conservation of mass and momentum for a control volume CV surrounded by a
control surface CS as (Liggett 1994; White 2009)
Fig. 2.51 Supercritical flow in rectilinear channel with parallel walls involving two steady
sources of perturbations at the inlet
ZZZ ZZ
d
qdv þ qðV nÞdA ¼ 0; ð2:158Þ
dt
CV CS
ZZZ ZZ
d
f¼ qVdv þ qVðV nÞdA: ð2:159Þ
dt
CV CS
Here, f is the force vector (resultant of body plus surface forces), V the velocity
vector, n the unit normal to the control surface, q water density, v the volume, and
A the area. Consider in this section steady supercritical flow in a horizontal and
frictionless channel. Further, the parallel-streamlined approach flow has depth h1
and velocity U1 deflected by a wall of angle h (Fig. 2.52a).
Shear forces are neglected, and pressures are assumed to be hydrostatic
(Fig. 2.52b). Due to the wall deflection, the streamlines downstream of the local
perturbation will adjust and become parallel to the tailwater wall direction.
However, this directional change is accompanied by an oblique shock front
crossing the supercritical stream at angle b (Fig. 2.52c).
Consider the control volume sketched in Fig. 2.53 to apply Eqs. (2.158)–
(2.159). Its extreme boundaries are parallel to the shock direction, and a unit width
is considered. The side faces of the control volume are by definition streamlines, so
there is no flow across them. Thus, Eq. (2.158) reduces to
Q1 ¼ Q2 ¼ Q: ð2:160Þ
Consider the velocity U decomposed in the normal (subscript n) and tangential

(subscript t) directions to the shock front. Equation (2.160) is therefore rewritten as
Q ¼ U1n h1 ¼ U2n h2 ¼ h1 U1 sin b ¼ h2 U2 sinðb hÞ: ð2:161Þ
Fig. 2.52 Oblique shock wave due to abrupt wall deflection: a plan view, b section along
streamline, c photograph of model test
Fig. 2.53 Application of (- - -) control volume approach to oblique shock wave: a plan and
b section
Equation (2.159) simplifies with fp as the resultant of the pressure forces acting on
the control surface to
f p ¼ qV2 Q qV1 Q: ð2:162Þ
Equation (2.162) is now projected in the normal direction, producing
1 1
qgh21 qgh22 ¼ qU2n Q qU1n Q: ð2:163Þ
2 2
Inserting Eq. (2.161) and re-arranging yields (Jain 2001)
1 2 1
gh1 þ h1 U12 sin2 b ¼ gh22 þ h2 U22 sin2 ðb hÞ: ð2:164Þ
2 2
Now, Eq. (2.162) is projected in the tangential direction. Note for the control
volume selected that pressure forces acting on each streamline in the tangential
direction are in equilibrium side by side, so there is no net pressure force acting in
this direction. The momentum balance reduces then to
0 ¼ qU2t Q qU1t Q; ð2:165Þ
or
U1 cos b ¼ U2 cosðb hÞ: ð2:166Þ

From Eqs. (2.161) and (2.164) follows (Ippen 1951)
h2 1 h 1=2 i U1
¼ 1 þ 8F21 sin2 b 1 ; F1 ¼ : ð2:167Þ
h1 2 ðgh1 Þ1=2
Equation (2.167) can be recast in the alternative form

1 1 h2 h2 1=2
sin b ¼ 1þ : ð2:168Þ
F 1 2 h1 h1
Note that for a finite shock h2/h1 > 1. Only for shocks of infinitesimal height,
namely for h2 h1 results sin b ¼ 1=F1 from Eq. (2.168) as the shallow water
wave solution [see Eq. (2.67)]. These disturbance lines or oblique wave fronts (or
shocks) are sometimes called oblique hydraulic jumps, which is not adequate. The
flow in the downstream portion of these waves is supercritical, so the oblique
standing waves represent transitions from super- to supercritical flow with different
directions of the streamlines. Given that a hydraulic jump is a transition from super-
to subcritical flows, the term “oblique hydraulic jump” should be avoided. Under
chocking conditions, it is possible to have an oblique hydraulic jump, but this is a
particular case. We will therefore denominate these waves as oblique standing or
shock waves. The term “shock wave in supercritical flow” is also a typical
nomenclature in dam hydraulics (Vischer and Hager 1998). The celerity c of the
oblique shock wave is (Ippen 1951)

1 h2 h2 1=2
c ¼ ðgh1 Þ1=2 1þ U1 sin b: ð2:169Þ
2 h1 h1
Given that c equals the velocity component normal to the shock front, the wave is
steady. From Eqs. (2.161) and (2.166), another identity to be satisfied by h2/h1 is
h2 tan b
¼ : ð2:170Þ
h1 tanðb hÞ
Equating Eqs. (2.167) and (2.170) results in
tan b 1 h 1=2 i
¼ 1 þ 8F21 sin2 b 1 : ð2:171Þ
tanðb hÞ 2
The angle b is a function of F1 and h, as indicated by Eq. (2.171). Its solution is

obtained numerically by any standard technique, as the Newton–Raphson method.
However, it is illustrative to represent graphically b = b(h) with F1 as a parameter,
as did Ippen (1951) (Fig. 2.54).
Fig. 2.54 Oblique shock

wave diagram of Ippen (1951)
The ratio of Froude numbers up- and downstream of the shock origin is from
Eq. (2.166) (Jain 2001)
1=2
F2 h1 cos b
¼ : ð2:172Þ
F1 h2 cosðb hÞ
It is of interest to plot in Fig. 2.54 the line defining critical flow in the tailwater
shock portion, e.g.,
1=2
h1
1=2
1 ¼ F1 cos b 1 þ tan2 ðb hÞ ; ð2:173Þ
h2
or, with the aid of Eq. (2.170),
1=2 " #1=2

h2 tan2 b
F1 cos b 1 þ ¼ 0: ð2:174Þ
h1 ðh2 =h1 Þ2
Using Eq. (2.167) for h2/h1, Eq. (2.174) was numerically solved and the result is
plotted in Fig. 2.54. For a constant value of F1 , b increases with h up to a maximum
value and then decreases. For b = 90° and h = 0, there is no oblique standing wave
but rather a hydraulic jump, given the change from super- to subcritical flow. This
steady wave is not related to the wall deflection, but rather the result of a tailwater
effect. Note from the diagram that for a given value of h two values of b are
possible, one for downstream supercritical flow (F2 [ 1) and the other for sub-
critical flow (F2 \1). The first case is relevant to the practical design of channels
with supercritical flows, given that the wave pattern is only determined by the
upstream flow condition, namely F1 . The angle of the cross-wave and its propa-
gation in the downstream direction by successive reflections at the channel walls
only depends on the upstream conditions. However, for F2 \1, there is a tailwater
effect, propagating a surge against the supercritical flow until a steady wave pattern
Fig. 2.55 Verification of

oblique shock wave diagram
is formed in conformity with Eq. (2.171). Therefore, this oblique shock with a
transition from super- to subcritical flow depends on a particular combination of up-
and downstream flow conditions. The oblique shock diagram is verified in Fig. 2.55
for supercritical tailwater conditions using the experiments by Ippen and Harleman
(1956).
Consider now supercritical flow in a channel where a wall is deflected by the
angle h, whereas the other wall remains fixed. As previously explained, an oblique
wave front starting at point A crosses the flow at angle b1 reaching the opposite
wall. The flow is subjected thus by a new perturbation at point B, so that the
incident shock is reflected, crossing the flow at a new angle b2. The flow beyond
shock AB becomes parallel to the deflected wall, with a value of F2 \F1 , along
with b2 > b1. This new shock reaches the opposite channel wall at point C and is
reflected with angle b3, with streamlines beyond the shock BC parallel to the
rectilinear wall, and F3 \F2 . The perturbation is further propagated downstream,
tending toward critical flow with each new reflection (Fig. 2.56).
Continuous curved wall with positive deflection
If the wall angle varies continuously as h(x), the continuous curve is divided into
portions of rectilinear walls of discrete deflections Δh to compute the corresponding
Δh across each oblique shock wave (Fig. 2.57). For an upstream value F1 [ 1, any
change Δh due to a flow disturbance occurs along wave fronts, crossing the stream
at angle b. This wave angle is linked to a finite change Δh in the direction of the
Fig. 2.56 Reflection of

oblique shock waves at
channel walls (adapted from
Ippen 1951)
Fig. 2.57 Supercritical flow

with continuous concave wall
deflection: a plan and
b longitudinal section along a
streamline. The wall
deflection is continuous but
might not be small, given that
the formation of shocks is
considered
boundary shape. Any continuous boundary of arbitrary shape is a streamline, and

the change of depth and velocity along it can therefore be computed. In a curved
channel, the lateral boundary streamline is the origin of all disturbance lines or
wave fronts. The complete characteristics of the supercritical flow are therefore
determined by the boundary shape (Fig. 2.57). For the concave wall shown, the
shock lines converge and the water depth increases along the streamlines.
Consider the trigonometric identity
tan b tan h
tanðb hÞ ¼ : ð2:175Þ
1 þ tan b tan h
Equation (2.170) is thus rewritten as
h2 tan2 b tan h þ tan b

¼ : ð2:176Þ
h1 tan b tan h
Defining h2 = Δh + h1, one obtains from Eq. (2.176)
Dh tan hð1 þ tan2 bÞ sec2 b tan h

¼ ¼ : ð2:177Þ
h tan b tan h tan b tan h
Continuous solutions for gradual wall deflections
It was already demonstrated that for a positive wall deflection (Fig. 2.53) the
streamlines converge and a discontinuous solution, namely the oblique shock wave,
is formed. In a negative wall deflection, however, streamlines diverge and dis-
continuous solutions are not formed, invalidating the former solutions. Rather, a
continuous depression wave is the correct flow solution (Liggett 1994; Montes
1998). An approximate method applicable to the analysis of negative wall deflec-
tions is now presented. The method is also applicable to positive wall deflections if
these are small and shocks are not formed. Consider infinitesimal wave jumps,
namely Δh ! dh, associated with small wall deflections, for which tan b
tan h,
tan h ! h, and h ! dh. Using these simplifications in Eq. (2.177) yields
dh h
¼ : ð2:178Þ
dh sin b cos b
For an infinitesimal oblique wave, Eq. (2.168) simplifies to

1 1 h2 h2 1=2 1 ðghÞ1=2
sin b ¼ 1þ ¼ : ð2:179Þ
F1 2 h1 h1 F U
Substitution into Eq. (2.178) yields (Ippen 1951)
dh U 2
¼ tan b: ð2:180Þ
dh g
For infinitesimal perturbations, the energy loss is negligible and the specific
energy head remains constant. The velocity head is thus
U2
¼ E h: ð2:181Þ
2g
Noting from Eq. (2.179) that

1=2
tan b ¼ F2 1 ; ð2:182Þ
and inserting Eqs. (2.181)–(2.182) into Eq. (2.180) yield the ODE
dh 2ðE hÞh1=2
¼ : ð2:183Þ
dh ð2E 3hÞ1=2
It describes the propagation of infinitesimal perturbations due to a continuous wall

curvature. The integral of Eq. (2.183) is (Ippen 1951)
2 !1=2 3 2 !1=2 3
h h
h ¼ 31=2 tan1 4 5 tan1 431=2 5 h1 ;
2E=3 2E=3
ð2:184Þ
1 2E=3
h
1 2E=3
h
or, with h1 as constant of integration determined by the condition that h = 0 for

F ¼ F1 ,
Fig. 2.58 Gradual wall

deflection h ¼ hðFÞ in
supercritical flow with
constant specific energy
head E
2 3 2 3
1=2
3 1
h ¼ 31=2 tan1 4 1=2 5 tan 4
1
1=2 5 h1 : ð2:185Þ
F 1
2
F 1
2
Equation (2.185) directly links h to F. For the solution produced here, the pertur-
bation lines are no more defining a discontinuous solution. Rather, these are lines of
constant flow depth determined by the boundary angle h(x) at the origin of them on
the wall. An additional relation based on constant specific energy head
E ¼ h½1 þ ð1=2ÞF2 = const. is
h F2 þ 2
¼ 12 : ð2:186Þ
h1 F þ 2
Equation (2.185) (Fig. 2.58) is of paramount interest in the design of channel

structures. For example, if the streamlines are given, the changes in h/E or F along
the streamlines are immediately determined by Eqs. (2.185)–(2.186). In the inverse
case, it is possible to determine the boundary shape h(x) needed to produce pre-
scribed changes in h/E and F. For large values of F, this relation is approximated by
(Ippen 1951)
2
h¼ h1 : ð2:187Þ
F
A comparison of Eqs. (2.185) and (2.187) in Fig. 2.58 indicates that the latter is an
accurate approximation of the former if F > 5.
For a continuous convex wall (Fig. 2.59), the streamlines are deflected away
from the original boundary alignment. Due to the negative wall deflection angle
(h < 0), the water depth decreases along the streamlines (Liggett 1994; Montes
Fig. 2.59 Supercritical

negative waves in continuous
convex wall along with
perturbation lines or constant
flow depth lines
1998). This case can be analyzed assuming E = const. and thus resorting to
Eq. (2.185). To apply Eq. (2.185) to a positive wall deflection (Fig. 2.57a), the
angle variation must be weak, such that the disturbance lines do not intersect
thereby producing a shock (Liggett 1994). The flow at an abrupt negative wall
deflection (Fig. 2.60a) can be considered as the limit of a convex wall deflection if
the length of the transition curve tends to zero. In this case, the origin of all the
disturbance lines is unique and located at the wall break. The continuous solution is
an expansion fan defined by the lines of constant water depth (Fig. 2.60b) (Liggett
1994). The up- and downstream perturbation angles are

1 1
b1 ¼ sin1 ; b2 ¼ sin1 : ð2:188Þ
F1 F2
For the known values of h1 and U1, the value of F1 ¼ U1 =ðgh1 Þ1=2 is fixed, and the
constant h1 is determined from Eq. (2.185) as
2 3 2 3
1=2
3 1
h1 ¼ 31=2 tan1 4 1=2 5 tan 4
1
1=2 5: ð2:189Þ
F21 1 F21 1
Fig. 2.60 Supercritical

abrupt wall deflection: a plan
view of expansion fan and
b water surface profile along a
streamline
Equation (2.185) is then used to compute F2 for the value of the wall deflection ho,
either numerically or with the aid of Fig. 2.58. Once F2 is known, the flow depth h2
is determined by resorting to Eq. (2.186) and the value of b2 computed by
Eq. (2.188)2. If F1 > 5, it is feasible to use the simpler Eq. (2.187).
Channel contraction
A positive wall deflection produces an oblique shock increasing the water depth,
whereas a negative wall deflection produces a centered depression wave with the
opposite effect. This fact is applied to the design of channel transition structures,
wherein positive and negative waves are superposed to minimize the transmission
of standing waves in the tailwater. The channel contraction (Fig. 2.61) in super-
critical flow is a special case where the wall deflections produce two positive
oblique wave fronts originating at points A and B that intersect and then are
propagated into the tailwater. The length of the channel contraction is selected so
that the wave fronts arrive just at its ends (points C and D). These positive waves
thus interact with the negative waves originating at the same corners, with a the-
oretical cancelation of the wave effects (Ippen and Dawson 1951). This design
approach would work only for the design discharge and is based on assuming that
the interaction of positive and negative waves exactly cancels each other (Vischer
and Hager 1998).
Fig. 2.61 Design of channel

contraction with minimum of
wave transmission: a plan
view, b free surface profiles
along centerline and channel
wall, c 3D view
2.4 Control Sections 117
2.4 Control Sections
2.4.1 General
A hydraulic control section involves a unique relation between the flow depth h and
the discharge Q and vice versa, e.g., Q = Q(h) or h = h(Q) (Henderson 1966).
A control section is used as boundary condition in water surface profile compu-
tations; thus, their discharge characteristics need to be known in advance (Jain
2001). Further, they are used at selected locations for water discharge measurement
purposes (Bos 1976). Here, the usual control sections in open channel flows are
described, namely uniform flow, critical flow, and the artificial channel control
(Rouse 1950; Jain 2001). The latter is a device inserted into a flow which generates
a head–discharge rating curve. It produces a large perturbation choking the flow,
resulting in a subcritical approach flow that changes to a supercritical flow
downstream of the structure. Examples are weirs and gates. Most artificial channel
controls involve complex flow phenomena as non-hydrostatic pressure, fluid
recirculation, and boundary layers. Here, only an introduction to typical channel
controls is given. Montes (1998) provides an extensive discussion on the fluid flow
fundamentals of head–discharge rating curves of hydraulic structures.
2.4.2 Uniform Flow
The steady gradually varied flow equation for prismatic channels is [see Eq. (1.110)]
dh So Sf
¼ ; ð2:190Þ
dx 1 F2
where So is the channel bottom slope and Sf the friction slope. Uniform flow is
regained by setting dh/dx = 0 into Eq. (2.190), resulting in
Sf ¼ So : ð2:191Þ
If a parametrization of the type Sf ¼ Cf F2 (valid for a wide rectangular channel) is

used in Eq. (2.191), with Cf as a boundary shear resistance coefficient, it is easy to
infer that a rating curve Q = Q(hN), or hN = hN(Q), emerges, where hN is the
uniform flow depth. Particular parametrizations for general cross sections are dis-
cussed in Chap. 3. Here, it is noted that uniform flow is a particular control section
where the discharge is related to the flow depth through a resistance coefficient and
the bottom slope.
2.4.3 Critical Flow
Critical flow is determined from Eq. (2.57), which generates a rating curve Q = Q(hc),
or hc = hc(Q). The frequent location of critical flow sections to start water surface
profile computations is at the maximum elevation of a round-crested weir (Fig. 2.6) or
at the maximum contraction in a non-prismatic channel section (Fig. 2.9). For a
channel with both variable elevation and cross section, the position of the critical flow
section is unknown in advance and must be determined by a singular point analysis
(Figs. 2.11 and 2.12) (Chow 1959; Montes 1998; Hager 2010). Once the critical depth
and its location are determined, the result is used as boundary condition to compute
sub- and supercritical free surface profiles. Note that for supercritical flow computations
in a non-prismatic channel the free surface is bi-dimensional due to standing waves, so
that the one-dimensional gradually varied flow equation breaks down. Another case
where the use of the critical flow theory needs careful attention relates to the flow in
zones of non-hydrostatic flow conditions, as in the vicinity of a free overfall in a
rectangular and horizontal channel, or at the transition from a mild to a steep bottom
slope (Castro-Orgaz and Hager 2017). In the former case, the critical flow section is
located (3–4)hc from the brink section (Fig. 2.62a), and the brink flow depth is hb =
0.715hc (Rouse 1938). The approach flow to the free fall is non-hydrostatic and
supercritical. To conduct gradually varied water surface profile computations, it is
assumed that the critical flow section is located at the brink section, given the small
reach with non-hydrostatic flow (Henderson 1966). Likewise, the flow near a slope
break is non-hydrostatic (Rouse 1938), with the critical flow section shifted upstream to
the small slope reach (Fig. 2.62b). For gradually varied flow computations, the critical
depth is taken at the slope break (Henderson 1966).
Fig. 2.62 Critical flow at a free overfall and b transition from mild to steep slope
2.4.4 Weir Flows

Broad-crested weir
A rectangular broad-crested weir is a hump that produces chocking of the upstream
flow and thus critical flow conditions along the weir crest. Under ideal hydrostatic
fluid flow, the streamlines are parallel along the crest, and the crest flow depth
equals the critical depth hc (Chanson 2004) (Fig. 2.63). The discharge equation is
then [see Eq. (2.31)]
3=2
3 1=2 2
q ¼ Cd gEmin ; Cd ¼ : ð2:192Þ
3
Under real fluid flow conditions, there is flow separation around the upstream
corner of the weir (Fig. 2.64a), with the corresponding recirculation of fluid. This
recirculation produces an energy loss and thus a reduction of Cd below its ideal
fluid flow value (2/3)3/2 (Castro-Orgaz and Hager 2017). Other weir types used in
practice are the round-nosed broad-crested weir, the embankment weir, and the
round-crested weir (Fig. 2.64).
Round-nosed broad-crested weir
If the upstream corner of the broad-crested weir is rounded (Fig. 2.64b), flow
separation is suppressed. The flow will pass above the weir smoothly, with parallel
streamlines, behaving like an inviscid fluid. The exception is the flow close to the
wall, which is influenced by fluid viscosity given the significant reduction in
velocity to comply with the nonslip wall condition. A boundary layer3 starts its
development close to the inlet section, but remains usually thin and rarely reaches
the free surface.
A generalization of Eq. (2.192) is (Montes 1998; Castro-Orgaz and Hager 2017)
1=2
q ¼ Cd gHo3 ; ð2:193Þ
where Ho is the upstream head over the weir. The discharge coefficient of the
round-nosed broad-crested weir is theoretically determined for hydrostatic flow
(Ho/L < 0.33) correcting the critical flow theory by viscous effects resorting to the
boundary layer theory, resulting in dL as the boundary layer displacement thickness
at the weir end section in (Ackers et al. 1978)
3
The boundary layer displacement thickness d* is the virtual shift of the potential flow from a solid
wall due to viscous effects (Montes 1998; White 2009).
Fig. 2.63 Ideal fluid flow over rectangular broad-crested weir
Fig. 2.64 Flow over weirs: a rectangular broad-crested, b round-nosed broad-crested (Ho/L < 0.33),
c embankment, d round-crested
3=2
2 d 3=2
Cd ¼ 1 L ; ð2:194Þ
3 Ho
where
q2
Ho ¼ ho þ w þ : ð2:195Þ
2gðh þ wÞ2
The boundary layer displacement thickness can be estimated with e as the rough-
ness height from (White 2009)
dL e 1=2
¼ 0:001 þ 0:2 : ð2:196Þ
L L
In reality, the flow over a broad-crested weir is surprisingly complex and depends
on the value of Ho/L. For 0.1 < Ho/L < 0.33, the streamlines are parallel over the
weir crest and the pressure distribution is hydrostatic, but viscous effects have to be
accounted for, e.g., by using Eq. (2.195) (Ackers et al. 1978). For 0.33 < Ho/L < 1.5,
the streamlines over the weir are curvilinear and the discharge characteristics are
governed by the non-hydrostatic pressure field. For Ho/L > 1.5, the flow separates
from the crest and the weir behaves like a sharp-crested weir (Montes 1998). For
Ho/L < 0.1, the pressure is non-hydrostatic and viscous effects are relevant, resulting
in an undular weir flow pattern attenuated by friction (Castro-Orgaz and Hager 2017).
Embankment weir
The embankment weir is a broad-crested weir with added slopes up- and down-
stream of the crest (Fig. 2.64c), forming a trapezoid (Hager 2010). The related flow
phenomena are complex, as already discussed, involving action of non-hydrostatic
pressure and bed friction depending on the dimensionless parameter Ho/L (Montes
1998, Hager 2010). From a practical viewpoint, the free flow discharge charac-
teristics are estimated from the empirical equation (Sargison and Percy 2009)
Cd ¼ 21=2 ð0:43 þ 0:06 sin½pðn 0:55Þ 0:0396 hðradÞ þ 0:0029Þ;

Ho ð2:197Þ
n¼ ;
Ho þ L
where the effect of the upstream slope angle h is accounted for. The predicted
discharge using Eqs. (2.193)–(2.197) is compared in Fig. 2.65 with data by
Sargison and Percy (2009) relating to trapezoidal weirs with w = 0.25 m and
L = 0.5 m.
The discharge characteristics for free flow are obtained by resorting to the energy
principle and assuming that critical flow takes place on the weir crest [see
Eq. (2.192)], tuning the equations with corrections due to fluid flow phenomena
like non-hydrostaticity or boundary friction. To apply these equations, it is a
requirement that the tailwater flow depth must not affect the control section (Ackers
et al. 1978). An illustrative example on how increased tailwater flow depth affects
weir flow is shown in Fig. 2.66. For a low tailwater depth (Fig. 2.66a), the flow is
critical somewhere on the crest and the discharge characteristics are as for free flow.
A hydraulic jump is formed at the toe of the weir. If the tailwater level is raised, the
hydraulic jump starts moving upstream along the downstream weir face
(Fig. 2.66b), stopping once a steady-state equilibrium in a control volume con-
taining the surge is reached. If the level is further increased (Fig. 2.66c), the
Fig. 2.65 Computation of rating curves of embankment weir for different upstream slopes
hydraulic jump disappears, and the flow in the vicinity of the slope break will be
non-hydrostatic and near-critical, producing a steady wave train in the tailwater
portion of the structure. For an even larger tailwater flow depth, the structure is
submerged and the flow is subcritical along the entire weir (Fig. 2.66d). The weir
acts as a “control section” for free flow conditions. In the other cases, it behaves like
a transition in bed elevation.
Round-crested weir
The round-crested weir (Fig. 2.64d) is another important type of weir used both as
dam overflow and as a water discharge measurement structure (Montes 1998;
Castro-Orgaz and Hager 2017). Its discharge characteristics are governed by the
non-hydrostatic flow field over the weir (Rouse 1938), with Cd given by
Fig. 2.66 Effect of increasing tailwater depth: a free flow over the crest with hydraulic jump at
weir toe, b hydraulic jump on downstream slope near slope break, c hydraulic jump disappears but
a train of standing waves typical of non-hydrostatic flow is formed, d submerged weir and
subcritical flow
3=2
2
Cd ¼ Co ; ð2:198Þ
3
where Co is a non-hydrostatic correction coefficient. It is determined with good

accuracy using potential flow methods (Liggett 1994; Montes 1998). An irrotational
and inviscid (potential) flow obeys the Laplace equation for the potential function / as
@2/ @2/
þ 2 ¼ 0; ð2:199Þ
@x2 @z
Fig. 2.67 Potential flow over

round-crested weir:
a photograph of model test
showing a streamline with
added colorant (Rouse 1961;
IIHR-Hydroscience &
Engineering, the University of
Iowa) and b potential flow net
with z as the vertical coordinate. For an incompressible fluid, the stream function w
satisfies also the Laplace equation,
@2w @2w
þ 2 ¼ 0: ð2:200Þ
@x2 @z
The lines w = const. are streamlines, and the velocity vector is tangent to them.
The equipotential lines / = const. are normal to the streamlines at intersection
points, forming the flow net of a potential flow (Fig. 2.67) (Rouse 1938; White
2009). While solving the elliptic problem posed by Eqs. (2.199)–(2.200) subject to
appropriate boundary conditions for a given value of Ho, the discharge q emerges as
part of the solution and thus Co. The disadvantage is that complex numerical
methods are required for this task (Montes 1998). Approximate methods to solve
Eqs. (2.199)–(2.200) are available based on the so-called Boussinesq approxima-
tion (Castro-Orgaz and Hager 2017) (see Chap. 11). If the ratio X = Ho/R is small,
a correction to the discharge coefficient of critical (non-hydrostatic) potential flow
over a round-crested weir is with R as the bottom curvature at the weir crest
(Matthew 1963; Castro-Orgaz and Hager 2017)
22 Ho 22
Co ¼ 1 þ ¼ 1þ X. ð2:201Þ
81 R 81
This equation is valid for Ho/R < 0.5 provided that the weir is of a minimum size as
to avoid scale effects originating from viscosity and surface tension (Castro-Orgaz
and Hager 2017). For higher heads, an empirical version of Eq. (2.201) is (Hager
1985)
Fig. 2.68 Flow over

round-crested weir:
a photograph of circular weir
model and b discharge
characteristics
3X
Co ¼ 1 þ : ð2:202Þ
11 þ 4:5X
Equations (2.201)–(2.202) are favorably compared in Fig. 2.68 with experimental

data on parabolic (Blau 1963) and circular (Castro-Orgaz 2010) weirs.
2.4.5 Gate Flows

Free gate flow
Consider a vertical sluice gate inserted into a channel where the discharge is q. The
gate opening is a, and the upstream energy head and flow depth are E and h1,
respectively. The approach flow is uniform and subcritical, whereas the flow issued
below the gate is supercritical with uniform flow depth h2. Note the flow con-
traction from the gate of opening a to the uniform flow depth h2 due to the
curvilinear flow (Rouse 1938; Montes 1998) (Fig. 2.69).
Fig. 2.69 Free flow from a

vertical sluice gate
If energy losses are neglected, conservation of energy head reads
q2 q2
E ¼ h1 þ 2
¼ h2 þ : ð2:203Þ
2gh1 2gh22
Note that Eq. (2.203) can be written as

q2 1 1 q2 ð h1 h2 Þ ð h1 þ h2 Þ
h1 h2 ¼ ¼ ; ð2:204Þ
2g h22 h21 2g h22 h21
resulting in
q2 h22 h21
¼ : ð2:205Þ
2g ðh1 þ h2 Þ
The contraction coefficient of sluice gate flow is Cc = h2/a; defining the dis-
charge coefficient Cd as
q
Cd ¼ ; ð2:206Þ
að2gh1 Þ1=2
the dimensionless form of Eq. (2.205) is (Rouse 1950)

C c a 1=2
Cd ¼ Cc 1 þ : ð2:207Þ
h1
The contraction coefficient Cc is a function of h1/a (Montes 1998), as observed also

from experiments by Roth and Hager (1999).
The theoretical value resulting from potential flow methods is Cc = 0.61
(Montes 1998). Equation (2.207) is compared in Fig. 2.70 with the experiments of
Roth and Hager (1999) using Cc = 0.61 and 0.59, resulting in a better match for the
latter value. Note that the sluice gate acts as a control section if the gate opening is
small enough to produce supercritical flow in the tailwater. It is immediate to obtain
2=3
the relations F21 ¼ 2Cd2 ða=h1 Þ2 and hc =a ¼ F1 h1 =a. From these identities, the
gate opening equals the critical depth roughly for a/h1 = 0.55. Once the gate
Fig. 2.70 Discharge

characteristics of free sluice
gate flow Cd = Cd(a/h1)
opening equals the critical depth, the downstream flow turns unstable due to the
presence of standing non-hydrostatic waves typical of near-critical flows (Montes
1998). In this case, there is no uniform flow downstream of the gate, and the
relevance of Cc is obsolete. For larger gate openings, the waves progressively
disappear, leading to a hydrostatic subcritical flow downstream of the gate. For
practical purposes, the gate can be considered to act as a control section for
a/h1 < 0.6, with Cc = 0.6. For larger gate openings, the gate acts as a local
obstruction to the flow.
The force F on the gate is now evaluated by application of the momentum
balance, producing after insertion of Eq. (2.205),
2
h21 q2 h2 q2 1 ðh1 h2 Þ3 1 ðh1 Cc aÞ3
F ¼ qg þ þ ¼ qg ¼ qg :
2 gh1 2 gh2 2 h1 þ h2 2 h1 þ Cc a
ð2:208Þ
This is another example on the complementary usage of the energy and momentum
principles; once the discharge q is determined from the energy principle, the
momentum principle is used to evaluate the force acting on the gate. In dimen-
sionless form, it is
F ð1 Cc a=h1 Þ3
¼ : ð2:209Þ
qg 12 h21 1 þ Cc a=h1
Gate flow with rejected hydraulic jump
Consider free flow past a vertical sluice gate in a rectangular channel of equal width
as the gate, with a tailwater flow depth that exactly equals the sequent depth of the
contracted flow depth h2 (Fig. 2.71). A hydraulic jump is therefore formed beyond
the sluice gate. If h3 is lower than the sequent depth h2, a hydraulic jump would still
be formed but further shifted into the tailwater. A water surface profile of super-
critical flow starting at the contracted section should be formed to increase the flow
depth and then equal the actual tailwater level through a hydraulic jump. These
cases are detailed in Chap. 4, given that the surface profile from the vena contracta
section is governed by friction.
It is common in the field monitoring of canal networks to install up- and
downstream of sluice gates ultrasonic sensors to measure h1 and h3. We will form
the resulting system of equations needed to compute the unknowns q and h2 for the
case depicted in Fig. 2.71a. The energy equation applied between the upstream
section and the contracted section is
q2 q2
E ¼ h1 þ 2
¼ h2 þ ; ð2:210Þ
2gh1 2gh22
with the discharge given by Eq. (2.205). The momentum equation applied between
the contracted and the tailwater sections reads
Fig. 2.71 Free flow across sluice gate with rejected hydraulic jump: a definition sketch and
b photograph of model test
h22 q2 h2 q2
S¼ þ ¼ 3þ : ð2:211Þ
2 gh2 2 gh3
It can be rewritten as

1 q2 h2 h3
ð h2 h3 Þ ð h2 þ h3 Þ ¼ ; ð2:212Þ
2 g h3 h2
from which the discharge is given by
q2 1
¼ ðh2 þ h3 Þh3 h2 : ð2:213Þ
g 2
Equating q from Eqs. (2.205) and (2.213) results in the quadratic equation for h2

1 1 2 h1 h3 1
h3 h2 þ h3 þ 2h1
2 2
h2 þ h1 h3 ¼ 0:
2
ð2:214Þ
2 2 2 2
Thus, once h2 is computed the discharge q follows using either Eq. (2.205) or
(2.213). This is an illustrative example highlighting how conservation of energy in
the sluice gate problem, and momentum in the hydraulic jump, is used simulta-
neously. In a final step, if needed, the force F on the gate and the energy loss ΔE of
the hydraulic jump are determined by application of the momentum and the energy
balances to the gate and the hydraulic jump, respectively. The use of the energy–
momentum equations in sluice gate problems is known as the EM method.
Gate flow with submerged hydraulic jump
Consider again the case depicted in Fig. 2.71a. The conjugate depth of the super-
critical contracted flow depth h2 = Cca is
1 h 1=2 i
h02 ¼ Cc a 1 þ 8F21 1 : ð2:215Þ
2
If h2′ = h3, the case of Fig. 2.71a occurs. If h2′ > h3, the jump is further rejected
into the tailwater and the gate operates under free flow conditions, with a backwater
curve starting from the vena contracta section. However, if the gate is operating
under free flow conditions and a tailwater depth is imposed such that h2′ < h3, a
moving hydraulic jump (surge) is shifted upstream, becoming drowned once
reaching the gate section; an alternative flow model is required in this particular
case. The basic assumption, in addition to the hydrostatic pressure distribution, is
that the flow velocity in the roller of the hydraulic jump is small and can be
neglected (Henderson 1966; Castro-Orgaz et al. 2013). Conservation of energy

between the approach flow section to the gate and the vena contracta section yields
q2 q2
E ¼ h1 þ ¼ h þ : ð2:216Þ
2gh21 2gh22
Observe that the kinetic flow energy is given by the velocity in the vena contracta.
Conservation of momentum between the vena contracta and the tailwater section
gives
h2 q2 h2 q2
S¼ þ ¼ 3þ : ð2:217Þ
2 gh2 2 gh3
Equations (2.216)–(2.217) are the EM equations for submerged flow below a sluice
gate (Henderson 1966). In field conditions, ultrasonic sensors are used to measure
h1 and h3. The unknowns are therefore q and h (Fig. 2.72).

q2 1 1
h1 h ¼ ; ð2:218Þ
2g h22 h21
Fig. 2.72 Submerged flow below sluice gate: a definition sketch and b photograph of model test
and Eq. (2.217) as

1 2 q2 1 1
h h3 ¼
2
¼ 0: ð2:219Þ
2 g h3 h2
Equating q from Eqs. (2.218)–(2.219) produces

1 2 2ðh1 hÞh22 h21 h2 h3
h h23 ¼ ; ð2:220Þ
2 ðh1 þ h2 Þðh1 h2 Þ h3 h2
which is a quadratic equation in h, e.g.,

" 1 # " 1 #
1 2 h2 h22 1 2 h2 h22
h þ 2h2 1 1 2 h h3 þ 2h2 h1 1 1 2
2 h3 h1 2 h3 h1
¼ 0:
ð2:221Þ
Using the definitions h2 = Cca, Y = h/h3, Eq. (2.221) is written in dimensionless

form as
" 1 # " 1 #
1 2 Cc Cc Cc2 1 C c r1 C c Cc2
Y þ 2 1 1 2 Y þ2 2 1 1 2
2 r3 r3 r1 2 r3 r3 r1
¼ 0;
ð2:222Þ
where r1 = h1/a and r3 = h3/a. This is written in compact form as
aY 2 þ bY þ c ¼ 0; ð2:223Þ
with the physically relevant solution
1=2
b þ ðb2 4acÞ
Y¼ : ð2:224Þ
2a
From Eq. (2.218), one can write using the definition of Cd given by Eq. (2.206)

h h21 h22 ð h1 þ h2 Þ ð h1 h2 Þ
1 ¼ Cd2 a2 ¼ Cd2 a2 ; ð2:225Þ
h1 h22 h21 h22 h21
Fig. 2.73 Discharge

characteristics Cd = Cd(h1/a,
h3/a) for a submerged vertical
sluice gate flow
from which Cd is given by

" 1 #1=2
r3 Cc2
Cd ¼ Cc 1 Y 1 2 : ð2:226Þ
r1 r1
For given values of h1/a and h3/a, Eq. (2.222) is solved for Y and then
Eq. (2.226) for Cd. This task was accomplished and is presented in Fig. 2.73 in the
so-called Henry (1950) diagram. As observed, the theoretical prediction is in fair
agreement with experiments. If a precise water discharge estimation is required,
however, a more advanced model is needed by relaxing some of the starting
assumptions (Castro-Orgaz et al. 2013), as that of zero roller flow velocities or
negligible energy losses across the gate. Note that these computations are to be
conducted only if the tailwater flow depth h3 is larger than the conjugate depth of
h2, given by Eq. (2.215). Computations in Fig. 2.73 were conducted until
Eq. (2.224) produced real solutions. In general, the result to identify a free flow
using Eq. (2.215) is not identical to that derived from Eq. (2.224), due to the special
form of the equations adopted (Jeppson 2011).
2.5 Application: Numerical Solution of Energy–

Momentum Equations for Gate Flow
Open channel flow applications frequently involve the simultaneous solution of

nonlinear systems of equations (Jeppson 2011). In this section, the Newton–
Raphson method is generalized for these cases using as practical example the EM
2.5 Application: Numerical Solution of Energy–Momentum Equations for Gate Flow 133
gate problem shown in Fig. 2.71. The numerical solution of this problem is for-
mulated defining two functions F1 and F2, for which the roots have to be deter-
mined, namely

q2 q2
F1 ¼ h1 þ h2 þ 0; ð2:227Þ
2gh21 2gh22
2 2
h2 q2 h3 q2
F2 ¼ þ þ 0: ð2:228Þ
2 gh2 2 gh3
As in the Newton–Raphson method for a single equation, iterations start with a

guessed solution for q and h2, generally not being correct. Using a Taylor series
development truncated to the first term, the new solution at iteration k + 1 is forced
to produce zero residuals, e.g.,

@F1 k k þ 1 @F1 k k þ 1
F1k þ 1 ¼ þ
F1k q q þ
k
h2 hk2 0;
@q @h2
ð2:229Þ
@F2 k k þ 1 @F2 k k þ 1
F2k þ 1 ¼ F2 þ
k
q q þ
k
h2 hk2 0:
@q @h2
These equations are rewritten as

k
@F1 @F1 k
F1k ¼ Dq þ Dh2 ;
@q @h2
ð2:230Þ
@F2 k @F2 k
F2k ¼ Dq þ Dh2 ;
@q @h2
or in matrix form
!k !k !
@F1 @F1
F1 @q @h2 Dq
¼ @F2 @F2 : ð2:231Þ
F2 @q @h2 Dh2
With F as the vector of residuals, J the Jacobian matrix, and X the vector of
unknowns,
! ! !
@F1 @F1
F1 @q @h2 q
F¼ ; J¼ @F2 @F2 ; X¼ ; ð2:232Þ
F2 @q @h2 h2
Equation (2.231) is written as
F ¼ J DX: ð2:233Þ
The solution of the linear system of equations yields
DX ¼ J1 F; ð2:234Þ
which can be accomplished using an LU decomposition (Hoffman 2001; Press et al.

2007). The Newton–Raphson vector formula for updating the solution is then
k
Xk þ 1 ¼ Xk J1 F : ð2:235Þ
Equation (2.235) must be iteratively applied until the residuals of X are below a
prescribed tolerance. For the present example, the elements of the Jacobian matrix
are
@F1 q q
¼ 2;
@q 2
gh1 gh2
@F1
¼ 1 F22 ;
@h2
ð2:236Þ
@F2 2q 2q
¼ ;
@q gh2 gh3
@F2
¼ h2 1 F22 :
@h2
A code to solve the problem is implemented in the file “EM_sluicegate.xls”,

available in Chap. 12. An LU decomposition is applied using a program adapted
from Press et al. (2007). For the test case h1 = 1 m and h3 = 0.5 m, the program
yields after five iterations h2 = 0.0778 m and q = 0.3322 m2/s, which is in full
agreement with the analytical solution [Eqs. (2.213)–(2.214)]. This numerical code
can be easily adapted to other channel cross sections for which no analytical
solutions exist, as well as for submerged flow conditions.
References
Ackers, P., White, W. R., Perkins, J. A., & Harrison, A. J. M. (1978). Weirs and flumes for flow
measurement. New York: Wiley.
Bakhmeteff, B. A. (1912). O нepaвнoмepнoм движeнии жидкocти в oткpытoм pycлe [Varied
flow of liquids in open channels]. Russia: St Petersburg (in Russian).
Bakhmeteff, B. A. (1932). Hydraulics of open channels. New York: McGraw-Hill.
References 135
Bélanger, J. B. (1849). Notes sur le Cours d’Hydraulique. [Notes on the Course in Hydraulics].
Mém. Ecole Nat. Ponts et Chaussées, Paris, France, session 1849–1850, 222 p (in French).
Blau, E. (1960). Die modellmässige Untersuchung von Venturikanälen verschiedener Grössen und
Form [Model examination of Venturi channels of different sizes and shapes]. Mitteilung 8 der
Forschungsanstalt für Schiffahrt, Wasser-und Grundbau. Berlin: Akademie-Verlag (in
German).
Blau, E. (1963). Der Abfluss und die hydraulische Energieverteilung über einer parabelförmigen
Wehrschwelle [Flow and the hydraulic energy distribution over a parabolic weir]. Mitteilung 7,
5–72. Forschungsanstalt für Schiffahrt, Wasser und Grundbau, Berlin (in German).
Bos, M. G. (1976). Discharge measurement structures. Publication 20. Wageningen, Netherlands:
International Institute for Land Reclamation (ILRI).
Bretz, N. V. (1988). Ressaut hydraulique forcé par seuil [Hydraulic jump forced by a sill].
Communication 2, Laboratoire de Constructions Hydrauliques, EPFL Lausanne, Switzerland
(in French).
Castro-Orgaz, O. (2010). Approximate modeling of 2D curvilinear open channel flows. Journal of
Hydraulic Research, 48(2), 213–224.
Castro-Orgaz, O., Giraldez, J. V., & Ayuso, J. L. (2008). Transcritical flow due to channel
contraction. Journal of Hydraulic Engineering, 134(4), 492–496.
Castro-Orgaz, O., & Hager, W. H. (2017). Non-hydrostatic free surface flows. Advances in
geophysical and environmental mechanics and mathematics (696p). Berlin: Springer. https://
doi.org/10.1007/978-3-319-47971-2.
Castro-Orgaz, O., Mateos, L., & Dey, S. (2013). Revisiting the Energy-Momentum method for
rating vertical sluice gates under submerged flow conditions. Journal of Irrigation and
Drainage Engineering, 139(4), 325–335.
Castro-Orgaz, O., & Sturm, T. W. (2018). Boris A. Bakhmeteff and the development of the
specific energy and momentum concepts. Journal of Hydraulic Engineering, 144(12),
02518004.
Chaudhry, M. H. (2008). Open-channel flow (2nd ed.). Berlin: Springer.
Hager, W. H. (1985). Critical flow condition in open channel hydraulics. Acta Mechanica, 54(3–
4), 157–179.
Hager, W. H. (2010). Wastewater hydraulics: Theory and practice. Berlin: Springer.
Henry, H. R. (1950). A study of flow from a submerged sluice gate (M.S. thesis). Department of
Mechanics and Hydraulics, State University of Iowa, Iowa City IA.
Hoffman, J. D. (2001). Numerical methods for engineers and scientists (2nd ed.). New York:
Marcel Dekker.
Ippen, A. T. (1951). Mechanics of supercritical flow. Transactions of ASCE, 116, 268–295.
Ippen, A. T., & Dawson, J. H. (1951). Design of channel contractions. Transactions of ASCE, 116,
326–346.
Ippen, A. T., & Harleman, D. R. F. (1956). Verification of theory for oblique standing waves.
Transactions of ASCE, 121, 678–694.
Jain, S. C. (2001). Open channel flow. New York: Wiley.
Jeppson, R. (2011). Open channel flow: Numerical methods and computer applications. Boca
Raton: CRC Press.
Khafagi, A. (1942). Der Venturikanal: Theorie und Anwendung [Venturi flume: Theory and
application]. Versuchsanstalt für Wasserbau, Eidgenössische Technische Hochschule Zürich,
Mitteilung 1. Leemann, Zürich, Switzerland (in German).
Matthew, G. D. (1963). On the influence of curvature, surface tension and viscosity on flow over
round-crested weirs. Proceedings of ICE, 25, 511–524; 28, 557–569.
Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. (2007). Numerical recipes:
The art of scientific computing (3rd ed.). Cambridge: Cambridge University Press.
Puertas, J., & Sánchez, M. (2001). Apuntes de Hidráulica de canales [Open channel hydraulics
lecture notes]. Civil Engineering School, University da Coruña, Spain (in Spanish).
Roth, A., & Hager, W. H. (1999). Underflow of standard sluice gate. Experiments in Fluids, 27(4),
339–350.
Rouse, H. (1938). Fluid mechanics for hydraulic engineers. New York: McGraw-Hill.
Rouse, H. (1950). Engineering hydraulics. New York: Wiley.
Rouse, H. (1961). Fluid motion in a gravitational field: IIHR Movies, University of Iowa.
Sargison, J. E., & Percy, A. (2009). Hydraulics of broad-crested weirs with varying side slopes.
Journal of Irrigation and Drainage Engineering, 135(1), 115–118.
Selby, S. M. (1973). Standard mathematical tables. Cleveland: CRC.
Subramanya, K. (1986). Flow in open channels. New Delhi: Tata McGraw-Hill.
Toman, E. M., Skaugset, A. E., & Simmons, A. N. (2014). Calculating discharge from culverts
under inlet control using stage at the inlet. Journal of Irrigation and Drainage Engineering,
140(2), 06013003.
Vischer, D. L., & Hager, W. H. (1998). Dam hydraulics. New York: Wiley.
Chapter 3
Computation of Steady
Gradually-Varied Flows
3.1 Introduction
Steady open channel flow is generally non-uniform. In gradually-varied flows, the

changes of depth and velocity in space are small, so that streamline curvature effects
can be neglected. If the channel bottom slope is small, then the hydrostatic vertical
pressure distribution prevails. This chapter presents water surface profile compu-
tations of steady gradually-varied flows in prismatic channels, including qualitative
sketches of the flow profiles by analyzing the nature of the governing equation and
numerical methods to produce accurate solutions. The chapter focuses on flows
with constant discharge, thereby overlooking the analysis of spatially varied flows
with either increasing or decreasing discharge (Montes 1998; Hager 2010).
3.2 Governing Equation of Non-uniform Flow
The momentum principle yields for a prismatic channel the system [Eq. (1.100)],
dM
¼ gA So Sf ; ð3:1Þ
dx
Q2
M¼ þ Ah: ð3:2Þ
gA
Here, M is the momentum function, Sf the friction slope, Q the discharge, A the flow
area, x the streamwise coordinate, g the gravity acceleration, h the depth of the

https://doi.org/10.1007/978-3-030-13073-2_3
138 3 Computation of Steady Gradually-Varied Flows
section centroid below the free surface, and So the bottom slope. The
gradually-varied flow equation resulting from Eqs. (3.1)–(3.2) is [Eq. (1.110)],
so
dh So qgRh So Sf
¼ ¼ ; ð3:3Þ
dx 1 B Q23
gA
1 F2
where so is the boundary shear stress, Rh = A/p the hydraulic radius, p the wetted
perimeter, B the free surface width, h the flow depth, q the water density, and F the
Froude number. Equation (3.3) was first presented by Bélanger (1828).
The energy principle yields, with E as the specific energy and Se the gradient of
dissipated energy (Montes 1998; Chanson 2004) the system
dE
¼ So Se ; ð3:4Þ
dx
Q2
E ¼ hþ : ð3:5Þ
2gA2
Equation (3.3) is obtained from Eqs. (3.4)–(3.5) if the gradient of dissipated energy Se
equals the friction slope Sf (Yen 1973, 1991, 2002; Montes 1998). Note that the
velocity distribution coefficients are neglected in the systems (3.1)–(3.2) and (3.4)–(3.5).
Given that the energy (Coriolis) and momentum (Boussinesq) velocity correction
coefficients are different theoretically, gradually-varied flows are defined by a unique
ODE [Eq. (3.3)] if and only if the velocity correction coefficients are assumed to be
unity and the friction slope equals the gradient of dissipated energy. Both slopes are
different from a theoretical point of view (Yen 1973, 2002): the friction slope is a
measure of the shear forces acting on the channel boundaries, whereas the gradient of
dissipated energy is a measure of all the losses accumulated by the entire mass of fluid.
For gradually-varied flows in a prismatic channel, both are equal once the velocity
distribution is assumed to be uniform. At abrupt variations of the channel
cross-section, e.g., expansions or contractions, local energy losses due to eddies
become significant. These additional losses have to be introduced empirically into
gradually-varied flow computations by using local energy-loss coefficients (Montes
1998). In this chapter, we limit the analysis to flow profiles along prismatic reaches.
3.3 Uniform Flow
3.3.1 Definition
Uniform flow is the simplest case of non-uniform flow, obtained by setting the
variation of depth with distance equal to zero in Eq. (3.3), that is,
3.3 Uniform Flow 139
dh
¼ 0 ) So Sf ¼ 0 ) so ¼ qgRh So : ð3:6Þ
dx
The constant flow depth satisfying Eq. (3.6) is defined as the normal or uniform
flow depth hN. In uniform flow, the friction slope is identical to the channel bottom
slope. To compute the uniform flow depth for a given So resorting to Eq. (3.6), a
parametrization of the bed-shear stress so as function of the flow variables is
required. This task is accomplished below.
3.3.2 Flow Resistance
Rouse (1965) parametrized the boundary shear stress in open channel flows by
defining the general relation
so
¼ HðR; F; e; C; N; UÞ: ð3:7Þ
qU 2
Here, R ¼ UD=m is the Reynolds number, U the mean velocity, D the hydraulic
diameter, m the kinematic viscosity, F the Froude number, e = ks/D the relative
roughness, ks the equivalent wall roughness height, C the effect of the
cross-sectional shape, N the effect of non-uniform flow, and U the unsteadiness
effect. While Eq. (3.7) is a general statement from dimensional analysis, it is dif-
ficult to find particular expressions in practice, given the nonlinear interactions of
the various hydraulic numbers (Yen 2002). For the specific case of steady uniform
flow, Eq. (3.7) reduces to (Yen 2002)
so f
¼ HðR; eÞ ¼ ; ð3:8Þ
qU 2 8
where f is the Darcy–Weisbach friction factor. The resistance factor C = (8g/f)1/2 is

called Chezy coefficient. Thus, the so-called Moody diagram is a particular case of
Eq. (3.7) (Rouse 1965; Yen 2002). Inserting Eq. (3.8) into Eq. (3.6) produces
f
qU 2 ¼ qgRh Sf ; ð3:9Þ
8
which is the Darcy–Weisbach equation (Chanson 2004),
1 U2 1 U2
Sf ¼ f ¼f : ð3:10Þ
4Rh 2g D 2g
Note that in open channel flows D = 4Rh. The problem is thus reduced to find a
suitable predictor for H(R, e) in Eq. (3.8). Open channel flow is generally turbulent
(R > 2000) (White 1991, 2003), and attention is therefore limited to this regime.
A Colebrook-White type formula suitable for open channel flows is (Yen 2002)

4e c
f 1=2 ¼ a log10 þ ; ð3:11Þ
b 4Rf 1=2
where the empirical coefficients (a, b, c) adopt different values, depending on the
literature source [see Yen (2002) for a detailed discussion]. Here, the values pro-
posed by Henderson (1966) (a = 2, b = 12, c = 2.5) are used. Equation (3.11) is
plotted in Fig. 3.1 using these coefficients, by numerically solving it for given
values of R and e.
The limiting form of Eq. (3.11) for turbulent smooth flows (e = 0) is

1=2 2:5
f ¼ 2 log10 ; ð3:12Þ
4Rf 1=2
whereas for turbulent rough flows (R ! ∞) one gets the explicit equation
2
1 1
f ¼ log10 e : ð3:13Þ
4 3
In hydraulic smooth flows, the wall roughness elements are submerged into the
laminar viscous sub-layer, and the flow resistance is fully determined by R (White
1991, 2003). In hydraulic rough flows, the roughness elements emerge above the
thin viscous sub-layer, and the flow resistance is a form drag, fully determined by e
(White 1991, 2003). Between these two limits, the flow is transitional, and the flow
resistance is a function of both R and e. Many practical open channel flow problems
are likely to be fully rough, resorting to Eq. (3.13) and greatly simplifying com-
putations, therefore. However, smooth flows are also usual in open channel laboratory
facilities. Thus, it is equally important to have a simple predictor for f in these flows.
Fig. 3.1 Moody-type

diagram for turbulent open
channel flows
As noted from Fig. 3.1, a good explicit approximation to Eq. (3.12) is (Yen
2002; Fenton 2010)

1 1:95 2
f ¼ log10 : ð3:14Þ
4 R0:9
By simply combining the limiting forms for rough and smooth flows, given by
Eqs. (3.13) and (3.14), the following explicit formula is obtained (Yen 2002,
Fenton 2010)

1 e 1:95 2
f ¼ log10 þ : ð3:15Þ
4 3 R0:9
This equation is plotted in Fig. 3.1 and compared to Eq. (3.11), resulting in an
excellent match. Applying Eq. (3.15) in combination with Eq. (3.10) provides a
simple and physically sound method to compute flow resistance in open channels.
Usual values of the roughness height are given in Table 3.1.
Another widely used flow resistance relation for turbulent rough open channel
flows is the Gauckler–Manning–Strickler (GMS) equation (Sturm 2001; Yen 2002)
n2 U 2
Sf ¼ 4=3
; ðSIÞ ð3:16Þ
Rh
where n is Manning’s dimensional resistance coefficient (units are L−1/3T in SI).

The value of n depends on the boundary roughness, but it lacks a direct physical
meaning (Yen 1991). Traditional methods for determining n include the use of
tables (see Table 3.2) or estimations based on photographs (Chow 1959; Chaudhry
2008). In practice, however, a method for computing n is based on fitting water
surface computations to observations, if available (Puertas and Sánchez 2001). The
compatibility of Eqs. (3.10) and (3.16) implies (Yen 2002)
f gn2
¼ 1=3 ; ð3:17Þ
8 R
h
which permits to express n as function of f. This relation indicates that n is a

function of R, e and the water depth h.
Table 3.1 Values for Material ks (mm)

roughness height (Jain 2001)
Steel 0.05–0.2
Cast iron 0.25–1
Cement 0.3–1.2
Concrete 0.5–3
Table 3.2 Typical values for Material n (sm−1/3)

Manning’s n (Chaudhry
2008) Steel 0.012
Cast iron 0.013
Corrugated metal 0.025
Lucite 0.009
Glass 0.01
Cement 0.011
Concrete 0.013
Wood 0.012
Clay 0.013
Masonry 0.025
Rock cuts 0.035
Clean natural stream 0.03
Natural stream with gravel 0.04
It is of interest to assess if Eq. (3.17) permits a reasonable estimate for n with

f given by Eq. (3.15). An empirical equation used to estimate n as function of ks is
(Jaeger 1956; Chaudhry 2008)
ks1=6
n¼ : ðSIÞ ð3:18Þ
21:1
Fenton (2010) took Eq. (3.18) as reference and conducted the corresponding
comparison, concluding that agreement between Eqs. (3.15)–(3.17) and Eq. (3.18)
results if the coefficient 21.1 in Eq. (3.18) is reduced by 10–20%. Chaudhry (2008)
quoted for computing n as function of ks the formula
ks1=6
n¼ ; ðSIÞ ð3:19Þ
8:25g1=2
which is considered here. Inserting Eqs. (3.19) into (3.17) yields
gn2 f g41=3 e1=3

¼ ¼ 0:0233e1=3 : ð3:20Þ
Rh
1=3 8 8:252 g
This is compared in Fig. 3.2 with Eq. (3.13) for the fully rough flow, rewritten as
2
gn2 1 1 1
1=3
¼ f ¼ log10 e : ð3:21Þ
Rh 8 32 3
Both equations reasonably agree, as observed in Fig. 3.2. From a theoretical

viewpoint, a drawback of using Manning’s n is its non-homogeneous form. Units are
L−1/3T in SI, with the temporal dimension due to the absence of g in Eq. (3.16).
A time dimension in a resistance coefficient lacks a physical meaning, because the
measure of surface roughness must be the same irrespective of the gravity field, e.g.,
a surface has the same roughness on the Earth and on the Moon (Yen 1991). Yen
(1991, 2002) offered a smart remedy defining a Manning’s coefficient n′ given by
n0 ¼ ng1=2 : ð3:22Þ
Manning’s equation is thus simply rewritten as
n0 2 U 2
Sf ¼ 4=3
: ð3:23Þ
gRh
The unit of n′ is L1/6 in SI, which is reasonable for the measure of a roughness; it is
only related in this way to the dimension length. Note that Eq. (3.22) is implicit in
the empirical Eq. (3.19), which relates n′ to k1/6 s , with ks as a physical-based
quantity. It permits to write Eq. (3.23) as (Chaudhry 2008)
g1=2 1=2 4=3

U ¼ 8:25 1=6
Sf R h : ð3:24Þ
ks
Equation (3.24) is a modified Manning equation suitable for river flows simply
replacing ks by a mean particle diameter of the bed surface layer.
3.3.3 Uniform Flow Depth
In this section, the computation of the normal depth will be presented using
Manning’s equation with a constant value of n. Similar computations are accom-
plished by adopting the Darcy–Weisbach equation or a variable n with flow depth
resorting to Eq. (3.17). Inserting the continuity equation Q = UA into Eq. (3.16)
produces, with Sf = So,
S1=2 2=3
Q¼ o
ARh : ð3:25Þ
n

Manning coefficient as a
function of relative roughness
Rewriting Eq. (3.25) as
nQ 2=3
1=2
¼ ARh ; ð3:26Þ
So
one can define the root-finding problem as
nQ
f ðhÞ ¼ AðhÞ½Rh ðhÞ2=3 1=2
0: ð3:27Þ
So
Therefore, for a given channel shape and values of Q, n and So, the flow depth by
which f(h) = 0 in Eq. (3.27) is the uniform flow depth hN. The computation must be
conducted using numerical methods. Given a known value of h at iteration “k” for
which f(hk) 6¼ 0, a better approximation to the root of Eq. (3.27) is given by the
Newton–Raphson iterative formula (Hoffman 2001; Jeppson 2011)
fk
hk þ 1 ¼ hk : ð3:28Þ
ðdf =dhÞk
df d 2=3 2=3 dA 2 1=3 dR

¼ ARh ¼ Rh þ AR ; ð3:29Þ
dh dh dh 3 h dh
or, after some manipulation, noting that B = dA/dh and R = A/p,

df 2=3 5 B 2 dp
¼ ARh : ð3:30Þ
dh 3 A 3p dh
Equation (3.28) is applied recursively until the difference between two successive
approximations is down to a prescribed tolerance e 10−6, e.g.,

kþ1

h hk

6

hk
10 : ð3:31Þ
As a practical example, consider uniform flow in a prismatic channel of sym-

metrical trapezoidal cross-section (Fig. 3.3). The flow area A, free surface width B,
wetted perimeter p, side slope 1:z, and hydraulic radius Rh are
A ¼ bh þ zh2 ;
B ¼ b þ 2zh;
p ¼ b þ 2hð1 þ z2 Þ1=2 ; ð3:32Þ
A
Rh ¼ :
p
Ven Te Chow was born on 14 August, 1919, at Hangchow, China, passing

away aged 62 years on 30 July, 1981, at Urbana IL, USA. He was educated at
the National Chiao Tung University, Hsingchu, Taiwan, from where he
received in 1946 his B.S. degree. In 1948, he graduated from Penn State
University with an M.S., and in 1950 obtained his Ph.D. degree from the
University of Illinois. He was there until 1955 assistant professor, until 1958
associate professor when being appointed professor of hydraulic engineering.
Chow was closely related to international congresses of the International
Association of Hydraulic Research IAHR, of Hydrological Sciences IAHS,
and the Intl. Union of Geodesy and Geophysics IUGG. He was the editor of
the series Advances in Hydroscience of which a total of 14 books were
published until 1986. He was also the founder and first editor of Water
International. He was elected in 1975 to Academia Sinica, the highest honor
offered to a scholar in China. He was a founder and the first president of the
International Water Resources Association (IWRA).
Chow was internationally known for his contributions to the sciences of

hydrology and water resources development. In the 1950s, Chow’s main
activity was a thorough development of hydraulics, mainly by his 1955 paper
on backwater curves, but particularly by his outstanding 1959 hydraulics
book. This work summarizes all aspects of hydraulics just when the computer
age began. It reads superbly because of conciseness, excellence, and
up-to-date information in all aspects of open channel flow. It may be con-
sidered one of the masterpieces in hydraulics through the ages. Chow added
in 1964 with the Handbook of applied hydrology substantially to the science
of hydrology and was editor of the Journal of Hydrology since 1969.
To apply Eq. (3.28), a plausible initial value of hN is required. Assuming a wide

rectangular channel (z = 0) of width b (Rh h), Eq. (3.26) yields the following
initial guess for the normal depth
!3=5
Qn
hk¼0 ¼ 1=2
: ð3:33Þ
bSo
The wetted perimeter derivative needed to apply Eq. (3.30) for the trapezoidal
section is
dp
¼ 2ð1 þ z2 Þ1=2 : ð3:34Þ
dh
A code with this scheme implemented is on the file “normaldepth_NR_Manning.

xls”, in Chap. 12. For a channel of base width b = 2 m, slope z = 5, n = 0.015
sm−1/3, So = 0.001 and Q = 15 m3/s, the program yields hN = 1.181 m after four
iterations.
Fig. 3.3 Uniform flow in trapezoidal channel a streamwise profile, b transverse section
3.4 Flow Profiles in Prismatic Channels 147
3.4 Flow Profiles in Prismatic Channels
3.4.1 Qualitative Analysis
In gradually-varied flows (GVF), the water surface profiles are determined by

solving the first-order ODE Eq. (3.3) (Bakhmeteff 1912, 1932; Woodward and
Posey 1941; Chow 1959; Henderson 1966). The typical solution is h = h(x), i.e., a
weak evolution of the flow depth along the x-spatial coordinate (Fig. 3.4). Given
that the flow is steady and discharge is constant, the velocity is computed using the
flow area A as U = Q/A once the water surface position is determined. This
information is of paramount relevance to understand the operation of rivers and
canals. Note that the velocity is assumed constant within the cross-section, equal to
its average value U. Further, streamline curvature is weak and the bottom slope
small, so that the vertical water depth equals the bottom pressure head. Due to the
variation of both h and U with x, all other relevant hydraulic quantities experience a
space variation, as for example, the friction slope Sf, which is no more equal to the
bottom slope So as for uniform flow.
While it should be possible to directly numerically solve Eq. (3.3), it is advisable
to first analyze the nature of the solutions. This is known as qualitative analysis,
while the numerical computation corresponds to a quantitative analysis
(Subramanya 1986; Jain 2001; Sturm 2001). In the qualitative analysis, we intend
to depict all possible variations of the flow profile for the case of constant discharge
in prismatic channels. One of the strongest reasons to conduct a qualitative analysis
is that the fundamental concepts of GVF emerge from it, and these could pass
unnoticed to the student by conducting a direct numerical integration. In fact, the
qualitative analysis provides the necessary tools to critically assess the results of a
numerical integration method, to be presented.
Before entering deeper into the problem, the fundamental concepts are rescued.
The normal depth hN is defined by the following implicit equation, assuming
Manning’s equation to quantity flow resistance effects,
Fig. 3.4 Gradually-varied

flow in open channels
S1=2 2=3
Q o
ARh ¼ 0 ) hðxÞ hN ) So ¼ Sf : ð3:35Þ
n
Given that the channel is prismatic, the normal depth is represented by a straight
line parallel to the bed along the channel, called normal depth line (NDL).
Further, the critical depth hc is determined by the implicit equation (Chap. 2)
A3
Q2 g ¼ 0 ) hðxÞ hc ) F ¼ 1: ð3:36Þ
B
Similar arguments reveal that the critical depth is also a constant along a prismatic
channel. If we plot the critical depth in the channel, it is represented by a straight
line parallel to the bed, referred to as the critical depth line (CDL).
As an example, the NDL and CDL are plotted in Fig. 3.5 for a channel with
constant Q for different values of the bottom slope. As the critical depth is inde-
pendent of the bottom slope, in the absence of cross-section variations, it remains
Fig. 3.5 Normal and critical

depth lines in a channel of
constant discharge Q for three
different values of the bottom
slope So1 < So2 < So3
[adapted from Puertas and
Sánchez (2001)]
unchanged. Note that for the cases (a) and (b), the normal depth is above the critical
depth, but for case (c), the bottom slope is large, so that the normal depth is below
the critical depth. This suggests that there is a particular value of the bottom slope
for which the normal depth exactly equals the critical depth. This value is called
critical slope Sc; it is determined as follows. The normal depth equation is
n2 Q 2
So ¼ 4=3
: ð3:37Þ
A2 Rh
Inserting the critical flow condition F = 1 produces

3
n2 g AB n2 g AB
Sc ¼ 4=3
¼ 4=3
: ð3:38Þ
A2 Rh Rh
Note that the critical slope is determined by n, the cross-sectional geometry, and the
flow depth hN = hc. As hc is a function of Q, Sc is also variable with it. However,
the critical slope Sc is independent of the actual bottom slope. The following
classification is therefore feasible:
Mild slope (M): So < Sc; hN > hc

Steep slope (S)1: So > Sc; hN < hc
Critical slope (C): So = Sc; hN = hc
Horizontal slope (H): So = 0; hN ! ∞
Adverse slope (A): So < 0; hN without physical meaning.
For classification purposes, the flow region is divided into three zones deter-
mined by the CDL and NDL (Fig. 3.6). The letters of each slope will be combined
with the numbers of the three zones to classify the possible flow profiles.
The first step in the qualitative analysis is to determine if the flow profile
increases (dh/dx > 0) or decreases (dh/dx < 0) in the flow direction. This task is
accomplished by studying the signs of the numerator (So − Sf) and denominator
ð1 F2 Þ of Eq. (3.3), which vary depending on the flow zone (1, 2, 3). If h > hN,
then U < UN and thus friction losses are less than for uniform flow, e.g., Sf < So.
Likewise, if h < hN, then Sf > So. As to the flow regime, if h > hc, then F < 1 and
the flow is subcritical. For h < hc, the flow is supercritical with F > 1.
As an example, consider the flow profile in zone 1 (h > hN > hc) for a mild
sloping channel (hN > hc) (Fig. 3.7). Given that h > hN, one gets Sf < So, Further,
the flow is subcritical F < 1. The sign of the GVF equation is thus
1
In the present context, a steep slope implies that the normal depth is below the critical depth.
However, a steep slope as used here shall be “mild” physically, that is, 1 + S2o 1. Otherwise
slope corrections are necessary in the GVF Eq. (3.3) [see Chap. 1, Eq. (1.151)]. A steep slope in
hydraulic structures implies that 1 + S2o > 1, to be discussed at the end of this chapter.
Fig. 3.6 Flow zones to

analyze gradually-varied flow
profiles
Fig. 3.7 Flow zones in mild

sloping channels [adapted
from Puertas and Sánchez
(2001)]
dh So Sf þ
¼ ¼ ¼ þ: ð3:39Þ
dx 1 F2 þ
The M1 type flow curve thus increases in the flow direction. Following the same
methodology, all the possible M-type flow profiles are summarized in Table 3.3.
The column dE/dx = So − Sf indicates the profiles with either an increase or
decrease of the specific energy.
Table 3.3 Classification of GVF profiles

Slope Zone Profile Depths So–Sf 1 F2 dh/dx
Mild 1 M1 h > hN > hc + + +
2 M2 hN > h > hc − + −
3 M3 hN > hc > h − − +
Steep 1 S1 h > hc > hN + + +
2 S2 hc > h > hN + − −
3 S3 hc > hN > h − − +
Critical 1 C1 h > hN = hc + + +
3 C3 hN= hc > h − − +
Horizontal 2 H2 h > hc − + −
3 H3 hc > h − − +
Adverse 2 A2 h > hc – + −
3 A3 hc > h − − +
Adapted from Montes (1998)
Fig. 3.8 Flow zones in steep sloping channels [adapted from Puertas and Sánchez (2001)]
For the steep slope channels, the flow zones are presented in Fig. 3.8 following the
same methodology. As example, the supercritical (F > 1) flow in zone 2 (hc > h > hN)
has Sf < So, resulting in dh/dx < 0, that is, the S2 curve is decreasing in the flow
direction. Following the same procedure, a classification of the remaining S-type
flow profiles entails no difficulty; all the possible profiles are summarized in
Table 3.3, including the critical, horizontal, and adverse flow profiles.
To plot the water surface profile, it is necessary not only to know the sign of
dh/dx in the flow zone, but also the tendency of the water surface profile toward the
reach boundaries (Jain 2001; Chaudhry 2008). The following trends prevail as the
flow depth approaches the critical depth, normal depth, or becomes very large:
dh
h ! hc ) ! 1 if So 6¼ Sf ;
dx
dh
h ! hN ) ! 0 if F 6¼ 1; ð3:40Þ
dx
dh
h!1) ! So :
dx
As the flow approaches the critical depth, the water surface slope becomes vertical.
It simply implies that the critical depth is not a mathematical solution of GVF in
prismatic channels, whereas it is for non-prismatic channels of variable slope, to be
discussed in Chap. 4. As the flow approaches the normal depth, the profile tends to
the NDL, e.g., the normal depth is an asymptotic value only. As the flow depth
becomes very large, the velocity tends to zero, as do both Sf and F. It implies
that the flow depth gradient equals So, i.e., the flow surface is horizontal,
d(h + zb)/dx = 0 (remember that So = –dzb/dx).
Fig. 3.9 M1 water surface profile
Fig. 3.10 M2 water surface

profile
Consider now the M1 type water surface profile. An ascending free surface
profile in zone 1 must approach to the left the uniform flow depth, and to the right
produce large water depths. From Eq. (3.40), the flow profile approaches in the
upstream direction asymptotically the uniform flow depth, and to the right it has a
horizontal asymptote (Fig. 3.9). For the M2 water surface profile, there is a trend to
uniform flow to the left, and to critical depth to the right, given that dh/dx < 0. The
corresponding flow profile is sketched in Fig. 3.10.
For the S2 curve in a steep slope, the flow starts close to the critical depth and
gradually approaches the uniform flow (Fig. 3.11).
Following the same methodology and combining the cases of Table 3.3 with
Eqs. (3.40), all possible free surface profiles are presented in Fig. 3.12. Several
real-life cases are presented in Fig. 3.13 following Chow (1959).
Fig. 3.11 S2 water surface profile
Fig. 3.12 Flow profiles in a mild, b steep, c horizontal, d critical, and e adverse slopes
Fig. 3.13 Real-life cases of water surface profiles [adapted from Chow (1959)]
Danny Lee Fread was born on 17 July, 1938, at Tuscola IL, passing away
aged 70 years on 5 February, 2009, at Huntington PA, USA. He worked a
plethora of jobs to pay his way through his Liberal Arts and Engineering
degrees from Carthage College, Kenosha WI, and the University of
Missouri-Rolla, Rolla MO. After working for Texaco, he returned to his
Alma Mater to continue his education, receiving the Ph.D. degree in civil
engineering in 1971. He joined the Office of Hydrology at the National
Weather Service then in Silver Spring MD as a research hydrologist. Inspired
by the tragedy of the failure of Grand Teton Dam in 1976, his research
focused on developing computer models to forecast the flow of flooding
rivers and dam failures. His computer models were used around the globe.
Fread received national awards for his work, including the Department of
Commerce Gold Medal, the Huber Research Prize from the American Society
of Civil Engineers ASCE, its 1976 J. C. Stevens Award, and the Association
of State Dam Safety Officials National Award of Merit. He also was a Fellow
of the American Meteorological Society. He ended his career as Director of
the Office of Hydrology. Following his retirement, he moved with his wife to
Pennsylvania to be near their daughter and family.
The 1973 ASCE paper presents a conceptual model to alleviate flood
damages due to overtopping failures of future small earthfill dams including
the erosion pattern. The potential reduction in the reservoir release due to the
proposed erosion-retarding layer is also investigated. A method to determine
the optimum layer location is provided to minimize the maximum possible
reservoir release due to a gradually-breached earth dam. The transient
reservoir flow is simulated by a numerical model based on the solution of the
one-dimensional Saint-Venant equations, solved by the method of charac-
teristics subjected to appropriate boundary conditions. The numerical simu-
lation provides the reduction in release discharge in terms of various
parameters.
3.4.2 Sketching of Flow Profiles
In general, a real channel presents reaches of different lengths, slopes, sections, and
roughness. The actual free surface profile will be a composite surface assembling
some of the elementary profiles depicted in Fig. 3.12. Before computing profiles
numerically, the sketching process is as follows (Chow 1959; Jain 2001; Chaudhry
2008):
1. Plot the bottom profile (use distorted scale in the vertical) and assign roughness
values to each reach.
2. Compute hN for each channel reach using Eq. (3.35) and plot the NDL.
3. Compute hc for each channel reach using Eq. (3.36) and plot the CDL.
4. Locate all possible control sections, namely a critical depth control (CDC),
normal depth control (NDC), and artificial channel control (ACC) (e.g., a weir
or a gate). In a CDC, the flow is subcritical upstream and supercritical down-
stream. The probable position of a NDC is upstream of the reach in subcritical
flow and downstream of the reach in supercritical flow. A NDC is an asymptotic
state. In an ACC, the flow is subcritical upstream of it and supercritical
downstream.
5. Starting at each control section, draw sub- and supercritical profiles in the
up- and downstream directions, respectively. Check possible interactions of
control sections.
6. If the flow is supercritical upstream of the reach and subcritical downstream, a
hydraulic jump may form within the reach. The jump may also not be formed,
resulting in fully sub- or supercritical flow along the reach. The position of the
jump can be plotted tentatively, though its formation will be discussed in depth
in Chap. 4.
As an example, two illustrative cases presented by Subramanya (1986) are
plotted in Fig. 3.14. In case (a), the three slopes of the channel are mild. Note that
the values of the NDL for each reach are above the CDL. The gate on the reach c–e
is an ACC that produces subcritical flow upstream and supercritical flow down-
stream. At the free overfall (section e), the critical depth is attained at a distance of
three to four times hc from the brink section (Rouse 1938). Given that this distance
is small, the critical depth control section is assumed to be established just at the
brink in GVF computations. Thus, the brink section is a control section interacting
with the vena contracta section (point d), which is the control section induced by the
gate. From the vena contracta, a supercritical M3 profile develops, whereas the
brink section will force a subcritical M2 profile. To have compatibility of both
profiles, a hydraulic jump is formed within the reach. Cases where the gate is
drowned or the jump rejected out of the reach will be discussed in Chap. 4. At point
c, the flow is subcritical as forced by the gate. Starting at this point, a M1 subcritical
profile is formed in the reach b–c. If this reach is long, as assumed here, the normal
depth will be asymptotically approached and established at point b. From point b,
an M2 profile is formed up to the upstream section a.
Fig. 3.14 Examples of GVF profiles with various controls [adapted from Subramanya (1986)]
In case (b), a mild slope is followed by two steep slopes. Note again the values
of the NDL as compared with the CDL. In the absence of any other control, it is
assumed that the normal depth (NDC) is established upstream in reach a–b, and
downstream in reach b–c. As the flow in the mild slope reach is subcritical, and
supercritical on the steep slope, a CDC is established, in this case just at the slope
break (point b). The flow profile upstream of b is therefore an M2 profile, whereas
downstream of it, an S2 profile develops. At point c, a NDC was assumed, so that
an S3 profile is formed within reach c–d.
3.4.3 Flow Profile for Unknown Discharge
The sketching methodology for the free surface profile previously explained is
based on the assumption of known discharge. However, sometimes, the discharge is
unknown and must be determined as part of the solution itself. The case of a
channel releasing water from a reservoir is an example (Fig. 3.15). Consider a
constant slope channel of rectangular cross-section of width b. The basic known
magnitude is the water surface elevation at the reservoir. If the elevation of the
channel invert intake is known, then the water depth at the channel inlet H is
available. In the forthcoming analysis, the velocity head in the reservoir and the
intake entrance losses are neglected.
The problem is thus to determine the discharge Q released from the reservoir by
the channel. As the discharge is unknown, it is not possible to compute hN and hc,
which are needed to determine the type of control section of the flow profile. Thus,
the solution must be conducted by iteration. Assume first that the channel is steep
(Fig. 3.15a). In this case, the critical depth is given by
2
hc ¼ H; ð3:41Þ
3
which is established at the channel inlet, with the released discharge given by
3=2
2 3 1=2
Q¼b gH : ð3:42Þ
3
Now, the steep slope assumption shall be checked. Therefore, the critical slope Sc is
computed by resorting to Eq. (3.38), and compared to the channel slope So. If
So > Sc, then the initial assumption is correct, the channel is thus steep and com-
putations are finished. The normal depth is not needed to compute Q, but it can be
determined from Manning’s equation to have the complete solution to the flow
conditions. If So < Sc, the channel is not steep, and computations need to be
reconsidered. For a mild slope channel (Fig. 3.15b), uniform flow is established
assuming that the channel is long. In this case, a normal depth control governs the
channel intake, resulting in the energy equation
Fig. 3.15 Discharge released from a reservoir for a mild and b steep slope channels
Q2
H ¼ hN þ : ð3:43Þ
2gb2 h2N
Using Manning’s equation to evaluate Q, Eq. (3.43) is rewritten as

4=3
So 4=3 So bhN
H ¼ hN þ R ¼ h N þ : ð3:44Þ
2gn2 h 2gn2 b þ 2hN
This is a nonlinear implicit equation to be solved iteratively to determine hN, for

example, by using the Newton–Raphson method. Once this task is accomplished,
the discharge is
2=3
S1=2 2=3 S1=2 bhN
Q¼ o
ARh ¼ o
ðbhN Þ ; ð3:45Þ
n n b þ 2hN
and the critical depth

1=3
Q2
hc ¼ : ð3:46Þ
gb2
Resorting to Eq. (3.38), the critical slope Sc can be updated for the correct discharge
Q. This solution process was implemented in the file Lake_discharge.xls available
from Chap. 12. As application example, consider a rectangular channel of b = 4 m,
n = 0.013 sm−1/3 releasing water from a reservoir with H = 1 m. For a slope
So = 0.01, the program yields hc = (2/3) m, Q = 6.82 m3/s, hN = 0.438 m and
Sc = 0.00278. The channel is therefore steep and the control section is a CDC at the
channel inlet (Fig. 3.15a). If the slope is So = 0.0001, a comparison with Sc shows
that the steep slope assumption is incorrect, and computations are conducted again
assuming mild slope with a NDC at the inlet. The program then yields the output
hc = 0.322 m, Q = 2.289 m3/s, hN = 0.983 m and Sc = 0.00295. Given that
0.0001 < 0.00295, the mild slope assumption is correct.
3.5 Computation of Steady Flow Profiles
3.5.1 Governing Integral and Differential Equations
The GVF equation is a first-order ODE describing the variation of h(x), written as
dh So Sf
¼ ¼ f ðhÞ: ð3:47Þ
dx 1 F2
For a prismatic channel So is a constant, whereas both Sf and F are functions of

h alone. For GVF computations, the friction slope Sf is determined from the pre-
viously explained formulae for uniform flow. In general, the f function depends on
the cross-sectional shape and the resistance law adopted. The values of So, n and
Q are assumed to be known for the solution. Unfortunately, there is not an exact
analytical solution of Eq. (3.47), given that f is a complex nonlinear function, so
that numerical methods are necessary.
The GVF equation for non-prismatic channels involving variations in bottom
slope and cross-sectional shape is [Eq. (1.107)]

Q2 @A 2
dh So Sf þ gA3 @x h¼const: So Sf þ FB @A @x h¼const:
¼ ¼ ; ð3:48Þ
dx Q2
1 gA 3 B 1 F2
which is rewritten as
dh
¼ f ðx; hÞ: ð3:49Þ
dx
For the sake of generality in the presentation of numerical methods, consider

Eq. (3.49) despite our function will be f = f(h) for prismatic channels. As Eq. (3.49)
is a first-order ODE, any numerical method to solve this kind of equation can be
adopted in principle. A variety of such techniques will be presented in this chapter.
There is an alternative method, however. Equation (3.49) is the differential form
of the energy (or momentum) equation in GVF. It is thus possible to use an integral
rather than a differential equation. The energy equation in integral form reads
[Eq. (2.11)]
Zx2
E2 E1 ¼ ðSo Se Þdx; ð3:50Þ
x1
where the specific energy is E = h + Q2/(2gA2). Equation (3.50) is applicable to

prismatic and non-prismatic channels. For a constant slope reach, Eq. (3.50)
produces
Zx2
E2 ¼ E1 þ So ðx2 x1 Þ Se dx: ð3:51Þ
x1
To apply Eq. (3.51), the integral of the gradient of dissipated energy (or energy
slope) is numerically approximated. Techniques to solve Eq. (3.49) or its integral
form, Eq. (3.50), will be presented in the next sections.
3.5 Computation of Steady Flow Profiles 161
3.5.2 Boundary Condition
The GVF equation is a first-order ODE subject to a boundary condition h(x =

xo) = ho, e.g., a position where the flow depth is known for a given discharge Q,
i.e., (Fig. 3.16)
dh
¼ f ðx; hÞ;
dx ð3:52Þ
hðx ¼ xo Þ ¼ ho :
The boundary condition is usually a control section, corresponding to a section

where the function h = h(Q) is available. Typical control sections in open channel
flow are a critical flow section, a uniform flow section, or an artificial channel
control, as a weir or a gate. Once the control section is identified, and ho determined
for Q, the nature of the control section determines the direction of integration of
Eq. (3.49), as briefly outlined while explaining the sketching process. The usual
case is to integrate a subcritical flow profile in the upstream direction, and a
supercritical flow profile in the downstream direction, as will be discussed. It should
be remarked that Eq. (3.49) is simply an ODE which can be integrated into both the
up- and downstream directions regardless of the value of F (Apelt 1971). However,
numerical errors are sensitive to the integration direction and care should be taken
(Fread and Harbaugh 1971; Jain 2001). The direction of integration is simply
marked by the position of the control section that is usually known in practice. We
illustrate this point by taking the critical flow section as an example. Consider the
transition from a mild to a steep slope (Fig. 3.17). The critical depth is a constant,
but there are two different normal depths, one for each value of the slope such that
hN2 > hc > hN1. The flow tends asymptotically to uniform flow far up- and
downstream from the slope transition, namely for x ! ± ∞. To pass from the
upstream uniform subcritical flow to the downstream uniform supercritical flow, a
varied flow profile is formed crossing eventually the critical depth. While in the
vicinity of the slope break the flow is curvilinear and the critical depth is formed
upstream of the break at a distance close to 3hc (Rouse 1938), this non-hydrostatic
transition is short and the critical section is taken at the slope break for GVF
Fig. 3.16 Boundary

condition to solve the GVF
equation
Fig. 3.17 Control and gauging sections in transition from mild to steep slope
computations. The critical flow section is thus a channel control producing

upstream subcritical flow and downstream supercritical flow. Therefore, the
directions of integration are in the up- and downstream directions for the sub- and
supercritical profiles, respectively, starting at a CDC. Note that at the critical depth
dh/dx ! ∞. Therefore, the CDC is taken in practice as
hðx ¼ xo Þ ¼ hc e; ð3:53Þ
where e is a small number. The inverse transition from super- to subcritical flow
occurs in the form of a hydraulic jump, and thus is not a control section.
From this discussion follows that the “usual” position of the control section is at
the downstream end in a subcritical reach. Likewise, the usual position of a control
section in a supercritical flow reach is at its upstream end. Accordingly, the GVF
computational rule of determining the profiles is in the upstream direction for
subcritical flow and in the downstream direction for supercritical flows. This is in
agreement with the physical fact that perturbations travel both up- and downstream
for subcritical flow, thus being influenced by what occurs at the end of the reach,
whereas supercritical flows are only influenced by perturbations induced upstream
of the reach. Subcritical flows are thus subject to a downstream control, whereas
supercritical flows to an upstream control. This will be further discussed once
unsteady flow is introduced in Chap. 5.
Consider now the setup of Fig. 3.17 in a laboratory facility. An experimentalist
may take a measurement at the gauging section xo and then get a known ho for Q,
which is different from the standard control section at the slope break. This section
could be used to integrate Eq. (3.49) in the up- and downstream directions, by
which a portion of the subcritical profile would be computed in the downstream
Fig. 3.18 Flow conditions

induced by artificial channel
controls (ACC) a weir, b gate
direction. This process is mathematically correct; it can be used instead of the

classical rule, if data are available.
A weir or a gate (Fig. 3.18) is an artificial channel control (ACC) producing
upstream subcritical flow (F < 1) and downstream supercritical flow (F > 1). The
normal depth is reached asymptotically at the upstream end in a subcritical reach
and at the downstream end in a supercritical reach. At the uniform depth
dh/dx ! 0, so that the NDC is taken in practice with e as a small number as
hðx ¼ xo Þ ¼ hN e: ð3:54Þ
3.5.3 Analytical Solution
Analytical solutions to gradually-varied flow problems are rare. A notable excep-

tion is the solution due to Bresse (1860) for the wide rectangular channel. In
addition to its historical interest, the main purpose of presenting such a solution is
that it is exact; therefore, it can be used to examine the accuracy of numerical
schemes. Using Chezy’s resistance coefficient C, Eq. (3.3) can be written for the
rectangular channel as
Q2 Q2
So S o
dh C 2 A2 Rh C 2 b2 h2 Rh
¼ 2
¼ : ð3:55Þ
dx Q Q2
1 3B 1 2 3
gA gb h
Assuming that the channel is very wide, the hydraulic radius is approximated by the
flow depth, resulting in the GVF equation
Q2 Q2 h3
So 1 1 N3
dh 2 2 3
C b h ¼S 2 2
So C b h 3
h :
¼ o ¼ So ð3:56Þ
dx Q2 Q2 h3
1 2 3 1 2 3 1 c3
gb h gb h h
In Eq. (3.56), hN and hc are the normal and critical depths, given, respectively, by
1=3
Q2
hN ¼ ;
So C 2 b2
2 1=3 ð3:57Þ
Q
hc ¼ :
gb2
Equation (3.56) can be integrated numerically using various methods. However, it

has an exact analytical solution, to be used to check the accuracy of any scheme, as
already stated. A dimensionless flow depth u = h/hN is defined. Inserting it into
Eq. (3.56) gives
du So 1 u3
¼ ; ð3:58Þ
dx hN 1 nu3
where n = (hc/hN)3. Separating the variables permits to write

So 1 nu3 1 u3
dx ¼ du ¼ n du
hN 1 u3 1 u3 1 u3
3 3
u 3 u3 u 1
¼ nu du ¼ n 3 du ð3:59Þ
u3 1 u3 1 u3 1 u 1

1 1 1n
¼ 1þ 3 n 3 du ¼ 1 þ 3 du:
u 1 u 1 u 1
Integrating the result produces

Zx Zu
So 1n
dx ¼ 1 du; ð3:60Þ
hN 1 u3
0 0
that is,
Zu
So 1
x ¼ u ð1 nÞ du ¼ u ð1 nÞUðuÞ; ð3:61Þ
hN 1 u3
0
where U(u) is the integral
Zu
1
UðuÞ ¼ du: ð3:62Þ
1 u3
0
Equation (3.61) can be recast in the more usual form

hN h3 hðxÞ
x¼ u 1 3c UðuÞ ; u¼ ; ð3:63Þ
So hN hN
where U(u) was first obtained by Bresse (1860, p. 248) as2

" #
1 u2 þ u þ 1 1 1 2u þ 1
UðuÞ ¼ ln þ tan þ C1 ; ð3:64Þ
6 ðu 1Þ2 31=2 31=2
or in the alternative form (Bresse 1860, p. 249) quoted by Chow (1959, p. 258) and
Jaeger (1956, p. 72)
" #
1 u2 þ u þ 1 1 1 2u þ 1
UðuÞ ¼ ln cot þ C1 : ð3:65Þ
6 ðu 1Þ2 31=2 31=2
The term C1 in Eq. (3.64) is an arbitrary constant. Equation (3.64) is plotted in

Fig. 3.19 for C1 = 0. Note the singularity at u = 1, implying that the uniform flow
is not a possible value. The specific value of C1 = 0 is of no concern, since
Eq. (3.63) is applied in practice between two successive sections xi+1 and xi, where
the flow depths are hi+1 and hi, respectively, such that the constant is eliminated,
e.g.,

hN h3
xi þ 1 ¼ xi þ ðui þ 1 ui Þ 1 3c ðUi þ 1 Ui Þ : ð3:66Þ
So hN
Applying Eq. (3.66) involves an inverse computation: starting with the known
point (xi, hi) an increment in the water depth is defined such that
hi+1 = hi ± Δh. With the known values of normalized water depths ui and ui+1,
U(u) is evaluated at both points, thereby applying Eq. (3.66) to obtain the coor-
dinate xi+1 where the depth is hi+1. This process is conducted as many times as
2
The readerhis warned i that Bresse’s
solution
appeared mistyped in many publications in the form
þuþ1
tan1 2u3 þ 1 þ C1 , which is obviously not the solution. Jan (2014,
2 1=2
UðuÞ ¼ 16 ln u ðu1Þ2 31=2
1
p. 24) detailed the integration process step-by-step. The typo appears to originate from
C. J. Posey in Rouse (1950, p. 613).
Fig. 3.19 Bresse’s (1860)

function for GVF profiles
desired to get the free surface profile. Note that although the computational
sequence looks like a numerical scheme, it is not. The solution is analytical and
therefore the value of Δh arbitrary, since Eq. (3.66) is only a relation between two
analytical points of the flow profile.
3.5.4 Implicit Integral Method
In this section, a numerical method based on the integral form of the energy
balance, namely Eq. (3.51), is presented. The technique is called standard step
method (Chow 1959) and consists in dividing the channel reach into a number of
segments with their boundaries located at selected positions, and then solve the
algebraic equation for the unknown flow depth at one of the extremes. This tech-
nique is widely used in river hydraulics, and it is implemented in program
HEC-RAS (Brunner 2016). Here, we follow the presentation by Fread and
Harbaugh (1971). Consider a channel reach divided for simplicity’s sake into
segments of equal length Dx (Fig. 3.20).
Fig. 3.20 Energy balance in gradually-varied flow (subcritical flow case)

Using the trapezoidal rule to evaluate the integral of the friction slope, one gets
xZþ Dx
1
Sf dx ¼ Sf 1 þ Sf 2 Dx; ð3:67Þ
2
x
which permits to write Eq. (3.51) as
1
E1 ¼ E2 So Dx þ ðSf 1 þ Sf 2 ÞDx; ð3:68Þ
2
or, in the alternative form
Q2 Q2 1
zb1 þ h1 þ 2
¼ zb2 þ h2 þ þ ðSf 1 þ Sf 2 ÞDx; ð3:69Þ
2gA1 2gA22 2
demonstrating the energy balance between Sections 1 and 2. Other approximations

to the average friction slope are discussed by Montes (1998). Consider the case of
computing a subcritical flow profile (F < 1). The probable location of the control
section is downstream of the reach, so flow conditions at point 2 are fully deter-
mined by a former computation. Therefore, flow conditions at point 1 are to be
determined. The problem consists of solving the nonlinear implicit equation (Fread
and Harbaugh 1971)
Q2 Dx n2 Q2
f ðh1 Þ h1 þ þ F ¼ 0; ð3:70Þ
2gA21 2 A2 R4=3
1 h1
where Manning’s equation was used and F is the known function
Q2 Dx n2 Q2
F ¼ So Dx h2 : ð3:71Þ
2gA22 2 A2 R4=3
2 h2
The solution of Eq. (3.70) is conducted here using the Newton–Raphson method as
fk
hk1 þ 1 ¼ hk1 ; ð3:72Þ
ðdf =dh1 Þk
with k as the iteration index. Since Dx is usually small, the initial value of h1 is
taken as h2. For the computation of a supercritical flow profile (F > 1), the upstream
water depth is h2 and the unknown water depth h1 is located at the downstream
boundary. The equations presented are thus applied just taking a negative value of
Dx to conduct computations in the flow direction.
!
df d Q2 Dx n2 Q2 d 1
¼ h1 þ 2
þF ¼ E1 DxSf 1 : ð3:73Þ
dh1 dh1 2gA1 2 A2 R4=3 dh1 2
1 h1
The derivative of E is
dE1 Q2
¼ 1 F21 ¼ 1 B1 3 ; ð3:74Þ
dh1 gA1
and one gets from Sf1

! !
dSf 1 d n2 Q 2 2 dA1 4 dR1
¼ ¼n Q
2 2
1 þ 4=3 dh
dh1 dh1 A2 R4=3 4=3
A31 Rh1 dh1 3A21 Rh1 1
1 h1
2 2

n Q 2B1 4 dR1
¼ 4=3

2
A1 Rh1 A1 3R h1 dh1

¼ Sf 1
2B1

4 1 dA1 A1 dp1
ð3:75Þ
A1 3Rh1 p1 dh1 p21 dh1

2B1 4 1 dA1 1 dp1
¼ Sf 1
A1 3 A1 dh1 p1 dh1

10B1 4 1 dp1
¼ Sf 1 þ :
3A1 3 p1 dh1
Inserting Eqs. (3.74) and (3.75) into Eq. (3.73), the final result is

df Q2 1 10B1 4 1 dp1
¼ 1 B1 3 DxSf 1 þ : ð3:76Þ
dh1 gA1 2 3A1 3 p1 dh1
The method can be used directly in non-prismatic channels with variations of

bottom slope and cross-sectional shape. If adequate flow resistance predictors and
energy velocity correction coefficients are introduced, it applies even to compound
channels (Brunner 2016). It permits also an easy incorporation of form losses.
3.5.5 Explicit Integral Method
For convenience, consider again Eq. (3.68), rewritten as
1
E2 E1 ¼ So Dx Sf 1 þ Sf 2 Dx: ð3:77Þ
2
Starting with the known point (xi, hi), an increment in the water depth is defined
such that hi+1 = hi ± Δh. Assume that E = E(h) and Sf = Sf(h) only, i.e., the
channel is prismatic. Thus, one can write
E2 E1
Dx ¼ x2 x1 ¼ : ð3:78Þ
So 12 Sf 1 þ Sf 2
From this equation, we can compute the position x2 where the depth will be h2,
given that the right-hand side of Eq. (3.78) is only a function of h1 and h2. This
method is called direct step method and was originally devised by Bélanger (1828).
A limitation is that it can only be used for prismatic channels, and that it does not
permit to get the solution for the flow depth at selected locations, which is the usual
case in practice. Thus, aside from its academic interest, is not a practical method.
3.5.6 Euler’s Method
For the solution of the GVF-ODE [Eq. (3.49)], the computational domain is divided
into a number of reaches of length Dx (Fig. 3.21). If the spatial derivative of the
flow depth is discretized using a forward finite-difference
dh hi þ 1 hi
¼ ; ð3:79Þ
dx Dx
Fig. 3.21 Discretization of

computational domain for
solution of GVF-ODE
Fig. 3.22 Sketch of the Euler

scheme
the algebraic expression approximating Eq. (3.49) reads (Chaudhry 2008)

So Sf
hi þ 1 ¼ hi þ f ðxi ; hi ÞDx ¼ hi þ Dx: ð3:80Þ
1 F2 i
Thus, the flow solution is computed on the basis of the tangent to the known point
(xi, hi) (Fig. 3.22). This equation has the advantage of being extremely simple to
apply, but the truncation error is O(Dx2), and it can destroy the solution if Dx is
large. It is noted that we are including in the solution only the terms of power Dx of
a Taylor series, e.g.,
2 3
dh dh 1 2 d h 1
hi þ 1 ¼ hi þ Dx þ 2
ðDxÞ þ ðDxÞ3 þ : ð3:81Þ
dx i dx i 2 dx3 i 6
The scheme is therefore first-order accurate only. It applies in practice, but extreme
care is needed to avoid an unacceptable growth of the numerical error in the
solution.
3.5.7 Fourth-Order Runge–Kutta Method
The family of Runge–Kutta schemes considers the Taylor series Eq. (3.81), sub-
stituting the use of derivatives of order higher than one by more evaluations of the
first derivative within the interval. These schemes use an updating explicit formula
of type
hi þ 1 ¼ hi þ f Dx; ð3:82Þ
where the average value of f is determined differently, depending on the order of the
scheme, and, thus, of the number of terms considered in the Taylor series.
A fourth-order scheme yields (Apelt 1971; Chaudhry 2008)
1
f ¼ ðk1 þ 2k2 þ 2k3 þ k4 Þ; ð3:83Þ
6
with
k1 ¼ f ðxi ; hi Þ;

1 1
k2 ¼ f xi þ Dx; hi þ k1 Dx ;
2 2
ð3:84Þ
1 1
k3 ¼ f xi þ Dx; hi þ k2 Dx ;
2 2
k4 ¼ f ðxi þ Dx; hi þ k3 DxÞ:
The computational process is depicted in Fig. 3.23, where the four evaluations of
the slope are seen. The truncation error of the scheme is O(Dx5) (Apelt 1971). This
scheme is suitable for open channel flows (Katopodes 2019). Apelt (1971) found
that this is an accurate and stable method to integrate Eq. (3.49) in the up- and
downstream directions regardless of whether the flow is subcritical or supercritical.
Equation (3.80) is simply the first-order Runge–Kutta method.
Fig. 3.23 Fourth-order

Runge–Kutta scheme
[adapted from Chapra and
Canale (2010)]
3.5.8 Predictor–Corrector Method
Another relevant method uses the integral form of the GVF-ODE, e.g.,
Zxi þ 1
hi þ 1 ¼ hi þ f ðx; hÞdx: ð3:85Þ
xi
The problem is to evaluate the integral numerically. Using the trapezoidal rule, one
gets the implicit equation in the depth hi+1 as (Apelt 1971; Chaudhry 2008)
1
hi þ 1 ¼ hi þ Dx½f ðxi ; hi Þ þ f ðxi þ 1 ; hi þ 1 Þ: ð3:86Þ
2
The following predictor–corrector scheme is therefore feasible (Apelt 1971):

Predictor step (explicit)
The predictor step is given by the Euler method, that is,
h0i þ 1 ¼ hi þ Dx f ðxi ; hi Þ: ð3:87Þ
This value is used to initiate the iterative corrector cycle, described below.
Corrector step (iterative)

With k as the recursion index, Eq. (3.86) can be written as the iterative formula
1
hki þþ11 ¼ hi þ Dx f ðxi ; hi Þ þ f xi þ 1 ; hkiþ 1 : ð3:88Þ
2
Iteration of hi+1 in the corrector cycle is stopped once the solution accuracy is
within a prescribed tolerance for two successive values of the iterated flow depths,
that is,

kþ1

hi þ 1 hkiþ 1

106 : ð3:89Þ

hkiþ 1
This predictor–corrector method is known as the trapezoidal method, and its

truncation error is O(Dx3) (Apelt 1971).
3.6 Applications 173
3.6 Applications
3.6.1 Test of High-Resolution Method
In this section, the fourth-order Runge–Kutta scheme is evaluated using Bresse’s

analytical solution. Thus, a wide rectangular channel is adopted and the flow in a
transition from mild to steep slope is used as test case (Fig. 3.17). The fourth-order
Runge–Kutta scheme for this test is implemented in the file
Bresse_RungeKutta4order.xls available from Chap. 12, whereas Bresse’s analytical
solution is available on the file Bresse_analytical.xls.
As application example consider a rectangular channel of width b = 5 m and
C = 100 sm−1/2 discharging Q = 15 m3/s on a slope So = 0.0001. An increase in the
slope to So = 0.1 is introduced abruptly. The critical depth is hc = 0.972 m,
whereas the normal depths are 2.08 m and 0.208 m for the mild and steep reaches,
respectively. Sub- and supercritical profiles were computed in the up- and down-
stream directions, respectively, using the fourth-order Runge–Kutta scheme with
Dx = 0.5 m. The boundary conditions at the slope break (x = 0 m) are taken
slightly above and below hc for the sub- and supercritical profiles, namely 1 m and
0.93 m, respectively. The results are compared in Fig. 3.24 with Bresse’s analytical
solution, showing an excellent match.
3.6.2 Comparative of Numerical Methods
Once the accuracy of the fourth-order Runge–Kutta scheme is established, it is used

in this section to produce reference solutions for channels of arbitrary Rh to test
other numerical schemes.
Consider as test case a rectangular channel of b = 2 m and n = 0.015 sm−1/3
discharging Q = 15 m3/s on a slope So = 0.001. The critical and normal depths are
Fig. 3.24 Evaluation of fourth-order Runge–Kutta scheme in transition from mild to steep slope
Fig. 3.25 Comparison of various numerical schemes to solve the GVF-ODE using an M2 profile
hc = 1.79 m and hN = 4.11 m. An M2 profile will be used to test various numerical

schemes taking as boundary condition ho = 1.85 m at xo = 0 m. The fourth-order
Runge–Kutta scheme was implemented in the file watersurfaceprofiles_Runge
Kutta4order.xls available from Chap. 12. It is compared in Fig. 3.25a with the
trapezoidal rule, available on the file watersurfaceprofiles_trapezoidalrule.xls,
using Dx = −0.001 m. Note that the match of the two schemes is excellent.
A comparison of the fourth-order Runge–Kutta scheme with Euler’s scheme,
available on the file watersurfaceprofiles_Euler.xls, is accomplished in Fig. 3.25b,
showing likewise a very good match. As previously stated, Euler’s scheme is
simple, producing good results if the truncation error is controlled with a small
Dx. The test was repeated using Dx = −1.5 m, with the corresponding results for
the trapezoidal rule and Euler’s schemes presented in Fig. 3.25c and d, respectively.
Note that the trapezoidal rule method is still producing good results, whereas
Euler’s scheme has notable deviations. This is due to the truncation error, which is
O(Dx2) for Euler’s scheme but O(Dx3) for the trapezoidal rule method.
The fourth-order Runge–Kutta scheme is compared with the standard step
method, available on the file Flowprofiles_NR.xls, in Fig. 3.26a using Dx = −1.5 m
for the latter. Note that there is an excellent match without any appreciable devi-
ation. The test was repeated using Dx = −2.5 m in Fig. 3.26b, and deviations are
Fig. 3.26 Evaluation of standard step method using an M2 profile
still small. This is one advantage of the standard step method, namely the possibility
of using larger space steps given its implicit character.
3.6.3 Flow on Mild Slope
The fourth-order Runge–Kutta scheme is compared in this section with experi-

mental data for mild slope channels (Lee et al. 1952). The test flume was rectan-
gular with b = 0.2032 m, So = 0.001, and n = 0.0092 sm−1/3. A subcritical M1
profile for Q = 0.004984 m3/s was numerically computed using as boundary
condition the experimental point (xo = 87.7824 m, ho = 0.14478 m). The profile
was computed from this section in the upstream direction using Dx = –0.05 m in
the Runge–Kutta solver. A comparison with experiments by Lee et al. (1952) in
Fig. 3.27a shows excellent agreement. A subcritical M2 profile for
Q = 0.009798 m3/s was computed using as boundary condition the experimental
point (xo = 34.1376 m, ho = 0.08128 m) with Dx = –0.05 m. A comparison with
experiments in Fig. 3.26b shows again a very good agreement.
3.6.4 Flow on Steep Slope
The GVF equation for a rectangular channel of steep slope is [Eq. (1.151)]
2
2 2
dh So Cf F2 ð1 þ S2o Þ2 So Cf gh3 ð1 þ So Þ
q
¼ ¼ ; ð3:90Þ
1 F2 ð1 þ S2o Þ
2
dx 1 gh
q
3 ð1 þ So Þ
2
where Cf is the skin friction coefficient. Note that under the small slope assumption
1 + S2o 1, so that Eq. (3.90) simplifies to Eq. (3.47). A steep physical slope is
Fig. 3.27 Evaluation of

fourth-order Runge–Kutta
scheme with experimental
data on mild slopes for the
a M1 curve, b M2 curve
thus different from a “hydraulically steep slope” as given by comparing with

Eq. (3.38). The main feature of a steep slope 1 + S2o > 1 is that the bottom fluid
pressure becomes non-hydrostatic, given by [Eq. (1.136)]
pb h
¼ : ð3:91Þ
qg 1 þ S2o
The test data of Hasumi (1931) for slopes of So = 1 and 1.732 were used to
check predictions by Eqs. (3.90) and (3.91). Equation (3.90) was solved with the
fourth-order Runge–Kutta scheme, available from the file steepslope_Runge
Kutta4order.xls in Chap. 12. The slope transition is composed of a horizontal reach
followed by a circular-shaped transition profile of R = 0.1 m that finishes in the
steep slope reach of 45° or 60° inclination angles. The channel width is
b = 0.402 m and the discharge Q = 0.04 m3/s. A value of Cf = 0.001 was used in
the simulations.
The first test for So = 1 is presented in Fig. 3.28a, where the boundary condition
taken to integrate Eq. (3.90) was the experimental point at the start of the flat
slope reach (xo = 0.070711 m, ho = 0.075 m). The elevation at this point is
zo = –0.02929 m. The computed free surface profile is compared with the
Fig. 3.28 Comparison of

gradually-varied flow theory
on steep slope with
experiments for So = a 1,
b 1.732
experiments by Hasumi (1931), showing a fair agreement down the slope. Once the
free surface is determined, the bottom pressure head was computed from Eq. (3.91),
showing again good agreement with experiments down the slope. The second test
for So = 1.732 is presented in Fig. 3.28b, with the boundary condition set as the
experimental point at the start of the flat slope reach (xo = 0.086602 m, ho =
0.081 m, zo = –0.05 m). A comparison of numerical predictions and experimental
observations in Fig. 3.28b shows again a good agreement down the slope.
The importance of the mild slope assumption 1 + S2o 1 is tested in Fig. 3.29,
where the tests are repeated this time including the solution of Eq. (3.90) neglecting
the slope corrections. The effect of the correction on the free surface for the 45°
slope is moderate (Fig. 3.29a), but obviously, the bottom pressure is poorly pre-
dicted, given that for the mild slope case it is
pb
¼ h: ð3:92Þ
qg
Fig. 3.29 Evaluation of

slope correction effect on
the free surface profile for
So = a 1, b 1.732
Bottom pressures are not considered in Fig. 3.29, therefore. The impact of the slope
correction on the free surface for the 60° slope is depicted in Fig. 3.29b, showing in
that case important deviations.
3.7 Compound Channels
In a compound channel, the cross-section is composed of a main channel and a

flood plain (Fig. 3.30). For low flows, the main channel operates and coveys all
discharge, whereas an overbank flow occurs for high floods and the flood plain
transports part of the stream. The roughness values are different in the main channel
and the flood plain, thereby inducing a lateral variation of the depth-averaged
velocity. Dividing the cross-section into N subsections of area Ai, each with a
different value of the roughness coefficient, and using Manning’s equation, one can
define the following velocity for each subsection
3.7 Compound Channels 179
Fig. 3.30 Compound channel flow
1=2
Sf 2=3
Ui ¼ Rhi : ð3:93Þ
ni
To compute the hydraulic radius of a subsection, only solid perimeters are con-
sidered. The free surface profile is determined from the integral energy balance as
(Jain 2001)
Q2 Q2 1
zb1 þ h1 þ a1 ¼ z b2 þ h 2 þ a 2 þ Sf 1 þ Sf 2 Dx; ð3:94Þ
2gA21 2gA22 2
where the friction slope Sf and Coriolis coefficient a shall be determined at each
section. The total discharge in the cross-section is (Chow 1959; Jain 2001)
X
N
1=2
X
N
1=2
Q¼ Ui Ai ¼ Sf Ki ¼ KSf ; ð3:95Þ
i¼1 i¼1
where
1 2=3
Ki ¼ Ai Rhi : ð3:96Þ
ni
Equation (3.95) permits to express Sf as a function of Q and flow depth in each

sub-section. Using Eq. (3.93), the energy velocity correction coefficient is (Sturm
2001; Jain 2001)
R PN 2 N
u3 dA i¼1 Ai
X K3
a¼ 3
A
¼ P 3
i
2
: ð3:97Þ
U A N
Ki i¼1 i A
i¼1
Inserting Eqs. (3.95) and (3.97) in Eq. (3.94), it can be solved by the standard step
method (Jain 2001). The determination of control sections, especially critical depth
controls, is a complex task, given the multiple critical depths in the specific energy
diagram (Sturm 2001). Further, various theoretically possible free surface profiles
in compound channel flow are not yet experimentally verified.
References
Apelt, C. J. (1971). Numerical integration of the equation of gradually varied and spatially varied
flow. In: Proceedings of 4th Australasian Conference on Hydraulics and Fluid Mechanics,
Monash University, Melbourne, Australia (pp. 146–153).
Bakhmeteff, B. A. (1912). O нepaвнoмepнoм движeнии жидкocти в oткpытoм pycлe [Varied
flow of liquids in open channels]. St Petersburg, Russia (in Russian).
Bakhmeteff, B. A. (1932). Hydraulics of open channels. New York: McGraw-Hill.
Bélanger, J. B. (1828). Essai sur la solution numérique de quelques problèmes relatifs au
mouvement permanent des eaux courantes [On the numerical solution of some steady water
flow problems]. Carilian-Goeury, Paris (in French).
Bresse, J. (1860). Cours de mécanique appliquée, 2ème partie: Hydraulique [Lecture notes on
applied mechanics, Part 2, Hydraulics]. Paris: Mallet-Bachelier (in French).
Brunner, G. W. (2016). HEC-RAS River Analysis System Hydraulic Reference Manual. Report
CPD69, Version 5.0, US Army Corps of Engineers, Hydrologic Engineering Center (HEC),
Davis, CA.
Chapra, S. C., & Canale, R. P. (2010). Numerical methods for engineers (6th ed.). New York:
McGraw-Hill.
Chaudhry, M. H. (2008). Open-channel flow (2nd ed.). New York: Springer.
Fenton, J. D. (2010). Calculating resistance to flow in open channels. Technical Report,
Alternative Hydraulics Paper 2, Vienna University of Technology, Vienna, Austria.
Fread, D. L., & Harbaugh, T. E. (1971). Open-channel profiles by Newton’s iteration technique.
Journal of Hydrology, 13, 78–80.
Hager, W. H. (2010). Wastewater hydraulics: Theory and practice (2nd ed.). Berlin: Springer.
Hasumi, M. (1931). Untersuchungen über die Verteilung der hydrostatischen Drücke an
Wehrkronen und -Rücken von Überfallwehren infolge des abstürzenden Wassers [Studies on
the distribution of hydrostatic pressure distributions at overflows due to water flow]. Journal of
the Department of Agriculture, Kyushu Imperial University 3(4), 1–97 (in German).
Henderson, F. M. (1966). Open channel flow. New York: MacMillan Co.
Marcel Dekker.
Jan, C.-D. (2014). Gradually-varied flow profiles in open channels: Analytical solutions by using
Gaussian hypergeometric function. Advances in Geophysical and Environmental Mechanics
and Mathematics. Berlin: Springer.
Jeppson, R. (2011). Open channel flow: Numerical methods and computer applications. Boca
Raton: CRC Press.
Katopodes, N. D. (2019). Free surface flow: Shallow-water dynamics. Oxford, UK:
References 181
Lee, M. T., Babbitt, H. E., & Baumann, E. R. (1952). Gradually varied flow in uniform channels
on mild slopes. Engineering Experiment Station Bulletin Nº 404, University of Illinois at
Urbana Champaign, College of Engineering, Engineering Experiment Station.
Montes, J. S. (1998). Hydraulics of open channel flow. Reston, VA: ASCE.
lecture notes]. Civil Engineering School, University da Coruña, Spain (in Spanish).
Rouse, H. (1950). Engineering hydraulics. New York: Wiley.
Rouse, H. (1965). Critical analysis of open-channel resistance. Journal of the Hydraulics Division,
ASCE, 91(HY4), 1–25.
Subramanya, K. (1986). Flow in open channels. New Delhi: Tata McGraw-Hill.
White, F. M. (1991). Viscous fluid flow. New York: McGraw-Hill.
Woodward, S. M., & Posey, C. J. (1941). Hydraulics of steady flow in open channels. New York:
Wiley.
Yen, B. C. (1991). Hydraulic resistance in open channels. In B. C. Yen (Ed.), Channel flow
resistance: Centennial of Manning’s formula (pp. 1–135). Highlands Ranch, USA: Water
Resources Publications.
Yen, B. C. (2002). Open channel flow resistance. Journal of Hydraulic Engineering, 128(1),
20–39.
Chapter 4
Computation of Steady Transcritical
Open Channel Flows
4.1 Introduction
There are two possible transitional open channels flows, as described in Chap. 2:
the transition from subcritical ðF\1Þ to supercritical ðF [ 1Þ flows, and the
transition from supercritical ðF [ 1Þ to subcritical ðF\1Þ flows. Both of these flow
types are fundamentally different. The transition from subcritical to supercritical
flows is smooth and continuous, as, for example, in weir flow (Fig. 4.1a). The
transition from supercritical to subcritical flows, however, is abrupt, encompassing
a highly turbulent free surface profile (Fig. 4.1b). This chapter details the compu-
tation of transitional flow profiles. Computations are then illustrated and compared
with experimental observations.
4.2 Transition from Sub- to Supercritical Flow
4.2.1 Formation of Singular Points in Free Surface Flows
The critical depth in a prismatic channel is a mathematically impossible value for

the flow profile, given that dh/dx ! ∞ (vertical flow profile) is regained from the
gradually varied flow equation (see Chap. 3). However, natural channels like rivers
are non-prismatic, for which the cross-sectional geometry and the bed slope vary
with the streamwise x-coordinate. Man-made channels, as weirs, may be also
non-prismatic. Consider the transition from a mild to a steep bottom slope with a
gradual bottom transition curve (Fig. 4.2), the problem investigated by Massé

https://doi.org/10.1007/978-3-030-13073-2_4
184 4 Computation of Steady Transcritical Open Channel Flows
Fig. 4.1 Transitions from a subcritical ðF\1Þ to supercritical ðF [ 1Þ flows over a weir,
b supercritical ðF [ 1Þ to subcritical ðF\1Þ flow in a hydraulic jump, with F as the Froude
number
Fig. 4.2 Transition from a mild to a steep bottom slope with a gradual bottom transition curve:
Formation of a critical point, where the flow depth equals the critical depth and the friction slope is
exactly equal to the bottom slope at this section
(1938), recently reviewed by Hager and Castro-Orgaz (2016). Between these two
reaches, the transitional bed profile is assumed to be smooth and continuous up to
second derivatives. The gradually varied flow equation is thus for a rectangular,
prismatic channel (Henderson 1966).
4.2 Transition from Sub- to Supercritical Flow 185
Pierre Massé was born on January 13, 1898, in Paris, passing away there
aged nearly 90 years old on December 15, 1987. He obtained in 1928 the
civil engineering degree from Ecole des Ponts et Chaussées, entering sub-
sequently in the electric industry of France, today’s Electricité de France
(EDF). He submitted in 1935 his Ph.D. thesis on variable regimes in fluvial
hydraulics, in which his mathematical knowledge was developed. Further
works are related to tidal flows and flows in mountainous rivers, in which
aspects of supercritical flow are fundamental, including the transition to
subcritical flow. His important work on the hydraulic jump and the free
surface profile in channels of variable bottom slope published in 1938 was
awarded the Prix Caméré from the French Académie des Sciences. He in
parallel was active in the hydropower industry, collaborating with the great
French dam designer André Coyne (1891–1960) on the Portillon Dam in the
French Pyrenees, or the Chastang Dam on Dordogne River in Southern
France. He was appointed in 1946 EDF Director of equipment, and in 1948
Vice-Director EDF, presiding from 1965 to 1969 the EDF Administrative
Council. He became in 1955 Commandeur of the French Légion d’Honneur
and was in 1977 elected Member of Académie des Sciences. The latter dis-
tinction is normally given to scientists only, shedding thus light on his great
career both in theoretical and applied engineering.
dzb
dh dx Sf So ð xÞ Sf
¼ ¼ ; ð4:1Þ
dx q2 1 F2
1 3
gh
where h is the flow depth, g the gravity acceleration, q the discharge per unit width,
F ¼ q=ðgh3 Þ1=2 the Froude number, Sf the friction slope, and So the bed slope,
defined as
dzb
So ¼ : ð4:2Þ
dx
If there are not gates, weirs, or, in general, any other channel control along the
channel reach, the only possible singularity provoking a variation in the flow
conditions is the variable slope reach zb = zb(x). Far up- and downstream along the
uniform slopes, asymptotic conditions will be settled, with the flow depth
approaching the corresponding uniform flow depth hN. The flow gradually adjusts
from the upstream subcritical uniform depth (hN > hc) to the downstream super-
critical uniform depth (hN < hc). The free surface profile will thus cut somewhere
the critical depth line to change from sub- to supercritical conditions. The flow
conditions at the critical flow section are investigated following Massé (1938) and
Henderson (1966). Rewrite Eq. (4.1) as
dh
1 F2 ¼ S o S f : ð4:3Þ
dx
If F ¼ 1 is set into Eq. (4.3), then So = Sf. As the flow gradually changes from
upstream subcritical flow to downstream supercritical flow, dh=dx 6¼ 0. Thus, at the
section where the flow depth equals the critical flow depth, the friction slope equals
the local bottom slope, and, obviously, the water surface slope at this section is
different from zero (Henderson 1966; Hager and Castro-Orgaz 2016). The point
satisfying these conditions is referred to as singular point, acting as a control section
of the flow. Starting at the singular point, the subcritical branch of the free surface
profile is computed in the upstream direction, whereas the supercritical portion is
computed in the downstream direction. A particular case of notable interest is the
transition from an adverse to a steep slope; it is the case occurring in weir flow
(Fig. 4.3) (Puertas and Sánchez 2001).
Fig. 4.3 Formation of a critical point in ideal fluid flow along the transition from an adverse to a
steep positive slope (adapted from Puertas and Sánchez 2001)
The fluid flow is assumed to be ideal given the short transitional length, and,
thus, Sf = 0. Equation (4.3) then produces
dh
1 F2 ¼ S o : ð4:4Þ
dx
Focus on the weir crest, where So = 0. Inserting this in Eq. (4.4) yields
dh
1 F2 ¼ 0: ð4:5Þ
dx
This identity is verified for the obvious case dh/dx = 0. This case is described in
Chap. 2 and corresponds to either fully sub- or supercritical flows over the weir
profile. However, the identity is also verified with F ¼ 1, corresponding to critical
flow at the weir crest with dh/dx < 0, given that the water accelerates from sub- to
supercritical flows, so that the flow depth must decrease in the flow direction
(Fig. 4.3).
4.2.2 Determination of Water Surface Slope at Critical

Point
If F ¼ 1, then Eq. (4.4) produces the indetermination dh/dx = 0/0 at the weir crest.
This singularity in the equation of motion is removed by applying L’Hospital’s rule
to Eq. (4.4) (Massé 1938; Escoffier 1958; Hager and Castro-Orgaz 2016). This
technique to remove flow depth gradients of the kind 0/0 on the shallow water
steady-state equations is known as the singular point method in open channel
hydraulics (Chow 1959; Montes 1998; Hager 1985, 2010). It originates from the
work of Poincaré (1881) on ODE equations and was applied to open channel
transition flow problems by Massé (1938), Escoffier (1958), Iwasa (1958), Wilson
(1969), and Chen and Dracos (1996). However, this method is questioned in the
literature given that the argument still prevails that Eq. (4.4) is invalid for h = hc
given the existence of non-hydrostatic pressure distribution. The gradually varied
flow model is mathematically valid at the critical depth, but, physically, it is
inaccurate if the free surface curvature is high, given that then the hydrostatic
pressure approximation is inadequate (Montes 1998; Castro-Orgaz and Chanson
2016). The singular point method is rarely explained in open channel flow books,
with Chow (1959), Henderson (1966) and Montes (1998) as notable exceptions.
However, mathematical books often describe it for general application in engi-
neering (i.e., von Kármán and Biot 1940).
Consider the weir flow depicted in Fig. 4.3. The boundary condition to integrate
Eq. (4.4) is the critical depth at the weir crest, h(x = 0) = hc = (q2/g)1/3, with the
origin of the x-coordinate at the weir crest. To integrate the ODE in the up- and
downstream directions using the Runge–Kutta method, the value of the free surface
slope dh/dx at the crest must be determined (Chen and Dracos 1996). However, this
value is unknown, given the mathematical indetermination existing there.
Equation (4.4) is a steady-state version of the SWE momentum equation. However,
in steady gradually varied flow, it is also obtained by differentiation of the total
energy head H = zb + h + q2/(2gh2) as

d q2
zb þ h þ ¼ 0; ð4:6Þ
dx 2gh2
where energy losses are neglected given the short transitional reach. Its first dif-
ferential is
dzb dh q2 dh
þ ¼ 0; ð4:7Þ
dx dx gh3 dx
which is obviously equivalent to Eq. (4.4), as stated. A second differential of H is

obtained from Eq. (4.7) as

d2 z b d2 h q2 d 2 h q2 dh 2
þ 2 3 2 þ3 4 ¼ 0: ð4:8Þ
dx2 dx gh dx gh dx
Inserting into Eq. (4.8) the singular point conditions at the weir crest
q2 dzb
¼ 1; ¼ 0; ð4:9Þ
gh3c dx
the following identity is regained [Eq. (2.37)]

d2 z b 3 dh 2
þ ¼ 0: ð4:10Þ
dx2 hc dx
The water surface slope at the weir crest is thus

1=2
dh hc d2 z b
¼ : ð4:11Þ
dx c 3 dx2
Note from this finding that the physically correct result implies a minus sign in front
of the square root function, corresponding to accelerating flow at the weir crest.
Further, given that hc is positive, imaginary numbers originate as a solution for
dh/dx if dz2b/dx2 is positive, meaning that a concave free surface profile is not
compatible with weir flow. Only a convex bottom profile (dz2b/dx2 < 0) produces a
transcritical free surface profile passing from sub- to supercritical flows, therefore.
The mathematical problem is thus the solution of the following ODE with critical
flow as the boundary condition:
2 8
>
> dzb
6 >
>
6 >
> dx if x 6¼ 0;
6 >
dh < 1 q
2
6
6 ODE: ¼
6 dx >> gh3
6 >
> 2 1=2 ð4:12Þ
6 >
> h d zb
6 >
:
c
if x ¼ 0;
6 3 dx2
6 2 1=3
6
4 q
Boundary condition: h ð x ¼ 0 Þ ¼ hc ¼ :
g
4.3 Transition from Super- to Subcritical Flows
4.3.1 Control Volume Equation for Hydraulic Jump
The transition from super- to subcritical flows is rapidly varied and, thus, cannot be
handled using the gradually varied flow theory. The pressure distribution can be
considered hydrostatic, but the velocity distribution is markedly non-uniform,
Reynolds stresses are extremely high and air entrainment provokes a two-phase
flow rather than a clear water flow (Fig. 4.1b). The flow is definitely not gradually
varied within a hydraulic jump, so Eq. (4.1) cannot be integrated to predict the free
surface profile of the hydraulic jump. Resort to more advanced approaches is
necessary for this task (Castro-Orgaz and Hager 2009), which is beyond the scope
of this book. The hydraulic jump is treated here using control volume equations,
with the purpose of relating the values of depth and velocity at the boundary
sections, where the flow is gradually varied (Fig. 4.4a).
Although the equation describing a hydraulic jump is presented in Chap. 2,
some developments are repeated here for convenience to introduce the hydraulic
jump as an element of free surface profile computations. Consider the hydraulic
jump in a horizontal and rectangular channel depicted in Fig. 4.4a; the control
volume with boundary sections 1 and 2, where the flow is super- and subcritical,
respectively, is used to formulate integral balances. At these boundaries, pressure is
hydrostatic and velocity uniform. Reynolds’ transport theorem states the conser-
vation of mass and momentum as (Liggett 1994; Sturm 2001; Chaudhry 2008),
ZZZ ZZ
d
dt
CV CS
X ZZZ ZZ
d
dt
CV CS
Here, V is the velocity vector, f a fluid force component, n the unit vector normal
to the control area A, v the volume, and q the water density. CV and CS are the
Fig. 4.4 Hydraulic jump in a horizontal channel a control volume for application of Reynolds’
transport theorem, b approximate representation of hydraulic jump as a local discontinuity
(Dx = 0) in the shallow water hydraulics theory
control volume and its surrounding control surface. The boundary limits of SC are
flow sections 1 and 2. For steady 1D flow, these reduce to
q ¼ U 1 h1 ¼ U 2 h2 ; ð4:15Þ
X
Fx ¼ qqðU1 U2 Þ; ð4:16Þ
where U is the depth-averaged velocity and Fx represents the x-component of any

external force acting on the control surface. The hydraulic jump involves highly
turbulent flow where a large amount of energy is dissipated. Further, the control
volume length Dx is of the order of only 6(h2 − h1), an experimentally determined
length of the jump profile. Therefore, shear forces are usually neglected, reducing
the external forces acting on the control surface to the hydrostatic forces.
Equation (4.16) reduces then to
1 2
qg h2 h21 ¼ qqðU1 U2 Þ: ð4:17Þ
2
After basic algebraic manipulation, Eq. (4.17) is alternatively written as
h1 h 1=2 i q2
h2 ¼ h1 ¼ 1 þ 1 þ 8F21 ; F21 ¼ : ð4:18Þ
2 gh31
4.3 Transition from Super- to Subcritical Flows 191
This relation is known as Bélanger’s equation for the hydraulic jump (Bélanger
1849). The water depth h1 is called sequent depth of h1. From Eq. (4.18), it is easily
verified that a decrease of h1 implies an increase of h1 . Note also that the sequent or
conjugate depth of the critical depth is itself.
Jean-Baptiste Bélanger was born on April 4, 1790, at Valenciennes, and

passed away aged 84 years on May 8, 1874, at Neuilly. He left Ecole
Polytechnique in 1813 as member of the Corps des Ponts et Chaussées. His
classmate Gaspard Coriolis (1792–1843) lectured later also with him. After
several years of service in the French army, Bélanger gradually taught at
Ecole Centrale, then at Ecole Polytechnique and finally at Ecole des Ponts et
Chaussées, i.e., the major French engineering schools, all located in Paris. He
retired in 1864 but continued to publish in scientific journals.
He worked in general mechanics and stressed the methods developed, the
technical language selected, and the correct style of presentation. In
hydraulics, he is particularly known for his Essai written in 1828 where he
presented the gradually varied flow equation and its first numerical solution.
The Bélanger equation correctly applies the momentum principle for the ratio
of the so-called sequent depths up- and downstream of a classical hydraulic
jump. His equation was systematically applied only about 100 years later
when verified by hydraulic experiments. He also presented a thorough
approach for backwater curves, as described by Adhémar Barré de
Saint-Venant (1797–1886), and subsequently developed by Pierre Vauthier
(1784–1847), Gaspard Coriolis (1792–1843), and Joseph Boussinesq
(1842–1929).
Within the context of the shallow water hydraulic theory, the effect of length of
the hydraulic jump is neglected (Dx = 0). The hydraulic jump is therefore treated as
a discontinuity determined by Eq. (4.18) (Fig. 4.4b). Physically, the difference
between the transition from sub- to supercritical and that from super- to subcritical
flows is easily appreciated (Fig. 4.1): The former is smooth and continuous,
whereas the latter is abrupt and highly turbulent. Mathematically, this is also stated
by the shallow water hydraulics theory: Ideal weir flow results in a mathematical
indetermination producing a smooth solution, whereas the hydraulic jump is
mathematically a discontinuity accompanied with significant energy losses.
For a given discharge q and supercritical flow depth h1, Eq. (4.18) defines the
required value of h2, namely h1 , to produce a steady hydraulic jump. If given h1, for
example, the actual tailwater level h2 is less than h1 , the shock front (h2 − h1) will
move in the downstream direction. The physical reason is that the momentum plus
pressure force of the incoming supercritical flow is larger than that at the tailwater
section. Thus, pressure forces and momentum are not in equilibrium, and the
unbalanced residual provokes an unsteady motion removing the shock in the
downstream direction. Only if h2 h1 , the jump is steady. The momentum equa-
tion of the hydraulic jump is rewritten as
h21 q2 h2 q2
þ ¼ 2þ ; ð4:19Þ
2 gh1 2 gh2
or using the definition of the momentum function S
h2 q2
S1 ¼ S2 ; S¼ þ : ð4:20Þ
2 gh
Equation (4.20) applies to physically explain the conditions triggering the

location of a hydraulic jump at a given channel position. Conceptually, the
supercritical inflow can be considered a jet-like flow, with the momentum qU1 as
the important component of S. In turn, the tailwater flow can be considered a
downstream water wall, with the hydrostatic force ð1=2Þgh22 as the important
component of S (Puertas and Sánchez 2001). These conceptual simplifications are
now used to explain the formation of surges, a topic to be described mathematically
in Chap. 5. Here, only a conceptual introduction is attempted. If the momentum of
the inflow jet is much larger than the hydrostatic force of the downstream water
wall, then the shock front will move to the right, forming a moving hydraulic jump,
that is, a traveling surge with h1 < h2 (Jain 2001) (Fig. 4.5a), which is an unsteady
flow. If the momentum of the jet is on the contrary much lower than the tailwater
hydrostatic force, then the traveling surge will move to the left (Fig. 4.5b). If S1 and
S2 are exactly in balance, the surge is not moving, forming a steady hydraulic jump
(Fig. 4.5c), satisfying Eq. (4.20). This relation therefore serves to determine the
position of a hydraulic jump. It is easily verified that the energy loss due to the
hydraulic jump is [Eq. (2.136)]

q2 q2 ðh2 h1 Þ3
DH ¼ h1 þ h2 þ ¼ : ð4:21Þ
2gh21 2gh22 4h1 h2
Fig. 4.5 Moving hydraulic jump, or surge with h1 < h2, displacing in: a the downstream
direction, b the upstream direction, c with no displacement, that is, under steady conditions
(adapted from Puertas and Sánchez 2001)
4.3.2 Determination of Hydraulic Jump Position
It is now explained how to use Eq. (4.18) to determine the position of a hydraulic
jump in a free surface profile using a practical example. Consider a sluice gate
upstream of a free overfall in a mild slope channel. Two flow profiles are possible,
each determined by a different channel control: subcritical flow profile computed in
the upstream direction assuming critical flow at the brink section; supercritical flow
profile computed in the downstream direction imposed by a gate opening below the
Fig. 4.6 Sub- and supercritical free surface profiles beyond a sluice gate
critical depth (Fig. 4.6). It must be determined which of the two profiles is the
actual physically valid solution, depending on which control is active, or, if both
controls are active (the brink section and the gate), the super- and subcritical
profiles must be assembled through a hydraulic jump. The computational process
encompasses the following steps:
(i) Compute the supercritical M3 curve h = h(x) starting at the gate section, or at
the vena contracta assuming Cc = 0.61, if contraction effects are accounted
for. The distance from the gate to the vena contracta is small and therefore
neglected in gradually varied flow computations. The computation must be
stopped ahead of reaching the critical depth line hc, given that dh/dx ! ∞.
(ii) After computation of the supercritical M3 profile, for each supercritical depth
h of this curve, compute the corresponding sequent depth from Bélanger’s
equation
hð x Þ h 1=2 i q2
hð x Þ ¼ 1 þ 1 þ 8F2 ; F2 ¼ : ð4:22Þ
2 g½ hð x Þ 3
Physically, this sequent depth is the tailwater depth necessary to produce a

hydraulic jump at section x where the supercritical depth is h(x). Note that the
hydraulic jump is treated as a local discontinuity.
(iii) Compute the subcritical M2 curve starting at the brink section, with the
boundary flow depth close but not identical to the critical depth, given that
dh/dx ! ∞ otherwise.
(iv) If the sequent depth curve M3 of the supercritical M3 profile intersects at any
point the subcritical profile M2 computed from the brink section, a hydraulic
jump appears at this section. The hydraulic jump is therefore a discontinuity
located at the intersection of the M2 profile with the M3 profile in this
example (Fig. 4.7).
If the sequent depth curve M3 of the M3 profile is below the M2 curve in all the
channel reach (Fig. 4.8), no hydraulic jump is formed; the momentum function of
the M2 profile is larger than that of the M3 profile for all the channel sections.
Fig. 4.7 Free surface profile beyond a sluice gate with the formation of a hydraulic jump
Fig. 4.8 Subcritical free surface profile beyond a sluice gate: the gate is drowned
Therefore, the gate is drowned and the real free surface profile is the M2 curve, with
subcritical flow along the channel. In this case, the only active channel control is the
brink section. This case may occur if the channel beyond the sluice gate is long and
the normal depth is relatively high.
If the conjugate depth curve M3 of the M3 profile is above the M2 curve along
the entire channel reach (Fig. 4.9), no hydraulic jump is formed; the momentum
function of the M3 profile is larger than that for the M2 profile for all the channel
sections. Therefore, the hydraulic jump is fully removed from the channel and the
real free surface profile is the supercritical M3 curve. In this case, the only active
channel control is the gate, given that the flow is forced to exit from the channel at
the brink section at a high speed, and critical flow conditions cannot be settled
there, deactivating this as control. This case may occur if the channel beyond the
gate is short and the M3 curve is not reaching the critical depth line.
Fig. 4.9 Supercritical free surface profile beyond a sluice gate: the hydraulic jump is rejected
from the channel reach
4.4 Computational Examples
4.4.1 Flow over Round-Crested Weirs
The steady transcritical free surface profile over a parabolic weir of bed shape
zb = 0.3216 − 0.54525x2 (m) was computed using Eq. (4.12). The numerical
model is implemented in the file parabolicweir_transcritical.xls of Chap. 12.
Equation (4.12) was solved using the fourth-order Runge–Kutta method (see
Chap. 3), and the corresponding sub- and supercritical branches of the free surface
profile were computed in the up- and downstream directions, respectively.
Theoretical predictions are compared in Fig. 4.10a, b with experimental data of
Blau (1963) for two runs corresponding to q = 0.51 and 0.151 m2/s, respectively.
In the first test, slight deviations between experiments and simulations are evident
given the effect of the non-hydrostatic pressure distribution, not accounted for by
the gradually varied flow theory. In the second run, however, predictions are in
excellent agreement with observations. At the weir crest, Eq. (4.11) was used to
remove the singularity of the equations of motion. This equation is compared in
Fig. 4.11 with the experimental measurements of Wilkinson (1974), confirming that
the result is not only theoretically sound, but also in agreement with experimental
observations (Castro-Orgaz and Chanson 2016).
Figure 4.12 contains the experimentalh data of Sivakumaran
i et al. (1983) for a
Gaussian hump of profile zb ¼ 20 exp 0:5ðx=24Þ2 (cm) for two test cases.
Simulations were conducted using Eq. (4.12) as previously described for the
parabolic weir. This test is implemented in the file Gaussianhump_transcritical.xls
in Chap. 12. The computed solution is presented for both cases and compared in
Fig. 4.12a, b with the experiments. The departure between simulations and
4.4 Computational Examples 197
Fig. 4.10 Transcritical flow

over parabolic weir profile.
Comparison with
experimental data (Blau 1963)
for q = a 0.51 m2/s,
b 0.151 m2/s
Fig. 4.11 Free surface slope

at the crest section of a
parabolic weir profile.
Comparison with
experimental data (Wilkinson
1974)
experiments for the test case of Fig. 4.12a (q = 0.111 m2/s) indicates that the effect
of the non-hydrostatic pressure is notable as the flow passes from sub- to super-
critical. For the test case of Fig. 4.12b (q = 0.0359 m2/s), deviations between
numerical results and experiments are small.
Fig. 4.12 Transcritical flow

over Gaussian weir profile.
Comparison with
experimental data
(Sivakumaran et al. 1983) for
q = a 0.111 m2/s,
b 0.0359 m2/s
4.4.2 Hydraulic Jump Beyond a Sluice Gate
Gharangik and Chaudhry (1991) conducted measurements of steady hydraulic

jumps in a horizontal flume 0.46 m wide and 14 m long. Consider a test conducted
for an inflow Froude number F1 ¼ 7. The upstream boundary section was located at
x = 0.305 m from the gate, and the depth and unit discharge there resulting from
experimentation were h = 0.031 m and q = 0.118 m2/s. The tailwater level was set
to h = 0.265 m. The experimental measurements for this test are displayed in
Fig. 4.13.
This test is available in the file sluicegate_hydraulicjump.xls of Chap. 12. The
predictions using Eq. (4.1) were conducted using Dx = 0.01 m and n = 0.008 m−1/3s.
The upstream boundary condition for the computation of the H3 supercritical free
surface profile was taken from the experiments, and the conjugate depth curve of the
H3 profile was determined applying Eq. (4.18). From the tailwater boundary condition,
a subcritical flow profile was computed. The intersection of this subcritical profile with
Fig. 4.13 Comparison of numerical simulation with experimental results (Gharangik and
Chaudhry 1991) for flow profile with steady hydraulic jump
the sequent depth curve fixes the position of the hydraulic jump. The computational
results are displayed in Fig. 4.13. A comparison of the transcritical free surface profile
with experimental data by Gharangik and Chaudhry (1991) in Fig. 4.13 shows good
agreement. Note that the finite length of the jump produced by turbulence is a feature
not accounted for in the theoretical predictions.
References
Bélanger, J. B. (1849). Notes sur le Cours d’Hydraulique (Notes on the Course in Hydraulics).
Mém. Ecole Nat. Ponts et Chaussées, Paris, France, session 1849–1850, 222 p (in French).
Wehrschwelle (Distributions of discharge and energy over a parabolic-shaped weir].
Mitteilung, 7, 5–72 Forschungsanstalt für Schiffahrt, Wasser- und Grundbau, Berlin (in
German).
Castro-Orgaz, O., & Chanson, H. (2016). Minimum specific energy and transcritical flow in
unsteady open channel flow. Journal Irrigation and Drainage Engineering, 142(1), 04015030.
Chen, J., & Dracos, T. (1996). Water surface slope at critical controls in open channel flow.
Journal of Hydraulic Research, 34(4), 517–536.
Escoffier, F. F. (1958). Transition profiles in nonuniform channels. Transactions ASCE, 123, 43–
56.
Gharangik, A. M., & Chaudhry, M. H. (1991). Numerical simulation of hydraulic jump. Journal of
Hydraulic Engineering, 117(9), 1195–1211.
Hager, W. H. (1985). Critical flow condition in open channel hydraulics. Acta Mechanica, 54(3–
4), 157–179.
Hager, W. H. (2010). Wastewater hydraulics: Theory and practice (2nd ed.). Berlin: Springer.
Hager, W. H., & Castro-Orgaz, O. (2016). Critical flow in open channel hydraulics: From Boess to
De Marchi. Journal of Hydraulic Engineering, 142(1), 02515003.
Iwasa, Y. (1958). Hydraulic significance of transitional behaviours of flows in channel transitions

and controls. Memoires of the Faculty of Engineering, Kyoto University, 20(4), 237–276.
Massé, P. (1938). Ressaut et ligne déau dans les cours à pente variable [Hydraulic jump and free
surface profile in channels of variable slope]. Revue Générale de l'Hydraulique, 4(19), 7–11;
4(20), 61–64 (in French).
Poincaré, H. (1881). Mémoire sur les courbes définies par une équation différentielle [Memoir on
the defined curves of a partial differential equation]. Journal de mathématiques pures et
appliquées, 7, 375–422 (in French).
lecture notes]. Spain: Civil Engineering School, University da Coruña (in Spanish).
Sivakumaran, N. S., Tingsanchali, T., & Hosking, R. J. (1983). Steady shallow flow over curved
beds. Journal of Fluid Mechanics, 128, 469–487.
von Kármán, T., & Biot, M. A. (1940). Mathematical methods in engineering: An introduction to
mathematical treatment of engineering problems. New York: McGraw-Hill.
Wilkinson, D. L. (1974). Free surface slopes at controls in channel flow. Journal of the Hydraulics
Division, ASCE, 100(8), 1107–1117.
Wilson, E. H. (1969). Surface profiles in non-prismatic rectangular channels. Water Power, 21
(11), 438–443.
Chapter 5
Unsteady Open Channel Flows:
Basic Solutions
5.1 Introduction
Flow conditions in rivers are usually unsteady (Cunge et al. 1980; Sturm 2001;
Chaudhry 2008), even though steady flow computations are conducted in engi-
neering applications, as for determining an inundation area in hydrological studies.
Further, the operation of man-made canals to control the water released from
reservoirs implies unsteady maneuvers resulting in transient open channel flows
(Chaudhry 2008). Therefore, real flow computations in canals and rivers require the
solution of the Saint-Venant equations or the shallow water equations (SWE). In
this chapter, these are presented and their basic continuous and discontinuous
unsteady flow solutions are discussed.
5.2 Shallow Water Equations
Unsteady free surface flow is mathematically determined by the two-dimensional

(2D) SWE, which are given by (see Chap. 1)
@h @ ðUhÞ @ ðVhÞ
þ þ ¼ 0;
@t @x @y
@ ðUhÞ @ 2 @ @zs 1
þ U h þ ðUVhÞ ¼ gh ðsxz Þb ; ð5:1Þ
@t @x @y @x q
@ ðVhÞ @ 2 @ @zs 1
þ V h þ ðVUhÞ ¼ gh syz b :
@t @y @x @y q

https://doi.org/10.1007/978-3-030-13073-2_5
202 5 Unsteady Open Channel Flows: Basic Solutions
Here, (U, V) are the depth-averaged velocity components in the (x, y) horizontal
directions, h is the flow depth, zs the free surface elevation, g the gravity acceler-
ation, q the water density, and s denotes here tangential stresses. Most river flows
can be considered one-dimensional (1D), given that the x-direction is predominant
(Fig. 5.1). Therefore, the transverse water surface slope can be assumed horizontal.
To obtain 1D flow equations for an arbitrary cross section, the vertically averaged
continuity and x-momentum equations are laterally integrated in the y-direction (see
Chap. 1). The 2D depth-averaged model is then reduced to a 1D, sectional inte-
grated formulation. The river is therefore basically treated as a 1D stream tube.
Using Leibniz’s rule and imposing that the free surface slope is independent of the
y-direction (horizontal water surface), the lateral integration of Eqs. (5.1)1 and
(5.1)2 produces the 1D Saint-Venant equations (De Saint-Venant 1871) for an
arbitrary cross section (see Chap. 1)
@A @Q
þ ¼ 0;
@t @x
2 ð5:2Þ
@Q @ Q @zs so
þ ¼ gA gA :
@t @x A @x qgRh
Here, Q is the discharge, A the flow area, Rh = A/p the hydraulic radius, p the
wetted perimeter, and so the average boundary shear stress acting on the wetted
perimeter. The free surface elevation zs(x, t) corresponds at any vertical line of the
cross section to the sum of the bed elevation and the local water depth. The standard
approach is to use the thalweg (lowest point) to define the flow depth h and bed
elevation of the longitudinal river profile (Cunge et al. 1980) (Fig. 5.1). If the
channel is prismatic, Eq. (5.2) can be recast as (Sturm 2001; Jain 2001; Chaudhry
2008)
Fig. 5.1 One-dimensional

unsteady river flow a profile
view, b cross section
5.2 Shallow Water Equations 203
@A @Q
þ ¼ 0;
@t @x ð5:3Þ
@Q @ @zb
þ QU þ gAh ¼ gA Sf ;
@t @x @x
where U = Q/A, h is the flow depth along the thalweg, Sf ¼ so =ðqgRh Þ the friction
slope, bars indicate the depth of centroid below the free surface, and @zb =@x is the
longitudinal slope taking as reference the thalweg.
The x-coordinate is usually curvilinear, following the thalweg path in the hori-
zontal plane (x, y). Therefore, a river must not have severe curvatures in plan to
apply Eq. (5.3). Otherwise, curvature terms must be introduced to transform the
Cartesian system to curvilinear coordinates following the river plan, e.g., at river
meanders. For the basic case of a rectangular cross section, and with q = Q/b as the
unit discharge and b as the channel width, Eq. (5.3) reduces to
@h @q
þ ¼ 0;
@t @x
2 ð5:4Þ
@q @ q h2 @zb
þ þg ¼ gh Sf :
@t @x h 2 @x
Equation (5.4) is widely used in this book to introduce numerical methods. Note
that the differential form of the SWE only applies to zones of the (x, t) plane where
the dependent variables U(x, t) and h(x, t) are single-valued and smooth (Cunge
et al. 1980).
5.3 Discontinuous Solutions: Basic Equations

of a Positive Surge
5.3.1 Introduction
A special feature of the Saint-Venant equations is that they admit both continuous
or gradually varied flow solutions, as well as discontinuous solutions (rapidly
varied flow zones), involving abrupt changes in the flow variables, even if initial
and boundary conditions are continuous (Cunge et al. 1980). Using shock-capturing
methods (Toro 2001), it is possible to simulate with accuracy both continuous and
discontinuous waves, where flow variables are not single-valued. To catch these
discontinuous solutions, the integral form of the SWE must be used (Cunge et al.
1980; Toro 2001). This is equivalent to apply the Reynolds transport theorem
(Chaudhry 2008, 2014; White 2009) to the fluid properties mass and momentum in
a finite control volume.
An important discontinuous unsteady open channel flow is the surge (Fig. 5.2),
corresponding to a traveling wave separating abruptly two open channel flow
portions of different depth and velocity. In this section, positive surges are
Fig. 5.2 Surge moving in a

hydropower canal (Seton
Canal, Vancouver, Canada)
a general view, b detail of
wave front (reprinted from
Chaudhry 2014). Note
secondary undulations (Favre
waves) linked to
non-hydrostatic pressure
distribution of non-breaking
central portion of the surge.
Wave is propagating in the
upstream direction after load
rejection
5.3 Discontinuous Solutions: Basic Equations of a Positive Surge 205
Fig. 5.3 Undular surge with

wave breaking (photograph
by late Prof. D. H. Peregrine)
considered, implying a zone of deeper waters propagating over shallower waters,1

either in the up- or downstream directions. A surge is generated by a sudden
increase of discharge in the channel (Favre 1935; Chow 1959; Henderson 1966).
Near the front of a positive surge, the fluid pressure is in reality non-hydrostatic, a
feature beyond the capabilities of the SWE (see Chaps. 1 and 11). The surge, in
fact, is undular due to non-hydrostatic pressure for small surge heights (Favre 1935;
Peregrine 1966; Cunge 1975; Cunge et al. 1980). This leads to the so-called sec-
ondary waves or Favre waves (Fig. 5.2). However, once the undular wave breaks
(Fig. 5.3), the SWE provide an excellent estimate of the shock height, celerity, and
energy dissipation (Cunge 1975; Toro 2001). If a surge breaks, the length of the
discontinuity is small and can be neglected. The positive surge of tidal origin is
called bore (Chanson 2004) (Fig. 5.4).
5.3.2 Reynolds Transport Theorem
The celerity of wave propagation in gradually varied flow zones, where solutions
are smooth and continuous, is c = (gh)1/2 (see Chap. 2). This is the celerity of an
elementary gravity wave in a hydrostatic field. However, in a surge, the wave height
is finite, resulting in a different propagation celerity. As will be demonstrated, a
surge or bore can even propagate into the upstream direction in a supercritical flow,
something impossible within the framework of small gravity waves. Here, the basic
equations governing the movement of a surge will be developed considering
1
In computational fluid dynamics (CFD), the analogous discontinuous wave stemming from the
Euler equations for gas dynamics is called shock wave (Toro 2001, 2009; LeVeque 2002),
although there are also contact or shear waves. The hydraulic shock wave is simply referred to as
positive surge (Jain 2001; Chanson 2004; Chaudhry 2008).
Fig. 5.4 A tidal bore on Qiantang River at Yanguan, China, on 7/Sept/2013. Note the breaking
front propagating upstream (Photo courtesy of Prof. H. Chanson)
unsteady flow computations. An alternative approach is possible by reducing the

unsteady flow to a steady flow problem in moving axes propagating with the
velocity of the surge Vw, which is a constant. This development will be presented in
Chap. 6.
Consider a positive surge conceptually simplified to a translation wave of per-
manent shape with a discontinuity in depth and velocity, as sketched in Fig. 5.5.
The channel is horizontal and the fluid ideal. The strategy is to insert this discon-
tinuous solution U(x, t) into the integral mass and momentum balances to obtain the
Fig. 5.5 Positive surge

moving in downstream
direction a profile view,
b cross section
relations defining the surge height and velocity. The movement of the surge is
considered with respect to a fixed control volume in space and time, where
Reynolds’ transport theorem (Chow et al. 1988; Chaudhry 2008) is applied. It states
the conservation of any system property within a finite control volume. For mass
and momentum, these statements are, respectively (Liggett 1994; Chaudhry 2008),
ZZZ ZZ
d
dt
CV CS
X ZZZ ZZ
d
dt
CV CS
Here, V is the velocity vector, f a fluid force component, n the unit vector normal to
the control area A, v the volume, and q the water density. CV and CS are the control
volume and its surrounding control surface. The boundary limits of SC are flow
sections 1 and 2. The rate of temporal variation of water mass stored in the control
volume is after time Dt
ZZZ ZZZ
d d dv
qdv ¼ q dv ¼ q ¼ qVw ðA2 A1 Þ; ð5:7Þ
dt dt dt
CV CV
where A1 and A2 are the flow areas at sections 1 and 2. The net mass flow across the
contour of the control volume is
ZZ ZZ
qðV nÞdA ¼ qUdA ¼ qU1 A1 qU2 A2 : ð5:8Þ
CS CS
Thus, from Eq. (5.5), one obtains
qVw ðA2 A1 Þ þ qU1 A1 qU2 A2 ¼ 0; ð5:9Þ
or
A1 ðU1 Vw Þ ¼ A2 ðU2 Vw Þ: ð5:10Þ
On inspecting Eq. (5.10), it is noted that the continuity equation across a surge is
simply obtained by superimposing a constant velocity (−Vw) to the unsteady flow
velocities. The unsteady translation of the surge is thus reduced to a steady problem
in moving axes at the absolute surge velocity. Likewise, the fluid force balance
projected in the x-direction of surge propagation in the control volume of Fig. 5.5
reads
X
fx ¼ qg h2 A2 h1 A1 : ð5:11Þ
This is simply a hydrostatic pressure balance. The temporal rate of momentum

change qV within the control volume, projected in the x-direction, is
ZZZ ZZZ
d d
qVdv ) q Udv ¼ qVw ðA2 U2 A1 U1 Þ; ð5:12Þ
dt dt
CV CV
whereas the net momentum flux across the control surface is

ZZ ZZ
qVðV nÞdA ) qU 2 dA ¼ qU12 A1 qU22 A2 : ð5:13Þ
CS CS
Inserting Eqs. (5.11)–(5.13) into Eq. (5.6) produces

qg h2 A2 h1 A1 ¼ qVw ðA2 U2 A1 U1 Þ þ qU12 A1 qU22 A2 : ð5:14Þ
This relation is rewritten as

g h2 A2 h1 A1 ¼ Vw A2 U2 U22 A2 Vw A1 U1 þ U12 A1
ð5:15Þ
¼ A1 U1 ðU1 Vw Þ A2 U2 ðU2 Vw Þ:
This identity demonstrates that the momentum balance across a surge can also be
reduced to a steady flow balance by superimposing the constant velocity (−Vw) to
the unsteady flow velocities. From Eq. (5.10), one gets
A1
ðU2 Vw Þ ¼ ðU1 Vw Þ: ð5:16Þ
A2
Inserting into Eq. (5.15) yields
A1
g h2 A2 h1 A1 ¼ A1 U1 ðU1 Vw Þ A2 U2 ðU1 Vw Þ
A2
ð5:16Þ
¼ A1 U1 ðU1 Vw Þ A1 U2 ðU1 Vw Þ
¼ A1 ðU1 Vw ÞðU1 U2 Þ:
From Eq. (5.10), the surge celerity is
U1 A1 U2 A2
Vw ¼ ; ð5:17Þ
A1 A2
which, inserted in Eq. (5.16), produces

2 h1 A 1 h2 A 2
ðU2 U1 Þ ¼ g ðA1 A2 Þ: ð5:18Þ
A1 A2
Using this relation to eliminate (U1 − U2) in Eq. (5.16) results in

2 A2 h2 A2 h1 A1
ðU1 Vw Þ ¼ g ; ð5:19Þ
A1 ðA2 A1 Þ
from which the surge velocity is

" #1=2
A2 h2 A2 h1 A1
Vw ¼ U1 g : ð5:20Þ
A1 ðA2 A1 Þ
A surge propagating in the downstream direction is generated by selecting the

sign “+” in Eq. (5.20). Jain (2001) discussed all the possible types of surges upon
algebraic analysis of Eq. (5.20). The surge celerity c is the velocity of the wave
relative to that of the fluid that is propagating, e.g., the body of water moving at
speed U1. Thus,
" #1=2
A2 h2 A2 h1 A1
c ¼ Vw U1 ¼ g : ð5:21Þ
A1 ðA2 A1 Þ
For a rectangular cross section, it reduces to

1=2
gh2
c ¼ Vw U1 ¼ ðh1 þ h2 Þ : ð5:22Þ
2h1
This result shows that a finite discontinuity travels with a celerity different from that
of small gravity waves
c ¼ ðghÞ1=2 : ð5:23Þ
Both equations only agree if h1 = h2 = h, e.g., if the discontinuity is an infinitesimal

perturbation. Thus, a positive surge of small height propagates at Vw U + (gh)1/2.
This issue is of paramount relevance and must be carefully accounted for when
developing open channel flow solutions involving both continuous and discontin-
uous portions. Equations (5.17) and (5.20) are a system of two equations deter-
mining the flow depth behind the surge h1 and its absolute velocity Vw if the initial
conditions (h2, U2) and the upstream discharge Q1 producing the surge are known.
The formation of a surge is physically explained by considering a smooth, gradually
varied wave profile implying a zone of deeper water invading shallower water,
progressively steepening due to a larger value of c = (gh)1/2 in deep water (Cunge
1975). As the wave evolves, the small hydrostatic waves progressively collapse,
until they form a finite surge. Mathematically, this process is described with the
theory of characteristics, explained in Sect. 5.5. For details, see Cunge (1975) and
Montes (1998).
5.3.3 Positive Surges Moving Upstream in Supercritical

Flow
An important feature of surges is that they can run up in a supercritical current. For
a surge propagating in the upstream direction (Fig. 5.6), the sign of Vw by appli-
cation of the fundamental Eq. (5.22) must be negative.
Consider Eq. (5.22) rewritten as
1=2
gh2
Vw ¼ U1 ð h1 þ h2 Þ : ð5:24Þ
2h1
To obtain values of Vw < 0, the “−” sign must be selected. Let us define the surge
jump as Dh = h2 − h1 and the dimensionless jump height as e = Dh/h1. Using these
definitions, Eq. (5.24) yields
Fig. 5.6 Surge moving in

upstream direction over a
supercritical current
1=2 1=2
h1 þ Dh 1þe
Vw ¼ U1 g ðh1 þ h1 þ DhÞ ¼ U1 gh1 ð2 þ eÞ ; ð5:25Þ
2h1 2
or
h e i1=2
Vw ¼ U1 ðgh1 Þ1=2 ð1 þ eÞ 1 þ : ð5:26Þ
2
This implies with F1 ¼ U1 =ðgh1 Þ1=2 as the approach flow Froude number
1=2
Vw 3 e2
¼ F1 1 þ e þ : ð5:27Þ
ðgh1 Þ1=2 2 2
Equation (5.27) shows that in a supercritical current with F1 [ 1, the surge will
move upstream (Vw < 0) if the surge jump height e is large enough. The minimum
height emin for a given F1 above which the surge begins to move in the upstream
direction is from Eq. (5.27), by setting Vw = 0, determined by
3 e2
F21 ¼ 1 þ emin þ min : ð5:28Þ
2 2
For an infinitesimal surge height e ! 0, Eq. (5.27) yields
Vw
¼ F1 1: ð5:29Þ
ðgh1 Þ1=2
This is always positive for F1 [ 1, implying that these infinitesimal surges cannot
travel in the upstream direction in supercritical flows. For a small amplitude e 1,
Eq. (5.27) reads

Vw 3 1=2
¼ F1 1 þ e : ð5:30Þ
ðgh1 Þ1=2 2
This equation was presented by De Saint-Venant (1870). Using it, surges propa-
gating in the upstream are generated. However, the resulting values of e are not of
small amplitude and thus nonlinear terms. In this case, better revert to the general
Eq. (5.27). The limiting case Vw = 0 corresponds to a steady surge or hydraulic
jump (Fig. 5.7).
Adhémar Barré de Saint-Venant was born on August 23, 1797, at Filliers,

France, and passed away aged 89 years at Saint-Ouen on January 6, 1886. He
was educated at Ecole Polytechnique, entering the Corps des Ponts et
Chaussées (Corps of Bridges and Roads), where he stayed for the next
25 years. He conducted works relating to harbors, navigation canals, and
roads. A memoir on agricultural hydraulics was honoured by the Agricultural
Society in 1849. He was named professor at Institut Agronomique de
Versailles in 1850 and lectured also at Ecole des Ponts et Chaussées. de
Saint-Venant left Paris city services in 1848 to devote himself exclusively to
sciences, starting in 1852. His main first works related to elasticity, following
Poncelet’s approach. In 1868, de Saint-Venant was elected Member of
Académie des Sciences, Paris.
The de Saint-Venant equations relate to one-dimensional unsteady flows;
these are currently also referred to as the shallow water equations. A set of
partial differential equations for flow depth and cross-sectional velocity as a
function of space and time was established in 1871. These equations are
based on hydrostatic pressure and uniform velocity distributions and serve
currently for the prediction of flood flows, either in one or in two dimensions.
de Saint-Venant also proposed in 1851 an equation for the loss of head in
pressurized conduit flows using a power function. This paper introduces early
ideas on the effect of fluid turbulence. It was recently proposed that the
current Froude number should be referred to the name of de Saint-Venant,
given the latter’s enormous developments in this field of hydraulics.
Fig. 5.7 Hydraulic jump or steady surge (photograph VAW, ETH Zurich)
From Eq. (5.28), emin is

1=2
h2 3 9 1=2 3 1
emin ¼ 1 ¼ 2 1 F21 ¼ þ 2F21 : ð5:31Þ
h1 2 4 2 4
Solving Eq. (5.31) for the sequent depth ratio h2/h1 produces Bélanger’s equation
for the hydraulic jump presented in Chap. 4 as [Eq. (4.18)]
h2 1 h 1=2 i
¼ 1 þ 8F21 1 : ð5:32Þ
h1 2
5.3.4 Formulation Used in Finite Volume Methods
The general equation for a shock wave [Eq. (5.22)] moving to the right direction is
1=2
gh2
Vw ¼ U1 þ ð h1 þ h2 Þ : ð5:33Þ
2h1
It will be rewritten in a form widely used in CFD applications based on finite

volume upwind methods with flux computations using Riemann solvers (Toro
2001; see Chap. 9). Equation (5.18) for a rectangular cross section is
g ð h1 h2 Þ
ðU2 U1 Þ2 ¼ ðh1 þ h2 Þ; ð5:34Þ
2 h 1 h2
which coupled with Eq. (5.33) yields Vw as function of the tailwater velocity U2
1=2
gh1
Vw ¼ U2 þ ð h1 þ h2 Þ : ð5:35Þ
2h2
Using the definition c2 = (gh2)1/2 one gets from this last expression
1=2
h1
Vw ¼ U2 þ ðgh2 Þ1=2 ð h1 þ h2 Þ ¼ U2 þ Kc2 : ð5:36Þ
2h22
Here, K is a dimensionless factor accounting for the finite surge height,

1=2
h1
K¼ ð h 1 þ h 2 Þ : ð5:37Þ
2h22
Equation (5.36) is a “signal” celerity used to propagate shocks in the finite volume
method, to be described in Chap. 9.
5.4 Methods of Solution
5.4.1 Shock-Capturing Methods
A shock-capturing method permits to compute the solution of a hyperbolic system

of conservation laws including continuous and discontinuous waves without any
special shock treatment (Cunge 1975; Toro 2001). The differential form of the 1D
SWE, Eq. (5.4), can be written in vector notation as
@U @F
þ ¼ S: ð5:38Þ
@t @x
Here, U is the vector of unknowns, F the flux vector, and S the source term vector,
given with Cf as a friction coefficient by

h Uh 0
U¼ ; F¼ ; S¼ ð5:39Þ
Uh U 2 h þ 12 gh2 gh @z
@x
b
Cf U jU j
This is called conservative form of the SWE (Toro 2001; Chaudhry 2008). This
system of PDEs is only valid for the computation of continuous solutions, where
h and q are smooth and single-valued in the (x, t) plane. The solution U(x, t) of
Eq. (5.38) must be differentiable with respect to x and t. To compute discontinuous
flows, the PDEs must be integrated over a control volume to obtain its integral form
(Cunge 1975). This fact was already used in Sect. 5.3.2 to obtain the basic relations
of a surge. Here, the integral form is reproduced by integrating the conservative
form of the SWE over a control volume. For simplicity, source terms are dropped,
resulting in the inviscid version of the SWE over horizontal terrain, that is
5.4 Methods of Solution 215
@U @F
þ ¼ 0: ð5:40Þ
@t @x
This equation is now integrated over an arbitrary control volume CV, resulting in
ZZZ
@U @F
þ dv ¼ 0; ð5:41Þ
@t @x
CV
or,
ZZZ
@U
þ r F dv ¼ 0: ð5:42Þ
@t
CV
Applying Gauss’s divergence theorem to the flux term in Eq. (5.42) results in
ZZZ ZZ
@U
dv þ ðF nÞdA ¼ 0: ð5:43Þ
@t
CV CS
After elementary manipulations, Eq. (5.43) becomes equivalent to Reynolds’

transport theorem as used in Sect. 5.3.2. While Eq. (5.40) only applies to compute
continuous flows, Eq. (5.43) describes any type of solution, both continuous and
discontinuous.2 The solution U(x, t) of Eq. (5.43) must be integrable, but it is not
necessary that it be continuous. Equation (5.43) is the basic relation used to con-
struct shock-capturing numerical methods (Toro 2001); it conserves mass and
momentum in a fluid system: If the system is discretized into control volumes
which are in contact, fluxes leaving a control volume enter into the adjacent control
volume across the contact surface. Overall, there is no net gain or loss of mass and
momentum in the entire system, corresponding to the so-called conservative
property (Roache 1972). The formulation of the SWE as presented in Eq. (5.40) is
the so-called conservative form, where the conserved variables in the fluid system
are h and (Uh) (Toro 2001). The non-conservative or primitive formulation of the
SWE is regained by formulating the PDEs in terms of h and U, yielding (Jain 2001)
2
In applied mathematics, the solutions of the integral form of a system of conservation laws are
called weak solutions of the differential form, even though they may or may not be the traditional
solutions of it (Cunge 1975; Macdonald 1995). This is simply a notation criterion to avoid having
to refer to the integral form, but, clearly, the integral form is linked to the concept of weak solution.
If a weak solution is continuous, then it is a solution of the differential form and it is called genuine
solution.
@h @U @h
þh þU ¼ 0;
@t @x @x ð5:44Þ
@U @U @h
þU þg ¼ 0:
@t @x @x
This form of the SWE generally applies only to compute continuous solutions.
A shock-capturing method based on Eq. (5.43) can resolve without any special
treatment surges and continuous waves, but, if applied to Eq. (5.44), will produce
shocks of erroneous celerity (Cunge 1975; Toro 2001) (see Sect. 5.8). An alter-
native is to use the so-called shock tracking methods: in these techniques,
Eq. (5.44) are used to solve the continuous portions of the solution (usually
resorting to the method of characteristics, see below), and the shocks are tracked
and characterized with the surge conditions as internal boundary conditions (Lai
1986; Cunge et al. 1980; Montes 1998). These techniques are not in use, given that
in the presence of multiple shocks computations are extremely tedious. Thus, shock
tracking methods are not further considered here.
5.4.2 Outline of Solution Methods
Method of Characteristics (MC): The SWE are transformed to ODEs, as will be

shown in the next section. In simple cases, it is possible to find analytical solutions,
but in most cases, a numerical solution is required using finite difference methods
(FDMs). It is an important method to understand the boundary conditions needed to
solve the SWE. It is also relevant for the development of analytical solutions while
solving the Riemann problem, the key ingredient of the finite volume method
(FVM). The numerical solutions of the SWE using the MC and finite differences
tend to be not used. For a detailed treatment, see Lai (1986) and Chaudhry (2008).
Analytical considerations emerging from the MC are described below.
Finite difference methods (FDMs): The SWE are solved in differential form
approximating the partial derivatives by finite differences using Taylor series
(Chaudhry 2008). These methods are described in Chap. 7.
Finite volume methods (FVMs): The SWE are solved in integral form, as given by
Reynolds’ transport theorem, in a number of control volumes on which the domain
is discretized (Toro 2001). It is a technique widely used. These methods are
described in Chap. 9.
Finite element methods (FEMs): The solution of SWE is approximated by assuming
functions in a number of finite elements on which the domain is discretized. For a
detailed treatment, see Hicks and Steffler (1990) and Katopodes (2019).
5.4 Methods of Solution 217
Discontinuous method of Galerkin: Hybrid method coupling finite volume com-

putation of fluxes resorting to Riemann solvers with discontinuous finite elements.
It is described and applied to open channel flows by Khan and Lai (2014).
5.5 Method of Characteristics
5.5.1 SWE in Characteristic Form
The MC is important to understand physically how a perturbation propagates in a

free surface flow. A perturbation over an initially steady flow generates a variation
in depth and velocity at a given position over its equilibrium values. The method is
also important to understand the boundary and initial conditions needed to solve the
SWE. Simple analytical solutions of the SWE are available using the MC.
The SWE in non-conservative form are, after expanding Eq. (5.3) (Jain 2001;
Chaudhry 2008),
@h @U @h
þD þU ¼ 0;
@t @x @x
ð5:45Þ
1 @U U @U @h
þ þ ¼ So Sf :
g @t g @x @x
Here, So ¼ @zb =@x is the bottom slope, Sf the friction slope and D = A/B the
hydraulic depth, with B as free surface width. Summing of the two equations, after
multiplying the continuity equation by an undetermined variable k yields, after
re-arrangement,

@U @U @h g @h
þ ðU þ kDÞ þk þ Uþ ¼ g So Sf : ð5:46Þ
@t @x @t k @x
The total differentials of the functions U = U(x, t) and h = h(x, t) are defined as
DU @U @U dx
¼ þ ;
Dt @t @x dt ð5:47Þ
Dh @h @h dx
¼ þ :
Dt @t @x dt
Comparing with Eq. (5.46), the terms inside the brackets are exact total differentials
only if
dx g
¼ U þ kD ¼ U þ : ð5:48Þ
dt k
From this identity, k is given by the expression

g 1=2
k¼ : ð5:49Þ
D
Therefore, using this value of k, one can write
dx
¼ U c; ð5:50Þ
dt
where c is the celerity of propagation of a small gravity perturbation
c ¼ ðgDÞ1=2 : ð5:51Þ
After mathematical manipulations, one finds
D
ðU þ xÞ ¼ g So Sf ; ð5:52Þ
Dt
provided that
dx
¼ U þ c; ð5:53Þ
dt
and
D
ðU xÞ ¼ g So Sf ; ð5:54Þ
Dt
if
dx
¼ U c: ð5:55Þ
dt
Here, Escoffier’s stage variable x is defined by (Jain 2001)
Zh
x¼ kdh: ð5:56Þ
0
5.5 Method of Characteristics 219
For a rectangular cross section, the reduced equations are after elementary
manipulations
D dx
ðU 2cÞ ¼ g So Sf along ¼ U c;
Dt dt ð5:57Þ
D dx
ðU þ 2cÞ ¼ g So Sf along ¼ U þ c:
Dt dt
Mathematically, this system of equations states that the total material derivatives of
(U ± 2c) are equal to g(So − Sf), but only along the paths in the x-t plane described
by the integrals of the ODEs dx/dt = U ± c. Therefore, the original SWE, a system
of PDEs valid at any point of the x-t plane, has been transformed into a system of
ODEs valid only along the paths given by the ODEs dx/dt = U ± c. Note that space
derivatives are no more in the equations, given that the x-coordinate is linked to
time by the integrals of dx/dt = U ± c.
Consider the case for which the source term is zero, that is, So = Sf = 0, resulting in
D dx
ðU 2cÞ ¼ 0 along ¼ U c;
Dt dt ð5:58Þ
D dx
ðU þ 2cÞ ¼ 0 along ¼ U þ c:
Dt dt
For inviscid flow over horizontal topography, the so-called Riemann invariants
(U ± 2c) are conserved along the paths in the x-t plane obtained upon integration of
the ODEs dx/dt = U ± c. These ODEs in the x-t plane are known as characteristic
curves. The family of curves given by the ODE dx/dt = U + c are known as for-
ward characteristics, while those given by dx/dt = U − c are known as backward
characteristics. In subcritical flow U < c (F\1), the forward and backward char-
acteristics curves in the x-t plane have positive and negative slopes, respectively
(Fig. 5.8a). In supercritical flow U > c (F [ 1), both the forward and backward
characteristics curves in the x-t plane have positive slopes (Fig. 5.8b). For critical
flow U = c (F ¼ 1), the backward characteristic in the x-t plane is a vertical line
(Fig. 5.8c).
Along the points on a forward characteristic,
D dx
ðU þ 2cÞ ¼ 0 ) U þ 2c ¼ const. along ¼ U þ c; ð5:59Þ
Dt dt
whereas along the points on a backward characteristic results
D dx
ðU 2cÞ ¼ 0 ) U 2c ¼ const. along ¼ U c: ð5:60Þ
Dt dt
Fig. 5.8 Characteristics in

a subcritical flow,
b supercritical flow, c critical
flow
Physically, this means that a perturbation generated at any point x at a given instant
of time t will propagate along the forward and backward characteristics curves
conserving along these paths the corresponding Riemann invariant. The propaga-
tion of any perturbation is therefore a transmission of information along the char-
acteristics: The non-equilibrium values h and U generated at the origin of the
perturbation will generate new non-equilibrium values at other sections at different
instants of time. In general, h and U are variable along the characteristics, which are
thus curved lines.
5.5.2 Initial and Boundary Conditions
Consider forward and backward characteristics curves in subcritical flow origi-

nating at points 1 and 4 and intersecting at point 10 (Fig. 5.9). Additional char-
acteristics may be drawn from other initial points, e.g., from points 2 and 3.
A curvilinear net of characteristics is formed. Consider first only points 1, 4, and 10.
Riemann invariants are conserved along the corresponding characteristics con-
necting 1–10 and 4–10. One may write along the forward characteristic
U1 þ 2c1 ¼ U10 þ 2c10 ; ð5:61Þ
and along the backward characteristic
U4 2c4 ¼ U10 2c10 : ð5:62Þ
Conditions at points 1 and 4 are assumed to be known (depth and velocity),

resulting in a system of two equations and two unknowns at point 10 (depth and
velocity). The same process may be applied to the other interior grid points if
additional characteristics are drawn, forming points 5, 6, 7, 8, and 9. Conditions at
5, 6, and 7 are first computed based on known conditions at 1, 2, 3, and 4. Then,
conditions at points 8 and 9 are determined based on known results at 5, 6, and 7.
Finally, conditions at point 10 are determined based on points 8 and 9. Note that the
flow conditions in the curvilinear domain 1–4–10 influence the conditions at point
10 itself. This zone is known as domain of dependence of point 10. In Fig. 5.10, the
domain of dependence of point 3 is plotted as a dark gray shadow. During the
computational process described above, it is as well noted that the flow conditions
at point 2 “are felt” in the curvilinear triangle between the backward and forward
characteristics emerging from it. This zone is known as domain of influence of
point 2 (Fig. 5.10).
The SWE are a system of 2 hyperbolic PDEs for the unknowns U(x, t) and h(x,
t). To solve the equations in a given channel reach, it is needed to prescribe an
initial condition. It is given by steady values of depth and velocity along the
Fig. 5.9 Characteristics net

for subcritical flow
Fig. 5.10 Domains of

influence and dependence for
subcritical flow
channel. Typically, the discharge is constant and the variation of water depth in
space is computed by solving the steady-state version of the momentum equation
(see Chap. 3)
dq
¼ 0 ) q ¼ qo ¼ const.,
dx
dh So Sf So Sf ð5:63Þ
¼ 2 ¼ :
dx 1 Uc2 1 F2
Here, F is the Froude number and qo the inflow discharge. Further, two boundary
conditions are needed. These are the evolution of depth or velocity with time at
boundary sections of the channel reach. The sections to apply the boundary con-
ditions are different depending on whether the flow is sub- or supercritical. Consider
first the channel reach A–B (Fig. 5.11) with subcritical flow conditions throughout.
At the initial instant of time tn, flow conditions at points An, 2, 3, 4, 5, and Bn are
known. Inside the reach A–B, the forward and backward characteristics emerging
from points 3 and 4, respectively, intersect at point P. Conditions there are
Fig. 5.11 Boundary

conditions for subcritical flow
determined mathematically based on points 3 and 4 on the tn time line, as previ-

ously explained. Now, consider point An+1 at time level tn+1. The backward char-
acteristic lies inside the physical reach A–B, but the forward characteristic is outside
the upstream boundary section of the channel reach. At point An+1, the unknowns
are UAn+1 and hAn+1, but only one equation is available, stating the conservation of
the Riemann invariant along C−, that is, UAn+1 − 2cAn+1 = U2 − 2c2. At point An+1,
therefore, a known value of either UAn+1 or hAn+1 has to be prescribed. This known
value needed is called boundary condition. Following the same arguments, at point
Bn+1, given that the backward characteristic is outside the physical channel reach,
we need to prescribe one boundary condition, either UBn+1 or hBn+1. In subcritical
flow, one boundary condition is required at the upstream section and another at the
downstream section of the reach. The subcritical flow is thus said to be controlled
from up- and downstream. A steady-state flow is determined based on suitable
unsteady computations, as explained by Stoker (1957). The basic idea is to abandon
the formulation of steady flow based on Eq. (5.63) in favor of the solution of the
SWE involving appropriate initial conditions and then allowing for the unsteady
flow to reach the limit in the solutions as t ! ∞. If a steady flow exits, then it
should be the result of unsteady computations for t ! ∞. Suitable boundary
conditions must be prescribed in the initial value problem compatible with the
existence of a steady flow solution (Macdonald 1995).
Consider now the channel reach A–B with supercritical flow conditions
throughout (Fig. 5.12). At a generic point P inside the reach, the forward and
backward characteristics intersect, so flow conditions there are determined mathe-
matically based on points 3 and 4 at the tn time level. Now, consider point An+1 at
time level tn+1. The backward and forward characteristic lines passing through this
point lie outside the upstream boundary section of the channel reach. At point An+1,
the unknowns are UAn+1 and hAn+1, but there are no equations available to undertake
computations. At point An+1, therefore, known values of UAn+1 and hAn+1 have to be
prescribed; that is, two boundary conditions are needed. At point Bn+1, the back-
ward and forward characteristics lie inside the physical channel reach, and its flow
conditions are therefore mathematically determined based on known conditions at
Fig. 5.12 Boundary

conditions for supercritical
flow
Fig. 5.13 Example of initial and boundary condition requirements for subcritical flow
points 5 and 6. Thus, we do not need to prescribe any boundary condition at that
section. In supercritical flow, two boundary conditions are required at the upstream
section of the reach, and the flow is said to be controlled from upstream.
For illustrative purposes of initial and boundary data requirements, consider an
initially steady subcritical flow profile of type M1 shown in Fig. 5.13, formed at
the transition from a canal to a reservoir at its downstream end. Note that normal
depth is larger than the critical depth in this flow profile, as explained in Chap. 3.
The boundary sections of the canal reach are points 1 and 4. The initial condition
needed to solve the SWE is given by the M1 backwater profile, computed by
integration of Eq. (5.63). The typical boundary condition upstream of a subcritical
stream is the inflow discharge hydrograph, e.g., the function q = q(t) at point 1.
Likewise, a typical boundary condition at the downstream section of a subcritical
stream (point 4) is the water depth h in the form of rating curve q = q(h). This
rating curve may be determined by critical flow conditions, uniform flow con-
ditions, or a hydraulic element, e.g., a gate or a weir (see Chap. 2 for details of
governing equations for each boundary condition), depending on the physical
characteristics of the boundary. In our case, the water depth function in the
reservoir is assumed to be known. This relation is inverted to produce h = h(q),
and, with computed values of q(t) at point 4, the function h4 = h4(t) is described.
With these initial and boundary conditions (discharge upstream, water depth
downstream), the SWE are numerically solved along the characteristics until a
new steady-state condition originates if the boundary conditions permit this
equilibrium solution.3
Thus, a subcritical flow is controlled from both up- and downstream, given that
perturbations at boundary sections propagate both up-and downstream of the reach.
The conditions at both boundary sections thus affect the flow inside of the compu-
tational reach. A typical boundary condition at the upstream section is the discharge,
e.g., a flood wave entering gradually into the reach or a sudden discharge increase
from one steady discharge to a new steady discharge. A typical boundary condition
at the downstream section is the water depth, controlled, for example, with
man-made structures, e.g., a gate or a weir. During unsteady flow, both boundary
conditions may depend on time, as explained. If a steady flow solution is physically
permitted by the boundary conditions, it is the asymptotic state produced by
unsteady computations based on the SWE (Macdonald 1995). For a steady sub-
critical flow, the upstream boundary condition q = q(t) reduces to qo = const. in the
entire channel reach (given that the continuity equation is dq/dx = 0), whereas the
downstream boundary condition h = h(t) transforms to h = const. at that section.
The steady form of the SWE is a first-order ODE, so only one boundary con-
dition is needed. This condition is, therefore, the downstream tailwater depth. The
upstream boundary condition of the SWE equations reduces simply to constant
discharge; i.e., the upstream constant discharge is conserved along the entire
channel reach. Based on this physical reasoning originating from the unsteady
SWE, it is possible to find a justification for the traditional rule stating “subcritical
steady flows are controlled from downstream”: Subcritical unsteady flows are
controlled from both up- and downstream, but in steady state, the upstream control
is “switched off.” Mathematically, the ODE for steady flow is a first-order equation
to be solved once a boundary condition is prescribed. From a mathematical
standpoint, this boundary condition is set either at the up- or downstream boundary
sections of the reach. Physically, a more logic choice in agreement with unsteady
SWE computations is to take the downstream boundary section. The same steady
computational rule states that “supercritical steady flows are controlled from
upstream”: Supercritical unsteady flows are controlled fully from upstream, but in
steady state, one of the upstream controls is “switched off.” At the upstream section,
both discharge and depth must be prescribed in the SWE. Following the former
discussion, in steady state, the discharge is a constant, so this is the boundary
condition that is automatically eliminated from the mathematical problem. The
water depth at the upstream boundary section of the reach remains as boundary
condition, in agreement with the steady-state computational rule.
3
Steady flow solutions can be generated using Eq. (5.63). To be of practical relevance, a steady
flow solution must be stable in time; that is, if the steady state is slightly perturbed, the flow
conditions must tend back again to the original steady state (Macdonald 1995). An example of
unstable steady flow solutions is presented when setting uniform flow on steep chutes, where roll
waves are generated (Stoker 1957).
Fig. 5.14 Effect of obstacle

in subcritical stream initially
as steady uniform flow
Fig. 5.15 Stone acting as

control section setting critical
flow conditions
The mathematical tools are now available to better understand the effect of a
stone (obstacle) inserted in an initially uniform and steady stream presented in
Chap. 2, given that it is in reality an unsteady gradually varied flow process. The
obstacle inserted into the initially uniform subcritical stream produces a gradual
variation of water levels (Fig. 5.14) (rapidly varied flow effects are overlooked for
the sake of simplicity). The upstream boundary condition at section A is the con-
stant discharge qo flowing initially in the uniform stream. The flow depths and
velocities within the reach evolve in time, but the discharge in the reach at section A
remains unaltered.
The downstream boundary condition at the vicinity of the stone (point B) is zero
discharge as long as the water depth is below the top of the stone. As the water
begins to pass over the stone, it acts like a round-crested weir, with critical depth
somewhere on the stone surface, close to the top of its round-crested surface
(Fig. 5.15).
Let q be the discharge at section B at any instant of time, the discharge equation
of the flow over the stone is with E as the specific energy over the crest
3=2
2 3 1=2
q¼ gE : ð5:64Þ
3
If h is the flow depth at point B and D the stone height, then
q2
E ¼ Dþhþ : ð5:65Þ
2gh2
Combining Eqs. (5.64) and (5.65), the nonlinear implicit equation is
3=2 " 3 #1=2

2 q2
q g Dþhþ ¼ 0: ð5:66Þ
3 2gh2
Thus, the boundary condition at point B is the function h = h(q) if h > D, obtained
by numerically inverting Eq. (5.66) using, e.g., the Newton–Raphson method. For
h < D, the boundary condition is simply q = 0.
With this reasoning, boundary conditions at points A and B are determined. The
net of characteristics is as shown in Fig. 5.14. Consider the backward characteristic
4-P. It transmits back in space, and forward in time, information on the perturbation
generated at point B. This information is transmitted back and back in space until at
a given instant of time the upstream water levels at A start to increase, in response
to this “information feedback.” After some time, a steady backwater profile is set
within the reach. Note that the discharge at section A was a constant during the
unsteady flow and that all the perturbation process was controlled by water levels at
B. Discharge at B is zero until the water begins to flow over the stone. Later, it
gradually increases until reaching the steady-state value qo that is the boundary
condition at the upstream section A. Therefore, the analogy with steady backwater
computations is evident: The discharge in the system is qo, and we set as boundary
condition at the tailwater section the depth h satisfying Eq. (5.66) for steady state,
that is,
3=2 " 3 #1=2

2 q2o
qo g Dþhþ ¼ 0: ð5:67Þ
3 2gh2
For the supercritical case (Fig. 5.16), the upstream boundary condition is the
constant discharge qo flowing initially in the uniform stream. The other upstream
boundary condition needed is the uniform flow depth of the initial stream. There is
not any mathematical way of transmitting to the inlet (back in space) the pertur-
bation generated by the stone at the tailwater section, as demonstrated by the
characteristics plotted in Fig. 5.16. The flow profile within the channel reach is
uniform and steady, until it impacts the stone, jumping above it.
Fig. 5.16 Effect of obstacle

for supercritical stream
initially in steady uniform
flow
5.5.3 Wave Celerity
Consider the propagation of a flood wave of hydrograph qo(t) over an initially

non-uniform flow in subcritical flow conditions at instant to (Fig. 5.17). The flow is
assumed to be frictionless and the bottom slope small, so that both effects are
neglected. The flow solutions are thus governed by Eqs. (5.59) and (5.60). In
general, given that the initial steady flow is non-uniform, all the characteristics will
be curved lines.
Consider the free surface profile at a time to: At sections 1 and 2 separated by a
distance Δx, the flow depths h(x, t) and velocities U(x, t) are known. After a time
Δt, the free surface profile will move to a new position in response to the flooding
produced by qo(t). The forward and backward characteristics originating from
points 1 and 2 at time t intersect in point P at time to + Δt, located within the
reach Δx. Physically, the perturbations originating at points 1 and 2 at time
t propagate along the characteristics during Δt and finally meet at point P. There,
h and U are determined on the basis of flow conditions at points 1 and 2 by using
the Riemann invariants. The derivative dx/dt is physically the absolute velocity of
propagation of the perturbations along the paths in the x-t plane where Riemann
invariants are conserved. This example shows that this derivative is in fact the rate
at which information is transmitted, but it is not a “wave velocity,” with the usual
meaning; there is no “visible wave profile” displacing in the physical plane at rate
dx/dt. Figure 5.17 highlights that point P of the free surface profile at time
to + Dt cannot be visualized by observing the temporal evolution of the free
surface during the interval Dt. Its spatial coordinate at time to + Dt can only be
determined mathematically, as explained above. Neither depth nor velocity is
constant along the characteristics in general, so there are not observable pertur-
bations of h or U traveling at rate ±(gh)1/2 relative to water flow. This is due to
Fig. 5.17 Physical meaning of dx/dt in method of characteristics for a perturbation propagating
over initially steady non-uniform flow
the fact that the initial steady flow was non-uniform; in this case, the magnitude
dx/dt is not giving the velocity of propagation of a “visual” wave. It is, however,
the mathematical velocity at which the information of the perturbations in depth
and velocity travels along the characteristic lines. In specific cases, dx/dt can yield
a “visual” wave. It was demonstrated that a surge of infinitesimal height propa-
gates with celerity (gh)1/2 relative to the water flow. Thus, a surge of small height
will be visually observable traveling at a velocity close to this value. In this case,
in fact, both h and U remain constant along the characteristic lines. For a negative
surge propagating over an initially uniform flow, to be described below, it will be
shown that dx/dt also yields the wave celerity which can be visually observed.
5.6 Simple Wave Problem: Basic Equations

of Rarefaction Waves
5.6.1 The Simple Wave
The method of characteristics applies to obtain closed-form analytical solutions for

an important type of continuous wave called simple wave. It is a smooth and
continuous wave (values of depth and velocity across the wave are single-valued)
solution of the SWE propagating over an initially steady and uniform stream of
depth ho and velocity Uo (Fig. 5.18). The channel is assumed to be horizontal and
the flow frictionless, leading to Eqs. (5.59) and (5.60) as governing equations. The
flow is perturbed at x = 0 in the form of a prescribed variation of the water depth
with time, or, equivalently, c(0, t) = g(t), where the function g(t) is decreasing with
time. The reduction of water depth with time at x = 0 will produce a perturbation in
the initial equilibrium values (ho, Uo), which propagates in the negative x-direction
at velocity
dx x
¼ U o co : ð5:68Þ
dt t
Fig. 5.18 Simple wave

solution a physical plane,
b characteristic plane (x, t)
5.6 Simple Wave Problem: Basic Equations of Rarefaction Waves 231
This characteristic is a straight line given that both depth and velocity in the
undisturbed flow zone are constants. It can be demonstrated that if a member of a
C-family is a straight line, then all members of this family are also straight lines
(Liggett 1994; Jain 2001). Thus, for all C− characteristics, it can be stated that
dx
¼ U c ¼ const. ð5:69Þ
dt
Along the C− lines, conservation of the Riemann invariant yields
U 2c ¼ const. ð5:70Þ
Thus, coupling Eqs. (5.69) and (5.70), it is seen that along the straight backward
characteristics both velocity and celerity are constants, e.g., (Fig. 5.18)
U ¼ const., c ¼ const. along BC: ð5:71Þ
Thus, one can write, taking as reference the point at the intersection of a C− line
with the t-axis (e.g., point B in Fig. 5.18), where time is s,
U ¼ UB ¼ U ð0; sÞ; ð5:72Þ
c ¼ cB ¼ cð0; sÞ: ð5:73Þ
Along a generic C+ characteristic, one can write based on conservation of the

Riemann invariant (Fig. 5.18)
UD þ 2cD ¼ UP þ 2cP : ð5:74Þ
But point D belongs to the undisturbed flow region, implying
UD ¼ Uo ;
ð5:75Þ
cD ¼ co :
Also, based on Eq. (5.71),
UP ¼ UB ;
ð5:76Þ
cP ¼ cB ;
resulting from Eq. (5.74) after consideration of Eqs. (5.75) and (5.76) in
Uo þ 2co ¼ U ð0; sÞ þ 2cð0; sÞ: ð5:77Þ
This relation states that the Riemann invariant U + 2c is a constant in the entire (x,
t) plane for the simple wave, that is,
U ðx; tÞ þ 2cðx; tÞ ¼ const. Uo þ 2co : ð5:78Þ
Along the C− characteristic BC, it is verified that
dx
¼ U ð0; sÞ cð0; sÞ; ð5:79Þ
dt
which, coupled with Eq. (5.77), produces
dx
¼ Uo þ 2co 3cð0; sÞ: ð5:80Þ
dt
As the C− lines are straight, it is possible to write
dx x
¼ : ð5:81Þ
dt t s
Inserting in Eq. (5.80) yields for the C− lines

x
¼ Uo þ 2co 3cð0; sÞ: ð5:82Þ
ts
As observed, dx/dt in this problem gives the velocity of propagation of a “visible”

wave at the interface of the uniform flow and the negative wave.
5.6.2 Rarefaction Wave: A Negative Surge
Consider a particular case of the simple wave problem. We seek a smooth wave
connecting two constant flow states A and B (Fig. 5.19). The simple wave solution
will be applied assuming the particular case where all C− characteristics are cen-
tered at the origin O, resulting in s = 0. The two edges of the wave will move at
velocities given by the corresponding backward characteristics. This wave is
referred to as rarefaction wave (Toro 2001), and it is a centered simple wave
solution of the SWE (Stoker 1957; Jain 2001). In hydraulics, it is also called
negative surge (Montes 1998; Chanson 2004), implying shallow waters invading
deeper waters. Its fundamental equation is thus
5.6 Simple Wave Problem: Basic Equations of Rarefaction Waves 233
Fig. 5.19 Left-going

rarefaction wave: a centered
simple wave a physical plane,
b characteristic plane (x, t)
U ðx; tÞ þ 2cðx; tÞ ¼ const. UA þ 2cA UB þ 2cB : ð5:83Þ
The free surface profile is easily determined as follows. Along the C−

characteristics
x
¼ U c ¼ const. ð5:84Þ
t
From Eq. (5.83), taking the left state as reference, one can write
U þ 2c ¼ UA þ 2cA : ð5:85Þ
The simultaneous solution of Eqs. (5.84) and (5.85) yields
1 x
c¼ UA þ 2cA ; ð5:86Þ
3 t
or
1 x 2
hðx; tÞ ¼ UA þ 2cA ; ð5:87Þ
9g t
and

1 2x
U ðx; tÞ ¼ UA þ 2cA þ : ð5:88Þ
3 t
5.7 Simplified Models: The Kinematic Wave
The SWE have simplified forms, upon neglecting specific terms of the momentum
equation, widely used in hydrology. Based on the terms retained, dynamic, diffu-
sive, and kinematic waves are defined as follows (Chow et al. 1988)
@U @U @h
þU þ g g So Sf ¼ 0: ð5:89Þ
@t @x @x |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl}
kinematic wave
|fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
diffusive wave
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
dynamic wave
With the actual power of computational methods, the dynamic wave model is by far
the dominant approach. Only the kinematic wave is, however, still used in
hydrological models. Given this fact and the conceptual interest of kinematic
waves, this simplified model is described here. The kinematic wave equations are,
for a rectangular channel, the continuity equation
@h @q
þ ¼ 0; ð5:90Þ
@t @x
and a simplified momentum equation stating balance of gravity and friction forces,
So Sf ¼ 0: ð5:91Þ
The kinematic wave implies that: (i) The flow is not significantly accelerated, and
(ii) the variation of flow depth with distance in a given instant of time is small with a
negligible pressure gradient. The friction slope is assumed to produce a relation of
the type
h ¼ aqb : ð5:92Þ
For example, using Manning’s equation for a wide rectangular channel, the fol-
lowing result in conformity with Eq. (5.92) is generated
5.7 Simplified Models: The Kinematic Wave 235
Fig. 5.20 Rating curve for

(—) dynamic and (- - -)
kinematic waves (adapted
from Henderson 1966)
2 3=10
n
h¼ q3=5 : ð5:93Þ
So
Here, n is Manning’s roughness coefficient. A basic hypothesis of the kinematic

wave theory is that the flow is from instant to instant locally uniform. Unsteadiness
is introduced by the continuity equation. This implies that the rating curve h-q is
singled-valued in kinematic waves (Fig. 5.20). However, in a dynamic wave, the
rating curves are looped, with different branches for the rising and falling limbs of
the flood wave (Henderson 1966).
While solving the SWE, a unique head–discharge rating curve can be prescribed
at the downstream section of a reach based on uniform flow. It is physically
equivalent to set a kinematic-like downstream boundary condition to solve a
dynamic wave model. Thus, from a computational viewpoint, the flow is considered
kinematic at the downstream end and dynamic in all other sections. To circumvent
this deficiency, this boundary condition is moved to far downstream from the
originally selected downstream boundary section, allowing thus in the entire reach
of interest for dynamic loop ratings (Abbott 1975).
@q @q
abqb1 þ ¼ 0: ð5:94Þ
@t @x
Accordingly, kinematic waves are fully determined by changes in discharge

q = q(x, t). The total differential of q is
@q @q
dq ¼ dt þ dx; ð5:95Þ
@t @x
which is rewritten as
Dq @q @q dx
¼ þ : ð5:95Þ
Dt @t @x dt
Comparing Eqs. (5.94) and (5.95), both are identical if
Dq
¼ 0; ð5:96Þ
Dt
dx 1
¼ : ð5:97Þ
dt abqb1
This is the characteristic form of the kinematic wave model. It implies that the
discharge q is a constant along the paths in the (x, t) plane defined by Eq. (5.97).
These paths are forward characteristics, thereby implying that in this type of wave,
information of a perturbation cannot be transmitted forward in time and back
in space. Basically, this means that only one boundary condition upstream of the
river reach is necessary for kinematic wave routing, e.g., the inflow hydrograph
q = q(0, t). Note by differentiation of Eq. (5.92) that
dh
¼ abqb1 : ð5:98Þ
dq
Thus, the kinematic wave characteristics are given with ck as the kinematic wave
celerity (Chow et al. 1988) by
dx dq
¼ ¼ ck : ð5:99Þ
dt dh
If the inflow discharge is known at the upstream end of the reach, a specific value
q = q(0, t) of the discharge at a time t will appear at a position L at the end of the
reach (Fig. 5.21) at time s given by
ZL
dx L
s ¼ tþ ¼ tþ : ð5:100Þ
c k ð qÞ ck ½qð0; tÞ
0
Note that information of the upstream perturbation (variation in discharge) is only

propagated in the downstream direction. The characteristic curves are straight,
given that q = const. along each line. The value of ck and hence the slope of a
characteristic is fully determined by the upstream hydrograph at time level t. A
kinematic wave does not subside, therefore, so the peak discharge qmax remains
5.7 Simplified Models: The Kinematic Wave 237
Fig. 5.21 Kinematic wave routing (adapted from Chow et al. 1988)
unaltered and the wave profile simply suffers a deformation (Fig. 5.21) (Henderson
1966). This deformation implies a steepening of the tailwater face, eventually
leading to the formation of a kinematic shock (Lighthill and Whitham 1955).
However, this issue is debatable, as shown by Henderson (1966), given that the
steepening is accompanied by an increase in the importance of free surface slope
terms neglected in the kinematic wave model. These terms may become important
at the threshold of formation of a kinematic surge, given that they introduce dif-
fusion and attenuation. If they are strong enough, the kinematic surge formation
may be counterbalanced and a smooth wave profile of permanent shape traveling at
constant speed finally forms. This smooth wave solution is called monoclinal wave
(Sturm 2001; Jain 2001). In the more complex case of flow with rainfall, a source
term must be added to the continuity equation and the characteristic lines are then
curved (Chow et al. 1988; Jain 2001).
The kinematic wave celerity can be written as
dq dðUhÞ dU
ck ¼ ¼ ¼ U þh : ð5:101Þ
dh dh dh
Using Manning’s equation for a wide rectangular channel
S1=2
U¼ o
h2=3 ð5:102Þ
n
results in
dU 2 So1=2 1=3
¼ h : ð5:103Þ
dh 3 n
Inserting into Eq. (5.101) produces
5
ck ¼ U: ð5:104Þ
3
Comparing Eq. (5.104) with the absolute celerity cd of the dynamic wave traveling
downstream
cd ¼ U þ ðghÞ1=2 ; ð5:105Þ
the ratio of both signals results in
ck 5
3U 5 F
¼ ¼ : ð5:106Þ
cd U þ ðghÞ1=2 3 F þ 1
Lighthill and Whitham (1955) demonstrated that in a flood wave propagating over
initially uniform flow, both kinematic and dynamic waves coexist. Consider a
smooth and continuous flood wave propagating over an initially steady and uniform
stream. They argued that the main body of the flood behaves, essentially, like a
kinematic wave, whereas the wave fronts are basically governed by the dynamic
wave model. If F\1:5, then Eq. (5.106) yields ck < c, implying that the main
body of the flood wave travels slower than the front advancing at velocity
Uo + (gho)1/2. In reality, close to, but slightly above F 1, the flow is
non-hydrostatic, with undular features, a characteristic beyond the scope of the
SWE. Thus, for a subcritical stream with F\1, which is clearly the most usual case
in river flows, the main body of the wave can in fact propagate at celerity ck as
found by Lighthill and Whitham (1955), whereas the leading positive front would
propagate at speed Uo+ (gho)1/2 (Fig. 5.22) implying that there is a “dynamic wave
ahead the main kinematic wave.” They also showed that in the domain of solution
where the dynamic wave is the governing model, there is no appreciable variation
of flow depth and velocity. Of course, this is likely to be linked to the assumption of
initially uniform flow. The strongest variations in both are accomplished once the
main (kinematic) bulk of the wave arrives (Henderson 1966) (Fig. 5.22).
5.8 Use of Non-conservative Form of SWE
As previously explained, Eq. (5.44) are the non-conservative form of the SWE.
Manipulation permits to write this system as (Cunge 1975; Toro 2001)
5.8 Use of Non-conservative Form of SWE 239
Fig. 5.22 Kinematic and dynamic waves (adapted from Stoker 1957; Henderson 1966)

@ h @ Uh 0
þ ¼ : ð5:107Þ
@t U @x 1 2
2 U þ gh 0
Mathematically, Eq. (5.107) is written in conservative form, yet physically, the

fluid velocity U is not a conserved quantity. Equation (5.107) state conservation of
mass and energy. Equation (5.107) apply to produce solutions of the SWE only if
they are smooth or gradually varied, e.g., as for a river flood wave or a negative
surge in a hydropower canal. However, shock waves are not well reproduced by
Eq. (5.107), to be shown. The celerity of a right-going positive surge is, based on
the conservative form of the SWE [Eq. (5.35)],
1=2
gh1
c¼ ð h1 þ h2 Þ : ð5:108Þ
2h2
This equation is generally in good agreement with experiments for breaking pos-
itive surges (Cunge 1975). The integral form of Eq. (5.107) yields as celerity of the
shock wave (Cunge 1975; Toro 2001)
Fig. 5.23 Ratio of

conservative to
non-conservative positive
surge celerity as function of
surge height
1=2
2gh21
cNC ¼ : ð5:109Þ
ð h1 þ h2 Þ
The ratio of both celerities is then (Fig. 5.23)

c 1 1þ h1
h2
¼ ; ð5:110Þ
cNC 2 ðh1 =h2 Þ1=2
This ratio is only equal to unity in the trivial case h1 = h2, e.g., for smooth solu-
tions. However, Eq. (5.110) shows in Fig. 5.23 that for h1/h2 < 2, the errors of
Eq. (5.107) predicting the celerity of a positive surge are small. Thus, the
non-conservative form of the SWE applies to compute smooth or continuous
solutions, as previously done using the method of characteristics, and for weak
shock waves. Physically, Eq. (5.109) is based on conservation of mass and energy
across a shock wave. Clearly, a shock wave is a highly dissipative phenomenon,
and, thus, a solution based on energy conservation is incorrect. For weak shocks,
however, energy dissipation is small, explaining why Eq. (5.109) is acceptable for
small surge heights. Thus, it is necessary to write the conservation laws in con-
servative form to obtain weak solutions, but, also, the conserved variables should
have a physical meaning in the problem being simulated (Cunge 1975).
A physical shock implies characteristics converging from both sides of the shock
path in the (x, t) plane (Cunge 1975; Toro 2001). It is possible to mathematically
obtain non-physical shocks due to the entropy-violating conditions, where char-
acteristics diverge (Toro 2001; Katopodes 2019). Obviously, these weak solutions
of the SWE must be discarded.
5.9 Limitations of SWE 241
5.9 Limitations of SWE
Limitations of SWE computational results are discussed with an illustrative

example. As previously explained, the integral form of the SWE applies to produce
steady flow solutions prescribing suitable initial and boundary conditions. For
unsteady 1D flow, the integral form of the SWE is
ZZ I ZZ
@U
dxdt þ ðF nÞds ¼ Sdxdt: ð5:111Þ
@t
CV CV
Here, the control volume CV is an area of the (x, t) plane, and the flux integral is
extended along the closed contour line which defines the CV, with s as the curvi-
linear arch length. Consider the spillway depicted in Fig. 5.24. Static water at the
upstream face of the weir is considered as initial condition. Boundary conditions are
physically a constant discharge supplied at the upstream inlet, and a prescribed
tailwater flow depth obtained by regulation of a gate. After transient flow, the
steady flow profile shown in Fig. 5.24 is generated.
SWE produce a continuous steady flow solution at the weir crest, where the flow
changes from sub- to supercritical conditions. Somewhere in the tailwater, a
hydraulic jump is formed to allow for transitional flow from super- to subcritical
conditions forced with a gate. Both types of transcritical flows are therefore
obtained as solutions of SWE [Eqs. (5.111)] (numerical computations to be detailed
in Chap. 9). However, these mathematical results might not be an accurate repre-
sentation of the physical phenomena being simulated: If the fluid pressure is in
reality non-hydrostatic, then the mathematical solution of the SWE is meaningless
(see Chap. 11). Roughly, for example, the continuous flow solution of the SWE at a
weir crest is only accurate if the ratio of crest specific energy to bottom curvature is
below 0.25 (Castro-Orgaz and Chanson 2016). The hydraulic jump is correctly
predicted by the SWE only if a roller is formed (Castro-Orgaz and Hager 2009). If
the hydraulic jump is undular, say for inflow Froude numbers below 1.7 (Montes
and Chanson 1998), then the prediction of the SWE is not in conformity with
experimental observations.
Fig. 5.24 Steady transcritical flow as solution of unsteady computations using SWE
Fig. 5.25 Dike erosion test showing continuous weir flow surface over eroded dike and
hydraulic jump in tailwater (test at experimental flume of IAS-CSIC, Córdoba; photograph by
O. Castro-Orgaz)
The continuous flow solution at a weir crest and a moving hydraulic jump
(positive surge propagating upstream) can be observed in many unsteady envi-
ronmental flows. A prominent example is the gradual erosion of dikes (Fig. 5.25).
These flows are simulated using extended versions of the SWE accounting for the
suspended and bed-load sediment transport modes (Wu 2008) (see Chap. 10).
Limitations of predictions based on these families of models are essentially those
described above; e.g., the fluid pressure shall be hydrostatic, in addition to these
related to empirical sediment transport formulae.
5.10 Hydrologic Routing
5.10.1 Reservoir Routing
There exist cases where the flood routing procedure can be accomplished by
ignoring dynamic effects, e.g., neglecting the momentum equation in the SWE and
solving the continuity equation alone. An important example involves the passage
of a flood wave across a reservoir (Fig. 5.26), given that the reservoir water surface
remains nearly horizontal. Consider here a reservoir with an overflow structure
consisting of a standard spillway crest without gates.
The continuity equation is
@A @Q
þ ¼ 0: ð5:112Þ
@t @x
Integrating between a section upstream of the reservoir (subscript e) and a section

on the spillway crest (subscript s) yields
5.10 Hydrologic Routing 243
Fig. 5.26 Reservoir routing
Zxs
d
Qe Qs ¼ Adx: ð5:113Þ
dt
xe
The volume stored in the reservoir 8 is defined as
Zxs
8¼ Adx; ð5:114Þ
xe
which permits to rewrite Eq. (5.113) as
d8
¼ Qe Qs : ð5:115Þ
dt
The inflow hydrograph is typically a function of time, e.g., Qe = Qe(t), whereas the
spillway rating curve is of the form
h i
Qs ¼ Cd Lð2gE 3 Þ1=2 Cd ð2gÞ1=2 Lh3=2 ¼ CLh3=2 : ð5:116Þ
Here, E is the energy head on the weir crest, approximated to the approach flow
depth above the spillway invert h, L is the spillway width, and Cd is the discharge
coefficient. Note that the C coefficient used in hydrological computations is not
dimensionless. For a known inflow hydrograph and given spillway, Eq. (5.115)
must be numerically solved to compute the outflow hydrograph.
First, the modified Puls method (Chow et al. 1988) is presented here. It is an
implicit finite difference method which is based on the integral form of Eq. (5.115)
over a time period Δt = ti+1 − ti, that is,
ZSi þ 1 Zti þ 1
dS ¼ ðQe Qs Þdt: ð5:117Þ
Si ti
Here, i represents the time subscript, and S is the storage volume above the spillway
crest. Using the trapezoidal rule, Eq. (5.117) yields
Dt
Si þ 1 S i ¼ ðQe Qs Þi þ 1 þ ðQe Qs Þi ; ð5:118Þ
2
or re-arranged
2Si þ 1 2Si
þ Qsi þ 1 ¼ ðQei þ Qei þ 1 Þ þ Qsi : ð5:119Þ
Dt Dt
To apply Eq. (5.119), the inflow hydrograph Qe = Qe(t) and discharge rating curve
Qs = Qs(h) are known in advance. However, to produce the output Qs = Qs(t) as
solution of Eq. (5.119), a function relating S to the actual value of h is additionally
needed. This information results from topographic data of the reservoir by evalu-
ating the integral
Zh
Sð hÞ ¼ F ðhÞdh: ð5:120Þ
0
Here, F(h) is the plan area of the water surface at elevation h relative to the spillway
crest. The function F = F(h) is known from constant elevation contour lines in the
reservoir, taken from available planimetric information. In general, the volume
integral must be numerically solved using a low order method, thereby introducing
errors in the estimation (Fenton 1992). Once S = S(h) is available, the numerical
procedure is as follows:
1. As initial condition, the water surface elevation is assumed to be at the spillway
invert when the inflow hydrograph reaches the upstream reservoir section. Then,
both S and Qs are zero at t = 0. Of course, other scenarios are possible. The
quantity 2S/Δt − Qs is thus zero. The sum Qei + Qei+1 is known from the inflow
hydrograph for all values of i.
2. A routing interval Δt is selected. The function 2S/Δt + Qs versus Qs is con-
structed using the functions Qs = Qs(h) and S = S(h): For a given h, S and Qs are
determined and thus 2S/Δt + Qs.
3. The variable 2S/Δt + Qs at i + 1 is computed from Eq. (5.119).
4. Using the function 2S/Δt + Qs versus Qs, the outflow discharge Qsi+1 is com-
puted from the known value (2S/Δt + Qs)i+1.
5. Apply the identity 2S/Δt − Qs = (2S/Δt + Qs) − 2Qs to generate the initial data
for the next time step.
6. Go back to Step 3 if the final time is not yet reached.
An alternative to use the integral Eq. (5.117) is to directly solve Eq. (5.115),
which is a first-order ODE (Chow et al. 1988; Ayuso 1990; Fenton 1992). The
change in water storage for an infinitesimal change in surface elevation dzs is
d8 ¼ Fdzs : ð5:121Þ
Inserting Eq. (5.121) in Eq. (5.115) produces
dzs Qe Qs
¼ ; ð5:122Þ
dt F
or,
dh Qe ðtÞ Qs ðhÞ
¼ f ðt; hÞ: ð5:123Þ
dt F ð hÞ
This is the equation describing the temporal variation of the reservoir level h = h(t).
Equation (5.123) can be solved using any method to solve ODEs. If the time
derivative is discretized using a forward in time finite-difference
dh hi þ 1 hi
¼ ; ð5:124Þ
dt Dt
the algebraic expression approximating Eq. (5.123) reads
Qe ðti Þ Qs ðhi Þ
hi þ 1 ¼ hi þ fi Dt ¼ hi þ Dt: ð5:125Þ
F ð hi Þ
This equation has the advantage of being extremely simple to apply. Given its
explicit character, the time step Δt must be limited to assure numerical stability.
However, as demonstrated by Fenton (1992), this is not a real limitation in practical
computations. Equation (5.125) is the so-called Euler equation, a first-order
Runge–Kutta method. Runge–Kutta schemes are explicit, using an updating for-
mula of the type (see Chap. 3)
hi þ 1 ¼ hi þ f Dt; ð5:126Þ
where the average value of f is determined differently, depending on the order of the
scheme. Chow et al. (1988) proposed a third-order Runge–Kutta method, whereas
Ayuso (1990) used a fourth-order scheme. For this last option, the average slope is
1
f ¼ ðk1 þ 2k2 þ 2k3 þ k4 Þ; ð5:127Þ
6
with
k1 ¼ f ðti ; hi Þ;

1 1
k2 ¼ f ti þ Dt; hi þ k1 Dt ;
2 2
ð5:128Þ
1 1
k3 ¼ f ti þ Dt; hi þ k2 Dt ;
2 2
k4 ¼ f ðti þ Dt; hi þ k3 DtÞ:
Following Fenton (1992), a comparison of the modified Puls method, the Euler
method, and the fourth-order Runge–Kutta method is shown in Fig. 5.27. The
inflow hydrograph considered is plotted in the figure; note that the peak discharge is
120 m3/s and the base time about 6 h. The reservoir is prismatic with a constant
horizontal area of 50,000 m2. The spillway for this test has a width L = 100 m with
a (hydrological) discharge coefficient C = 2.037 m1/2/s. A time step Δt = 0.1 h was
adopted. The codes prepared for the computations are implemented in the files
“Euler.xls,” “Puls.xls,” and “rk4.xls,” available in Chap. 12. The code for the
fourth-order Runge–Kutta method was adapted from the Fortran code PRAVEM
available from Ayuso (1990).
Fig. 5.27 Comparison of numerical methods for reservoir routing

Comparing the implicit (Puls) method with the explicit fourth-order Runge–
Kutta (rk4) method in Fig. 5.27a shows excellent agreement, thereby highlighting
that either of the two is a good selection. The Puls and the Euler methods are
compared in Fig. 5.27b, indicating reasonable agreement of the two. Note the small
distortion in the ascending outflow hydrograph branch predicted by Euler method,
as well as the overprediction of the peak discharge. However, despite these inac-
curacies, the simple implementation of this method supports its use at minimum for
teaching purposes (Fenton 1992).
Hager et al. (1984) experimentally determined the flood routing process in a
laboratory scale reservoir. The inflow hydrographs followed the smooth equation
q ¼ T n exp½nð1 T Þ; ð5:129Þ
Fig. 5.28 Comparison of reservoir routing solution with experiments involving smooth inflow
hydrographs with (Q*, t*, n, Cd) = a (60 l/s, 60 s, 2, 0.463), b (60 l/s, 60 s, 5, 0.46), c (150 l/s,
120 s, 10, 0.48), d (240 l/s, 60 s, 10, 0.49)
Fig. 5.29 Comparison of reservoir routing solution with experiments involving real inflow
hydrographs with Cd = a 0.471, b 0.475
where the normalized variables are q = Q/Q* and T = t/t*; * denotes a reference
quantity, and n is a parameter. Experiments for four tests are plotted in Fig. 5.28,
along with the values for (Q*, t*, n, Cd) in each run. The reservoir is prismatic in
elevation, with a constant plan area of F = 39.7 m2, and the spillway width is
L = 0.54 m.
The rk4 solver was run using Δt = 0.0005 h (1.8 s); the results are presented in
Fig. 5.28 superimposed with the inflow hydrograph and the measured outflows.
Note that the agreement is good, supporting the accuracy of Eq. (5.123) to describe
reservoir routing processes.
To further test the rk4 solver, an additional comparison with experiments is
made in Fig. 5.29, where the inflow hydrographs are irregular, aimed at describing
real flood events. The numerical simulation produces again a fair approximation to
the experiments, faithfully following the shape of the measured outflow discharge
curves.
5.10.2 Muskingum Channel Routing
The Muskingum method is a hydrologic routing technique for canals based on the
solution of the continuity equation. To match the backwater effects in a reach of a
canal or river, the storage function is composed of two terms, one representing the
backwater storage and the other representing a level (reservoir) storage, e.g. (Chow
et al. 1988)
S ¼ kxðI QÞ þ kQ : ð5:130Þ
|fflfflfflfflfflffl{zfflfflfflfflfflffl} |{z}
backwaterstorage levelstorage
Here, I is the inflow hydrograph and Q the outflow hydrograph. The parameters
k and x are empirical and must be therefore calibrated with data. Parameter x is a
weighting factor ranging from 0 to 0.5, controlling the wave attenuation, while
k has the dimension of time. It is usually estimated as the travel time of the peak
discharge. Using Eq. (5.130),
Si ¼ k½xIi þ ð1 xÞQi ;
ð5:131Þ
Si þ 1 ¼ k ½xIi þ 1 þ ð1 xÞQi þ 1 :
Qei þ Qei þ 1 Qsi þ Qsi þ 1

Si þ 1 S i ¼ Dt Dt ð5:132Þ
2 2
or,
Qsi þ 1 ¼ C0 Qei þ 1 þ C1 Qei þ C2 Qsi ð5:133Þ
where the routing coefficients are
Dt 2kx
C0 ¼ ;
2kð1 xÞ þ Dt
Dt þ 2kx
C1 ¼ ; ð5:134Þ
2kð1 xÞ þ Dt
2k ð1 xÞ Dt
C2 ¼ :
2kð1 xÞ þ Dt
Note that C0 + C1 + C2 = 1. For Δt = k and x = 1/2 results, C0 = 0, C1 = 1 and

C2 = 0, e.g., Qsi+1 = Qei; that is, the flood is moving with a pure translation without
any attenuation.
References
Abbott, M. B. (1975). Computational hydraulics: A short pathology. Journal of Hydraulic

Research, 14(4), 271–285.
Ayuso, J. L. (1990). Circulación de flujos: métodos de cálculo usuales en el diseño de canales y
embalses en pequeñas cuencas [Flood routing: Usual methods in the design of canals and
reservoirs in small drainage basins]. Córdoba, Spain: University of Córdoba Press (in Spanish).
unsteady open channel flow. Journal of Irrigation and Drainage Engineering, 142(1),
04015030.
Chaudhry, M. H. (2014). Applied hydraulic transients (3rd ed.). New York: Springer.
Chow, V. T., Maidment, D. R., & Mays, L. W. (1988). Applied hydrology. New York:
McGraw-Hill.
Cunge, J. A. (1975). Rapidly varying flow in power and pumping canals (Chap. 14). In K.
Mahmood & V. Yevjevich (Eds.), Unsteady flow in open channels, 2, 539–586. Fort Collins,
CO, USA: Water Resources Publications.
De Saint-Venant, A. B. (1870). Démonstration élémentaire de la formule de propagation d’une
onde ou d’une intumescence dans un canal prismatique; et remarques sur les propagations du
son et da la lumière sur les ressauts, ainsi que sur la distinction des rivières et des torrents
[Elementary demonstration of the formula of propagation of a wave or a disturbance in a
prismatic channel; and remarks on the propagations of sound, light, and bores, as also on the
distinction between a river and a torrent: Part 1]. Comptes Rendus de l’Académie des Sciences,
71, 186–195 (in French).
De Saint-Venant, A. B. (1871). Théorie du mouvement non permanent des eaux, avec application
aux crues des rivières et à l’introduction des marrées dans leur lit [Theory of unsteady water
movement, applied to floods in rivers and the effect of tidal flows: Part 2]. Comptes Rendus de
l’Académie des Sciences, 73, 147–154 (in French).
Favre, H. (1935). Etude théorique et expérimentale des ondes de translation dans les canaux
découverts [Theoretical and experimental study of travelling surges in open channels]. Paris,
France: Dunod (in French).
Fenton, J. D. (1992). Reservoir routing. Hydrological Sciences Journal, 37(3), 233–246.
Hager, W. H., Sinniger, R., & Regamey, J.-M. (1984). Reservoir storage equation experimentally
verified. Water Power and Dam Construction, 36(11), 44–48.
Hicks, F. E., & Steffler, P. M. (1990). Finite element modelling of open channel flow. Water
Resources Engineering Report 90-6. Canada: University of Alberta.
Katopodes, N. D. (2019). Free surface flow: Computational methods. Oxford, UK:
Khan, A. A., & Lai, W. (2014). Modeling shallow water flows using the discontinuous Galerkin
method. New York: CRC Press, Taylor and Francis.
Lai, C. (1986). Numerical modelling of unsteady open-channel flow. Advances in Hydroscience,
14, 161–333.
LeVeque, R. J. (2002). Finite volume methods for hyperbolic problems. New York: Cambridge
University Press.
Lighthill, M. J., & Whitham, G. B. (1955). On kinematic waves: Flood movement in long rivers.
Proceedings Royal Society London, A, 229, 281–345.
Macdonald, I. (1995). Analysis and computation of steady open channel flow (Ph.D. thesis).
University of Reading, UK.
Montes, J. S., & Chanson, H. (1998). Characteristics of undular hydraulic jumps: Results and
calculations. Journal of Hydraulic Engineering, 124(2), 192–205.
References 251
Peregine, D. H. (1966). Calculations of the development of an undular bore. Journal of Fluid

Mechanics, 25, 321–330.
Roache, P. J. (1972). Computational fluid dynamics. Albuquerque: Hermosa publishers.
Stoker, J. J. (1957). Water waves: The mathematical theory with applications. New York:
Interscience publishers.
Toro, E. F. (2001). Shock-capturing methods for free-surface shallow flows. Singapore: Wiley.
Toro, E. F. (2009). Riemann solvers and numerical methods for fluid dynamics. London: Springer.
Wu, W. (2008). Computational river dynamics. London, U.K.: Taylor & Francis.
Chapter 6
Ideal Dam Break Waves
6.1 Introduction
The dam break of an ideal fluid stored in an infinite reservoir in a rectangular,

prismatic, and horizontal channel will be considered in this chapter. Solutions for
both dry (Ritter 1892) and wet (Stoker 1957) tailwater conditions will be consid-
ered. Though these analytical solutions are ideal and too simplistic, they are rele-
vant for several reasons. These analytical solutions are illustrative examples of
computations involving the shallow water equations (SWE) where both continuous
and discontinuous waves are formed (Henderson 1966; Jain 2001; Chanson 2004).
Thus, this is an important material to learn basic aspects of unsteady open channel
flows. Further, these are reference solutions to test the behavior of numerical
schemes for the solution of the SWE (Toro 2001). The dam break problem is a
particular case of a more general initial value problem in fluid dynamics, known as
the “Riemann problem” (see Chap. 8) (Toro 2001, 2009). It represents the key
physical ingredient in the development of numerical solutions using the finite volume
method (FVM) for solving hyperbolic equations, to be presented in Chap. 9. Thus, it is
fundamental to master the dam break problem to properly understand the basis of the
FVM.
Historically, the first solution to the ideal dam break wave was given by Ritter
(1892) for dry tailwater conditions. However, it was only until World War II that its
study gained impulse (Ré 1946) given the consideration of the Allied Command to
anticipate the effects of a possible destruction of the large dams on Rhine River in
northwestern Germany (Chow 1959). Examples of catastrophic dam breaks by
bombs of the Allied Forces are the Möhne and Eder Dams on Ruhr Valley
(Kirschmer 1949) (Fig. 6.1).

https://doi.org/10.1007/978-3-030-13073-2_6
254 6 Ideal Dam Break Waves
Fig. 6.1 Dam breaks of a Möhne Dam, b Eder Dam. Photographs taken from the webpage http://
www.thedambusters.org.uk/
6.2 Dam Break Wave Under Dry Tailwater Conditions
Consider the dam break generated by instantaneously removing a vertical gate at

location “O” in a dry channel containing an initially motionless ideal fluid upstream
of the gate (Uo = 0) of depth ho (Fig. 6.2). The basic equations to study these waves
of an inviscid fluid propagating over a horizontal channel are, in characteristic form
(Henderson 1966; Jain 2001) (see Chap. 5),
6.2 Dam Break Wave Under Dry Tailwater Conditions 255
Fig. 6.2 Dry-bed dam break

wave a physical plane,
b (x, t) plane
D dx
ðU 2cÞ ¼ 0 if ¼ U c;
Dt dt ð6:1Þ
D dx
ðU þ 2cÞ ¼ 0 if ¼ U þ c:
Dt dt
Here, U is the water velocity, c = (gh)1/2 the shallow water wave celerity, g the
gravity acceleration, h the water depth, x the space coordinate, and t time.
After dam break initiation, a negative wave spreads back into the reservoir, while
a wave front of zero depth travels forward over the dry channel bed. Line OB
represents the border of the negative wave, which is always in contact with still
water where the water depth equals ho. Thus, this is a straight backward C−
characteristic; all other C− characteristics are also straight lines (Stoker 1957; Jain
2001). Its equation is
dx x
¼ Uo co ¼ ¼ ðgho Þ1=2 : ð6:2Þ
dt t
The dry front is a discontinuity of zero water depth. Therefore, it is simultaneously

a backward and forward characteristic, determined with UF as the absolute wet–dry
front velocity by
dx
¼ UF : ð6:3Þ
dt
Consider a forward characteristic connecting a point on the backward characteristic

representing the negative wave edge, given by Eq. (6.2), and the dry front. The
following identity applies, based on conservation of the corresponding Riemann
invariant,
Uo þ 2co ¼ UF þ 2cF : ð6:4Þ
Given that Uo = 0 (fluid at rest), and that cF = 0 (zero depth at dry front), Eq. (6.4)
yields
2co ¼ UF : ð6:5Þ
When inserted into Eq. (6.3), it produces
dx x
¼ ¼ 2co ¼ 2ðgho Þ1=2 : ð6:6Þ
dt t
Accordingly, the dry front propagating over dry terrain moves at double speed of the
negative wave edge. As the slope determined by Eq. (6.6) is constant, the dry front path
is a straight line passing across origin O. From the simple wave problem, the solution
obtained from Eq. (6.1) for the C− characteristics is (see Chap. 5) [Eq. (5.82)]
x
¼ Uo þ 2co 3½ghð0; sÞ1=2 : ð6:7Þ
ts
Here, h(0, s) is the flow depth at the point of intersection of the C− characteristic
with the t-axis. In the dam break problem Uo= 0, simplifying Eq. (6.7) to
x
¼ 2co 3½ghð0; sÞ1=2 : ð6:8Þ
ts
At the dam axis, x = 0. Thus, given that (t − s) > 0, one must have for all values of t
2co 3½ghð0; sÞ1=2 ¼ 0: ð6:9Þ
From this relation, the depth at the dam axis is a constant, given by
4 c2o 4
h¼ ¼ ho ; ð6:10Þ
9g 9
or,
2
c ¼ co : ð6:11Þ
3
The basic equation of the simple wave (Chap. 5) is written for the dam break
problem as
U ðx; tÞ þ 2cðx; tÞ ¼ const: 2co : ð6:12Þ
Applying Eq. (6.12) to the dam axis, the corresponding velocity is a constant given by
4 2
Uþ co ¼ 2co ) U ¼ co : ð6:13Þ
3 3
The slope of the C− characteristic at the dam axis is thus determined by

dx 2 2
¼ U c ¼ co co ¼ 0: ð6:14Þ
dt 3 3
Equation (6.14) reveals that the t-axis is a C− characteristic corresponding to the

critical flow condition. It further implies that the other C− characteristics cannot
cross the t-axis, given that otherwise an intersection of members of the same family
would result in the development of a surge, which is excluded here. This reasoning
is analogous to set s = 0 in Eq. (6.8). The instantaneous free surface profile is,
using Eq. (6.12)
x
¼ U c ¼ ð2co 2cÞ c ¼ 2co 3c ¼ 2ðgho Þ1=2 3ðghÞ1=2 ; ð6:15Þ
t
or, with h = h(x, t) as the flow depth at an arbitrary point P
1 h xi2
hðx; tÞ ¼ 2ðgho Þ1=2 : ð6:16Þ
9g t
Note also that along the C− characteristics both h and U are constant by virtue of the
simple wave features (Chap. 5). The SWE along the wet–dry interface degenerates
from the hyperbolic to the parabolic type (Dressler 1952). Given that all the C−
characteristics are straight lines passing across the origin O, the C+ lines are of
necessity curved (Stoker 1957; Jain 2001). The network of resulting characteristic
curves is depicted in Fig. 6.3.
Fig. 6.3 Net of

characteristics for dry-bed
dam break wave (adapted
from Dressler 1952)
August Ritter was born on 11 December, 1826 at Lüneburg, in today’s

Germany, and there passed away aged 82 years on 26 February, 1908. He
studied at the Hannover Polytechnic School and at Göttingen University.
Once having obtained the degree as mechanical engineer and the PhD title in
1853, he worked at Hannover, Chemnitz, Leipzig, and Lüneburg in
machinery workshops and lectured at the Hannover Polytechnic School from
1859. Ritter was appointed professor of mechanics at Aachen Technical
University in 1870. He authored books in analytical and technical mechanics,
including chapters relating to hydraulics. He explored the laws of heat flow,
based on the theory of Robert Julius Mayer. After retirement in 1899, he
moved to his native city, and was active in the local society for natural
sciences.
Ritter is well known in hydraulics for two papers published in 1892 and
1895, the first relating to the dam break wave and the second to the hydraulic
jump. Both works are not in his general field of work, yet the dam break
paper was the first approach to a phenomenon that has currently much impact
on dam safety. The so-called Ritter dam break wave describes perfect fluid
flow once a body of water in a prismatic rectangular channel is released by
instantaneously removing a vertical gate. Ritter introduced both the negative
wave front propagating into the reservoir, and the positive wave front moving
downstream over a frictionless horizontal bed. His approach is fundamental,
although the effects of friction and bottom slope are excluded. Ritter was
awarded the honorary doctorate from Technical University of Dresden in
1903.
In the dry-bed dam break problem for x < 0, the backward characteristics has a
negative slope, so the flow is subcritical (F\1; with F ¼ U=ðghÞ1=2 as the Froude
number). At the dam axis, the flow is critical with a backward characteristic
coincident with the t-axis. For x > 0, the backward characteristics has a positive
slope, so the flow is supercritical ðF [ 1Þ: Consider a forward characteristic con-

necting a point on the backward characteristic corresponding to the leading edge of
the negative wave, and an arbitrary point P on the x-t plane (Fig. 6.3). Conservation
of the corresponding Riemann invariant along the forward characteristic gives
UP ¼ 2co 2cP : ð6:17Þ
Inserting the result for the free surface profile, Eq. (6.15), produces for the velocity
as function of time and space
2 hx i 2 hx i
UP ¼ U ðx; tÞ ¼ þ co ¼ þ ðgho Þ1=2 : ð6:18Þ
3 t 3 t
Equations (6.16) and (6.18) are the original solutions to the dry-bed dam break
problem of Ritter (1892). The local and convective accelerations from his solution
are, respectively,
@U 2x
¼ 2; ð6:19Þ
@t 3t
and
@U 4h xi
U ¼ ðgho Þ1=2 þ : ð6:20Þ
@x 9t t
Further, the pressure force gradient for this wave is
@h 2h xi
g ¼ 2ðgho Þ1=2 : ð6:21Þ
@x 9t t
An interesting issue on inspecting Eqs. (6.19)–(6.21) is that the local acceleration

exactly balances the sum of the convective acceleration plus the pressure force
gradient. It implies that Ritter’s unsteady dam break wave is largely governed by
local acceleration effects, a matter rarely emphasized. It can be easily verified by the
reader that the use of Eqs. (6.19)–(6.21) satisfies the non-conservative or primitive
formulation of the momentum balance
@U @U @h
þU þg ¼ 0: ð6:22Þ
@t @x @x
An analogous verification can be conducted by the reader for the continuity

equation. At the origin of the dam axis, the flow is critical and steady in Ritter’s
solution. Thus, with hc as the critical depth and q the unit discharge (see Chap. 2)
2 1=3
U q
F¼ ¼ 1 ) h ¼ hc ¼ : ð6:23Þ
ðghÞ1=2 g
Using Eqs. (6.10) and (6.13), the steady discharge is

2 4 8
q ¼ Uh ¼ co o :
ho ¼ g1=2 h3=2 ð6:24Þ
3 9 27
Inserting Eq. (6.24) into Eq. (6.23) produces for the critical depth
1=3 " 2 #1=3

82 3 23 4
hc ¼ h ¼ ho ¼ ho : ð6:25Þ
272 o 3 3 9
This verifies that the water depth at the dam axis corresponds to the critical flow
depth. Given that the solution at the dam axis is steady, the specific energy concept
applies. There, E is a minimum given by
q2 3
E ¼ hþ ¼ hc : ð6:26Þ
2gh2 2
Inclusion of hydraulic resistance notably modifies the flow behavior in the vicinity
of the dry front,1 yet the conditions near the dam axis and the negative front are not
greatly affected.
6.3 Dam Break Wave Under Wet Tailwater Conditions
A finite water depth is now considered in the downstream channel (Fig. 6.4a).
Ritter’s parabolic profile cannot be simply connected to the stillwater layer, given
that the velocity is then discontinuous at the assembling point B (finite at the
parabolic edge, zero at the still water). Physically, the only possibility to account for
a velocity discontinuity is the formation of a surge of finite height, where both
velocity and depth are discontinuous (Fig. 6.4b) (Stoker 1957; Henderson 1966).
The structure of a dam break wave under wet tailwater conditions implies, there-
fore, a rarefaction wave connected to a surge advancing over the still water
(Fig. 6.5).
1
The effect of hydraulic resistance on dam break waves is a complex problem which was notably
tackled using analytical developments by Dressler (1952, 1954) and Whitham (1955), and
experimentally by Schoklitsch (1917). However, no general analytical solution is yet known to the
problem. Thus, this topic will be presented within a numerical framework in Chap. 9 using finite
volume solvers of the SWE.
6.3 Dam Break Wave Under Wet Tailwater Conditions 261
Fig. 6.4 Wet-bed dam break

wave physically a erroneous,
b correct solution

wave for hd/ho < 0.138
a physical plane, b (x, t) plane.
Critical flow ðF ¼ 1Þ is at the
t-axis (this wave produces an
instantaneous transcritical
flow profile with both
continuous and discontinuous
portions)
Thus, we have to determine the surge front and ensemble it to Ritter’s parabolic
wave profile. The new unknowns are the water depth and velocity in the surge zone
(UB, hB), and the surge front velocity Vw (Montes 1998). At this point, it is illus-
trative to detail an alternative method to the Reynolds transport theorem (see Chap. 5)
to obtain the surge control-volume equations. The surge is a discontinuity in depth and
velocity that propagates at constant speed Vw. Therefore, its shape is not deformed as it
moves, so it is possible to transform the SWE PDEs (see Chap. 5),
@U @F
þ ¼ 0; ð6:27Þ
@t @x
to moving axes with the surge using a Galilei coordinate transformation, where
U = (h, Uh)T and F = (Uh, gh2/2 + U2h)T are the vectors of unknowns and fluxes,
respectively. The space coordinate X in moving axes is related to the absolute
longitudinal coordinate x by X = x − Vwt. Therefore, basic calculus permits to write
the identities
@U @U @X @U
¼ ¼ ðVw Þ; ð6:28Þ
@t @X @t @X
@F @F @X @F
¼ ¼ ð þ 1Þ: ð6:29Þ
@x @X @x @X
The SWE are now transformed for a wave traveling with constant speed and shape to
@U @F @U @F
þ ¼ Vw þ ¼ 0; ð6:30Þ
@t @x @X @X
that is, to the system of ODEs
d
ðF Vw UÞ ¼ 0: ð6:31Þ
dX
These ODEs do not apply at the discontinuity of a surge. Therefore, its integral
form is
ZXd
d
ðF Vw UÞdX ¼ 0; ð6:32Þ
dX
XB
which yields the following solution as a function of the states ahead and behind the
shock
FB Fd
¼ Vw : ð6:33Þ
UB Ud
This vector equation states the so-called Rankine–Hugoniot jump conditions for a
shock wave. Therefore, the two equations to be satisfied are the mass and
momentum balances across the surge, given, respectively, from the Rankine–
Hugoniot jump conditions Eq. (6.33)
V w ð hB hd Þ ¼ U B hB ; ð6:34Þ
2
h2B h
Vw ðUB hB Þ ¼ g þ UB2 hB g d : ð6:35Þ
2 2
After algebraic manipulation, these equations are rewritten as
hB ðVw UB Þ ¼ Vw hB ; ð6:36Þ
and
h2B ðVw VB Þ2 h2d hd Vw2

þ hB ¼ þ : ð6:37Þ
2 g 2 g
An alternative form of Eq. (6.37) presented in Chap. 5 is

1=2
g hB
Vw ¼ ð hB þ hd Þ : ð6:38Þ
2 hd
Equations (6.36) and (6.38) are identical to those found by applying Reynolds’
transport theorem in a fixed reference system (see Chap. 5). The three unknowns to
fully determine the wave structure are UB, hB, and Vw. A third equation needed to
mathematically close the system originates from conserving the Riemann invariant
along a forward characteristic C+ connecting the joint point of the parabolic and
surge portions (point B in Fig. 6.5), with a point on the negative front. The resulting
equation is given by
UB þ 2cB ¼ 2co : ð6:39Þ
The computational process involves the solution of a system of three non-linear

equations for the unknowns UB, hB, and Vw [Eqs. (6.36), (6.38), and (6.39)]. First,
these are coupled to produce a single equation in the depth hB inserting Vw from
Eq. (6.38) and UB from Eq. (6.39) into Eq. (6.36) (Jain 2001)
h i
1=2 h 2ðgh Þ1=2 2ðgh Þ1=2
g hB B o B
ð hB þ hd Þ ¼ 0: ð6:40Þ
2 hd ð hB hd Þ
Its solution to find the root hB is easily implemented using the Newton–Raphson
method as follows. Equation (6.40) is rewritten for convenience as
1=2 h i
g ð hB þ hd Þ
f ðhB Þ ðhB hd Þ þ 2 ðghB Þ1=2 ðgho Þ1=2 ¼ 0: ð6:41Þ
2 hd hB
A better approximation to the root of Eq. (6.41) is based on the Newton–Raphson

method with k as the recursion index by (Hoffman 2001)
fk
hkBþ 1 ¼ hkB : ð6:42Þ
ðdf =dhB Þk
The derivative term needed in Eq. (6.42) is

df g ðhB þ hd Þ 1=2 ðhB hd Þ g ðhB þ hd Þ 1=2
¼ gðghB Þ1=2 þ g : ð6:43Þ
dhB 2 hd hB 4h2B 2 hd hB
A good initial value for hB to initialize the algorithm is

" #2
1 ðgho Þ1=2 þ ðghd Þ1=2
hB ¼ : ð6:44Þ
g 2
Once hB is computed, Vw follows from the momentum balance applying Eq. (6.38),
and UB is given from Eq. (6.39) stating the Riemann invariant along the C+
characteristic by
UB ¼ 2ðgho Þ1=2 2ðghB Þ1=2 : ð6:45Þ
The computational sequence is as follows:

(1) For given values of ho and hd compute UB, hB, and Vw as described above.
(2) For a given instant of time t, the position of the negative front is xback = −cot, of
the surge front is xfront = Vwt, and of the assembling point of the parabolic
profile with the constant state behind the surge is xjoin = (UB− cB)t.
(3) Compute the free surface and velocity profiles applying the rarefaction wave
Eqs. (6.16) and (6.18) if xback < x < xjoin. The constant state values behind the
surge apply if xjoin < x < xfront.
As the tailwater depth hd increases, the velocity UB decreases. For a particular
value of hd, the limit UB = cB can occur, implying that point B becomes the critical
flow section, coincident thus with the t-axis (Fig. 6.6). In this case, the entire
constant flow zone behind the surge is under critical flow. From Eq. (6.10), the
critical flow depth is hc = (4/9)ho. To find this limiting value of hd, UB = cB is set
into Eq. (6.40), resulting in
1=2
g hB hB c B
ðhB þ hd Þ ¼ 0; ð6:46Þ
2 hd ð hB hd Þ

wave for hd/ho = 0.138
a physical plane,
b (x, t) plane. Entire
water body behind the
surge is under critical flow
condition. This case is likely
to be theoretical but not
realistic. Under critical or
near-critical flow conditions,
waves exhibit undular flow
features in response to
non-hydrostatic fluid pressure
ðF ¼ 1Þ
or,

1 1 hd 1=2 1
1þ ¼ 0: ð6:47Þ
2 hd =hB hB ð1 hd =hB Þ
The solution of this implicit equation produces hd/hB 0.3105. Thus, the limiting
ratio of water depths is hd/ho = 0:3105 ð4=9Þ ¼ 0:138. For hd > 0.138ho, the wave
structure is as depicted in Fig. 6.7, with subcritical flow throughout.

wave for hd/ho > 0.138
a physical plane, b (x, t) plane.
Subcritical flow occurs
through the wave ðF\1Þ
The solution for the ideal dry and wet dam break waves described above is
implemented in a source code available in the sheet DamBreakAnalytical.xls
(Chap. 12). An example using this code is presented here, applying it to simulate
the ideal dam break of the Möhne Dam. The maximum water depth in the reservoir
was ho = 32 m (Kirschmer 1949). It was observed that a surge of about 10 m
height propagated in the tailwater. Simulations were conducted assuming hd =
1.5 m. The numerical solution gave hB = 9.7 m, UB = 15.93 m/s, and
Vw = 18.84 m/s. The computed free surface and velocity profiles 1 h after the dam
break are plotted in Fig. 6.8. Note that the surge height is close to that observed.
The relative wave celerity is cw = Vw− UB = 2.9 m/s, which is also in fair agree-
ment with estimations available; an average value of 2.88 m/s is quoted by
Kirschmer (1949). This computation gives an impression of the tremendous mag-
nitude of the flow variables after breaking of a real dam.
Fig. 6.8 Ideal dam break wave of Möhne Dam for 1.5 m in the tailwater, after 1 h of routing
Fig. 6.9 Ideal dam break wave of Möhne Dam for dry tailwater after 1 h of routing
Fig. 6.10 Ratio Vw/(2co) of

wet to dry front celerities as
function of depth ratio hd/ho
The same computation was repeated for the dry-bed condition (hd = 0), resulting
in the profiles plotted in Fig. 6.9. As observed, the increase of velocity at the
positive front is considerable. Comparing these two sets of simulations for the
Möhne Dam results in Vw/(2co) = 18.84/35.43 = 0.531. This shows the tremen-
dously decelerating effect of wetting the tailwater. A plot of Vw/(2co) as a function
of the depth ratio hd/ho is presented in Fig. 6.10, to show with more generality the
behavior of the front celerity as a function of tailwater wetting conditions. Note that
the deceleration effect occurs fast just after a minimal wetting of the tailwater (see
detail in Fig. 6.10). The positive front velocity ratio then becomes nearly constant
with values around Vw/(2co) = 0.47, and then slightly increases when hd/ho raises
above 0.34.
The quality of Ritter’s solution to predict dry-bed dam break waves is examined
in Fig. 6.11, where the free surface predictions using Eq. (6.16) are compared with
the experimental data of Ozmen-Cagatay and Kocaman (2010) at different nor-
malized times T = t(g/ho)1/2 starting at abrupt gate removal. The shape of the dam
break wave curves of Fig. 6.11 is similar to those previously measured by Dressler
(1954) and by Lauber and Hager (1998). The upstream water depth in the exper-
iments by Ozmen-Cagatay and Kocaman (2010) was ho = 0.25 m. Note that the
parabolic shape predicted by Ritter is in fair agreement with observations for this
dataset. For a more general prediction of dry-bed dam break waves, non-hydrostatic
and friction effects have to be accounted for. Castro-Orgaz and Chanson (2017)
found that the negative wave celerity and the wave profile are accurately predicted
if non-hydrostatic effects are considered in a depth-averaged model, whereas fric-
tion effects for laboratory data are confined to the narrow front tip which behaves
essentially like a diffusive wave.
Computations for wet-bed dam break waves are compared with laboratory data
for hd/ho < 0.138 and hd/ho > 0.138 in Figs. 6.12 and 6.13, respectively. During
wave initiation, the wave profile is affected by the non-hydrostatic flow condition, a
feature beyond the capabilities of the SWE, and, thus, agreement with experiments
is not expected during this process (see Chap. 11). For a wet-bed dam break wave
of shallow tailwater (Fig. 6.12), supercritical flow occurs in a part of the wave
profile. As shown by experiments, the turbulence and wave breaking are strong in
these flows, producing an irregular free surface that is not in precise agreement with
the theoretical prediction. However, the overall position of the surge front is rea-
sonably predicted.
For the subcritical dam break wave (hd/ho > 0.138), turbulence effects are less
stringent, and once the wave breaks, the prediction of the SWE is in excellent
agreement with experiments (see Fig. 6.13 at T = 6.51 and 6.89).

theoretical dry-bed dam break
prediction h/ho[x/ho] at
various times T with
laboratory data

theoretical wet-bed dam break
various times T for
hd/ho < 0.138 with laboratory
data

theoretical wet-bed dam break
various times T for
hd/ho > 0.138 with laboratory
data
References
Castro-Orgaz, O., & Chanson, H. (2017). Ritter’s dry-bed dam-break flows: Positive and negative
wave dynamics. Environmental Fluid Mechanics, 17(4), 665–694.
Dressler, R. F. (1952). Hydraulic resistance effect upon the dambreak functions. Journal of
Research, National Bureau of Standards, 49(3), 217–225.
Dressler, R. (1954). Comparison of theories and experiments for the hydraulic dam-break wave.
Proceedings of the International Association Scientific Hydrology, Assemblée Générale, 3(38),
319–328, Rome, Italy.
Marcel Dekker.
Kirschmer, O. (1949). Zerstörung von Talsperren und Dämmen [Destruction of dams and dikes].
Schweizerische Bauzeitung, 67(20), 277–281 (in German).
Lauber, G., & Hager, W. H. (1998). Experiments to dambreak wave: Horizontal channel. Journal
Montes, J. S. (1998). Hydraulics of open channel flow. Reston VA: ASCE.
Ozmen-Cagatay, H., & Kocaman, S. (2010). Dam-break flows during initial stage using SWE and
RANS approaches. Journal of Hydraulic Research, 48(5), 603–611.
Ré, R. (1946). Étude du lacher instantané d’une retenue d’eau dans un canal par la méthode
graphique [Study of instantaneous release of water in a channel using the graphical method].
La Houille Blanche, 1(3), 181–188 (in French).
Ritter, A. (1892). Die Fortpflanzung von Wasserwellen [Propagation of water waves]. Zeitschrift
Verein Deutscher Ingenieure, 36(2), 947–954 (in German).
Schoklitsch, A. (1917). Über Dammbruchwellen [On dam break waves]. Kaiserliche Akademie
der Wissenschaften, Wien, Mathematisch-Naturwissenschaftliche Klasse. Sitzungberichte IIa,
126, 1489–1514 (in German).
Stoker, J. J. (1957). Water waves: The mathematical theories with applications. New York: Wiley.
Whitham, G. B. (1955). The effects of hydraulic resistance in the dam-break problem. Proceedings
of the Royal Society of London, A, 227, 399–407.
Chapter 7
Finite Difference Methods
7.1 Introduction
The SWE are a system of two nonlinear hyperbolic PDEs that must be numerically
solved to describe the time evolution of the fluid velocity and water depth in the
entire computational domain. Finite-difference methods to obtain approximate
numerical solutions are described in this chapter. First, basic numerical aspects are
presented. The implementation of boundary conditions for continuous and dis-
continuous flows is then discussed, and various schemes widely used are explained
in detail. The performance of these schemes is assessed using analytical solutions
and experimental data for selected test cases.
7.2 Basic Numerical Aspects
7.2.1 Remark on Basic Numerical Concepts
Some basic concepts linked to the numerical solution of hyperbolic conservation

laws to be used in this chapter are (Hoffman 2001):
Accuracy: The truncation errors are an accuracy measure of the finite-difference
(FD) scheme. These result from approximating the derivatives by FDs. In general,
the accuracy of a numerical scheme is impaired by both truncation errors and
(machine) round-off errors. However, the latter are usually negligible, given their
smallness.
Convergence: Reference is made to the relation between the solution produced by
the FD scheme and the solution of the original differential equation. Convergence is
reached if the FD solution approaches the solution of the differential equation as the
spatial and temporal steps, Dx and Dt, tend to zero.

https://doi.org/10.1007/978-3-030-13073-2_7
274 7 Finite Difference Methods
Consistency: This refers to the relation between the algebraic equation produced by
the FD scheme and the original differential equation. Consistency is reached if the
FD equation approaches the original differential equation as Dx and Dt tend to zero.
Consistency is needed to have a convergent solution, but it is not a sufficient
condition.
Stability: A numerical scheme is stable if the solution remains bounded as it evolves
in time. An unstable solution becomes unbounded due to the accumulation of errors
and may result in a crashing of computations.
7.2.2 Finite-Difference Approximations
The SWE for a rectangular cross section, written in conservative form, are
@U @F
þ ¼ S: ð7:1Þ
@t @x
Here, U is the vector of unknowns, F the flux vector, and S the source term vector,
with h as the water depth, U the fluid velocity, zb the bed elevation, x the longi-
tudinal coordinate, t the time, g the gravity acceleration, and Cf as a friction
coefficient, given by

h Uh 0
U¼ ; F¼ ; S¼ : ð7:2Þ
Uh U 2 h þ 12 gh2 gh @z
@x Cf U jU j
b
To compute the variables U(x, t) and h(x, t), a discretization of the governing
equations is required. The partial derivatives (or integrals) in the governing equa-
tions are then replaced by discrete numbers. Discretization of the partial differential
equations, Eq. (7.1), leads to finite difference methods (FDM). In turn, the dis-
cretization of integral equations as
ZZ I ZZ
@U
dxdt þ ðF nÞds ¼ Sdxdt; ð7:3Þ
@t
CV CV
leads to the so-called finite volume methods (FVM) (Anderson 1995; Hoffman
2001; Toro 2009), with CV as the control volume in the (x, t) plane, n as the unit
vector normal to the closed contour of CV, and s as the curvilinear coordinate. Here,
FDM are considered, where the PDEs are approximated by algebraic relations
obtained by replacing the partial derivatives by finite-differences.
7.2 Basic Numerical Aspects 275
Fig. 7.1 Finite-difference

grid in the (x, t) plane
Consider a finite-difference grid in the x-t plane (Fig. 7.1), where nodes are
defined by uniform steps Dx and Dt in the x- and t-directions, respectively. The
objective is to replace the original PDEs, valid at any point on the x-t plane, by
algebraic expressions that apply only to the grid points (i, k) of the mesh, with i and
k as the node indices in the x- and t-directions, respectively.
Finite-difference representations of derivatives are based on Taylor series
expansions (Sturm 2001; Chaudhry 2008). For example, for vector F, a Taylor
expansion in the x-direction at point i + 1 and time level k reads (Hoffman 2001)
k k 3 k
@F @ 2 F ðDxÞ2 @ F ðDxÞ3
Fkiþ 1 ¼ Fki þ Dx þ þ þ : ð7:4Þ
@x i @x2 i 2 @x3 i 6
Solving for the first derivative of F from the former expression results in
k 2 k 3 k
@F Fkiþ 1 Fki @ F Dx @ F ðDxÞ2
¼ þ : ð7:5Þ
@x i Dx
|fflfflfflfflfflffl{zfflfflfflfflfflffl} @x2 i 2 @x3 i 6
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Finite difference approximation Truncation error
If the first derivative in Eq. (7.5) is taken as the finite-difference approximation, the
remaining part of the equation is the so-called truncation error, thus
k
@F Fk Fki
iþ1 : ð7:6Þ
@x i Dx
The partial derivative of a continuous magnitude, F, is replaced by the algebraic

difference. As the neglected truncation error is proportional to Δx, this approxi-
mation is first-order accurate. Mathematically, this is expressed as the so-called
forward finite-difference
k
@F Fkiþ 1 Fki
¼ þ OðDxÞ: ð7:7Þ
@x i Dx
A Taylor expansion at point (i − 1, k) about the point (i, k) reads

2 k 3 k
@F k @ F ðDxÞ2 @ F ðDxÞ3
Fki1 ¼ Fki þ ðDxÞ þ þ þ : ð7:8Þ
@x i @x2 i 2 @x3 i 6
From similar arguments, one finds for the so-called backward finite-difference
k
@F Fk Fki1
¼ i þ OðDxÞ: ð7:9Þ
@x i Dx
In applications, the first-order accuracy is often insufficient. Simply subtracting the

two Taylor series, Eqs. (7.4) and (7.8), results in (Hoffman 2001)
3 k
@F k @ F ðDxÞ3
Fkiþ 1 Fki1 ¼ 2 Dx þ 2 þ : ð7:10Þ
@x i @x3 i 6
From Eq. (7.10), the so-called central finite-difference approximation becomes

k
@F Fk Fki1
¼ iþ1 þ O(Dx2 ), ð7:11Þ
@x i 2Dx
which is second-order space accurate. Forward, backward, and central

finite-differences can also be formed for the time derivative @U=@t.
Remark: A logic choice for the finite-difference quotient to use in the discretization
of spatial gradients is based on the concept of upwinding. Most field equations in
environmental physics contain terms representing advection and diffusion pro-
cesses, as
@f @f @2f
þ u ¼ D 2 : ð7:12Þ
@t @x
|{z} @xffl}
|fflffl{zffl
convective term diffusion term
Here, f is a property advected and diffused, e.g., concentration of a scalar, D is a

diffusion coefficient, and u is the advection speed (Roache 1972). Neglecting dif-
fusion, Eq. (7.12) reduces to
@f @f
þ u ¼ 0: ð7:13Þ
@t @x
While the obvious choice for discretization of the time derivative to advance in
time starting at known time level is the forward finite-difference
@f fik þ 1 fik
¼ þ O(DtÞ; ð7:14Þ
@t Dt
the best choice for the spatial gradient is not so evident. The following options are
mathematically possible:
Forward finite-difference
@f fikþ 1 fik
¼ þ O(DxÞ; ð7:15Þ
@x Dx
Backward finite-difference
@f fik fi1
k
¼ þ O(DxÞ; ð7:16Þ
@x Dx
Central finite-difference
@f fikþ 1 fi1
k
¼ þ O(Dx2 Þ: ð7:17Þ
@x 2Dx
These finite-difference quotients take information on the perturbation propagation

in different ways. The backward difference takes information at a node from its
upstream side, the forward difference from the downstream side, and the central
difference from both. An upwind model uses information available about the wave
propagation to select the proper numerical discretization of the spatial derivatives in
the PDE. An upwind approximation for Eq. (7.13) is designed by taking infor-
mation at a node from the side to which the perturbation approaches while prop-
agating. The simple first-order upwind model for the convection equation is then
fik þ 1 fik fik fi1

k
Dt þu Dx ¼ 0 if u [ 0,
fik þ 1 fik f k 1 fik
ð7:18Þ
Dt þ u i þDx ¼0 if u \0.
This scheme, due to Courant et al. (1952), is conditionally stable. Other choices
based on central [Eq. (7.17)], or one-sided finite-differences in the same direction as
u (down-winding schemes), are unconditionally unstable (Roache 1972; Anderson
1995). A useful one-sided scheme is only obtained by accounting for the sign of the
wave propagation speed u. This is the utility of up-winding. The centered scheme
based on Eq. (7.17) is not good, given the unstable behavior. Note that the upwind
differences are related to the central difference by
fik fi1
k fikþ 1 fi1
k Dx fikþ 1 2fik þ fi1
k
Dx ¼ 2Dx 2 ðDxÞ2
if u [ 0,
fikþ 1 fik fikþ 1 fi1
k fikþ 1 2fik þ fi1 k ð7:19Þ
Dx ¼ 2Dx þ Dx
2 ðDxÞ2
if u \0.
Therefore, upwind methods introduce numerical diffusion into the centered scheme
to make it stable. The diffusion is numerical with D = Dx/2.
Returning to the SWE, a possible model for the discretization of Eq. (7.1)
written without source terms,
@U @F
þ ¼ 0; ð7:20Þ
@t @x
is with accuracy O(Dt, Dx2)
Uki þ 1 Uki Fk Fki1

þ iþ1 ¼ 0: ð7:21Þ
Dt 2Dx
In spite of the second-order spatial accuracy, this finite-difference model is not

suitable, given its instability (Liggett and Cunge 1975; Cunge et al. 1980; Montes
1998; Sturm 2001; Chaudhry 2008), as also explained with Eq. (7.13). Therefore,
the proper selection of an FD model should not rely on producing small truncation
errors simply by increasing the order of the scheme. The lesson to be learnt is: Not
all mathematically permissible discretizations are necessarily a good approximation
for the physical problem to be simulated.
7.2.3 Shock-Capturing Schemes
Conservative and non-conservative forms of the SWE apply to simulate continuous

and smooth wave fields. However, the celerity of positive surges is poorly predicted
by the non-conservative form (Chap. 5). Thus, in this chapter, only the conservative
form is considered. The SWE in differential form are valid only for continuous
flows, excluding the generation of discontinuities. Thus, logic would dictate to
abandon Eq. (7.1) and any FD approximation of this model when dealing with
discontinuous flows. Practice, however, reveals that shock-capturing models can be
constructed based on Eq. (7.1) (Cunge 1975). The reason for this apparently con-
tradictory success is as follows: The finite-difference algebraic equations repre-
senting the original PDEs have shock-capturing ability because they mimic (or
Fig. 7.2 Computed shock

solutions contaminated by
a diffusion errors and
b dispersion errors
approximate) the discrete form of the integral conservation laws (to be derived for
the FVM in Chap. 9). Therefore, some finite-difference schemes approximating the
original PDEs as DU/Dt + DF/Dx = S are interpreted as particular discretizations of
the integral Eq. (7.3). In this chapter, only shock-capturing finite-difference
schemes are considered.
Two typical numerical errors appear in shock-capturing schemes. First-order
finite-difference schemes introduce numerical diffusion, smearing the shock front
(Fig. 7.2a). Second-order schemes introduce numerical dispersion errors. These
produce high-frequency oscillations near sharp gradients, as close to a shock
(Fig. 7.2b). In general, the second-order schemes are preferred in practice, given the
sharp resolution of shocks. The dispersion errors are locally controlled by adding
limited numerical diffusion.
Some finite-difference schemes mimic the integral form of the conservation laws
and thus produce shocks in the solution, as explained. However, not all
shock-capturing schemes produce physical shocks in the solution. To reproduce a
physical shock in the solution, the following entropy condition must be satisfied
(Lax 1954; Lax and Wendroff 1960; Toro 2001)
kðUL Þ S kðUR Þ: ð7:22Þ
Here, k(U) = U + (gh)1/2 are the eigenvalues associated with the C+ characteristics
for the right-going shock depicted in Fig. 7.3, and L and R denote states at the left
and right sides of the shock displacing with velocity S. Equation (7.22) implies that
the characteristic curves converge to the path of the shock in the (x, t) plane
(Fig. 7.3). If Eq. (7.22) is not satisfied, the finite-difference scheme may produce a
non-physical jump (Bhallamudi 2002). Upwind models for the SWE use the
eigenvalues to detect the direction of signal propagation and thus the adequate
finite-difference representation of spatial gradients. At a critical point [U = (gh)1/2],
the eigenvalue corresponding to the backward characteristic vanishes. Then, some
Fig. 7.3 Sketch of

discontinuity in the physical
plane and conditions for a
physical shock in (x, t) plane
Fig. 7.4 Unphysical shock in

transcritical dam break wave
schemes are unable to distinguish a rarefaction wave crossing smoothly a critical

point from a shock (Bhallamudi 2002; Zoppou and Roberts 2003). These models
need an entropy fix to produce meaningful results. Without this correction, a pre-
diction with an unphysical jump as in Fig. 7.4 is produced.
7.2.4 Explicit and Implicit Schemes: Stability
In Eq. (7.21), the vector U at the new time level k + 1 is computed based on data at
time level k, where all information is known. It permits to evaluate F at all nodes of
the mesh. Values of U at each node i for time t + Dt are therefore explicitly solved.
This type of model is referred to as explicit. However, it is possible to evaluate F at
time level k + 1, resulting in
Uki þ 1 Uki Fk þ 1 Fki1

þ1
þ iþ1 ¼ 0: ð7:23Þ
Dt 2Dx
Note that F = F(U), e.g.,
@U @FðUÞ
þ ¼ 0; ð7:24Þ
@t @x
with
! !
h U1 Uh U2
U¼ ¼ ; F¼ ¼ U22 : ð7:25Þ
Uh U2 U 2 h þ 12 gh2 U1 þ 12 gU12
Thus, Eq. (7.23) cannot be directly solved for U at the new time level k + 1. Rather,
it is an implicit equation to be solved numerically to compute the vector U at time
level k + 1. The equation at a node is thus
k þ 1
Uki þ 1 Uki Fk þ 1 Uk þ 1 Fki1
þ1
Ui1
þ iþ1 iþ1 ¼ 0: ð7:26Þ
Dt 2Dx
This type of model is said to be implicit. The implicit method results in a system of
nonlinear implicit equations, to be simultaneously solved for all nodes. They are
unconditionally stable (Chaudhry 2008). The selection of the type of scheme,
implicit or explicit, is by no means a rigid decision (Anderson 1995). The implicit
methods are applied to large time steps, given the unconditional stability. However,
the solution at each time level involves a more complex numerical method, which
obviously implies more coding work. Additionally, even if computations being
stable for any time step are an advantage, the use of a large step may deteriorate the
quality of the computed transient wave solution due to its contamination by trun-
cation errors. The explicit methods involve simple computations for each new time
level, with the corresponding simplicity of coding. However, to be stable, the time
step cannot be arbitrary. Rather, a physical constraint must be added, as explained
below, to limit the largest permissible time step. Due to coding simplicity, and if the
main interest is the computation of an accurate wave pattern which is likely to
evolve fast, as in a dam break wave, explicit schemes are a good practical choice
(Zoppou and Roberts 2003). In this chapter, the attention is thus restricted to these
techniques. To implement implicit schemes, see Sturm (2001) or Chaudhry (2008).
In an explicit method, the time step Dt for a size of cell Dx is limited to the time
needed by perturbations originating at adjacent nodes to propagate and interact.
Consider a node i, surrounded by nodes i − 1 and i + 1 in a finite-difference mesh
(Fig. 7.5). Assume that the flow is subcritical in the entire computational domain
and that Δx is fixed. If Δx is small, it is permissible to assume that the forward and
backward characteristics originating at nodes i − 1 and i + 1, respectively, are
Fig. 7.5 Stable time

stepping: analytical domain of
dependence at the
computational point d of the
finite-difference mesh lies
inside the numerical domain
of dependence [Adapted from
Anderson (1995)]
approximately straight lines, intersecting at point P. For simplicity’s sake, it is

assumed that flow conditions at time level k are uniform. The numerical domain of
dependence is the triangle defined by points P, i + 1, and i − 1. Flow conditions at
point P are influenced by these in the area defining the numerical domain of
dependence. In this domain, with c = (gh)1/2 as the shallow water wave celerity, the
maximum time step (Δt)max to stick inside it is
Dx
ðDtÞmax ¼ : ð7:27Þ
Uc
Consider that the time step Δt selected for computations in our finite-difference mesh
is less than this value, resulting in point d at the new time level. The points on the
x-axis at the actual time level k influencing this node lie between i − 1 and i + 1.
Computationally, the analytical domain of dependence defined by the characteristics
at point d (i, k + 1) (shaded area in Fig. 7.5) thus lies inside the numerical domain of
dependence (triangle i − 1, d, i + 1) produced by the mesh. In this case, computations
are stable (Anderson 1995).
Consider now that the time step selected in the finite-difference mesh is greater
than the maximum value fixed by the numerical domain of dependence of point P
(Fig. 7.6). Therefore, the computational point d at the new time level k + 1 is above
point P. The points on the x-axis at the actual time level k influencing this node lie
outside the interval (i − 1, i + 1). The analytical domain of dependence at point d,
triggering the state of node i at the new time level k + 1, then lies outside the
numerical domain of dependence. This implies that at the new time level our
finite-difference model is demanding information about the propagation of pertur-
bations to nodes located outside the computational portion (i − 1, i + 1). This
information is not actually arriving at node d, so computations will become
unstable. The production of computations is thus forced at time Dt after the meeting
of perturbations originating from nodes i − 1 and i + 1, occurring at time (Dt)max.
Fig. 7.6 Unstable time

stepping: analytical domain of
dependence at the
computational point d of the
finite-difference mesh lies
outside the numerical domain
of dependence [Adapted from
Anderson (1995)]
As a consequence, the “computations arrive late in time,” by which the model loses
the information generated at nodes i − 1 and i + 1 “in the immediate past.” That is,
we demand our model to do computations at a time when all the relevant infor-
mation regarding perturbation propagation already passed through this spatial
location just a moment ago. From this, reasoning originates the stability condition
for explicit methods, stating that the Courant–Friedrichs–Lewy number CFL must
satisfy (Courant et al. 1928)
Dt
CFL ¼ ðU cÞ 1: ð7:28Þ
Dx
It assures that the analytical domain of dependence at a node lies within the
numerical domain of dependence (Anderson 1995). In other words, let a time step
be small enough to advance in time “catching” the information of perturbations
before it leaves a portion of the computational domain. From a practical viewpoint,
a common Δt for all nodes to “jump” to the new time level k + 1 is determined.
Therefore, the smallest Δt possible is selected, that is (Toro 2001; Chaudhry 2008),
CFL Dx
Dt ¼ : ð7:29Þ
maxðjUi j þ ci Þ
For practical computations, CFL ¼ 0:9 is typically used (Toro 2001; Chaudhry
2008).
Richard Courant was born on January 8, 1888, at Lublinitz, formerly in

Germany and today in Poland, passing away at age 84 on January 27, 1972, at
New Rochelle, NY, USA. He started studies in mathematics at Breslau
(currently Wroclaw) University, moving in 1905 to University of Zurich and
later to University of Göttingen, Germany, becoming a collaborator of David
Hilbert (1862–1943) and earning his Ph.D. title in 1910. He continued his
research during a two-year period at the University of Münster as Professor of
mathematics, founding there its Mathematical Institute, which he directed
from 1928 to 1933. He left Germany in 1933 due to his Jewish background,
moving to University of Cambridge in the UK. In 1936, he accepted a pro-
fessorship at New York University, New York City, NY, founding an
institute for graduate studies in applied mathematics. It was renamed in 1964
Courant Institute of Mathematical Sciences, gradually developing in the
foremost centers of applied mathematics .
Courant and Hilbert authored in 1924 the textbook Methods in
Mathematical Physics, several times revised and widely used by mathe-
maticians, physicists, and engineers. The book deals mainly with methods to
numerically solve both ordinary and partial differential equations using
finite-differences. Courant also is the namesake for the Courant–Friedrichs–
Lewy (CFL) condition in fluid mechanics, establishing a necessary condition
for convergence of hyperbolic partial differential equations, as are, e.g., the
shallow water equations, arising in the numerical analysis of explicit time
integration schemes. He also was strongly related to the advances in the finite
element method, a numerical method for solving problems of engineering and
mathematical physics. Given that most equations in fluid mechanics and
hydraulics do not have analytical solutions, the role of Courant in advancing
the solution methods is outstanding.
Remark: Lax equivalence theorem

This fundamental theorem in the analysis of finite-difference equations was for-
mulated by Lax and Richtmeyer (1956) stating that, for a well-posed linear initial
value problem, a consistent finite-difference approximation will produce a con-

vergent solution if there is stability. This theorem is important because if stability
and consistency are satisfied (which are simple to analyze), convergence is assured
(which is difficult to analyze directly). Obviously, the SWE are nonlinear and most
of the practical applications are not amenable for simplified linearized solutions, but
the theorem is also accepted for more general problems.
7.3 Boundary and Initial Conditions
7.3.1 Initial Conditions
To compute the transient wave solution of the SWE, it is necessary to know the
values of depth and velocity in all the spatial nodes of the mesh. These values
cannot be arbitrary: If these data are not compatible with the SWE, then spurious
waves will be generated in the system after initiation of the transient computations,
masking the real waves. Thus, physically realistic initial conditions must be
determined. Typically, one of the two following options is possible:
System initially static: The free surface is horizontal, and the fluid velocity is zero in
all nodes. Dry areas may be included in the computational domain.
System initially under steady flow: Depth and velocity must be determined using the
steady-state techniques (Chaps. 2–4). The solution involves the numerical solution
of the ODE for gradually varied or continuous steady flows (Henderson 1966; Jain
2001)
dh @z b
Sf
¼ @x q2 ; ð7:30Þ
dx 1 3 gh
where h is water depth, q unit discharge, zb bed elevation, and Sf the friction slope.
For rapidly varied or discontinuous steady flows (hydraulic jumps), Bélanger’s
equation is used (Henderson 1966; Jain 2001)
h2 1 h 1=2 i
¼ 1 þ 8F21 1 : ð7:31Þ
h1 2
Here, h1 and h2 are the sequent depths and F1 ¼ ðq2 =gh31 Þ1=2 is the inflow Froude
number at the supercritical section.
7.3.2 Boundary Conditions for Continuous Flows
The flow state at the up- or downstream ends of the reach determines the nature of
the boundary conditions required there. For subcritical flow, one boundary condi-
tion is needed at each end of the reach, either the flow depth or the discharge or a
function relating both (rating curve). In continuous flows, the SWE in characteristic
form are applied to determine the other unknown flow variable there. For a rect-
angular cross section, and with So as the bottom slope, the SWE in characteristic
form are (Jain 2001; Chaudhry 2008)
D dx
ðU 2cÞ ¼ g So Sf if ¼ U c; ð7:32Þ
Dt dt
D dx
ðU þ 2cÞ ¼ g So Sf if ¼ U þ c; ð7:33Þ
Dt dt
or,
DU g Dh dx
¼ g S o Sf if ¼ U c; ð7:34Þ
Dt c Dt dt
DU g Dh dx
þ ¼ g So Sf if ¼ U þ c: ð7:35Þ
Dt c Dt dt
At an upstream boundary section with subcritical flow, discretization of Eq. (7.34)

yields (Fig. 7.7a)
g k
U1k þ 1 Upk k hk1 þ 1 hkp ¼ g So Sf p Dt: ð7:36Þ
cp
Assuming that flow conditions at point P are approximated by those of node 2, to

avoid interpolations, the result is (Chaudhry 2008)
g kþ1 k
U1k þ 1 ¼ U2k þ k
h1 hk2 þ g So Sf 2 Dt: ð7:37Þ
c2
For a more rigorous computation, the exact coordinate of point P is determined

by the backward characteristic (García-Navarro and Saviron 1992). At a down-
stream boundary section with subcritical flow, discretization of Eq. (7.35) yields
(Fig. 7.7b)
g k
UNk þ 1 Upk þ k hkNþ 1 hkp ¼ g So Sf p Dt; ð7:38Þ
cp
7.3 Boundary and Initial Conditions 287
Fig. 7.7 Characteristics in

subcritical flow at a upstream
boundary and b downstream
boundary
which is similarly simplified to (Gharangik and Chaudhry 1991; Chaudhry 2008)

g k
UNk þ 1 ¼ UN1
k
hkNþ 1 hkN1 þ g So Sf N1 Dt: ð7:39Þ
ckN1
If the flow is supercritical at the upstream boundary section, two boundary

conditions must be supplied. If the flow is supercritical at the downstream boundary
section, both Eqs. (7.34) and (7.35) are discretized to determine (h, U) at point
(N, k + 1) based on data at point (N–1, k).
7.3.3 Boundary Conditions for Discontinuous Flows
If the flow at a boundary section is discontinuous, Eqs. (7.34) and (7.35) cannot be
used (Terzidis and Strelkoff 1970; García-Navarro and Saviron 1992). Then, the
Rankine–Hugoniot jump conditions stating the mass and the momentum conser-
vation equations across a shock wave apply, namely [Eqs. (5.17) and (5.35)]
U1 h1 U2 h2
Vw ¼ ; ð7:40Þ
h 1 h2

1=2
gh1
Vw ¼ U2 þ ð h1 þ h2 Þ : ð7:41Þ
2h2
Here, Vw is the absolute surge velocity and subscripts 1 and 2 indicate conditions
up- and downstream of the shock front. Equations (7.40)–(7.41) contain five
variables (h1, h2, U1, U2, and Vw), so they apply to produce boundary conditions
provided that three are known. If a flow starts at i = 1 as discontinuous supercritical
flow due to a sudden (known) increase of discharge, the equations can be used to
generate the value of the unknown water depth there if the tailwater conditions (h2,
U2) are given.
7.4 Explicit Schemes
7.4.1 FTCS Scheme
The forward in time, centered in space scheme is possibly the simplest FD scheme,
but, as demonstrated by Liggett and Cunge (1975), it is unstable. It is therefore
called “unstable scheme.” The fluxes F are evaluated at the time level k where the
vector U is completely known, using central finite differences. The conservative
formula is then
Dt k
Uki þ 1 ¼ Uki F Fki1 þ Ski Dt: ð7:42Þ
2Dx i þ 1
This model appears to work under extremely low values of CFL in combination
with artificial viscosity to dampen oscillations. However, the scheme is of no
practical value.
7.4 Explicit Schemes 289
Peter David Lax was born on May 1, 1926, in Budapest, Hungary. His
Jewish parents both were physicists. They left Hungary in 1941, traveling via
Lisbon to New York, NY, USA. He started his studies in mathematics in
1944 at New York University (NYU). During World War II, the army sent
him to Texas A&M University, then to the Oak Ridge National Laboratory,
and soon afterward to the Manhattan Project at Los Alamos, NM, where he
began working as a calculator operator, but eventually moved on to
higher-level mathematics. After World War II, he remained with the army at
Los Alamos for another year, while taking courses at the University of New
Mexico, and then studied at Stanford University for a semester with the
Hungarian mathematicians Szegő and Pólya. He returned to NYU in 1946,
obtaining his graduation and earning the Ph.D. title in 1949 under the
supervision of Kurt O. Friedrichs (1901–1982) .
Lax made remarkable contributions early in his career continuing to
produce research, which changed the direction of many areas of mathematics.
In 1957, he published the important paper Asymptotic solutions of oscillating
initial value problems, laying the beginnings of the theory of Fourier integral
operators. Asked what was so novel about the viewpoint that made the ideas
able to enjoy such wide application, he replied: “It is a micro-local description
of what is going on. It combines looking at the problem in the large and in the
small. The numerical implementation of the micro-local point of view is by
wavelets and similar approaches, which are powerful numerically.” He
thrived in the Courant Institute of Mathematical Sciences, NYU, where
applied mathematics was studied alongside relevant pure mathematics in an
exciting mix of ideas, which led to great progress. He was appointed Institute
Director in 1972, continuing until 1980. It was a particularly difficult time to
take on this role since NYU had just closed down their School of
Engineering, moving the mathematicians from there into the Courant
Institute. This produced friction when these people wanted to set up their own
computing department while a new Computer Science Department had just
been founded. He succeeded in ensuring that there were not two rival
departments in the Institute, but the politics involved was difficult. His most
important works include Nonlinear partial differential equations in applied
science (1983), Hyperbolic systems of conservation laws and the mathe-
matical theory of shock waves (1987), or his Hyperbolic partial differential
equations (2006). He has received numerous awards and honors for his
mathematical researches and his advances in the field of computational
mathematics.
7.4.2 Lax’ Diffusive Scheme
The Lax diffusive method is a one-step finite-difference method with

shock-capturing ability, but numerically diffusive. It is a variant of the unstable
scheme. The scheme is given by the updating formula

Uki1 þ Ukiþ 1 Dt k
Uki þ 1 ¼ F Fki1 þ Ski Dt: ð7:43Þ
2 2Dx i þ 1
It is conditionally stable, and the time step Dt is restricted by the CFL condition
(Hirsch 1988, 1990; Hoffman 2001). The Lax diffusive method depends on the
value of CFL, as highlighted below. An FD scheme is said to be consistent if the
finite-difference form tends to the original PDE if both Dx and Dt ! 0. The fol-
lowing space and time Taylor expansions around the node (i, k) are formed
k 2 k
@U @ U ðDtÞ2
Uki þ 1 ¼ Uki þ Dt þ þ ; ð7:44Þ
@t i @t2 i 2
k 2 k
@U @ U ðDxÞ2
Ukiþ 1 ¼ Uki þ Dx þ þ ; ð7:45Þ
@x i @x2 i 2
k 2 k
@U @ U ðDxÞ2
Uki1 ¼ Uki Dx þ þ ; ð7:46Þ
@x i @x2 i 2
k 2 k
@F @ F ðDxÞ2
Fkiþ 1 ¼ Fki þ Dx þ þ ; ð7:47Þ
@x i @x2 i 2
k k
@F @ 2 F ðDxÞ2
Fki1 ¼ Fki Dx þ þ ð7:48Þ
@x i @x2 i 2
Inserting Eqs. (7.44)–(7.48) into Eq. (7.43) produces

2 k " 2 k #
@U k @ U ðDtÞ2 1 @ U 2
Uki þ Dt þ ¼ 2Ui þ
k
ðDxÞ
@t i @t2 i 2 2 @x2 i
" # ð7:49Þ
Dt @F k
2 Dx þ Ski Dt;
2Dx @x i
or
@U @F 1 ðDxÞ2 @ 2 U 1 @ 2 U
þ ¼ Dt þ S: ð7:50Þ
@t @x 2 Dt @x2 4 @t2
For Dx and Dt ! 0, it reduces to
@U @F 1 @ 2 U
þ ¼ D þ S: ð7:51Þ
@t @x 2 @x2
This is the equation modeled by the Lax scheme, where D is a numerical mesh
diffusion-like coefficient given by
ðDxÞ2
D¼ : ð7:52Þ
Dt
The Lax scheme is thus not consistent. Therefore, for a fixed value of Dx, if the
value of CFL is reduced, computations are performed with a smaller value of
Dt. Accordingly, D increases and the solution experiences greater diffusion. This
diffusive scheme has mesh-dependent features, therefore.
7.4.3 MacCormack Predictor–Corrector Scheme
The MacCormack (1969) method is a two-step predictor–corrector finite-difference

scheme with shock-capturing ability. In the predictor step, the fluxes F are evalu-
ated at the time level k using forward finite-differences
Dt k
Upred ¼ Uki Fi þ 1 Fki þ Ski Dt: ð7:53Þ
i
Dx
This step gives an estimation of the flow conditions at the new time level k + 1,
denoted as predicted values. In the corrector step, the fluxes F are evaluated with
the estimated vector U at time level k + 1, using backward finite-differences,
resulting in
Dt pred
Ucorr ¼ Uki Fi Fpred pred
i1 þ Si Dt: ð7:54Þ
i
Dx
The final value is taken as average of estimations in predictor and corrector steps,
e.g.,
Upred þ Ucorr
Uki þ 1 Uave
i ¼ i i
: ð7:55Þ
2
It is possible to construct a variant of the scheme using backward

finite-differences in the predictor step and forward finite-differences in the corrector.
Both variants of the scheme can also cyclically be applied during computations
(Chaudhry 2008). Practice indicates that the best results occur if the discretization
in the predictor step is in the direction of shock wave propagation. Accordingly, the
scheme presented here would be more adequate for a right-going shock wave in the
positive x-direction. MacCormack’s scheme is the second-order accuracy in both
space and time (Hoffman 2001), and is conditionally stable and thus constrained by
the CFL condition.
Remark: Gibbs phenomenon and Godunov’s theorem
The Godunov theorem states that any numerical scheme of order higher than one
cannot be monotone (Hirsch 1988, 1990). The immediate consequence is that
higher-order schemes will produce non-physical oscillations in zones of high gra-
dients of U, as near a shock wave, e.g., the surge front of a dam break wave. These
oscillations are known as the Gibbs phenomenon and are the result of dispersive
errors contaminating the solution.
Given that the MacCormack scheme is second-order accurate in time and space
O(Dx2, Dt2), it exhibits spurious oscillations near a shock. First-order schemes do
not suffer from unphysical oscillations, but at the price of excessive diffusion
smearing the shock front along a large number of computational nodes. In the
second-order schemes, there is no numerical diffusion, and the shock front is
sharply resolved within only a few computational nodes. Thus, in practice, the
second-order schemes are used and the strategy to control non-physical oscillations
is to introduce locally an amount of numerical diffusion if needed (Hirsch 1988,
1990).
7.4.4 MacCormack Scheme with Calibrated Artificial

Viscosity
Unphysical oscillations in the MacCormack scheme are reduced by introducing

artificial diffusion in the scheme. Consider Eq. (7.51) as the governing equation.
After application of predictor and corrector steps, incorporation of artificial vis-

cosity is given by solving the additional problem
@U ðDxÞ2 @ 2 U
¼e : ð7:56Þ
@t Dt @x2
Here, e is a diffusion coefficient introduced to control the amount of dissipation in

the scheme. If e is taken constant, and the spatial derivative of the diffusive term is
discretized with a second-order central finite-difference, the updating formula is
ave
Uki þ 1 ¼ Uave
i þ e Ui þ 1 2Ui þ Ui1 :
ave ave
ð7:57Þ
The method by Jameson et al. (1981) follows this idea, and it is simple to imple-
ment. The variant of Eq. (7.57) used is
ave
Uki þ 1 ¼ Uave
i þ ei þ 1=2 Ui þ 1 Ui
ave
ei1=2 Uave
i Ui1 ;
ave
ð7:58Þ
where
ei þ 1=2 ¼ Kmaxðei þ 1 ; ei Þ; ð7:59Þ
and ei is a parameter computed based on the water surface profile as
jhi þ 1 hi þ hi1 j
ei ¼ : ð7:60Þ
jhi þ 1 j jhi j þ jhi1 j
It is closely related to the water surface curvature at node i, discretized with a

second-order central finite-difference. K is the coefficient of artificial viscosity,
which must be calibrated. It varies typically between 0.5 and 3 (Zoppou and
Roberts 2003). This scheme has been used by Fennema and Chaudhry (1986) and
Gharangik and Chaudhry (1991). It introduces numerical diffusion only in zones of
the solution where oscillations develop, but leaves other portions of the profile
unaffected.
7.4.5 TVD MacCormack Scheme
The “total variation diminishing (TVD)” schemes are of high order, thereby pro-
ducing a sharp resolution of shocks, but introduce a bounding of the solution to
avoid spurious oscillations. The total variation TV of a function f is defined as
(Hirsch 1988, 1990)
X
1
TV ð f Þ¼ jfi þ 1 fi j; ð7:61Þ
i¼1
and a scheme for the conservation laws that satisfies

TV f k þ 1 TV f k ð7:62Þ
is said to be TVD. A TVD scheme is second-order accurate in smooth regions and

reduces to first-order accurate near discontinuities to suppress spurious oscillations
where sharp gradients are detected. The numerical dispersion of the second-order
scheme is then suppressed by the numerical diffusion introduced where necessary
by switching to a first-order accurate scheme. Here, the TVD version of the
MacCormack scheme developed by García-Navarro et al. (1992) is presented. The
final values are computed adding a dissipation step given by
Dt k
Uki þ 1 ¼ Uave
i þ Di þ 1=2 Dki1=2 ; ð7:63Þ
Dx
where the diffusive term Di+1/2 is given by

1X 2
Dt n

Dkiþ 1=2 ¼ aniþ 1=2 wniþ 1=2 1 ki þ 1=2 1 /niþ 1=2 eniþ 1=2 : ð7:64Þ
2 n¼1 Dx
The various terms in Eq. (7.64) are systematically presented below for imple-
mentation. First, the average values of velocity and celerity are determined from
pffiffiffiffi pffiffiffiffiffiffiffiffiffi
Ui hi þ Ui þ 1 hi þ 1 hi þ hi þ 1 1=2
Ui þ 1=2 ¼ pffiffiffiffi pffiffiffiffiffiffiffiffiffi ; ci þ 1=2 ¼ g : ð7:65Þ
hi þ hi þ 1 2
The eigenvalues k1;2

i and average eigenvalues k1;2
i þ 1=2 are given by
k1;2
i ¼ U i ci ; k1;2
i þ 1=2 ¼ Ui þ 1=2 ci þ 1=2 : ð7:66Þ
The eigenvectors are determined from

1
e1;2
i þ 1=2 ¼ k1;2 : ð7:67Þ
i þ 1=2
The function w is an entropy fix to avoid unphysical jumps in rarefaction waves,

given by
8
< k1;2 1;2 1;2
ki þ 1=2 ei þ 1=2
i þ 1=2 if
w1;2
i þ 1=2 ¼ ; ð7:68Þ
: e1;2 else
i þ 1=2
where

e1;2
i þ 1=2 ¼ max 0; k 1;2
i þ 1=2 k 1;2 1;2
i ; k iþ1 k 1;2
i þ 1=2 : ð7:69Þ
The jump projections in the values of U between nodes i and i + 1 onto the
eigenvectors e1;2
i þ 1=2 are
1 h i
a1;2
i þ 1=2 ¼ k2;1
i þ 1=2 ðhi þ 1 hi Þ ðqi þ 1 qi Þ : ð7:70Þ
2c1i þ 1=2
The function / is a flux limiter taken here as (Hseng and Chu 2000)
2
ri1;2
þ 1=2 þ r 1;2
i þ 1=2
/1;2
i þ 1=2 ¼ 2 ; ð7:71Þ
1 þ ri1;2
þ 1=2
where
a1;2
i þ 1=2s
ri1;2
þ 1=2 ¼ ; ð7:72Þ
a1;2
i þ 1=2
and

s ¼ sign k1;2
i þ 1=2 : ð7:73Þ
A useful feature of this scheme is that it is free from calibration parameters.
7.4.6 Upwind Scheme
In upwind methods, the discretization of the flux gradient @F=@x is sensitive to the
sign of the eigenvalues. This aims at capturing numerically the propagation of
perturbations by correctly identifying transmission up- or downstream. To illustrate
concepts, consider subcritical flow (Fig. 7.8). For perturbations propagating
Fig. 7.8 Upwind

discretization of fluxes in
subcritical flow
downstream, the flux contribution is denoted by F+, whereas for those propagating
upstream the associated flux is denoted by F−. These are linked to the eigenvalues
U + (gh)1/2 and U–(gh)1/2, respectively. The flux F is then split as (Hirsch 1988,
1990)
F ¼ F þ þ F : ð7:74Þ
This definition permits to formulate the SWE as1
@U @F þ @F
þ þ ¼ 0: ð7:75Þ
@t @x @x
An upwind discretization of @F þ=@x uses backward finite-differences to detect the

propagation of perturbations in the downstream direction along the C+ character-
istics. Likewise, an upwind discretization of @F=@x uses forward finite-differences
to detect the propagation of perturbations in the upstream direction along the C−
characteristics (Fig. 7.8) (Steger and Warming 1981). The result is thus
Dt þ Dt
Uki þ 1 Uki þ þ
Fi Fi1 þ Fi þ 1 F ¼ 0: ð7:76Þ
Dx Dx i
1
For simplicity's sake, source terms are not considered here. It is possible to treat S using an
upwind method (Toro 2001).
It is necessary to develop the expressions determining F+ and F−. Following the

Bermudez and Vazquez-Cendon (1994) upwind scheme, consider matrix G that
satisfies the identity
F ¼ GU: ð7:77Þ
Zoppou and Roberts (2003) demonstrated that the desired matrix is

0 1
G ¼ PKP1 ¼ ; ð7:78Þ
2 gh U
1 2
2U
where

1 1
P¼ U þ pcffiffi2 U pcffiffi2 ; ð7:79Þ
and
pffiffiffi pffiffiffi
1 1 c Upffiffi2ffi p2ffiffiffi :
P ¼ ð7:80Þ
2c cþU 2 2
The matrix K contains the eigenvalues k1 and k2 of G as

!
k1 0 U þ pcffiffi2 0
K¼ ¼ : ð7:81Þ
0 k2 0 U pcffiffi2
Now, K is decomposed in positive and negative components as
K ¼ K þ þ K ; ð7:82Þ
where

þ k1þ 0
K ¼ ; ð7:83Þ
0 k2þ
and

k 0
K ¼ 1
: ð7:84Þ
0 k2
In these matrices, the positive and negative components of the eigenvalues are
defined by the expressions

þ k1;2 þ k1;2 k1;2 k1;2
k1;2 ¼ ; k1;2 ¼ : ð7:85Þ
2 2
From these relation results the identity
jKj ¼ K þ K ; ð7:86Þ
where
0
1
U þ pcffiffi2 0
jK j ¼ @ A: ð7:87Þ
cffiffi
0 U p
2

The following transformation of the flux F is then conducted

F ¼ GU ¼ PKP1 U ¼ P K þ þ K P1 U ¼ PK þ P1 þ PK P1 U
¼ ½G þ þ G
U ¼ F þ þ F ;
ð7:88Þ
thereby demonstrating the flux decomposition stated in Eq. (7.74). Also,

F þ ¼ G þ U ¼ PK þ P1 U ¼ PðjKj þ K ÞP1 U ¼ PjKjP1 þ PK P1 U
¼ ½jGj þ G
U ¼ jFj þ F :
ð7:89Þ
From this result emerge the definitions
jFj ¼ jGjU; ð7:90Þ
and
jGj ¼ PjKjP1 : ð7:91Þ
Using Eqs. (7.88) and (7.89) produces

F þ jFj
Fþ ¼ ; ð7:92Þ
2
Similarly, one obtains for F
F jF j
F ¼ : ð7:93Þ
2
After elementary matrix operations, jGj is given by

G11 G12
jGj ¼ PjKjP1 ¼ ; ð7:94Þ
G21 G22
where
pffiffiffi
2 c c
c c
G11 ¼ U þ pffiffiffi U pffiffiffi U pffiffiffi U þ pffiffiffi ; ð7:95Þ
2c 2 2 2 2
p ffiffiffi
2 c c
G12 ¼ U þ pffiffiffi þ U pffiffiffi ; ð7:96Þ
2c 2 2
pffiffiffi
2 c c c
G21 ¼ U þ pffiffiffi U pffiffiffi U þ pffiffiffi
2c 2 2 2

ð7:97Þ
c c c
U pffiffiffi U pffiffiffi U þ pffiffiffi ;
2 2 2
pffiffiffi
2 c c c c
G22 ¼
U þ pffiffiffi U þ pffiffiffi þ U pffiffiffi U pffiffiffi : ð7:98Þ
2c 2 2 2 2
Inserting Eqs. (7.92) and (7.93) into Eq. (7.76) yields

Dt F þ jFj F þ jFj
Uki þ 1 Uki þ
Dx 2 2
i
i1 ð7:99Þ
Dt F jFj F jFj
þ ¼ 0;
Dx 2 iþ1 2 i
or
Dt k Dt k
Uki þ 1 ¼ Uki Fi þ 1 Fki1 þ jFji þ 1 2jFjki þ jFjki1 : ð7:100Þ
2Dx ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl 2Dx
unstable scheme diffusive term
Here, jFj is given by

!
G11 U1 þ G12 U2
jFj ¼ : ð7:101Þ
G21 U1 þ G22 U2
Note that the upwind scheme introduces a diffusive term into the updating equation
of the unstable scheme [see Eq. (7.18) for the advection of a scalar]. The method
presented here is called flux splitting.
7.5.1 Subcritical Dam Break Wave with Wet Tailwater
Consider a subcritical dam break wave of depth ratio hd/hu = 0.4 and upstream
water depth hu = 1 m. Simulations conducted with the unstable scheme using
CFL ¼ 0:1 are displayed at time t = 2 s in Fig. 7.9a, b, and compared with the
analytical solution of Stoker (1957). The numerical model is implemented in the
code available on the file “Dambreak_Unstable.xls” (Chap. 12). Note the large
instabilities generated, showing the poor performance of the scheme for resolving
rarefaction and shock waves in free surface flows. The small value of CFL used was
needed to obtain a computational result. For larger values, computations failed.
Attempts to use this scheme are found in the literature. A possible way to find an
“apparent” stable result is to introduce a large amount of dissipation in the scheme.
This is revealed in Fig. 7.9c, d, where the artificial viscosity method by Jameson
et al. (1981) was applied. Note that oscillations are suppressed, but this is no
indication that the scheme correctly works; simply, a numerical artifact was used to
force a solution.
Simulations conducted for the same test using the Lax diffusive scheme and
CFL ¼ 1 are presented in Fig. 7.10a, b. As expected, the shock and rarefaction
waves are smeared due to the diffusion term introduced by this scheme. The
numerical model is implemented in the code available on the file
“Dambreak_Diffusive.xls” (Chap. 12). The same simulation was conducted using
CFL ¼ 0:2, with the results shown in Fig. 7.10c, d. The low CFL value introduced
an unacceptable amount of numerical diffusion.
The simulation results from the MacCormack scheme are shown in Fig. 7.11a, b.
Spurious oscillations in the shock front and at the toe of the rarefaction wave are
present, albeit more visible in the velocity profile. These oscillations result from
dispersive errors. Applying the artificial viscosity method of Jameson et al. (1981)
as presented by Chaudhry (2008) with a calibrated K = 2, the oscillations are
almost suppressed, as noted from Fig. 7.11c, d. The numerical model is imple-
mented in the code available from “Dambreak_MacCormack_Jameson.xls”
(Chap. 12). Applying the MacCormack TVD scheme results in Fig. 7.12a, b.
Fig. 7.9 Subcritical dam break wave for hd/hu = 0.4 computed with the unstable scheme using
CFL ¼ 0:1 showing effect of artificial viscosity coefficient K
Fig. 7.10 Subcritical dam break wave for hd/hu = 0.4 computed with diffusive scheme
highlighting effect of CFL
Fig. 7.11 Subcritical dam break wave for hd/hu = 0.4 computed with MacCormack scheme using
CFL ¼ 0:9 showing effect of artificial viscosity coefficient K
Fig. 7.12 Subcritical dam break wave for hd/hu = 0.4 computed with TVD MacCormack scheme
using CFL ¼ 0:9
Spurious oscillations in the shock front and at the toe of the rarefaction wave are
suppressed without the need of using any calibration parameter, which is a
remarkable advantage. The numerical model is implemented in the code available
on the file “Dambreak_MacCormack_TVD.xls” (Chap. 12). The simulation results
obtained using the upwind scheme are shown in Fig. 7.13a, b. Note that the dif-
fusion of the scheme, as in other first-order schemes, is present. The numerical
model is implemented in the code available on the file “Dambreak_Upwind.xls”
(Chap. 12).
Fig. 7.13 Subcritical dam break wave for hd/hu = 0.4 computed with upwind scheme using
CFL ¼ 0:9
7.5.2 Transcritical Dam Break Wave with Wet Tailwater
The subcritical dam break wave is not a severe test case (Zoppou and Roberts
2003). A transcritical dam break wave was thus generated using a depth ratio
hd/hu = 0.001 and hu = 1 m. Critical flow establishes at the dam axis in a point of
the rarefaction wave. Only the TVD MacCormack, upwind, and Lax diffusive
schemes were able to generate computational results, which are displayed in
Fig. 7.14 for t = 2 s.
The results of the MacCormack TVD scheme are excellent, as shown in
Fig. 7.14a, b. The upwind method, however, produces an unphysical jump at the
dam axis related to a violation of the entropy condition. The Lax diffusive scheme
produces results not in precise agreement neither for the rarefaction nor for the
shock waves.
7.5.3 Subcritical Surge
In this section, the propagation of a subcritical positive surge is considered. Its

analytical solution is with Vw as the absolute surge velocity and subscripts 1 and 2
denoting conditions behind and in front of the surge,

U1 if x Vw t
Uðx; tÞ ¼ : ð7:102Þ
U2 if x Vw t
The discontinuity is located at x = 0 for t = 0. The surge mass conservation states

[Eq. (7.40)]
U1 h1 U2 h2
Vw ¼ ; ð7:103Þ
h 1 h2
Fig. 7.14 Transcritical dam break wave for hd/hu = 0.001 computed using CFL ¼ 0:9 with
(a, b) MacCormack TVD scheme, (c, d) upwind scheme, (e, f) Lax diffusive scheme
whereas the momentum equation is [Eq. (7.41)]

1=2
g h1
Vw ¼ U2 þ ð h1 þ h2 Þ : ð7:104Þ
2 h2
Combining Eqs. (7.103) and (7.104), and using the definition of unit discharge
q = Uh, results in

1=2
q1 q2 g h1
¼ U2 þ ð h1 þ h2 Þ ; ð7:105Þ
h1 h2 2 h2
or

1=2
g h1
q1 ¼ U2 h1 þ ðh1 h2 Þ ð h1 þ h2 Þ : ð7:106Þ
2 h2
This is an implicit equation to be solved for h1, once the values of q1, h2, and U2 are
prescribed, e.g.,

1=2
g h1
f ð h1 Þ U 2 h1 þ ð h1 h2 Þ ð h1 þ h2 Þ q1 ¼ 0: ð7:107Þ
2 h2
A better approximation to the root of Eq. (7.107) is based on the Newton–

Raphson method with z as the recursion index by Hoffman (2001)
fz
hz1 þ 1 ¼ hz1 : ð7:108Þ
ðdf /dh1 Þz
The derivative term needed in Eq. (7.108) is

1=2
1=2
df 2h1 þ h2 g h1 g h1
¼ U2 þ gðh1 h2 Þ ð h1 þ h2 Þ þ ð h1 þ h2 Þ :
dh1 4h2 2 h2 2 h2
ð7:109Þ
A good initial value for h1 to initialize the algorithm is

q1 q2 q1 q2
h1 ¼ h2 þ ¼ h2 þ : ð7:110Þ
Vw U2 þ ðgh2 Þ1=2
Once h1 is determined, Vw follows from Eqs. (7.103) or (7.104), and the profile
given by Eq. (7.102) is determined. The solution is implemented in a code available
on the file “SurgeAnalytical.xls” (Chap. 12).
The MacCormack TVD scheme was used to simulate a subcritical surge prop-
agating over initially still water of depth 0.205 m. A sudden increase of discharge at
the inflow section of 0.028 m3/s was supplied in a channel 0.42 m wide. These are
the test conditions for surge experiments conducted by Favre (1935). Consider now
ideal surges (n = 0 m1/3/s). At the first time step, the water depth at the upstream
section was determined with the boundary conditions for discontinuous flows, to
apply then the method of characteristics for the remaining time steps with
Eq. (7.37) as
g kþ1 k
U1k þ 1 ¼ U2k þ k
h1 hk2 þ g So Sf 2 Dt: ð7:111Þ
c2
Equation (7.111) yields the quadratic function for hk+1

1
g k þ 1 2 g k k
h þ U k
h þ g S o S f 2 Dt hk1 þ 1 qk1 þ 1 ¼ 0: ð7:112Þ
ck2 1 2
ck2 2
The numerical solution is implemented in the code “Favre_FDM_TVD_

BoundaryDiscont_Subcrit.xls” (Chap. 12). The program output at t = 40 s is shown
in Fig. 7.15a, b. Note the excellent agreement with the analytical solution.
Computations were repeated implementing the boundary condition at the upstream
section using only the method of characteristics. The numerical solution is
Fig. 7.15 Subcritical surge computed with MacCormack scheme

implemented in the code “Favre_FDM_TVD” (Chap. 12). Results are displayed in

Fig. 7.15c, d, not showing an evident variation of the former results. However, it is
better to properly treat the boundary conditions as discontinuous when they involve
jumps. The use of the classic MacCormack scheme with the boundary condition at
the upstream section determined using the method of characteristics yields the
results of Fig. 7.15e, f. A spurious oscillation appears at the shock front, but,
overall, this computation is also acceptable. The numerical solution is implemented
in the code “Favre_FDM_Jameson” (Chap. 12).
7.5.4 Supercritical Surge
A test suggested by García-Navarro and Saviron (1992) is considered.

A supercritical surge is generated with a discharge increase of 1400 m3/s in a
horizontal and frictionless flume 1 m wide and 2 m of initial static water. As the
flow is supercritical, both depth and velocity must be prescribed at the inflow
section. The water depth is thus computed using the jump conditions, and the values
of h and Q are fixed along the simulation. The numerical solution is implemented in
the code “Favre_FDM_TVD_ BoundaryDiscont_Super.xls” (Chap. 12).
Computational results are displayed in Fig. 7.16a, b, showing good agreement with
Fig. 7.16 Supercritical surge computed with MacCormack scheme

the analytical solution except for a small hump numerically generated during the
initial simulation instants. To evidence the importance of correct implementation of
boundary conditions, we demonstrate the result of implementing naive approxi-
mations. It is tempting to assume that a boundary section is just another compu-
tational point, treating @F=@x using either backward of forward finite-differences. In
our case, the idea may be to generate the unknown water depth at the inflow section
by discretizing the continuity equation using forward finite-differences. That is,
Dt k
hk1 þ 1 ¼ hk1 q2 qk1 : ð7:113Þ
Dx
This naive determination of the unknown water depth at the inflow section was
implemented, and the poor results generated are shown in Fig. 7.16c, d, as found by
García-Navarro and Saviron (1992). This demonstrates the importance of the cor-
rect implementation of boundary conditions. The numerical solution is imple-
mented in the code “Favre_FDM_TVD_ BoundaryDiscont_Super_Cont.xls”
(Chap. 12).
7.5.5 Positive Surge with Friction
Favre (1935) measured a subcritical surge propagating over initially still water of
depth 0.205 m. The discharge increase at the inlet section was 0.028 m3/s, and
the channel width was 0.42 m. Manning’s coefficient for this flume was
n = 0.01 m1/3/s (Favre 1935; Terzidis and Strelkoff 1970). Computational results
generated with the TVD MacCormack scheme along with a discontinuous–con-
tinuous implementation of the upstream boundary condition were generated at
several instants of time. Results are displayed in Fig. 7.17 and compared with
experimental data. The time needed to switch on the pumps and generate into the
flume the discharge pulse was computationally accounted for as a time lag of 1 s,
given that the generation of the discharge pulse is instantaneous in the mathematical
model. Comparing profiles at two different instants of time, the absolute velocity of
the surge predicted by the numerical model is in excellent agreement with that
indicated by the experiments. The numerical solution is implemented in the code
“Favre_FDM_TVD_ BoundaryDiscont_Subcrit.xls” (Chap. 12).

computational results with
Favre’s subcritical surge
experiments
In general, the MacCormack TVD scheme by García-Navarro et al. (1992) was

found to work well in all test cases; given its simple implementation, it is thus
recommended.
References
Anderson, J. D. (1995). Computational fluid dynamics: The basics with applications.

McGraw-Hill, New York.
Bermudez, A., & Vazquez, M. E. (1994). Upwind methods for hyperbolic conservation laws with
source terms. Computers & Fluids, 23(8), 1049–1071.
Bhallamudi, S. M. (2002). Computation of open-channel flows with shocks: an overview. In R.

Prasad & S. Vedula (Eds.), Research perspectives in hydraulics and water resources
engineering (pp. 89–130). London, UK: World Scientific.
Courant, R., Friedrichs, K., & Lewy, H. (1928). Über die partiellen Differenzengleichungen der
mathematischen Physik [On the partial differential equations in Mathematical physics].
Mathematische Annalen, 100(1), 32–74 (in German).
Courant, R., Isaacson, E., & Rees, M. (1952). On the solution of nonlinear hyperbolic differential
equations by finite differences. Communications on Pure and Applied Mathematics, 5(3), 243–
255.
Cunge, J. A. (1975). Rapidly varying flow in power and pumping canals. In K. Mahmood & V.
Yevjevich (Eds.), Unsteady flow in open channels (Chap. 14, pp. 539–586). Fort Collins, CO:
Water Resources Publications.
découverts. [Theoretical and experimental study of travelling surges in open channels]. Dunod,
Paris, France (in French).
Fennema, R. J., & Chaudhry, M. H. (1986). Explicit numerical schemes for unsteady free-surface
flows with shocks. Water Resources Research, 22(13), 1923–1930.
Garcia-Navarro, P., Alcrudo, F., & Saviron, J. M. (1992). 1-D open-channel flow simulation using
TVD-McCormack scheme. Journal of Hydraulic Engineering, 118(10), 1359–1372.
Garcia-Navarro, P., & Saviron, J. M. (1992). McCormack’s method for the numerical simulation
of one-dimensional discontinuous unsteady open-channel flow. Journal of Hydraulic
Research, 30(1), 95–105.
Hirsch, C. (1988). Numerical computation of internal and external flows, vol. 1: Fundamentals of
numerical discretization. Chichester: Wiley.
Hirsch, C. (1990). Numerical computation of internal and external flows, vol. 2: computational
methods for inviscid and viscous flows (Vol. 2). Chichester: Wiley.
Marcel Dekker.
Hseng, M., & Chu, C. R. (2000). The simulation of dam-break flows by an improved
predictor-corrector TVD scheme. Advances in Water Resources, 23(6), 637–643.
Jameson, A., Schmidt, W., & Turkel, E. (1981). Numerical solutions of the Euler equations by
finite volume methods using Runge-Kutta time-stepping schemes. American Institute of
Aeronautics and Astronautics, Fluid and Plasma Dynamics Conf., Palo Alto CA,
AIAA-81-1259.
Lax, P. (1954). Weak solutions of nonlinear hyperbolic equations and their numerical
computation. Communications on Pure and Applied Mathematics, 7, 159–193.
Lax, P. D., & Richtmyer, R. D. (1956). Survey of the stability of linear finite difference equations.
Communications on Pure and Applied Mathematics, 9, 267–293.
Lax, P., & Wendroff, B. (1960). System of conservation laws. Communications on Pure and
Applied Mathematics, 13(2), 217–237.
Liggett, J. A., & Cunge, J. A. (1975). Numerical methods of solution of the unsteady flow
equations. In K. Mahmood & V. Yevjevich (Eds.), Unsteady flow in open channels (Chap. 4,
pp. 89–179). Fort Collins, CO: Water Resources Publications.
MacCormack, R. W. (1969). The effect of viscosity in hypervelocity impact cratering. American
Institute Aeronautics and Astronautics, Paper 69–354.
Roache, P. J. (1972). Computational fluid dynamics. Albuquerque, NM: Hermosa Publishers.
References 311
Steger, J. L., & Warming, R. F. (1981). Flux vector splitting of the inviscid gas dynamic equations
with application to finite-difference methods. Journal of Computational Physics, 40(2),
263–293.
Interscience Publishers.
Terzidis, G., & Strelkoff, T. (1970). Computation of open channel surges and shocks. Journal of
the Hydraulics Division, ASCE, 96(HY12), 2581–2610.
Zoppou, C., & Roberts, S. (2003). Explicit schemes for dam-break simulations. Journal of
Chapter 8
The Riemann Problem
8.1 Introduction
In fluid dynamics, the classical Riemann problem (Riemann 1860) is an

initial-value problem for a set of homogeneous PDEs in which the initial data
consist of two constant states forming a discontinuity (Toro 1997, 2001; LeVeque
2002; Guinot 2003) (Fig. 8.1). It is a generalization of the dam break problem
(Stoker 1957) described in Chap. 6. The Riemann problem is important for a
number of reasons. First, it is the simplest initial-value problem for the shallow
water equations (SWEs), with a solution including both smooth and discontinuous
waves. The understanding of this fundamental wave problem is of paramount
importance to assimilate how waves propagate in shallow water. Despite its
apparent simplicity, it represents the practical case of instantaneous gate maneuvers
in open channels (Henderson 1966; Jain 2001; Jeppson 2011; Katopodes 2019).
Another important aspect is that the Riemann problem yields an exact solution of
the SWE; thus, it applies to check the performance of numerical schemes. Finally, it
is a fundamental component of a class of numerical schemes known as finite
volume Godunov-type methods (Godunov 1959), to be described in Chap. 9.
Let subscripts L and R denote conditions at the left- and right-hand sides of the
discontinuity at time t = 0; the initial conditions then are

UL if x\0;
Uðx; 0Þ ¼ ð8:1Þ
UR if x [ 0;
where x is the spatial coordinate and the discontinuity is located initially at x = 0.

The solution U(x, t) subject to Eq. (8.1) is to be determined from
@U @F
þ ¼ 0. ð8:2Þ
@t @x

https://doi.org/10.1007/978-3-030-13073-2_8
314 8 The Riemann Problem
Fig. 8.1 Initial conditions of Riemann problem for the SWE a physical plane, b x-t plane
Here, U is the vector of unknowns and F is the flux vector given, with h as the
water depth and U as the depth-averaged velocity, by

h Uh
U¼ ; F¼ : ð8:3Þ
Uh U 2 h þ 12 gh2
The fundamental difference with the dam break problem stems from the possi-
bility of having nonvanishing initial values of U at each side of the discontinuity.
The basic wave solutions are those described in Chaps. 5 and 6, namely shock
waves and rarefaction waves. Previous relations are then first generalized to
accommodate arbitrary initial data, and then, the solution strategy of the wave
structure for wet-bed conditions is detailed. The computation of the complete wave
profile is presented. The dry-bed Riemann solver is subsequently explained, and
finally, the instantaneous gate operations in open channels are analyzed in detail
using the theory presented.
8.2 Wet-Bed Exact Riemann Solver 315
8.2 Wet-Bed Exact Riemann Solver
8.2.1 Wave Relations
As previously discussed, the basic wave solutions of Eqs. (8.1), (8.2) are shocks
and rarefaction waves. The type of wave generated at each side of the discontinuity
for t > 0 is not known in advance, demanding for an iterative solution of the wave
field starting with an assumed wave pattern. In general, four cases may arise:
(1) Left wave is a rarefaction, and right wave is a shock.
(2) Left wave is a rarefaction, and right wave is a rarefaction.
(3) Left wave is a shock, and right wave is a shock.
(4) Left wave is a shock, and right wave is a rarefaction.
Consider case 1, with a left-going rarefaction wave and a right-going shock
(Fig. 8.2). The basic equation for a left rarefaction wave results from the relation
stating conservation of the Riemann invariant U + 2c within the wave, that is, with
c = (gh)1/2 [Eq. (5.78)]
U ðx; tÞ þ 2cðx; tÞ ¼ const: ð8:4Þ
Direct application yields
U þ 2c ¼ UL þ 2cL ; ð8:5Þ
Fig. 8.2 Wave pattern of

Riemann problem for
left-going rarefaction wave
and right-going shock wave
a physical plane, b x-t plane
(gray shadow depicts the
rarefaction wave x-t domain,
and the thick black line
represents the shock front
path)
or,
h i
U ¼ UL 2ðc cL Þ ¼ UL 2 ðgh Þ1=2 ðghL Þ1=2 : ð8:6Þ
This equation states the conditions at the star region as function of initial conditions
at the left side of the discontinuity. It is therefore an equation “connecting” the
constant state zone with the left-side initial conditions. Following Toro (2001), we
define a left-side function fL as
U ¼ UL fL ;
h i ð8:7Þ
fL ¼ 2 ðgh Þ1=2 ðghL Þ1=2 :
The propagation speed SR of a right-going shock wave is given from the Rankine–
Hugoniot jump conditions by [Eq. (5.35)]
1=2
gh
SR ¼ UR þ ð h þ hR Þ : ð8:8Þ
2hR
This relation is now expressed as a function of the depth-averaged velocity at the

star region. Using the mass conservation equation in moving axes, namely
hR ðUR SR Þ ¼ h ðU SR Þ; ð8:9Þ
it is possible to write
ðh hR ÞðSR UR Þ ¼ h ðU UR Þ: ð8:10Þ

1=2
g
U ¼ U R þ ð h hR Þ ð h þ hR Þ : ð8:11Þ
2h hR
This relation is also written as
U ¼ UR þ fR ;
1=2 ð8:12Þ
g
f R ¼ ð h hR Þ ð h þ hR Þ :
2h hR
Here, fR is a function “connecting” the constant state zone with the right-side
conditions. Equating the velocity in the star region from Eqs. (8.7) and (8.12)
produces
f ðh Þ ¼ UR UL þ fR ðh ; hR Þ þ fL ðh ; hL Þ ¼ 0: ð8:13Þ
This is an equation that must be numerically solved to determine the water depth at
the star region h*. Once done, U* is computed either from Eqs. (8.7) or (8.12). The
important aspect at this stage is that the Riemann problem produces an equation
where the constant state zone is linked to initial data via left- and right-hand
functions. Note that the type of function is different for shock and rarefaction waves,
thereby indicating that it is necessary to discriminate in advance which type of wave
originates at each side of the discontinuity. Note further that the left-going rar-
efaction wave produces a depression of water depths (h* < hL), whereas the
right-going shock involve an increase in water level (h* > hR). An important issue
is that Eq. (8.13) is general, as verified by repeating this analysis for the other three
types of wave patterns. It implies formulating the equations for a left-going shock
using the Rankine–Hugoniot jump conditions, and the continuous solution of the
right-going smooth wave, basically implying conservation of the Riemann invariant
U(x, t) − 2c(x, t) across the rarefaction (Jain 2001; Toro 2001). The repetition
process is elementary and is left to the reader. The results for fR and fL, depending
on whether they are linked to rarefactions of shocks, are summarized as (Toro 2001)
8 h i
< 2 ðgh Þ1=2 ðghL Þ1=2 if h hL ðrarefaction waveÞ
fL ¼ h i ; ð8:14Þ
: ðh h Þ g ðh þ h Þ 1=2 if h [ hL ðshock waveÞ
L 2h hL L
8 h i
< 2 ðgh Þ1=2 ðghR Þ1=2 if h hR ðrarefaction waveÞ
fR ¼ h i : ð8:15Þ
: ðh h Þ g ðh þ h Þ 1=2 if h [ hR ðshock waveÞ
R 2h hR R
Note that in a given Riemann problem, once the value of h* is adopted, the types of
waves generated at the right and left sides are automatically determined.
Bernhard Riemann was born on September 17, 1826, at Breselenz, in

today’s Germany, passing away almost aged 40 years on July 20, 1866, at
Selasca, Italy. He initiated studies in 1846 in theology and philology at
Göttingen University, but soon turned to mathematics. One year later, he
moved to Berlin University submitting, from Göttingen in 1850, a thesis on
the Theory of functions with a complex variable. This work was praised by
Gauss and led the foundations for his work in the following decade. In 1859,
he moved to Göttingen, where Gauss had been Professor until 1855. The
struggle to survive on the casual fees of a private teacher had undermined his
health such that he passed away at young age.
Riemann introduced in mathematics the many-leafed Riemann surfaces
whose sheets are connected around the branch points of the function, on
which the latter can be defined as single value. He demonstrated how to treat
the function as a conformal map of a simply connected domain, described
essentially by the point singularities of the function and the shape of the map
along the boundary of the domain. He also laid the foundations to the
non-Euclidian geometry. Throughout this subject, he was motivated by his
studies in mathematical physics. Riemann computed in hydromechanics
two-dimensional outflow from a large tank by the hodograph method and
determined the contraction coefficient from an infinitely small orifice. His
method was generalized by Richard von Mises (1883–1953) some fifty years
later and served as a starting point of plane inviscid flow analyses in the
twentieth century. Riemann also worked on the propagation characteristics of
waves in air.
8.2.2 Solution of Star Region
The solution process is iterative, starting with an initial guess of h*. This value,
upon comparison with hL and hR, determines which types of waves generate at the
left- and right-hand sides. In general, the initial guess ho does not satisfy Eq. (8.13),
producing

UR UL þ fR ho ; hR þ fL ho ; hL 6¼ 0: ð8:16Þ
A better approximation to the root of Eq. (8.13) involves the Newton–Raphson

method with k as the recursion index as (Hoffman 2001; Toro 2001)
fk
ðh Þk þ 1 ¼ ðh Þk : ð8:17Þ
ðdf =dh Þk
With this new value of h*, the types of waves at L or R may change, therefore. Thus,
the wave field is iteratively updated in the solution process. The derivative term
needed in Eq. (8.17) is
df dfL dfR
¼ þ ; ð8:18Þ
dh dh dh
where
8
dfL < hgðgh Þ
1=2
if h hL ;
¼ i1=2 h i1=2 ð8:19Þ
ðh hL Þ
dh : 2hg h ðh þ hL Þ g 4h 2
g

2h hL ð h
þ h L Þ if h [ hL ;
L
and
8
dfR < gh ðgh Þ
1=2
if h hR ;
¼ i1=2 h i1=2 ð8:20Þ
ðh hR Þ
dh : 2hg h ðh þ hR Þ g 4h 2 2h
g
h ð h þ hR Þ if h [ hR :
R R
Following Toro (2001), a good initial estimate of h* is given by the so-called

two-rarefaction approximation. For two rarefaction waves, inserting the corre-
sponding fL and fR functions into Eq. (8.13), results in
h i h i
f ðh Þ ¼ UR UL þ 2 ðgh Þ1=2 ðghR Þ1=2 þ 2 ðgh Þ1=2 ðghL Þ1=2 ¼ 0; ð8:21Þ
whose analytical solution is (Toro 2001)

" #2
1 ðghL Þ1=2 þ ðghR Þ1=2 1
h ¼ ðUR UL Þ : ð8:22Þ
g 2 4
The wave relation across the left rarefaction implies conservation of the Riemann
invariant U + 2c, resulting in
U þ 2c ¼ UL þ 2cL : ð8:23Þ
Across the right rarefaction wave, U − 2c is conserved, from which
U 2c ¼ UR 2cR : ð8:24Þ
Summing the two equations yields for U*
1
U ¼ ðUR þ UL Þ þ ðcL cR Þ: ð8:25Þ
2
8.2.3 Complete Wave Profiles
Left shock wave
For a left shock wave, the signal speed is (Toro 2001)
SL ¼ UL kL ðghL Þ1=2 ;
h i
1=2 ð8:26Þ
kL ¼ 12 ðh þh2hL Þh :
L
This is needed to track the position of the shock front at any instant of time. The
depth and velocity within the shock wave are simply h* and U*.
Left rarefaction wave
For a left rarefaction wave, the signal speeds of its edge and tail are, respectively,
SL ¼ UL ðghL Þ1=2 ;
ð8:27Þ
SL ¼ U ðgh Þ1=2 :
Between the edge and the tail of the rarefaction, the spatial variations of depth and
velocity for any instant of time are given by solving x/t = U–c and U + 2c = const.
for U(x, t) and c(x, t) by
1h xi
ðghÞ1=2 ¼ UL þ 2ðghL Þ1=2 ;
3 t ð8:28Þ
1h 1=2 xi
U ¼ UL þ 2ðghL Þ þ 2 :
3 t
Right shock wave
For a right shock wave, the signal speed is, from the Rankine–Hugoniot jump
conditions
SR ¼ UR þ kR ðghR Þ1=2 ;
ð8:29Þ
1 ðh þ hR Þh 1=2
kR ¼ ;
2 h2R
which is similarly needed to track the position of the shock front at any instant of
time. The depth and velocity within the shock wave are simply h* and U*.
Right rarefaction wave
For a right rarefaction wave, the signal speeds of its edge and tail are, respectively,
SR ¼ UR þ ðghR Þ1=2 ;
ð8:30Þ
SR ¼ U þ ðgh Þ1=2 :
Between the rarefaction edge and tail, the spatial variations of depth and velocity
for any instant of time are given by
1h xi
ðghÞ1=2 ¼ UR þ 2ðghR Þ1=2 þ ;
3 t ð8:31Þ
1h 1=2 xi
U ¼ UR 2ðghR Þ þ 2 :
3 t
8.2.4 Possible Wave Patterns and Computation of Fluxes
For the four cases previously reported, there are a number of subcases depending on
the relative position of each wave with respect to the t-axis, determined by the signs
of SL, SL , SR and SR . The 16 possible cases are plotted in Figs. 8.3, 8.4, 8.5, and 8.6
(Zoppou and Roberts 2003; Roberts 2013).
As stated at the start of this chapter, the Riemann problem is a key ingredient
used in Godunov-type finite volume methods. Here, it is advanced that a funda-
mental quantity to be determined is the value of the flux vector F at coordinate
x = 0, i.e., the flux crossing the plane where the discontinuity in U was initially
generated. Using the definitions of U and F, it is elementary to demonstrate that
Fig. 8.3 Wave patterns of Riemann problem for left rarefaction wave and right shock wave
(case 1) (adapted from Roberts 2013)

h U1 U2
U¼ ¼ ; F ¼ FðUÞ ¼ U22 : ð8:32Þ
Uh U2 U1 þ 12 gU12
This means that the value of F is fully determined by U. Thus, the value of F(x = 0)
is given by the solution of the Riemann problem at x = 0, namely U(x = 0).
Depending on the specific wave pattern, U(x = 0) may equal UL, UR, U*, or a value
Uc to be determined using the rarefaction wave equations, corresponding to critical
flow. The computation of the value of F(x = 0) for the 16 possible cases depicted in
Figs. 8.3, 8.4, 8.5 and 8.6 is summarized in Table 8.1. Note that some cases are
equivalent in terms of computation of F(x = 0), despite the different complete wave
patterns. If a finite volume scheme is constructed using the exact solution of the
Riemann problem, it is said to use an “exact Riemann solver.” This nomenclature
should be understood as the use of the numerical solution originating from the
consideration of the complete Riemann problem. “Exact Riemann solver” should
not be confused with having closed-form analytical solutions in a Riemann solver.
In fact, this is only true for the two-rarefaction case. To the authors’ knowledge,
exact analytical solutions are unknown for the rest of wave patterns.
Fig. 8.4 Wave patterns of Riemann problem for left and right rarefaction waves (case 2) (adapted
from Roberts 2013)
8.2.5 Computational Examples
In this section, computational examples are presented to illustrate the propagation

of waves originating from the solution of the Riemann problem under wet-bed
conditions. Test conditions for the examples are summarized in Table 8.2.
The first example is presented in Fig. 8.7, simulating a flume with initially static
water and a gate closed at left. The gate is suddenly opened at time t = 0, and the
wave motion shown in Fig. 8.7 is generated after 2 s of routing. The wave pattern is
composed of two shocks propagating to the right. Given that SL < SR, the width of
the star region progressively increases with time. The computational output is
summarized in Table 8.3. The results were generated with a code in which the
described solution of the wet-bed Riemann problem was implemented. This code
was inserted into the file “ExactRiemannSolver_wetbed.xls”, available in Chap. 12.
The program was also used to run the test presented in Fig. 8.8, originally due to
Toro (2001). It consists of two identical rarefaction waves traveling in opposite
Fig. 8.5 Wave patterns of Riemann problem for left and right shock waves (case 3) (adapted from
Roberts 2013)
Fig. 8.6 Wave patterns of Riemann problem for left shock wave and right rarefaction wave
(case 4) (adapted from Roberts 2013)
Table 8.1 Evaluation of the flux F in the Riemann problem (see Figs. 8.3, 8.4, 8.5, and 8.6)
Case Subcase F(x = 0)
1 a F(Uc); critical flow at left rarefaction
1 b F(U*)
1 c F(UL)
1 d F(UR)
2 a F(Uc); critical flow at right rarefaction
2 b F(U*)
2 c F(UL)
2 d F(UR)
2 e F(Uc); critical flow at left rarefaction
3 a F(U*)
3 b F(UR)
3 c F(UL)
4 a F(Uc); critical flow at right rarefaction
4 b F(UR)
4 c F(UL)
4 d F(U*)
Table 8.2 Test conditions for examples of wave solutions of the Riemann problem
Test Time (s) hL (m) hR (m) UR (m/s) UL (m/s)
Figure 8.7 2 0.2 0.4 0 7
Figure 8.8 2 1 1 5 –5
Figure 8.9 2 1 0.2 2.5 0.5
Fig. 8.7 Riemann solution of test with two right-going shocks for a free surface profile h(x) and
b depth-averaged velocity U(x)
Table 8.3 Test results for examples of wave solutions of Riemann problem. The discontinuity is
located at x = 0, t = 0
Test Time (s) SR (m/s) SL (m/s) h* (m) U* (m/s)
Figure 8.7 2 4.213 1.477 1.02 2.56
Figure 8.8 2 8.132 −8.132 0.041 0
Figure 8.9 2 4.5 −2.632 0.317 3.238
Fig. 8.8 Riemann solution of test with left- and right-going rarefaction waves for a free surface
profile h(x) and b depth-averaged velocity U(x)
directions. The interest of this type of test is that the interaction of the rarefaction
waves produces a star region of static water where the bed may become dry under
certain conditions. The mathematical conditions for two identical rarefaction waves
traveling in opposite directions are hR = hL and UL = −UR. Inserting these into
Eq. (8.22) produces
" #2 2
1 2ðghR Þ1=2 1 1 1
h ¼ ð2UR Þ ¼ ðghR Þ1=2 UR : ð8:33Þ
g 2 4 g 2
This is rewritten with the Froude number FR ¼ UR =cR as

2
1
h ¼ hR 1 F R : ð8:34Þ
2
Thus, for the limiting case FR ¼ 2, a dry-bed condition h* = 0 is generated. The

velocity at the star region is from Eq. (8.25)
1
U ¼ ðUR UR Þ þ ðcR cR Þ ¼ 0: ð8:35Þ
2
The water there is thus stagnant for all values of FR .

The last example is presented in Fig. 8.9, where the results include a rarefaction
wave traveling to the left, and a shock to the right. The tailwater depth is small, so
that the rarefaction wave crosses the initial position of the discontinuity (x = 0), and
critical flow F ¼ 1 is thus settled there. Transcritical flow is established along the
rarefaction wave, which is connected to a supercritical bore along the right portion
of the wave profile.
8.3 Dry-Bed Exact Riemann Solver 327
Fig. 8.9 Riemann solution of test with left-going rarefaction wave and right-going shock for
a free surface profile h(x) and b depth-averaged velocity U(x)
8.3 Dry-Bed Exact Riemann Solver
8.3.1 Wet–Dry Interface
The wave structure of the Riemann problem changes drastically if a portion of the
computational domain is dry. Before presenting the corresponding wave solutions,
consider Toro (2001), who demonstrated that a shock wave cannot connect the wet–
dry portions. Let UL represent the initial wet-bed data, and UR the data of the
dry-bed region. Obviously, hR = 0. Assume that a shock wave propagating with
velocity S connects the two states. The Rankine–Hugoniot jump conditions are then
hL UL hR UR hL UL2 þ 12 gh2L hR UR2 12 gh2R

S¼ ¼ : ð8:36Þ
hL hR hL UL hR UR
Setting hR = 0, the first identity yields S = UL. Substitution in the second identity
yields
hL UL2 þ 12 gh2L
S¼ UL ) hL ¼ 0; ð8:37Þ
hL U L
which is obviously in contradiction with the initial assumption hL > 0. Thus, a

shock wave cannot physically exist at the wet–dry interface. At this interface, the
depth is simply zero, and there is a jump in velocity, which is zero in the dry terrain,
and finite at the wave edge.
8.3.2 Possible Wave Patterns
Consider first wet-bed conditions on the left, corresponding to the initial data
(Fig. 8.10)
Fig. 8.10 Dry-bed

conditions of the right a initial
data, b free surface profile
h(x), and c x-t plane

UL if x\0
Uðx; 0Þ ¼ : ð8:38Þ
0 if x[0
A rarefaction wave is formed, with the left edge propagating at velocity
SL ¼ UL ðghL Þ1=2 : ð8:39Þ
The velocity of propagation of the right-going wet–dry interface is determined

using the basic equation of the rarefaction wave, namely
UR þ 2cR ¼ UL þ 2cL : ð8:40Þ
At the dry front hR = 0, so that from Eq. (8.40)

Fig. 8.11 Dry-bed

conditions of the left a initial
data, b free surface profile
h(x), and c x-t plane
SR ¼ UR ¼ UL þ 2cL : ð8:41Þ
The depth and velocity within the rarefaction wave are thus given by (Toro 2001)
1h xi
ðghÞ1=2 ¼ UL þ 2ðghL Þ1=2 ;
3 t ð8:42Þ
1h 1=2 xi
U ¼ UL þ 2ðghL Þ þ 2 :
3 t
For wet-bed conditions on the right side (Fig. 8.11), the corresponding initial
data are

0 if x\0
Uðx; 0Þ ¼ : ð8:43Þ
UR if x[0
A rarefaction wave is formed, with the right edge propagating at velocity
SR ¼ UR þ ðghR Þ1=2 : ð8:44Þ

Fig. 8.12 Dry-bed conditions along center of two rarefactions a free surface profiles h(x) and
b x-t plane
The propagation velocity of the left-going wet–dry interface is determined using the
basic equation of the rarefaction wave, namely
UR 2cR ¼ UL 2cL : ð8:45Þ
At the dry front hL = 0, so that from Eq. (8.45)
SL ¼ UL ¼ UR 2cR : ð8:46Þ
The depth and velocity within the rarefaction wave are thus given by (Toro 2001)
1h xi
ðghÞ1=2 ¼ UR þ 2ðghR Þ1=2 þ ;
3 t ð8:47Þ
1h 1=2 xi
U ¼ UR 2ðghR Þ þ 2 :
3 t
A third case occurs in relation to the two-rarefaction wave case previously

studied for the wet-bed Riemann solver. The solution of the wet-bed Riemann
problem yields Eq. (8.33). If FR [ 2, a dry bed is generated within the two rar-
efaction waves (Fig. 8.12). The wave solution then changes; the celerity of the dry
fronts is given by
SL ¼ UL þ 2cL ð8:48Þ

for the left rarefaction wave and
SR ¼ UR 2cR ð8:49Þ
for the right rarefaction wave.

From Eq. (8.21), the celerity in the star region is
ðghL Þ1=2 þ ðghR Þ1=2 1

c ¼ ðgh Þ1=2 ¼ ðUR UL Þ: ð8:50Þ
2 4
Given that c* must be positive to generate real values of h*, the depth-positivity
condition is thus, for arbitrary data (Toro 2001),
1h i 1
c ¼ ðghL Þ1=2 þ ðghR Þ1=2 ðUR UL Þ 0: ð8:51Þ
2 4
For given initial data in a Riemann problem, a preliminary check is conducted to

decide whether a wet-bed or dry-bed Riemann solver applies. If hR = 0, or hL = 0,
or c* < 0, the dry-bed Riemann solver is selected, whereas the wet-bed Riemann
solver must be used otherwise.
Eleuterio Francisco Toro was born on July 16, 1946, at Capitan Pastene,
Chile. He started, in 1967, his career as Teacher of primary education at
Victoria, Chile, continuing studies of mathematical pedagogy, obtaining the
B.Sc. degree in pure mathematics in 1977 at the University of Warwick, UK,
and in 1978 the M.Sc. degree in applied mathematics from University of
Dundee, UK. He obtained the PhD degree in 1982 in computational mathe-
matics from the University of Teesside, Middlesbrough UK. From 1983 until
2001, he was Lecturer and Senior Lecturer in computational fluid mechanics
at Cranfield University, UK, and Professor of applied mathematics at the
Manchester Metropolitan University. From 2002 to 2016, he was Full
Professor of numerical analysis at DICAM, University of Trento, Italy, from

when he there is Emeritus Professor.
His research was and still is focused on the construction of computational
methods for solving partial differential equations, thereby particularly dealing
with the hyperbolic equations, as these of the shallow water equations.
The ADER framework launched in 2001 allows for the construction of
one-step, nonlinear numerical methods of arbitrary order of accuracy in both
space and time. This method results in a huge efficiency gain, thereby
increasing the power of mathematical modeling and simulation for solving
practical problems in science and engineering. This technique is relevant in
many applications including aero-acoustics, shock waves, tsunami wave
propagation, meteorology, biology, and medicine. He is famous in the sci-
entific community for having written two books dealing with Riemann sol-
vers and the finite volume technology; the first, Riemann solvers and
numerical methods in fluid dynamics, is dedicated to the solution of the Euler
equations in gas dynamics. The second, Shock-capturing methods for shallow
free surface flows, is devoted to the solution of the Saint Venant equations for
water flows. The two books are supported by the NUMERICA library, a
collection of source codes for teaching, research and practice, where the
details of the methods presented in the books are explained in detail. The
book readers are encouraged to download this library from the Web site of
Prof. Toro: https://eleuteriotoro.com/.
8.4 Application: Gate Maneuvers in Open Channels
8.4.1 Complete Gate Opening
Consider steady flow at a gate (Fig. 8.13), where subcritical approach flow (state L)
is transformed into supercritical flow1 (state R). Contraction effects are overlooked
in the present shallow water analysis (Jain 2001). These were accounted for by
Cozzolino et al. (2015) for both complete and partial gate openings.
If the gate is fully opened, the initial data correspond to the Riemann problem
stated in Fig. 8.1, with ULhL = URhR = const. as the initial discharge. Figure 8.2
shows the wave profile for a transcritical rarefaction after full gate opening. The
computational example presented in Fig. 8.9 represents thus a full gate opening.
1
We will analyze instantaneous (partial or full) sluice gate openings and closures for supercritical
conditions downstream of the gate. Submerged flows are not considered; the gate acts thus as a
control section in all cases.
8.4 Application: Gate Maneuvers in Open Channels 333
Fig. 8.13 Planar gate in open channel forming a discontinuity between two uniform flow zones.
Contraction effects are overlooked in the SWE analysis
8.4.2 Partial Gate Opening
Consider initially steady gate flow as sketched in Fig. 8.14a, with subscripts u and
R referring to the up- and downstream zones of the gate. For given values of hu and hR,
the values of Uu and UR are determined with the known initial discharge. For partial
gate opening (w > hR and w < hL) (Fig. 8.14b), a depression of depth hd and velocity
Ud is formed upstream of the gate, propagating in the upstream direction via a rar-
efaction wave. The unknowns in the upstream flow portion are thus hd, Ud and the
velocity at the gate opening UL. Across the rarefaction, one may write the identity
Uu þ 2cu ¼ Ud þ 2cd : ð8:52Þ
Two additional equations are needed. One involves energy conservation across the
gate
Ud2 U2
hd þ ¼ wþ L ; ð8:53Þ
2g 2g
while the other is the mass conservation equation
UL w ¼ Ud hd : ð8:54Þ
Using Eqs. (8.54), (8.52) is rewritten as

w
Uu þ 2cu ¼ UL þ 2ðghd Þ1=2 : ð8:55Þ
hd
From Eq. (8.53), one gets with Eq. (8.54)
UL2 U2 w2 U 2
¼ hd w þ d ¼ hd w þ 2 L ; ð8:56Þ
2g 2g hd 2g
Fig. 8.14 Partial gate opening a initial steady state, b generation of transient motion (flow
downstream of the gate corresponds to case c of Fig. 8.3) (adapted from Montuori and Greco
1973)
or
1=2
1=2 w2
UL ¼ ½2gðhd wÞ 1 2 : ð8:57Þ
hd
Inserting Eq. (8.57) into Eq. (8.55) yields the nonlinear implicit equation for hd
f ðhd Þ ¼ wð2gÞ1=2 ðhd þ wÞ1=2 þ 2ðghd Þ1=2 Uu 2cu ¼ 0: ð8:58Þ
This is easily solved using the Newton–Raphson method. Once hd is computed,

UL is determined from Eq. (8.57) and Ud from Eq. (8.54). In the downstream gate
zone, a wave pattern as shown in Fig. 8.14b may be generated, involving a rar-
efaction and a shock wave. The solution of this flow portion corresponds essentially
to the computation of a Riemann problem with a right-going rarefaction wave and a

right-going shock. Obviously, other wave patterns are in principle possible
depending on the initial data of the Riemann problem posed on the right gate side.
Montuori (1968) presents a notable discussion of possible wave solutions. The
solution to the partial gate opening was implemented in a code, inserted in the file
“Partialgateopening.xls”, available in Chap. 12. To use this code:
(1) There must be a gate opening w > hR because results are meaningless other-
wise, given that it is assumed that a rarefaction wave is formed upstream of the
gate.
(2) The left wave in the Riemann problem downstream of the gate must be right
going.
A computational example is summarized in Table 8.4, and the profiles downstream
of the gate are plotted in Fig. 8.15. This type of solution was described by Jain
(2001).
If the flow depth hR is very shallow, the right wave may be a rarefaction wave
instead of a shock (Fig. 8.16a). There is still another possible flow profile down-
stream of the gate, composed of two shocks (Fig. 8.16b). This situation occurs if
w is not large as compared to hR; we consider it here for illustrative purposes.
Suppose the limiting case hL = hR, with UL > UR. Equation (8.13) yields for a
two-shock case if hL = hR,
Table 8.4 Example of partial gate opening. Initial discharge is 0.5 m2/s and new gate opening
w = 0.4 m
Flow hu (m) Uu (m/s) hd (m) Ud (m/s) UL (m/s)
Upstream 1 0.5 0.645 1.733 2.795
hR (m) UR (m/s) h* (m) U* (m/s)
Downstream 0.2 2.5 0.316 3.234
Fig. 8.15 Riemann solution of partial gate opening test for flow downstream of the gate plane
with a free surface profile h(x) and b depth-averaged velocity U(x). Details see Table 8.4
Fig. 8.16 Partial gate opening a transient motion with two rarefactions beyond the gate,
corresponding to case c in Fig. 8.4, b transient with two shocks beyond the gate, corresponding to
case c in Fig. 8.5 (adapted from Montuori and Greco 1973)
Table 8.5 Example of partial gate opening. Initial discharge is 0.5 m2/s and new gate opening
w = 0.25 m
Flow hu (m) Uu (m/s) hd (m) Ud (m/s) UL (m/s)
Upstream 1 0.5 0.827 1.066 3.53
hR (m) UR (m/s) h* (m) U* (m/s)
Downstream 0.2 2.5 0.307 3.185
1=2
g
2ð h hR Þ ð h þ h R Þ ¼ UL UR : ð8:59Þ
2h hR
Note that the right-hand side of Eq. (8.59) is positive; to have a compatible
positive left-hand side, of necessity then h* > hR, implying that the two-shock
Fig. 8.17 Riemann solution of partial gate opening test for flow downstream of the gate plane
with a free surface profile h(x) and b depth-averaged velocity U(x). Details see Table 8.5
assumed wave pattern, is the correct solution. Another simulation was conducted
based on the example presented in Table 8.4, but using the smaller gate opening
w = 0.25 m. Results are summarized in Table 8.5, with the flow profiles down-
stream of the gate presented in Fig. 8.17.
8.4.3 Complete Gate Closure
Consider the case of a complete gate closure (Fig. 8.18). Upstream of the gate, a
shock wave traveling to the left, is formed. The unknowns in this portion of the
computational domain are the velocity Ud, the flow depth hd, and the surge velocity
S (Jain 2001). By definition of gate closure, Ud = 0. Thus, the two remaining
unknowns are determined using the mass and momentum conservation equations
across the surge, e.g., the Rankine–Hugoniot jump conditions, which are,
respectively,
hu ðUu SÞ ¼ hd S; ð8:60Þ

1=2
ghd
S ¼ Uu ð hu þ hd Þ : ð8:61Þ
2hu
The solution of this system of equations entails no difficulty. On the downstream
gate side, the flow solution admits two possible wave patterns. If the downstream
Froude number is FR \2, the channel is wet beyond the gate and the water is static
there, with a rarefaction wave traveling to the right (Fig. 8.18a). This case is
computationally equivalent to the two-rarefaction Riemann problem for wet-bed
conditions. However, if FR [ 2, the channel is dry beyond the gate, while there is
still a rarefaction wave traveling to the right, yet of nonvanishing celerity at the
wet–dry interface (Fig. 8.18b).
A computational example is presented in Fig. 8.19. The test data and the results
are shown in Table 8.6. The first test case corresponds to a two-rarefaction
Riemann problem with wet-bed conditions. The profiles for h(x) and U(x) at t = 3 s
are plotted in Fig. 8.19a, b only for x > 0, given that the rest of the profiles are
Fig. 8.18 Complete gate closure a wet bed beyond the gate, b dry bed beyond the gate
meaningless. The simulations were conducted with the wet-bed Riemann solver
available on the file “ExactRiemannSolver_wetbed.xls”. A second test was conducted
by increasing the velocity at the tailwater section to produce a Froude number in excess
of 2. A dry-bed two-rarefaction approach was used to model the flow, with the profiles
for x > 0 plotted in Fig. 8.19c, d. The results were generated with a dry-bed Riemann
solver available on the file “ExactRiemannSolver_drybed.xls”.
8.4.4 Partial Gate Closure
For the case of a partial gate closure, a surge propagates in the upstream direction
within the approach flow to the gate (Fig. 8.20). The unknowns in this flow portion
are S, hd, Ud, and UL. The equations available are the continuity equation across the
surge
Fig. 8.19 Riemann solution for full gate closure test: flow downstream of the gate with a, b free
surface profile h(x) and depth-averaged velocity U(x) for wet-bed conditions, c, d idem for dry-bed
conditions. Details see Table 8.6
Table 8.6 Examples of complete gate closure using the two-rarefaction Riemann problem
Test hL (m) hR (m) UL (m/s) UR (m/s) h* (m) U* (m/s)
Wet-bed 1.2 1.2 −5 5 0.088 0
Dry-bed 1.2 1.2 −10 10 0 3.138
hu ðUu SÞ ¼ hd ðUd SÞ; ð8:62Þ
the momentum equation across the surge

1=2
ghd
S ¼ Uu ð hu þ hd Þ ; ð8:63Þ
2hu
conservation of energy at the gate
Ud2 U2
hd þ ¼ wþ L ; ð8:64Þ
2g 2g
and conservation of mass at the gate
UL w ¼ Ud hd : ð8:65Þ
Fig. 8.20 Partial gate closure a shock–rarefaction waves, see case c in Fig. 8.6, b two shock
waves, see case c in Fig. 8.5, c two rarefaction waves, see case c in Fig. 8.4 (adapted from
Montuori and Greco 1973)
Once this system of four equations and four unknowns is solved (Jain 2001), the
flow downstream of the gate is computed as a Riemann problem. The case
involving a shock wave followed by a rarefaction wave is plotted in Fig. 8.20a.
Depending on the initial data of the Riemann problem, a number of wave profiles
may be generated beyond the gate. The cases of two shocks and two rarefaction
waves are sketched in Fig. 8.20b, c, respectively.
These scenarios are illustrated with a numerical example using the code avail-
able in “ExactRiemannSolver_wetbed.xls”. Data for this example are taken from
Jain (2001, example 7-6). The initial discharge is qo = 7.92 m2/s, hR = 1 m,
UR = qo/hR = 7.92 m/s and the gate opening is set to w = 0.7 m. Solving the
equations upstream of the gate yields UL = 8.63 m/s (Jain 2001). The downstream
flow profiles for this test are plotted in Fig. 8.21a, b, and computations are sum-
marized in Table 8.7. Note that the wave patterns involve a left shock and right
rarefaction. Henceforth, we keep w fixed. Given that the gate acts as a control, flow
conditions upstream of the gate remain unaltered for fixed w. Now, raise hR up to
1.2 m, computing the corresponding UR to satisfy the initial data. As observed from
Figs. 8.21c, d and Table 8.7, the wave patterns in this case correspond to two
shocks. If the analysis is repeated this time lowering hR to 0.8 m, two rarefactions
waves travel to the right of the gate (Fig. 8.21e, f and Table 8.7). As illustrated in
this example, the treatment of the gate maneuvers within the context of a Riemann
problem permits to identify all possible wave patterns without the need of pre-
liminary assumptions.
8.4.5 Comparison with Experiments
The experimental investigation by Montuori and Greco (1973) is used to test the
overall quality of the theory presented against experiments. Two experiments are
selected, with experimental data for the Riemann problem posed beyond the gate
listed in Table 8.8, as well as visual observations reported with photographs by
Montuori and Greco (1973). Computations of the star region values using the
wet-bed Riemann solver are included in the same table. In the comparison under-
taken below, the left conditions of the Riemann problem are taken as the experi-
mental conditions at the new vena contracta after the gate movement. Thus,
contraction effects are accounted for here.
The first experiment is reproduced computationally in Fig. 8.22. Wave results
are plotted in Fig. 8.22 after 5 s of routing. Note that the experimental data for the
Riemann problem produce computationally a two-shock wave pattern, in concor-
dance with the visual observations by Montuori and Greco (1973). Table 8.9
compares the theory with available experimental measurements, indicating fair
agreement.
The second experiment is reproduced computationally in Fig. 8.23. Wave results
are plotted in Fig. 8.23 after 5 s of routing. The experimental data for the Riemann
problem in this test produce computationally a shock–rarefaction wave pattern, in
Fig. 8.21 Partial gate closure. Generation of different wave patterns keeping w fixed and varying
hR (and UR = qo/hR). See Table 8.7 for test conditions of each plot
Table 8.7 Examples of Riemann problems downstream of gate after partial gate closure
Test Waves hR (m) UR (m/s) h* (m) U* (m/s)
Figure 8.21a, b Shock–rarefaction 1 7.92 0.95 7.76
Figure 8.21c, d Two shocks 1.2 6.6 1.26 6.77
Figure 8.21e, f Two rarefactions 0.8 9.9 0.58 9.08
Test conditions are: qo = 7.92 m2/s, hL = w = 0.7 m, UL = 8.63 m/s, qnew = ULw = 6.041 m2/s
Table 8.8 Experiments of Montuori and Greco (1973): data of Riemann problem beyond the gate
and computed values of star region using wet-bed Riemann solver
Test Wave pattern hL (m) hR UL UR h* U*
experimentally observed (m) (m/s) (m/s) (m) (m/s)
Figure 8.22 Two bores 0.057 0.1 1.844 0.11 0.165 0.685
Figure 8.23 A bore followed by smooth 0.0248 0.093 2.513 1.955 0.073 1.733
wave
Subscript L indicates conditions at the new vena contracta after gate movement, and R indicates
tailwater conditions existing before moving the gate
Fig. 8.22 Computational reproduction of a two-shock wave pattern beyond a gate generated
experimentally by Montuori and Greco (1973). For details, see Table 8.9
Table 8.9 Comparison of theory and experiments by Montuori and Greco (1973) for the
Riemann problem presented in Fig. 8.22
SL (m/s) SR (m/s) h* (m)
Simulation 0.072 1.573 0.165
Experiment 0.1 1.45 0.165
Fig. 8.23 Computational reproduction of a shock–rarefaction wave pattern beyond a gate

generated experimentally by Montuori and Greco (1973). For details, see Table 8.10
Table 8.10 Comparison of theory and experiments by Montuori and Greco (1973) for the
Riemann problem presented in Fig. 8.23
SL (m/s) h* (m)
Simulation 1.329 0.073
Experiment 1.45 0.065
concordance with the visual observations by Montuori and Greco (1973).

A comparison of theory with the available experimental measurements is reported
in Table 8.10, showing again fair agreement of the Riemann solution with
experiments.
Remark: Lax entropy condition for shocks

While presenting surges and depression waves in Chap. 5, the usual case was
explained, namely the shock as a discontinuous increase in depth and the rarefac-
tion wave as a gradual depression in depth. In the literature, the depression shock
(see wave propagation at rate SL in Fig. 8.20a, for example) tends to be overlooked.
While it is certainly true that it is not the usual case while describing surges of tidal
origin, it is a realistic flow profile in the context of gate maneuvers in open
channels. It was first described by Montuori and Greco (1973) based on conser-
vation of mass and momentum and verified experimentally. In recent open channel
flow books, it is an accepted solution (e.g. Fig. 8.20a) (Jain 2001 page 222; Jeppson
2011 page 920; Katopodes 2019 page 768). We will now demonstrate that the
experiments of Montuori and Greco (1973) satisfy Lax’ entropy condition, and
thus, the depression shock of Fig. 8.23 is physically realistic.
Let the eigenvalues be
k1 ðUÞ ¼ U ðghÞ1=2 ; k2 ðUÞ ¼ U þ ðghÞ1=2 ; ð8:66Þ
a discontinuity separating states UL and UR and propagating at speed S satisfies Lax

entropy condition if there is a p index such that (LeVeque 2002, page 268)
kp ðUL Þ [ S [ kp ðUR Þ: ð8:67Þ
In this case, the p-characteristics are impinging on the discontinuity, while the other
characteristics are crossing it,
ki ðUL Þ\S and ki ðUR Þ\S for i\p;

ð8:68Þ
ki ðUL Þ[S and ki ðUR Þ[S for i [ p:
In our case, p can be 1 or 2, so one of the following set of identities shall be verified
by the depression shock:
Conditions for p = 2
k2 ðUL Þ [ S [ k2 ðUR Þ;
k1 ðUL Þ\S; ð8:69Þ
k1 ðUR Þ\S:
It can be easily verified that this set of statements is coherent with Fig. 7.3 and
Eq. (7.22).
Conditions for p = 1
k1 ðUL Þ [ S [ k1 ðUR Þ;
k2 ðUL Þ [ S; ð8:70Þ
k2 ðUR Þ [ S:
Table 8.11 Verification of Lax entropy condition for depression shock in Fig. 8.23
(SL = 1.329 m/s), using data in Table 8.8
Test k1(UL) (m/s) k1(U*) (m) k2(UL) (m/s) k2(U*) (m/s)
Figure 8.23 2.02 0.889 3.007 2.578
This is the set of identities verified by the depression shock in Fig. 8.23. To analyze
this shock, the left state is that of the Riemann problem and the right state corre-
sponds to the star region. Therefore, we will check the identities
k1 ðUL Þ [ SL [ k1 ðU Þ;
k2 ð U L Þ [ SL ; : ð8:71Þ
k2 ð U Þ [ SL :
This work is accomplished in Table 8.11 using previous information of Table 8.8,
resulting that Eq. (8.71) is verified, and therefore, the depression shock in Fig. 8.23
is a correct physical solution, as also observed experimentally by Montuori and
Greco (1973).
The described exact solutions apply to analyze instantaneous gate operations. If
the gate maneuvers are not instantaneous, or if source terms are introduced into the
SWE to account, for example, for frictional effects, the theory described is invalid
and a numerical solution is then required.
References
Cozzolino, L., Cimorelli, L., Covelli, C., Della Morte, R., & Pianese, D. (2015). The analytic
solution of the shallow-water equations with partially open sluice-gates: The dam-break
problem. Advances in Water Resources, 80(6), 90–102.
Godunov, S. K. (1959). A difference method for numerical calculation of discontinuous solutions
of the equations of hydrodynamics. Matematicheskii Sbornik, 47(3), 271–306 (in Russian).
Guinot, V. (2003). Godunov-type schemes: An introduction for Engineers. Amsterdam, Boston:
Elsevier science.
Marcel Dekker.
Jeppson, R. (2011). Open channel flow: Numerical methods and computer applications. CRC
Press, Taylor and Francis, New York.
Katopodes, N. D. (2019). Free surface flow: Computational methods. Oxford, UK:
University Press.
Montuori, C. (1968). Brusca immissione di una corrente ipercritica a tergo di altra preesistente
[Sudden perturbation of a supercritical flow over the pre-existing flow]. L’Energia Elettrica,
45(3), 174–187 (in Italian).
Montuori, C., & Greco, V. (1973). Fenomeno di moto vario a valle di una paratoia piana [Varied
flow phenomena beyond a plane gate]. L’Energia Elettrica, 50(2), 73–88 (in Italian).
Riemann, B. (1860). Über die Fortpflanzung ebener Luftwellen von endlicher Schwingungsweite
[On the propagation of plane air waves of finite amplitude]. Abhandlungen der Königlichen
Gesellschaft der Wissenschaften zu Göttingen, 8, 43–65 (in German).
Roberts, S. (2013). Numerical solution of conservation laws applied to the shallow water
equations. Lecture notes. Australia: Mathematical Sciences Institute, Australian National
University.
Interscience publishers.
Chapter 9
Finite Volume Methods
9.1 Introduction
The one-dimensional shallow water equations (SWE), or Saint-Venant equations,

are a system of nonlinear hyperbolic conservations laws (Toro 2001). The mathe-
matical meaning behind these “surnames” linked to the development of
Saint-Venant is clearly elucidated by the definitions (Karni 2011; Vazquez-Cendón
2015):
Conservation is a fundamental principle of the physical world. Matter may move around and
redistribute but it does not appear or disappear. Hyperbolicity means that news that happen at
a given point A, take time before they may affect affairs at another point B. Nonlinear means
the manner in which news propagate depends on what type of news it is. This makes the
subject of nonlinear conservations laws fascinating, rich and challenging to study.
The SWE are solved in this chapter using finite volume methods for hyperbolic
conservation laws, where the space–time plane is divided into control volumes or
simply finite volumes. The presentation focuses on Godunov-type methods, in
which the solution of a number of local Riemann problems between each pair of
adjacent cells is used. The constructed numerical method satisfies locally the
conservation laws, so that wave propagation information is generated at the cell
interfaces. This wave information is used to construct the numerical scheme and
then evolve the solution in time, representing therefore an upwind scheme. First,
Godunov-type methods are stated for the homogeneous SWE, after which
approximate Riemann solvers are presented. The treatment of dry beds is discussed,
as well as the inclusion of source terms including the discretization of the bed slope
term to produce a well-balanced scheme. The treatment of bed friction is explained
in detail. The one-sided first-order method is presented, as well as the second-order
Total Variation Diminishing (TVD) MUSCL (Monotonic Upstream Centered
Schemes for Conservation Laws)-Hancock scheme. Numerical computations for
selected test cases are finally compared with exact solutions and experimental
observations to highlight the quality of computations.

https://doi.org/10.1007/978-3-030-13073-2_9
348 9 Finite Volume Methods
9.2 Godunov-Type Schemes
9.2.1 Conservative Formula
The SWE are written in vector conservative form as
@U @F
þ ¼ S: ð9:1Þ
@t @x
Here, U is the vector of the conserved variables, F the flux vector and S the source
term vector, given for a rectangular channel by

h hU 0
U¼ ; F¼ ; S¼ ; ð9:2Þ
hU hU 2 þ 12 gh2 gh @z
@x ghSf
b
with h as the water depth, U the depth-averaged velocity, zb the bed elevation, g the
gravity acceleration and Sf the friction slope. The differential Eq. (9.1) is valid in
zones of the computational domain with smooth or continuous solutions, but it does
not apply at discontinuous portions as shocks. Therefore, Eq. (9.1) is integrated
over a control volume in the x-t plane, resulting in
ZZ ZZ
@U @F
þ dxdt ¼ Sdxdt: ð9:3Þ
@t @x
The integral Eq. (9.3) allows for the computation of both continuous and discontinuous
solutions. It is the fundamental relation of the finite volume method. For the rectangular
control volume in the x-t plane depicted in Fig. 9.1, one can write (Toro 2001, 2009)
Fig. 9.1 Control volume in

x-t plane
9.2 Godunov-Type Schemes 349
tZþ Dt xZ
i þ 1=2 Z
xi þ 1=2 tZþ Dt tZþ Dt Z
xi þ 1=2
Z Z
xi þ 1=2 t þ Dt
@U @F @U @F
þ dxdt ¼ dx dt + dt dx ¼ Sdxdt:
@t @x @t @x
t xi1=2 xi1=2 t t xi1=2 xi1=2 t
ð9:4Þ
Here, i is the cell index in the x-direction, and i + 1/2 the interface between cells
i and i + 1. Equation (9.4) generates after elementary integration (Toro 2001, 2009)
Z
xi þ 1=2 tZþ Dt

½Uðx; t þ DtÞ Uðx; tÞdxþ F xi þ 1=2 ; t F xi1=2 ; t dt
xi1=2 t
ð9:5Þ
Z Z
xi þ 1=2 t þ Dt
¼ Sdxdt:
xi1=2 t
Let the temporal and spatial average values be defined as
tZþ Dt
1
Fi þ 1=2 ¼ F xi þ 1=2 ; t dt;
Dt
t
tZþ Dt
1
Fi1=2 ¼ F xi1=2 ; t dt;
Dt
t
xi1=2 þ Dx
Z
1
Uki þ 1 ¼ Uðx; t þ DtÞdx; ð9:6Þ
Dx
xi1=2
xi1=2 þ Dx
Z
1
Uki ¼ Uðx; tÞdx;
Dx
xi1=2
Z Z
xi þ 1=2 t þ Dt
1
Si ¼ Sdxdt:
DxDt
xi1=2 t
Inserting Eq. (9.6) into Eq. (9.5) results in the exact conservative formula
Dt
Uki þ 1 ¼ Uki Fi þ 1=2 Fi1=2 þ DtSi : ð9:7Þ
Dx
This is the fundamental equation of the schemes to be presented here. The com-
putational domain in the finite volume method is divided into a number of control
Fig. 9.2 Intercell fluxes
volumes in the x-t plane (Fig. 9.2), where Eq. (9.7) is applied. Here, Fi+1/2 is the
numerical flux across the interface between cells i and i + 1. No numerical
approximations are introduced in Eq. (9.7), although it looks like a finite-difference
equation. Until stating the contrary, the homogeneous version of the SWE is
focused by dropping the source terms, e.g., setting Si = 0 in Eq. (9.7).
9.2.2 Conservative Property: Definition of Numerical

Discharge
Note that in the conservative formula (9.7), the flux leaving cell i through the
interface i + 1/2 is identical to the flux entering into cell i + 1 across this common
interface. Let us write the conservative formula for updating the conserved variables
at cell i − 1,
þ1 Dt
Uki1 ¼ Uki1 Fi1=2 Fi3=2 : ð9:8Þ
Dx
Summing it with Eq. (9.7) produces, with Si = 0,
Dt
Uki þ 1 þ Uki1
þ1
¼ Uki þ Uki1 Fi þ 1=2 Fi3=2 : ð9:9Þ
Dx
It indicates that the total variation of U is determined by the fluxes entering and
leaving the extreme interfaces of the block formed by the adjacent cells i − 1 and
i. If the computational domain is divided into N cells, the summation of the
updating equation for all the cells yields
X
N X
N
Dt
Uki þ 1 ¼ Uki FN þ 1=2 F1=2 : ð9:10Þ
i¼1 i¼1
Dx
In this case, the total variation of U in the computational domain is determined by

the fluxes entering and leaving the boundary sections. This is the so-called con-
servative property (Roache 1972; Toro 1992, 2001; LeVeque 1992, 2002;
Vazquez-Cendón 2015).
The numerical discharge is the first component of the numerical flux vector, e.g.,
tZþ Dt
1
qi þ 1=2 ¼ Fi þ 1=2 ð1Þ ¼ q xi þ 1=2 ; t dt; ð9:11Þ
Dt
t
corresponding to a time average of the discharge crossing the interface i + 1/2. This
definition of discharge satisfies the conservative property, for the reasons stated
above. However, the numerical discharge can also be defined as the second com-
ponent of the vector U; it is a spatial average value at the time level k + 1,
xi1=2 þ Dx
Z
1
qki þ 1 ¼ Uki þ 1 ð2Þ ¼ qðx; t þ DtÞdx: ð9:12Þ
Dx
xi1=2
This definition of the numerical discharge does not satisfy the conservative prop-
erty, however. Whether Eq. (9.11) or (9.12) for the numerical discharge is used to
represent the true physical discharge is open to debate. In some steady flow
problems (q = const.), like the hydraulic jump, the steady-state solution generated
by the SWE using certain numerical schemes for transient computations experience
unphysical jumps in qki þ 1 , whereas qi þ 1=2 attains a constant steady value due to the
conservative property (Ying and Wang 2008). This last definition of numerical
discharge may be then more useful to approximate the true physical discharge
(Ying and Wang 2008). In practice, computational schemes are not affected by the
choice; it is only a matter on which variable is taken as the output numerical
discharge once the solution is determined.
9.2.3 Godunov Upwind Numerical Flux
The conservative Eq. (9.7) with Si = 0 applies to update the solution of the
space-averaged values Ui once the numerical flux Fi+1/2 is estimated. This step
transforms the exact Eq. (9.7) approximate for a particular computation. The exact
solution U(x, t) within a cell can be written in the form of a Taylor series as

@U ðx xi Þ2 @ 2 U
UðxÞ ¼ Uðxi Þ þ ðx xi Þ þ þ :::: ð9:13Þ
@x i 2 @x2 i
In the Godunov upwind method, the actual solution U(x) at time t is approximated
by the space-averaged values within each cell (Fig. 9.3)
UðxÞ ¼ Uðxi Þ ¼ Ui : ð9:14Þ
The piecewise constant data representations produce a scheme first-order

accurate in space and time. Between each two adjacent cells, a discontinuity in U is
formed generating a sequence of local Riemann problems along the computational
domain (Fig. 9.4). Godunov-type methods incorporate the solution of the Riemann
problem (exact or approximate) into the conservative formula
Dt
Uki þ 1 ¼ Uki Fi þ 1=2 Fi1=2 : ð9:15Þ
Dx
Fig. 9.3 Cell-averaged

values Ui as approximation of
the exact solution U(x) at time
t. The plot represents the
typical distribution of a single
component of U = (U1, U2)T
Fig. 9.4 Godunov upwind

method. a Piecewise constant
data representation and
b local Riemann problems
formed along computational
domain
Consider two initial states UL and UR at a generic interface i + 1/2 between cells
i and i + 1. This is by definition a local Riemann problem, from which a number of
wave patterns may result. The typical case with a left-going rarefaction wave and
right-going shock is presented in Fig. 9.5. The shock and rarefaction waves
propagate with signal speeds SR and SL, respectively. The exact Riemann solution
presented in Chap. 8 thus applies here, taking the role of a component of the
numerical scheme. The complete Riemann solution is constructed assembling rar-
efaction and shock waves. Thus, it is noted that the solution of the Riemann
problem is self-similar, e.g., Ui+1/2(x/t). At the original position of the discontinuity
(x = 0), the solution of the Riemann problem is thus independent of t and therefore
steady. The constant state region behind the shock front (star region) depicted in
Fig. 9.5 is denoted as U*. The numerical flux crossing the original discontinuity at
x = xi+1/2 is Fi+1/2, which is needed to apply Eq. (9.15). The star region in the
Riemann problem is a steady-state zone where the conserved variables are U*. This
is part of the total (local) Riemann solution Ui+1/2. Therefore, the numerical flux
crossing the t-axis in the Riemann problem is also a constant, so that the intercell
flux is exactly evaluated based on the Riemann solution at x = xi+1/2 as
Ztk þ 1 ZDt
1 1
Fi þ 1=2 ¼ F x ¼ xi þ 1=2 ; t dt ¼ Fðx ¼ 0; tÞdt
Dt Dt
tk
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
0
|fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl} ð9:16Þ
global system of reference local system of reference

¼ Fðx ¼ 0Þ ¼ F Ui þ 1=2 ðx ¼ 0Þ :
Fig. 9.5 Riemann problem

solution at interface i + 1/2
for left-going rarefaction
wave and right-going shock
under subcritical flow. a Free
surface profile, b x-t plane
This is the Godunov upwind numerical flux, namely the physical flux function
evaluated using the solution of the Riemann problem at the location of the interface
in question. Note that in local coordinates, the origin of the discontinuity is taken at
x = 0.
For the case depicted in Fig. 9.5, the vector Ui+1/2(x = 0) is U*. Thus,
Fi þ 1=2 FðU Þ: ð9:17Þ
The value of U* is determined solving the exact Riemann problem for wet-bed
conditions. If the left rarefaction wave crosses the t-axis, critical flow occurs along
it, producing the values of c and U after setting dx/dt = U − c = 0 (Toro 2001),
1
cc ¼ ðghc Þ1=2 ¼ Uc ¼ ðUL þ 2cL Þ: ð9:18Þ
3
The critical depth hc is thus from Eq. (9.18)
1
hc ¼ ðUL þ 2cL Þ2 : ð9:19Þ
9g
Using Eq. (9.18), the numerical (critical) flux is then

hc Uc g1=2 h3=2
Fi þ 1=2 ¼ FðUc Þ ¼ ¼ c : ð9:20Þ
hc Uc2 þ 12 gh2c 3
2 ghc
2
This is also a steady vector. For a right-going rarefaction the flow is simply neg-
ative. Thus, critical flow is obtained setting dx/dt = U + c = 0, resulting in (Toro
1992, 2001)
1
cc ¼ ðghc Þ1=2 ¼ Uc ¼ ðUR þ 2cR Þ; ð9:21Þ
3
with the critical depth hc given by
1
hc ¼ ðUR þ 2cR Þ2 : ð9:22Þ
9g
The numerical (critical) flux is then

hc Uc g1=2 h3=2
Fi þ 1=2 ¼ FðUc Þ ¼ ¼ c ; ð9:23Þ
hc Uc2 þ 12 gh2c 3 2
2 ghc
which is obviously identical to Eq. (9.20) but with a flow reversal. If the left-state
values reach the t-axis, the numerical flux is

hL UL
Fi þ 1=2 ¼ FðUL Þ ¼ : ð9:24Þ
hL UL2 þ 12 gh2L
Likewise, if the right-state values are propagated to the t-axis

hR UR
Fi þ 1=2 ¼ FðUR Þ ¼ : ð9:25Þ
hR UR2 þ 12 gh2R
The solution of the Riemann problem may be conducted exactly, as described

here, or approximately, based on a simplified representation of the wave structure.
A summary of all possible wave patterns is presented in Chap. 8 (see Figs. 8.3-8.6
and Table 8.1), as well as a discussion of the possible dry-bed cases. This method,
first-order accurate in both space and time, is essentially the Godunov first-order
upwind method (Godunov 1959). In this method, the waves computed at each
interface satisfy locally the conservation equations, and this wave propagation
information generated is used to evolve the solution in time, and thus, the numerical
method constructed is of upwind type. High-order accuracy is sometimes needed
(Harten 1983; Sweby 1984), and to obtain second-order accuracy, data represen-
tation within a cell must be changed, using linear approximations.
9.2.4 Stability Condition
Consider the right wave from the Riemann problem formed at interface i − 1/2 and
the left wave from Riemann problem formed at interface i + 1/2, both propagating
within cell i (Fig. 9.6). Consider that the wave of maximum speed Smax within the
data is the left wave emanating from interface i + 1/2. For stability of the explicit
Godunov scheme, the Courant–Friedrichs–Lewy number CFL must be limited. The
time step Dt is determined at time level k using the equation
Dx
Dt ¼ CFL: ð9:26Þ
Smax
At first glance, stability of the scheme may be thought to be imposed by

avoiding the interaction of waves within a cell (Fig. 9.6a). The limiting time step is
thus given by CFL ¼ 1=2, resulting in
1 Dx
Dt ¼ : ð9:27Þ
2 Smax
This time step avoids that during time Δt, the fastest signal crosses more than half of
the cell width (see point A in Fig. 9.6a). However, a careful inspection of the
conservative Eq. (9.15) and the Godunov numerical flux Eq. (9.16) reveal that
computations are only a function of the numerical flux Fi+1/2, which, in turn, is not a
Fig. 9.6 Propagation of

waves in Godunov scheme
with CFL = a 1/2 there is no
wave interaction within the
cell, b 1; there is wave
interaction within the cell, yet
the Riemann problem at
interface i + 1/2, from which
the wave of maximum speed
Smax ensues, does not
influence the state at interface
i − 1/2, and, thus, the
numerical flux Fi−1/2 resulting
from the Riemann problem
there is not contaminated
function of the entire Riemann solution, but rather of the particular value at xi+1/2.
Thus, the detailed Riemann solution is not important at all, and wave interaction
within a cell can be permitted as long as the waves originating from a discontinuity
are not affecting the solution of the Riemann problem at the adjacent discontinu-
ities. The stability condition of the scheme is thus CFL ¼ 1, providing a maximum
time step of (Fig. 9.6b, point B)
Dx
Dt ¼ : ð9:28Þ
Smax
In practice, CFL ¼ 0:9 is usually adopted. One may estimate Smax using the
eigenvalues available at time k, e.g.,
Dx Dx
Dt ¼ CFL ¼ h i CFL: ð9:29Þ
Smax 1=2
max jU j þ ðghk Þ
k
i i
Under the presence of dry fronts, whose speeds are much faster than the eigen-
values, the scheme may turn unstable, so care is required. In these cases, a lower
value of CFL may be necessary to produce stable results.
Sergei Konstantinovich Godunov was born on July 17, 1929, at Moscow.

He was from 1946 to 1951 at the Department of Mechanics and Mathematics
of the Moscow State University, from where he obtained in 1951 the M.S.
diploma. He became in 1954 Candidate (Ph.D.), and in 1965 Doctor of
Physical and Mathematical Sciences D.Sc. He was elected Corresponding
Member of the USSR Academy of Sciences in 1976, and Member of the
Russian Academy of Sciences in 1994. In 1997, he was awarded Honorary
Professor at the University of Michigan, Ann Arbor, MI. He was further
awarded the State Lenin Prize in 1959, the A. N. Krylov Prize in 1972, and in
1993 the M. A. Lavrentiev Prize.
His most influential work is in the area of applied and numerical mathe-
matics, with a major impact on science and engineering, particularly in the
development of methodologies used in computational fluid dynamics (CFD).
His scheme is a conservative method for solving partial differential equations.
Therein, the conservative variables are considered as piecewise constant over
the mesh cells at each time step, and the time evolution is determined by the
exact solution of the Riemann (shock tube) problem at the intercell
boundaries. His theorem formulated in 1959 states that linear numerical

schemes for solving partial differential equations, having the property of not
generating new extrema (monotone scheme), can be at most first-order
accuracy.
He was from 1951–1953 Scientific Researcher at the Steklov Institute of
Mathematics (Moscow), from 1953–1962 Scientific Researcher at the
Keldysh Institute of Applied Mathematics (Moscow), followed from 1962–
1969 as Head of Laboratory, Keldysh Institute of Applied Mathematics, then
from 1969–1980 Head of Laboratory, Computing Center of the Siberian
Branch of the USSR Academy of Sciences (Novosibirsk), 1980–2000 Head
of Laboratory and Vice-President (1981–1986) of the Sobolev Institute of
Mathematics, the Siberian Branch of the Russian Academy of Sciences
(Novosibirsk), from when he is their Counsellor.
9.2.5 Computational Sequence for Godunov First-Order

Scheme
The implementation of Godunov’s first-order scheme encompasses the following

steps:
1. Start at time level k with known data of the cell-averaged values Ui in the entire
computational domain
2. Solve exactly the Riemann problem R(Ui, Ui+1) formed at interface i + 1/2
between cells i and i + 1, and compute the numerical flux Fi+1/2. Repeat for all
computational cells. Approximate solutions of the Riemann problem are applied
also, as presented below
3. Use Eq. (9.29) to determine a stable time step Δt, with CFL\1
4. Apply the conservative formula Eq. (9.15) to obtain the cell-averaged values Ui
at the new time level k + 1 for all computational cells
5. Go to step 1, and repeat the computational cycle until reaching the final target
time.
It is necessary for this computational sequence to set the boundary conditions at the
extremes of the computational domain. This issue is detailed in Sect. 9.8.1.
The scheme previously described is the version of Godunov’s method that
prevails in practice. However, the original scheme (Godunov 1959) is a variant in
which the complete Riemann solution is used to evolve the solution. The version of
the scheme is given by the following steps (Toro 2001, 2009):
1. Start at time level k with known data of the cell-averaged values Ui in the entire
computational domain
2. Solve exactly the Riemann problem R(Ui, Ui+1) formed at interface i + 1/2
between cells i and i + 1. The complete solution Ui+1/2(x/t) of each Riemann
problem will be used. The numerical flux is not determined. Repeat for all
computational cells
3. Use Eq. (9.29) to determine a stable time step Δt, with CFL\1=2, given that
the complete solution of the Riemann problem will be used. This limit avoids
that the waves emanating from an interface interact with waves originated at
adjacent interfaces. Interaction of Riemann problems is not considered in the
scheme, given that each interface is treated in an isolated way
4. At each cell i, determine the new cell-averaged values at time level k + 1
averaging the solutions Ui+1/2 and Ui−1/2, e.g.,
Z2Dx Z0
1
1 1
Uki þ 1 ¼ Ui1=2 ðx; DtÞdx þ Ui þ 1=2 ðx; DtÞdx: ð9:30Þ
Dx Dx
0 12Dx
Note that x is the local coordinate for the interface in question. The updated
cell-averaged conservative variables at the new time level are thus a spatial average
of the solution of two Riemann problems solved locally at the two interfaces of each
finite volume. Note that the solution of each Riemann problem is evaluated at time
Δt, and that each solution is determined within half of the cell width. For this
reason, interaction of the waves from the Riemann problems is prohibited and CFL
cannot exceed 1/2 in this version of the scheme.
5. Go to step 1, and repeat the computational cycle until reaching the final target
time.
Figure 9.7 presents schematically a snapshot of computational cell i after time
Dt. Within it, three waves are propagated: a right-going shock generated at interface
i − 1/2 and two left-going shocks coming from interface i + 1/2. The value of
CFL\1=2, so that waves are not interacting within the cell. This simple case
illustrates how to evaluate Eq. (9.30); solving the integrals at time level k + 1 from
point A to B (Fig. 9.7), one finds

1
Uki þ 1 Dx ¼ Ui1=2 SRi1=2 Dt þ Uki Dx SRi1=2 Dt
2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Integration of Riemannproblem from i1=2
1
þ Ui SLi þ 1=2 Dt Dx þ Ui þ 1=2 SRi þ 1=2 Dt SLi þ 1=2 Dt Ukiþ 1 SRi þ 1=2 Dt :
k
2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Integration of Riemann problem from i þ 1=2
ð9:31Þ
Fig. 9.7 Riemann solutions at cell i for CFL\1=2: with left-going rarefaction and right-going
shock generated at interface i – 1/2, and two left-going shocks generated at interface i + 1/2.
a Free surface profile at local time Dt, b x-t plane
As shown with this simple example, the updating process may be tedious, espe-
cially if a rarefaction wave is present in the solution at time level k + 1. Further, the
restrictive stability condition CFL\1=2 is another disadvantage of this version of
Godunov’s scheme. It is thus not further considered.
Applying Godunov’s scheme based on Eq. (9.15), the numerical fluxes are

Fi1=2 ¼ F Ui1=2 ; ð9:32Þ
and

Fi þ 1=2 ¼ F Ukiþ 1 : ð9:33Þ
Equation (9.15) yields then
Dt h k
i
Uki þ 1 ¼ Uki F Ui þ 1 F Ui1=2 : ð9:34Þ
Dx
As shown, Eq. (9.34) is simpler to apply than Eq. (9.31), and the maximum time
step Dt could be multiplied by two.
9.3 Approximate Riemann Solvers 361
9.3 Approximate Riemann Solvers
9.3.1 HLL Riemann Solver
Harten et al. (1983) developed an approximate Riemann solver, named Harten–

Lax–van Leer (HLL). The fundamental idea to simplify the Riemann problem is to
assume that the rarefaction waves (continuous waves of finite extension) can be
replaced by discontinuous waves (infinitely thin waves). Consider the exact
Riemann problem depicted in Fig. 9.8a, with a left-going rarefaction wave and a
right-going shock. The simplified wave structure according to the HLL principle is
as shown in Fig. 9.8b: The constant initial states UL and UR are separated by a
single constant state UHLL. The first issue is thus to find an equation describing this
constant state zone.
Following Toro (2009), consider the control volume sketched in Fig. 9.9 with a
gray shadow. The integral form of the homogeneous conservation laws yields there
[Eq. (9.5)]
ZxR ZT
½Uðx; T Þ Uðx; 0Þdx þ ½FðxR ; tÞ FðxL ; tÞdt ¼ 0: ð9:35Þ
xL 0
For the simplified HLL wave structure, Eq. (9.35) yields after elementary
integration
ZxR
Uðx; T Þdx ¼ xR UR xL UL þ T ðFL FR Þdt: ð9:36Þ
xL
The undetermined integral in Eq. (9.36) is expanded as follows
ZxR ZTSL ZTSR ZxR

Uðx; T Þdx ¼ Uðx; T Þdx þ Uðx; T Þdx þ Uðx; T Þdx: ð9:37Þ
xL xL TSL TSR
Two of the integrals in Eq. (9.37) are immediate, resulting in
ZxR ZTSR
Uðx; T Þdx ¼ ðTSL xL ÞUL þ ðxR TSR ÞUR þ Uðx; T Þdx: ð9:38Þ
xL TSL
Fig. 9.8 Riemann problem

with left-going rarefaction
wave and right-going shock.
a Exact wave structure and
b simplified HLL wave
structure
Equating Eqs. (9.36) and (9.38) generates
ZTSR
Uðx; T Þdx ¼ T ðSR UR SL UL þ FL FR Þ; ð9:39Þ
TSL
from which the average state is

Fig. 9.9 Control volume for

the development of HLL
Riemann solver
ZTSR
1 SR UR SL UL þ FL FR
UHLL ¼ Uðx; T Þdx ¼ : ð9:40Þ
T ð SR SL Þ SR SL
TSL
Both the rarefaction waves and shocks are discontinuous waves in the HLL
approximation (see example in Fig. 9.8b). Thus, the Rankine–Hugoniot jump
conditions apply across both, resulting in
FHLL FL
¼ SL ; ð9:41Þ
UHLL UL
and
FHLL FR
¼ SR : ð9:42Þ
UHLL UR
Using the value of UHLL in either of the two jump conditions, the HLL numerical
flux is
SR FL SL FR þ SR SL ðUR UL Þ
FHLL ¼ : ð9:43Þ
SR SL
The possible cases for the numerical flux are thus (Toro 2001)
8
>
> FL if SL 0
>
<
Fi þ 1=2 ¼ if SL 0 SR : ð9:44Þ
>
> SR SL
>
:
FR if SR 0
Fig. 9.10 Example of possible wave cases in HLL approximate Riemann solver [corresponding
to Eq. (9.44)]. a Two right-going waves, b left- and right-going waves, c two left-going waves
These are sketched in Fig. 9.10. To apply Eq. (9.44), a reliable estimate of both SR
and SL is needed. For wet-bed conditions, these are given as used in the exact
Riemann solver by (Toro 2001)
SL ¼ U L c L k L ; SR ¼ U R þ c R kR : ð9:45Þ
Here, c = (gh)1/2 is the celerity, and the corrector factor to distinguish between the
celerity of shock propagation, and that of a rarefaction wave given by the corre-
sponding eigenvalue, is (K = L, R)
(h
i1=2
h ðh þ hK Þ
kK ¼
1
2 h2K
h [ hK ðshock wave on K Þ; ð9:46Þ
1 h hK ðrarefaction wave on K Þ:
Obviously, an estimate of the flow depth in the star region of the exact Riemann
problem is needed. The two-rarefaction result is used here following Toro (2001)
[Eq. (8.22)]
2
1 1 1
h ¼ ð cL þ cR Þ þ ð U L U R Þ : ð9:47Þ
g 2 4
For the dry-bed case on the right (hR = 0)
SL ¼ UL cL ; SR ¼ UL þ 2cL ; ð9:48Þ
whereas for the dry-bed case on the left (hL = 0)
SL ¼ UR 2cR ; S R ¼ U R þ cR : ð9:49Þ
9.3.2 Lax Numerical Flux
The Lax diffusive method (Lax 1954; Lax and Wendroff 1960) is a finite-difference
scheme with artificial viscosity (see Chap. 7). It is a variant of the unstable scheme,
given by the updating formula (Cunge 1975; Cunge et al. 1980)

Uki1 þ Ukiþ 1 Dt k
Uki þ 1 ¼ Fi þ 1 Fki1 : ð9:50Þ
2 2Dx
It is pertinent at this stage to provide an interpretation of the Lax diffusive method

within the context of finite volume methods and the solution of the Riemann
problem. Consider a finite-difference mesh where the nodes are coincident with the
cell centers in a finite volume setup. Nodal values of the finite difference scheme are
thus equal to the cell-averaged values. The equation modeled by the Lax scheme is
(Chaudhry 2008) [see Eq. (7.51)]
@U @F 1 @ 2 U
þ ¼ D ; ð9:51Þ
@t @x 2 @x2
with D as an artificial viscosity coefficient given by
ðDxÞ2
D¼ : ð9:52Þ
Dt
An issue of the Lax scheme is its inconsistency, given that a source term with
artificial viscosity is added to obtain numerical stability. The integral solution of
Eq. (9.51) will be worked out as follows; rewrite it using basic calculus as

@U @ 1 @U
þ F D ¼ 0; ð9:53Þ
@t @x 2 @x
or with f as a flux function affected by artificial viscosity
@U @f
þ ¼ 0: ð9:54Þ
@t @x
Its integral form over a control volume in the x-t plane is thus
Dt
Uki þ 1 ¼ Uki f i þ 1=2 f i1=2 : ð9:55Þ
Dx
The numerical flux fi+1/2 at interface i + 1/2 is exactly
tZþ Dt
1
f i þ 1=2 ¼ f xi þ 1=2 ; t dt
Dt
t
tZþ Dt
1 1 @U
¼ F D dt
Dt 2 @x i þ 1=2
t ð9:56Þ
tZþ Dt tZþ Dt
1 1 1 @U
¼ F xi þ 1=2 ; t dt D dt
Dt Dt 2 @x
t t

1 @U
¼ Fi þ 1=2 D :
2 @x i þ 1=2
Approximations are now introduced into the general Eq. (9.56). The numerical flux
Fi+1/2 is simply approximated by averaging the physical flux function evaluated at
the initial states of the Riemann problem at interface i + 1/2, whereas the derivative
of U is approximated by a finite difference, resulting in
k
Fki þ Fkiþ 1 1 U Uki Fk þ Fkiþ 1 1 Dx k
f i þ 1=2 ¼ D iþ1 ¼ i Ui þ 1 Uki :
2 2 Dx 2 2 Dt
ð9:57Þ
Developing the corresponding expression for fi−1/2, and inserting the results into
Eq. (9.55), Eq. (9.50) is regained. In the finite volume interpretation of the Lax
scheme, the solution of the local Riemann problems is avoided by adding artificial
viscosity, thereby forcing numerical stability. The numerical flux of the Lax method
is usually defined as (Toro 2001)
FL þ FR 1 Dx
Fi þ 1=2 ¼ ðUR UL Þ; ð9:58Þ
2 2 Dt
to be used into
Dt
Uki þ 1 ¼ Uki Fi þ 1=2 Fi1=2 : ð9:59Þ
Dx
The price of avoiding the solution of the local Riemann problems is to add artificial
viscosity. If the artificial viscosity term is dropped, the numerical flux is
Fki þ Fkiþ 1
Fi þ 1=2 ¼ ; ð9:60Þ
2
resulting in the unstable formula (Hoffman 2001; Hirsch 1988, 1990; LeVeque 2002)
Dt
Uki þ 1 ¼ Uki ðFi þ 1 Fi1 Þ: ð9:61Þ
2Dx
Thus, not solving the Riemann problem and taking the numerical flux as an average
of estimates based on initial data is not a good choice.
An alternative interpretation of Eq. (9.58) is given by Toro (2001) as follows.
Consider the HLL numerical flux given by Eq. (9.44) with the signal speeds
approximated as
SR ¼ SL ¼ Smax ; ð9:62Þ
where Smax is to be determined. The numerical flux is then
Smax FL þ Smax FR S2max ðUR UL Þ

Fi þ 1=2 ¼
2Smax ð9:63Þ
FL þ FR 1
¼ Smax ðUR UL Þ:
2 2
Using Eq. (9.29) with CFL ¼ 1 yields
Dx
Smax ¼ ; ð9:64Þ
Dt
which inserted into Eq. (9.63) yields Eq. (9.58). Therefore, the Lax–Friedrichs
scheme can be alternatively understood as a Godunov-type method with the HLL
Riemann solver using the simplest possible choice for the signal speeds.
9.3.3 Roe Riemann Solver
Roe (1981) developed an approximate Riemann solver for the Euler equations later
applied by Glaister (1987) to the SWE. The homogeneous version of Eq. (9.1) can
be written as
@U @U
þ AðUÞ ¼ 0; ð9:65Þ
@t @x
where A is the Jacobian matrix
@F
A¼ : ð9:66Þ
@U
With A~ as a matrix of constant coefficients used to approximate matrix A, Roe

transformed the nonlinear Eq. (9.65) into the linear system
@U ~ @U
þA ¼ 0: ð9:67Þ
@t @x
This matrix is defined locally in terms of the initial data of the Riemann problem at
a generic interface, e.g., A~ ¼A ~ ðUL ; UR Þ: Using the solution to the Riemann
problem for the linear system and resorting to the integral form of the conservation
laws result in the Roe numerical flux (Toro 2001, 2009; Khan and Lai 2014)
FL þ FR 1 X 2
i þ 1=2 ¼
FRoe aj ~
~ kj K
~ j: ð9:68Þ
2 2 j¼1
Here, the wave strengths are

~hDU ~
1 1 hDU
~a1 ¼ Dh ; ~a2 ¼ Dh þ ; ð9:69Þ
2 ~c 2 ~c
whereas the average eigenvalues are
~k1 ¼ U
~ ~c; ~k2 ¼ U
~ þ ~c; ð9:70Þ
and right eigenvectors

~1 ¼ 1 ~2 ¼ 1
K ~ ~c ;
U
K ~ þ ~c :
U
ð9:71Þ
The jumps in h and U are defined by
Dh ¼ hR hL ; DU ¼ UR UL ; ð9:72Þ
and the Roe averages are given by
~h ¼ ðhR hL Þ1=2 ; ð9:73Þ

hg i1=2
~c ¼ ð hR þ hL Þ ; ð9:74Þ
2
1=2 1=2
~ ¼ hL UL þ hR UR :
U ð9:75Þ
1=2 1=2
hL þ hR
The Roe solver as stated above is not entropy satisfying. Thus, an entropy fix is
needed to avoid unphysical jumps in transcritical rarefactions.
9.4 Dry-Bed Treatment
Computationally we must decide whether cells are wet or dry, even though con-
ceptually they could be partially filled with water. This is a practical case when
solving the SWE, given that the position of a shoreline usually lies between cell
centers. Thus, cells can be fully wet, partially wet (partially dry) or dry, but
computationally we must define only wet and dry cells (Brocchini and Dodd 2008).
If the numerical flux is computed using the exact Riemann solver, the dry-bed
cases are automatically considered. The HLL Riemann solver previously exposed
also applies to dry-bed conditions if the correct signal speeds SL and SR for the
dry-bed problem are accounted for, as given by Eqs. (9.48) and (9.49). This permits
reasonably well to compute the numerical flux for dry-bed problems (Toro 2001).
To identify a dry bed in the computational domain, consider a small threshold depth
e (Sanders 2001; Khan and Lai 2014). For a generic interface i + 1/2, the following
cases are possible:
(i) If hL [ e and hR e, the numerical flux Fi+1/2 is computed assuming a dry
bed on the right
(ii) If hR [ e and hL e, the numerical flux Fi+1/2 is computed assuming a dry
bed on the left
(iii) If hR e and hL e, the numerical flux Fi+1/2 is set to zero.
After each time loop, a check of computed cell-averaged water depths hki þ 1 is
conducted to identify the formation of new dry-bed conditions within the compu-
tational domain. Two methods are used in practice:
(a) Dry nodes are defined with zero water depth: The following conditions are
implemented in the code
If hki þ 1 e ) Uik þ 1 ¼ 0: ð9:76Þ
If hki þ 1 \0 ) Uik þ 1 ¼ 0; hki þ 1 ¼ 0: ð9:77Þ
In this method, the depth is clearly zero at the dry nodes. Thus, physical functions
divided by h must be also checked in the code to reset the values to zero under
dry-bed conditions, e.g., the function q2/(gh) or the friction slope Sf = n2q2/(h10/3),
with n as Manning’s roughness coefficient.
(b) Dry nodes are not defined with zero water depth: The following conditions are
implemented in the code
If hki þ 1 e ) Uik þ 1 ¼ 0; hki þ 1 ¼ e: ð9:78Þ
This method can generate unphysical fluxes over uneven beds, so that method (a) is
preferred (Khan and Lai 2014). A typical value used is e = 10–16 m.
Care is needed during assignation of U values at the interface of a wet and a dry
cell, e.g., during the tracking of a shoreline. Suppose an interface with a wet cell on
the left and a dry cell on the right, that is, the shoreline is inside the (computa-
tionally) dry cell, which is in reality a partially-filled cell. Now assume water is
static on a slope. At this interface hL 6¼ 0 from the wet cell, but hR = 0 from the dry
cell, thereby generating unphysical numerical flux. Thus, in this event hR, is set
equal to hL and qR ¼ qL to avoid unphysical flux (Bradford and Sanders 2002).
9.5 Source Terms
9.5.1 Splitting Technique
The problem posed now is the solution of the SWE with a source term S = S(U),
e.g.,
@U @F
þ ¼ SðUÞ: ð9:79Þ
@t @x
9.5 Source Terms 371
The friction slope is parametrized for a wide channel with Cf as a friction

coefficient as
U jU j
Sf ¼ Cf ; ð9:80Þ
gh
so that the source term of the SWE is
! !
0 0
S ¼ SðUÞ ¼ @zb ¼ @zb
U 2 U 2 ; ð9:81Þ
gh ghSf gU1 Cf
@x @x U 1 U 1
where U = (U1, U2) = (h, q). This source term is reaction-like, given that terms
with U derivatives are missing. Here, a splitting scheme following Toro (2001) is
presented. An alternative to splitting is up-winding the source terms (Bermudez and
Vazquez-Cendón 1994). Our problem is thus to find the vector U at time level k + 1
including the effect of the source terms, that is,
9
@U @F =
PDEs: þ ¼ SðUÞ
@t @x ) Uki þ 1 xi ; tk þ 1 : ð9:82Þ
Initial cond:: Uðx; tÞ ¼ U k ;
i
In the splitting approach, the problem given by Eq. (9.82) is solved in two con-
secutive steps as follows:
Step 1 Solve the homogeneous part of Eq. (9.82) using the Godunov-type scheme,
9
@U @F = k þ 1
PDEs: þ ¼0
@t @x ) Uadv xi ; t : ð9:83Þ
Initial cond:: Uðx; tÞ ¼ Uki ;
i
In this first step, the effects of advection are accounted for. This step may be
regarded as a predictor part to compute an approximate solution at time level k + 1
overlooking source terms effects. From the viewpoint of open channel hydraulics,
this advection step forces the so-called “pseudo-uniform flow condition” (bed-slope
effects in equilibrium with friction). The solution of the advection step is given by
Dt
Uadv ¼ Uki Fi þ 1=2 Fi1=2 ; ð9:84Þ
i
Dx
which was already described in the preceding sections.

Step 2 Update the solution including the effect of the source terms,
9
dU =
ODEs: ¼ SðUÞ
dt ) Uki þ 1 xi ; tk þ 1 : ð9:85Þ
Initial cond:: Uðx; tÞ ¼ Uadv ;
i
A variety of ODE solvers apply in this step. It may be regarded as a correction step
to deviate the advection solution from pseudo-uniformity. An integral form of the
ODEs is
tZþ Dt
kþ1
Ui ¼ Ui þ
adv
Si ðUÞdt: ð9:86Þ
t
This equation highlights that the vector Si(U) may be evaluated at different states
during the time integration depending on how the integral is discretized. This fact is
important and may affect the quality of the computed solution. Obviously, this
source term Si(U) shall be regarded as a cell-averaged value.
9.5.2 ODE Solvers
There is a great variety of ODE solvers to be used in Eq. (9.86); see, for example,
Hoffman (2001). Here, the most widely employed are summarized.
First-order forward Euler scheme
The integral in Eq. (9.86) is evaluated taking the value of S(U) at a specified time
level, resulting in an explicit scheme. Using the value of S(Uk) yields the not best
possible choice (Toro 2001)
k
Uki þ 1 ¼ Uadv
i þ S Ui Dt: ð9:87Þ
A better approximation is
adv
Uki þ 1 ¼ Uadv
i þ S Ui Dt; ð9:88Þ
where the source terms include the effects of advection.

First-order backward Euler scheme
The integral in Eq. (9.86) is evaluated taking the value of S(U) at the unknown time
level k + 1, resulting in an implicit scheme. The updating formula is then
k þ 1
Uki þ 1 ¼ Uadv
i þ S Ui Dt: ð9:89Þ
Second-order TVD Runge–Kutta scheme

The updating sequence is given by the two explicit steps (Gottlieb and Shu 1998;
Khan and Lai 2014)
ð1Þ adv
Ui ¼ Uadv
1 1 ð1Þ 1 h ð1Þ i
Uki þ 1 ¼ Uadv þ U þ S Ui Dt: ð9:91Þ
2 i 2 i 2
Third-order TVD Runge–Kutta scheme

The updating sequence is given by the three explicit steps (Gottlieb and Shu 1998;
Khan and Lai 2014)
ð1Þ adv
Ui ¼ Uadv
ð2Þ 3 1 ð1Þ 1 h ð1Þ i

Ui ¼ Uadv
i þ Ui þ S Ui Dt; ð9:93Þ
4 4 4
1 2 ð2Þ 2 h ð2Þ i
Uki þ 1 ¼ Uadv
i þ Ui þ S Ui Dt: ð9:94Þ
3 3 3
9.5.3 Well-Balanced Schemes
The discretization applied to the bed-slope source term has an important impact on
the quality of the steady-state solutions generated by the unsteady flow solver, an
Fig. 9.11 Static water over variable topography (zs = const., q = 0)

issue demonstrated as follows. Consider static water over a variable bed profile
zb = zb(x) (Fig. 9.11), for which the conservation laws reduce to
@F
¼ S; ð9:95Þ
@x
or

@ gh2 @zb
¼ gh : ð9:96Þ
@x 2 @x
The integral form of Eq. (9.95) over a cell is

Fi þ 1=2 Fi1=2 DxSi ¼ 0; ð9:97Þ
or,

h2i þ 1=2 h2i1=2 @zb
g ¼ Dx gh : ð9:98Þ
2 @x i
To preserve this identity, the average source term contribution within a cell must be
compatible with the discretization scheme of the numerical fluxes. Otherwise,
artificial waves originating from the numerical scheme may appear, breaking down
static conditions. Topography is defined at cell interfaces, and within a cell, the bed
profile is assumed to vary linearly. The bed source term may be discretized as

@zb hi þ 1=2 þ hi1=2 zbi þ 1=2 zbi1=2
gh ¼g : ð9:99Þ
@x i 2 Dx
The imbalance of numerical fluxes at cell i is given by the hydrostatic forces as

h2i þ 1=2 h2i1=2 hi þ 1=2 þ hi1=2 hi þ 1=2 hi1=2

g ¼g : ð9:100Þ
2 2
Given that the water is static, e.g.,

hð xÞ þ zb ð xÞ ¼ const: ð9:101Þ
Equation (9.100) transforms into

h2i þ 1=2 h2i1=2 hi þ 1=2 þ hi1=2 zbi þ 1=2 zbi1=2
g ¼ g : ð9:102Þ
2 2
Inserting Eqs. (9.99) and (9.102) into Eq. (9.98), the identity is preserved, so the
scheme is said to be well-balanced, or to satisfy the C-property (Bermudez and
Vazquez-Cendón 1994). As noted in this illustrative example, the hydrostatic forces
(gh2/2) contained inside the flux gradient @F=@x are responsible for the potential
imbalance with the bed-slope source term if Eq. (9.99) is not used.
For partially filled cells, e.g. with static water on a slope, the bed jump zbi þ 1=2 zbi1=2
in Eq. (9.99) shall be modified (Brufau et al. 2002). If hi þ 1=2 ¼ 0 then zbi þ 1=2
zbi1=2 is replaced by hi1=2 , whereas for hi1=2 ¼ 0 it is substituted by hi þ 1=2 .
An alternative to circumvent this problem is to reformulate the SWE as follows
@U @F
þ ¼ S; ð9:103Þ
@t @x
where U is the vector of the conserved variables, F the flux vector, and S the source
term vector, given in this case by (Ying et al. 2004)
!
h hU 0
U¼ ; F¼ ; S¼ @zs : ð9:104Þ
hU hU 2 gh ghSf
@x
As noted, the hydrostatic forces are no more contained in F; in the source term,
the bed-slope source term is substituted by a free surface elevation
(zs = zb + h) gradient term. For static water, the integral form of Eq. (9.103)
reduces to

@zs
Dx gh ¼ 0: ð9:105Þ
@x i
Using Eq. (9.101), the scheme is seen to be automatically well-balanced.
9.5.4 Treatment of Friction
Consider the bed-slope source term overlooked temporarily. In this case, and with
f as the Darcy–Weisbach friction factor (Cf = f/8) (White 2009), Eq. (9.85) is
reduced to the scalar ODE, assuming a wide rectangular channel and positive U
dq f
¼ ghSf ¼ U 2 : ð9:106Þ
dt 8
For wet-bed conditions, the forward Euler discretization is acceptable, resulting in

the updating formula
f 2
qki þ 1 ¼ qadv
i Uiadv Dt: ð9:107Þ
8
However, near wet–dry fronts the explicit discretization may generate instabilities,
so that an alternative method is explained here. Using backward Euler discretization
[Eq. (9.89)], Eq. (9.106) yields
f k þ 1 2
qki þ 1 ¼ qadv
i U Dt; ð9:108Þ
8 i
or
qadv f k þ 1 2
Uik þ 1 ¼ i
U Dt: ð9:109Þ
hki þ 1 8hki þ 1 i
Equation (9.109) is a quadratic function solved at each time step to obtain the
cell-averaged velocity Uk+1
i accounting for frictional effects. Note that the friction
force might stop the flow in the extreme; e.g., a flow reversal cannot be numerically
permitted. Thus, the following condition must be verified by numerical
computations
qki þ 1 qadv
i 0: ð9:110Þ
Other flow resistance formulae like Manning’s equation are frequently used in river
flow computations (Henderson 1966; Cunge et al. 1980; Khan and Lai 2014).
A semi-implicit approach is also used in some models.
9.6 One-Sided First-Order Upwind Scheme
In this section, the first-order upwind finite volume method of Ying et al. (2004) is
described. In this model, the SWE are used in the form given by Eqs. (9.103) and
(9.104), so the scheme is well-balanced. The updating formula is given by the
integral equation
Dt
Uki þ 1 ¼ Uki Fi þ 1=2 Fi1=2 þ DtSi : ð9:111Þ
Dx
The numerical flux is evaluated with the one-sided upwind method as (Ying et al.
2004)
2 3
qkiþ j

6 k 27
Fi þ 1=2 ¼6
4 qi þ j 5 :
7 ð9:112Þ
hkiþ j
9.6 One-Sided First-Order Upwind Scheme 377
Here j = 0 if qi and qi+1 > 0, j = 1 if qi and qi+1 < 0, and j = 1/2 for any other case;
subscript i + 1/2 means in that case average values between i and i + 1
cell-averaged values. The water surface gradient cannot be evaluated with a central
difference, given that non-physical jumps are then formed within rarefaction waves
(Ying et al. 2004). Thus, an average between up- and downwind gradients is
formed as (Ying et al. 2004)
!
@zs kþ1
zks;iþþ11j zks;ij
þ1
zks;iþþ1j zks;i1
þ1
þj
gh ¼ ghi w1 þ w2 : ð9:113Þ
@x Dx Dx
Note that the water depths used in Eq. (9.113) are those at time k + 1, after solving
the advection step for the continuity equation. The weighting factors are
Dt Uikþ 1j + Uij

k
Dt Uikþ j + Ui1
k
þj
w1 ¼ 1 and w2 ¼ : ð9:114Þ
Dx 2 Dx 2
The computational sequence to apply the scheme encompasses the following steps:
(1) Start with known cell-averaged values of U at a time level k.
(2) Determine a stable time step Dt using the CFL condition, e.g., typically with
CFL ¼ 0:9, using Eq. (9.29).
(3) Compute the numerical fluxes Fi+1/2 using Eq. (9.112).
(4) Compute the water depths solving the advection step for the continuity
equation
Dt
hki þ 1 ¼ hki qi þ 1=2 qi1=2 : ð9:115Þ
Dx
(5) Evaluate the water surface gradient source term using Eq. (9.113).
(6) Compute the discharge at the new time level including the effects of source
terms as
qki þ 1 ¼ qki
" # !
Dt q2i þ j q2ij kþ1
zks;iþþ11j zks;ij
þ1
zks;iþþ1j zks;i1
þ1
þj
ghi w1 þ w2 Dt
Dx hi þ j hij Dx Dx
f
Uik Uik Dt:
8
ð9:116Þ
(7) Go to step (1), and repeat the cycle until reaching the final time of
computations.
9.7 MUSCL-Hancock Second-Order TVD Scheme
9.7.1 MUSCL Reconstruction
van Leer introduced a second-order version of the Godunov (1959) scheme (van
Leer 1979). The scheme is called MUSCL (Monotonic Upstream Centered
Schemes for Conservation Laws). Second-order accuracy is regained by recon-
structing the solution U(x) within the cells. The idea is to replace data representation
using the constant cell-averaged value Ui by a function U(xi−1/2 < x < xi+1/2) that
approximates the exact solution within a cell. The function used for the recon-
struction must preserve the cell-averaged value regained from the conservative
formula. That is, the latest cell-averaged values available obtained by applying the
time stepping are used in the reconstruction of the solution at the actual time level.
In van Leer’s scheme, the reconstruction of the solution within a cell is linear
(Fig. 9.12a, b), but parabolic approximations also apply (Colella and Woodward
1984). The linear reconstruction of the solution U(x) within a cell gives
DUi
Ui ð xÞ ¼ Uki þ ðx xi Þ : ð9:117Þ
Dx
Note that the slope vector is constant for a given cell, but each conserved variable
will have a different scalar slope contained therein. To compute the slope vector for
the reconstructed solution, DUi =Dx; up- and downstream jumps are first defined as
DUi1=2 ¼ Uki Uki1 ; DUi þ 1=2 ¼ Ukiþ 1 Uki : ð9:118Þ
Taking an average slope, with the averaging coefficient −1 < x < 1, produces
(Toro 2001)
1
DUi ¼ ð1 xÞDUi þ 1=2 þ ð1 þ xÞDUi1=2 : ð9:119Þ
2
A typical value is x = 0. The boundary extrapolated values at each cell face are
thus (Fig. 9.12b)
1 1
Uki1=2 ¼ Uki DUi ; Ukiþ 1=2 ¼ Uki þ DUi : ð9:120Þ
2 2
Thus, the solution of a Riemann problem at a generic interface i + 1/2 is now

conducted based on extrapolated values at the left and right sides of each interface
(Fig. 9.12c)
1 1
ULiþ 1=2 ¼ Uki þ DUi ; URiþ 1=2 ¼ Ukiþ 1 DUi þ 1 : ð9:121Þ
2 2
9.7 MUSCL-Hancock Second-Order TVD Scheme 379
Fig. 9.12 MUSCL

reconstruction. a Linear
approximation within cells,
b definition of extrapolated
values at cell faces [the figure
represents the typical
reconstruction of a component
of the vector U = (U1, U2)T],
and c formation of local
Riemann problem at interface
i + 1/2
Bram van Leer was born on November 26, 1942, at Surabaya, Netherlands East
Indies. He obtained the Ph.D. degree from Leiden State University in 1970. He
made seminal contributions to computational fluid dynamics (CFD) in his 5-part
article series “Towards the ultimate conservative difference scheme” published
from 1972 to 1979, where he extended Godunov’s finite volume scheme to
second order (MUSCL), developed non-oscillatory interpolation using limiters, an
approximate Riemann solver, and discontinuous Galerkin schemes for unsteady
advection. Since joining the University of Michigan’s Aerospace Engineering
Department in 1986, he has worked on convergence acceleration by local pre-
conditioning and multigrid relaxation for Euler and Navier–Stokes problems,
unsteady adaptive grids, space-environment modeling, atmospheric flow model-
ing, extended hydrodynamics for rarefied flows, and discontinuous Galerkin
methods. He retired in 2012. His research interests were in CFD, fluid dynamics,
and numerical analysis.
He obtained in 1996 the College of Engineering Research Award
(University of Michigan), was elected in 1995 AIAA Fellow, was awarded in
1990 and 1992 the NASA Langley Group Achievement Award, in 1992 the
Department of Aerospace Engineering Research Award (University of
Michigan), the Honorary Doctorate from the Vrije Universiteit Brussels in
1990, and the C. J. Kok Prize of Leiden State University in 1978. He is a
Fellow of the American Institute of Aeronautics and Astronautics (AIAA)
and Member of the Society for Industrial and Applied Mathematics (SIAM).
9.7.2 Slope Limiting
The reconstruction process previously exposed produces spurious numerical

oscillations in zones with sharp variations of U, as at a shock wave front (Toro
2009). Gibbs’ phenomenon must be controlled, therefore. This control is achieved
using the so-called total variation diminishing (TVD) schemes. They suppress
Gibbs’ phenomenon by degradation of the scheme to first-order accuracy in zones
with sharp variations of U, while keeping second-order accuracy in smooth portions
of the solution. Spurious oscillations are suppressed by limiting the slope in the
reconstruction stage. Thus, the reconstruction of the solution reads now with
DUi =Dx as the limited slope vector
DUi
Ui ð xÞ ¼ Uki þ ðx xi Þ ;
Dx ð9:122Þ
1 1
ULiþ 1=2 ¼ Uki þ DUi ; URiþ 1=2 ¼ Ukiþ 1 DUi þ 1 :
2 2
The idea is to limit the slopes obtained in the reconstruction stage by applying a
limiter function. A widely used limiting function is the so-called minmod limiter,
which for two numbers a and b is
Minmodða; bÞ ¼ signðaÞmax½0; minðjaj; signðaÞ bÞ: ð9:123Þ
Expanding the function, the possible cases are summarized as:

8
>
<a if jaj\jbj and a b [ 0
Minmodða; bÞ ¼ b if jaj [ jbj and a b [ 0 : ð9:124Þ
>
:
0 if a b \0
The minmod function gives the smallest modulus if all arguments have the
same sign and is zero otherwise. For the MUSCL scheme, the minmod limiter
is thus

DUi ¼ sign DUi þ 1=2 max 0; min DUi þ 1=2 ; sign DUi þ 1=2 DUi1=2 :
ð9:125Þ
The three possible cases are depicted in Fig. 9.13. If the jumps have the same sign
(Fig. 9.13a, b), the solution is monotone. Therefore, the limited jump is nonzero, and
equal to the smaller one in absolute value. If the jumps change their signs, the up- and
downslopes have different signs (Fig. 9.13c), thereby implying a solution U(x) of
certain curvature (oscillation). The propagation of these non-physical undulations is
constrained turning to first-order accuracy the scheme by setting the limited jump to
zero. Thus, the linear reconstruction is not done in these cases. Therefore, the TVD
scheme is adaptive; it decides the regions of the computational domain where the
scheme has to be second-order accurate and those where the linear reconstructions must
Fig. 9.13 Effect of minmod slope limiter. a Limited jump is the upstream jump, b Limited jump
is the downstream jump, and c limited jump is zero [the figure represents a component of the
vector U = (U1, U2)T]
be switched off, thereby reducing to Godunov’s first-order scheme. A family of limiters

is formulated as (Bradford and Sanders 2005)
DUi
(
sign DUi þ 1=2 min min DUi þ 1=2 ; DUi1=2 ; b min DUi þ 1=2 ; DUi1=2 if DUi þ 1=2 DUi1=2 [ 0
¼
0 else:
ð9:126Þ
For b = 2 results the Superbee limiter; for b = 1, it simplifies to the minmod

limiter, whereas b = 0 reduces the scheme to first order. The Superbee limiter is less
diffusive than minmod, thus producing a sharper resolution of shocks. The recon-
structed flow depths at the cell interface shall not become negative, to be assured
prior to compute the numerical flux (Sanders 2001).
9.7.3 Hancock Step
The MUSCL scheme (van Leer 1979) described is second-order accurate in space,
but only first-order accurate in time. It was simplified and improved by Hancock
(1980) while developing the PISCES industrial simulation code (Van Leer 2006).
The method would have remained unnoticed in the manual of this code, but it was
rescued by van Albada et al. (1982) and applied for the simulation of cosmic gas
dynamics. The idea of Hancock’s predictor–corrector scheme is remarkably simple:
A first-order upwind method transforms into second-order time accuracy by
evolving the cell boundary values used to compute the numerical flux, and the
source terms, by half the time step (Sweby 1999). Thus, to regain second-order
accuracy in both time and space, the vector U is expanded using a Taylor series
k þ 1=2
within a cell as follows. The cell interface value evolved a time Dt/2, Ui þ 1=2 ¼
1
U 2 Dx; 12 Dt ; can be expressed as a function of the cell-averaged value at time
k using a Taylor series for two variables as (Sweby 1999)
k k
k þ 1=2 1 @U 1 @U
Ui þ 1=2 ¼ Uki þ Dx þ Dt þ O Dx2 ; Dt2 : ð9:127Þ
2 @x i 2 @t i
Note that the value of U at the cell center at time k is the cell-averaged value, as
assured by the MUSCL linear reconstruction at time level k. The time derivative of
U is from the differential form of the SWE
@U @F
¼ þ S: ð9:128Þ
@t @x
Inserting Eq. (9.128) into Eq. (9.127) produces

k k
k þ 1=2 1 @U 1 @F
Ui þ 1=2 ¼ Uki þ Dx þ Dt þS
2 @x i 2 @x
k k
i
ð9:129Þ
1 @U 1 @F 1
¼ Uki þ Dx Dt þ DtSki :
2 @x i 2 @x i 2
Based on the MUSCL reconstruction, the slope is

k
@U Ukiþ 1=2 Uki1=2
¼ : ð9:130Þ
@x i Dx
Approximating the gradient of fluxes by a first-order finite-difference as

k F Uk
i þ 1=2 F Ui1=2
k
@F
; ð9:131Þ
@x i Dx
and noting that

Ski ¼ S Uki ; ð9:132Þ
the relations needed in Eq. (9.129), namely Eqs. (9.130)–(9.132), are available.
Inserting these, one gets

1 U k
i þ 1=2 U k
1 F Uk
i þ 1=2 F Uk
i1=2 1
k þ 1=2 i1=2
Ui þ 1=2 ¼ Uki þ Dx Dt þ DtS Uki
2 Dx
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} 2 Dx 2
MUSCL reconstruction
Dt h k

i Dt
¼ Ukiþ 1=2 F Ui þ 1=2 F Uki1=2 þ S Uki :
2Dx 2
ð9:133Þ
This is Hancock’s step to regain second-order time accuracy: The MUSCL reconstructed
interface values Ukiþ 1=2 at time k are evolved half the time step to obtain an improved
k þ 1=2
data representation given by Ui þ 1=2 . Looking to an interface rather than to a single cell,
Hancock’s step amounts to evolve the interface values ULi þ 1=2 and URi þ 1=2 to regain second-
L R
order accuracy in time to Ui þ 1=2 and Ui þ 1=2 (Fig. 9.14a). These are given by Hancock’s
scheme from Eq. (9.133) as
Fig. 9.14 Hancock’s

step. a Evolution of boundary
extrapolated value at interface
i + 1/2 half the time step and
b formation of new local
Riemann problem at interface
i + 1/2 with evolved values at
the interface
L Dt h L

i Dt
Ui þ 1=2 ¼ ULi þ 1=2 F Ui þ 1=2 F URi1=2 þ Si ;
2Dx 2 ð9:134Þ
R Dt h L

i Dt
Ui þ 1=2 ¼ Ui þ 1=2
R
F Ui þ 3=2 F URi þ 1=2 þ Si þ 1 ;
2Dx 2
forming a new local Riemann problem. With these evolved boundary extrapolated
L R
variables Ui þ 1=2 and Ui þ 1=2 defining states L and R, the numerical flux is computed
using the HLL approximate Riemann solver (Fig. 9.14b). Once this flux is deter-
mined, the computation of the transient flow is conducted as described previously.
Note that this predictor step considers the source terms. The bed-slope effect is
typically accounted for (Zhou et al. 2001; Aureli et al. 2008). It is evaluated based
on interface data at cell i as [see Eq. (9.99)]

hL R
@zb i þ 1=2 þ hi1=2 zbi þ 1=2 zbi1=2
gh ¼g : ð9:135Þ
@x i 2 Dx
Friction is treated in a final splitting step using an implicit treatment (Aureli et al.
2008) to increase stability in the vicinity of wet–dry interfaces.
9.7.4 Computational Sequence
The computational process is summarized in the following steps:

(1) Start at time level k with the cell-averaged values Ui.
(2) Apply the CFL condition [Eq. (9.29)] to select a stable time step.
(3) Reconstruct the solution using the TVD MUSCL scheme: Get a limited slope
applying the minmod limiter [Eq. (9.125)] to suppress spurious oscillations.
Other limiters are possible.
(4) Compute values at cell interfaces using the limited slopes [Eq. (9.122)].
(5) Evolve the interface values half the time step using Hancock’s method
[Eq. (9.134)].
(6) Solve the Riemann problem at each interface using the evolved interface values
typically with the HLL approximate Riemann solver. Other methods to com-
pute the numerical flux are possible, as the exact Riemann solver.
(7) Apply the conservative equation to resolve the advection step
Dt h
L R

L R

i
Uadv ¼ Uki Fi þ 1=2 Ui þ 1=2 ; Ui þ 1=2 Fi1=2 Ui1=2 ; Ui1=2 : ð9:136Þ
i
Dx
(8) Include the effect of the bed-slope source terms using the values obtained at
time k + 1/2 with Hancock’s method, e.g.,

T
~ i ¼ Uadv 0; Dt g hL
U þ h
R
z bi þ 1=2 z : ð9:137Þ
i i þ 1=2 i1=2 bi1=2
2Dx
Combining Eqs. (9.136) and (9.137) yields (Mingham and Causon 1998; Aureli
et al. 2008)
h

i
~ i ¼ Uk Dt Fi þ 1=2 UL
U ; U
R
F U
L
; U
R
i þ 1=2 i þ 1=2 i1=2
i
Dx i1=2 i1=2

T ð9:138Þ
Dt L R
0; g hi þ 1=2 þ hi1=2 zbi þ 1=2 zbi1=2 :
2Dx
This is only a function of the predicted values at k + 1/2 obtained with Hancock’s
method.
(9) Include the effect of friction using an implicit Euler scheme

T
~ i Dt 0; f U k þ 1 U k þ 1 :
Uki þ 1 ¼ U ð9:139Þ
8 i i
A semi-implicit trapezoidal rule also applies (Aureli et al. 2008).

(10) Go to step (1), and repeat the cycle until reaching the target time.
9.7.5 Surface Gradient Method
The MUSCL reconstruction process of the vector U = (h, q)T (step 3 in the
sequence above) is called depth gradient method (DGM) (Zhou et al. 2001); it may
lead to unphysical discharges over uneven topography; given that for static water,
the reconstruction of variable h will produce jumps in water depths at cell interfaces
and thus a water movement even if the water is initially at rest. Figure 9.15 depicts
the application of DGM to an uneven profile, from where is seen that
hki 6¼ hkiþ 1 6¼ hkiþ 2 : ð9:140Þ
A linear reconstruction yields a local Riemann problem at each cell involving left
and right states
Fig. 9.15 Generation of jump in water depths at cell interfaces by the depth gradient method over
uneven topography
! !
hLiþ 1=2 hRiþ 1=2
ULiþ 1=2 ¼ ; URiþ 1=2 ¼ : ð9:141Þ
qLiþ 1=2 ¼ 0 qRiþ 1=2 ¼ 0
Assume to compute now the first time step of the transient flow, starting with zero
discharge in the system and a horizontal free surface. Reconstruction of the depth
profile yields the momentum equation

2
2
< <
Dt Fi þ 1=2 Fi1=2 hLiþ 1=2 þ hRi1=2 zbi þ 1=2 zbi1=2
qki þ 1 ¼ g gDt 6¼ 0:
Dx 2 2 Dx
ð9:142Þ
In this equation, Fi<þ 1=2 is the momentum function obtained at the interface i + 1/2
solving the Riemann problem formed there. Obviously, the discharge is in general
not zero, indicating the generation of an unphysical discharge. If
hLiþ 1=2 ¼ hRiþ 1=2 ¼ hi þ 1=2 , Eq. (9.142) reduces to

h 2
Dt i þ 1=2 h 2
i1=2 hi þ 1=2 þ hi1=2 zbi þ 1=2 zbi1=2
qki þ 1 ¼ g gDt ¼ 0;
Dx 2 2 Dx
ð9:143Þ
which is zero by virtue of the condition zs = const. Thus, to avoid unphysical

discharge it is necessary to block the formation of jumps in water depths at cell
faces during the reconstruction stage.
Based on this result, an alternative is to reconstruct the free surface elevation zs

and discharge q, and, once done, the water depths are deduced at the intercell faces
(Nujic 1995; Zhou et al. 2001). The reconstructed interface values are then
1 1
Qki1=2 ¼ Qki DQi ; Qkiþ 1=2 ¼ Qki þ DQi ; ð9:144Þ
2 2
where

zs zb þ h
Q¼ ¼ : ð9:145Þ
q q
Once the interface values are determined with the limited slopes, the water depths at
the cell faces are
hki1=2 ¼ ðzs Þki1=2 ðzb Þi1=2 ; hkiþ 1=2 ¼ ðzs Þkiþ 1=2 ðzb Þi þ 1=2 : ð9:146Þ
With this technique, the identity given by Eq. (9.143) is preserved, and thus,
unphysical discharge is not generated, given that the flow depths at the cell faces are
unique for static conditions. This technique is referred to as the surface gradient
method (SGM). This reconstruction, however, can lead to small, even negative
water depths near dry–wet fronts (Aureli et al. 2008). DGM, in contrast, is more
robust for bore front tracking. Aureli et al. (2008) propose a hybrid reconstruction
combining the good capabilities of the SGM and DGM, called weighted
surface-depth gradient method.
9.8 Boundary and Initial Conditions
9.8.1 Boundary Conditions
The computational domain is typically divided into N cells (Fig. 9.16a) of width
Δx, e.g., with the initial data
Uki ; i ¼ 1; 2; . . .. . .. . .. . .; N 1; N: ð9:147Þ
The conservative Eq. (9.7) is used to evolve the vector U as
Dt
Uki þ 1 ¼ Uki Fi þ 1=2 Fi1=2 þ DtSi ; i ¼ 1; 2; . . .. . .. . .. . .; N 1; N:
Dx
ð9:148Þ
On inspecting Eq. (9.148), two issues are noted:

Fig. 9.16 Boundary conditions. a Definition of ghost cells in the computational domain, ghost
cells in (b) weir flow, c gate flow
(i) The updating equation for cell i = 1 is
Dt
Uk1 þ 1 ¼ Uk1 F3=2 F1=2 þ DtS1 : ð9:149Þ
Dx
In this formula, the numerical flux entering into the left face of cell i = 1, namely
F1/2, is needed. However, there is no initial data defined to form a Riemann problem
at this interface and thus compute the numerical flux.
(ii) The updating equation for cell i = N is
Dt
UkNþ 1 ¼ UkN FN þ 1=2 FN1=2 þ DtSN : ð9:150Þ
Dx
In this formula, the numerical flux leaving the right face of cell i = N, namely FN+1/2,
is needed. However, there is no initial data defined to form a Riemann problem at
this interface and thus compute the numerical flux.
From the above arguments, it is necessary to incorporate two additional cells at
the left and right ends of the computational domain, called ghost cells (LeVeque
2002; Toro 2009) (Fig. 9.16a). Physical boundary conditions are transmitted to the
computational model through these cells, which in turn are used to form two
additional Riemann problems needed to compute the missing numerical fluxes and
thus evolve in time cells i = 1 and N. Here, the simplest way of implementing
boundary conditions is exposed using ghost cells with a zero-order extrapolation
from the interior solution. Care should be taken to ensure that spurious waves are
not generated when implementing boundary conditions (LeVeque 2002).
Subcritical inflow section
At a subcritical inflow section, one boundary condition must be prescribed, as
known from the method of characteristics, explained in Chap. 5 (Henderson 1966;
Jain 2001). This is typically the inlet discharge for weir flow (Fig. 9.16b). The other
variable needed at the ghost cell is obtained approximately by extrapolation from
the interior solution (LeVeque 2002). The vector U at the ghost cell i = 0 is thus
given by
qko þ 1 ¼ qinlet ;
ð9:151Þ
hko þ 1 ¼ hk1 þ 1 :
For a more rigorous computation of hko þ 1 the characteristic C - shall be used (see
Chaps. 5 and 7; Eq. 7.37).
Subcritical outflow section
At a supercritical outflow section, boundary conditions are not needed (see Chap. 5).
The other variable required at the ghost cell is obtained approximately by extrapolation
from the interior solution (LeVeque 2002). This is typical for supercritical flow at the
tailwater portion of weir flow (Fig. 9.16b). The vector U at the ghost cell i = N + 1 is
thus given by
qkNþþ11 ¼ qkNþ 1 ;
ð9:152Þ
hkNþþ11 ¼ houtlet :
For a more rigorous computation of qkNþþ11 the characteristic C + shall be used (see
Chaps. 5 and 7; Eq. 7.39).
Supercritical inflow section

At a supercritical inflow section, two boundary conditions must be prescribed (see
Chap. 5). This is typical for supercritical flow issued from a sluice (Fig. 9.16c). The
vector U at the ghost cell i = 0 is thus given by
qko þ 1 ¼ qinlet ;
ð9:153Þ
hko þ 1 ¼ hinlet :
Here, inlet conditions refer to the vena contracta conditions.

Supercritical outflow section
At a supercritical outflow section, boundary conditions are not needed (see Chap. 5).
The two variables required at the ghost cell are obtained by extrapolation from
the interior solution (LeVeque 2002). This is typical for supercritical flow at the
tailwater portion of weir flow (Fig. 9.16b). The vector U at the ghost cell i = N + 1
is thus given by
ð9:154Þ
hkNþþ11 ¼ hkNþ 1 :
After application of the conservative equation in a given time loop, the following
data are generated:
Uki þ 1 ; i ¼ 1; 2; . . .. . .. . .; N 1; N: ð9:155Þ
Once this step finished, U at the ghost cells is determined depending on the type
of flow section as exposed above and summarized here as:
k þ 1
h
Uko þ 1 ¼
q o

one boundary condition; one value extrapolated from i ¼ 1 if F1 \1
¼
two boundary conditions if F1 [ 1;
k þ 1
h
UkNþþ11 ¼
q N þ1

one boundary condition; one value extrapolated from i ¼ N if FN \1
¼
two values extrapolated from i ¼ N if FN [ 1:
ð9:156Þ
9.8.2 Initial Conditions
The initial conditions are the values of vector U at the computational cells at the
initial instant at which transient computations start, e.g.,
0
h
U0i ¼ ; i ¼ 1; 2; . . .. . .. . .. . .; N 1; N: ð9:157Þ
q i
If the water is initially static over a variable bottom profile of constant elevation zs,
the vector is
0
h zs zbi
U0i ¼ ¼ ; i ¼ 1; 2; . . .. . .. . .; N 1; N: ð9:158Þ
q i 0
This is a wet-bed initial vector. For dry-bed portions of the computational domain,
the vector reads
0
h 0
U0i ¼ ¼ ; i ¼ 1; 2; . . .. . .. . .; N 1; N: ð9:159Þ
q i 0
The water may be also under steady-state conditions, once changes in boundary
conditions are introduced. Then, the initial discharge is a constant qo in the absence
of sources/sinks, e.g., infiltration, side weirs, or precipitation. The water depth
profile is varied, ho = ho(x), and it has to be determined with the steady-state
techniques of Chaps. 3 and 4, or taken from the steady asymptotic state of a
previous unsteady flow computation. The initial vector then reads
0
h ho ¼ ho ð x Þ
U0i ¼ ¼ ; i ¼ 1; 2; . . .. . .. . .. . .; N 1; N: ð9:160Þ
q i qo ¼ const:
9.9.1 Ideal Dam Break Waves
The MUSCL-Hancock scheme using the HLL approximate Riemann solver was
coded and implemented in the file DamBreakWave.xls available in Chap. 12. The
code can be run with friction, and the selected value of b in the limiter [Eq. (9.126)]
fixes the use of minmod (b = 1), Superbee (b = 2) or changes the second-order
scheme to the Godunov first-order scheme if b = 0. In this section, ideal fluid flow
computations in a horizontal channel using the minmod slope limiter are conducted.
If the limiter is deactivated inside the code, spurious oscillations remain unsup-
pressed contaminating numerical solutions.
A subcritical dam break wave of depth ratio hd/hu = 0.4 and upstream water
depth hu = 1 m is considered in Fig. 9.17a, b. Numerical simulations were con-
ducted using CFL ¼ 0:9 and Dx = 0.1 m, with computational results displayed at
time t = 3 s. In the same figure, the exact solution of Stoker (1957), detailed in
Chap. 6, is plotted. The MUSCL-Hancock scheme is shown to produce an excellent
oscillation-free solution of the wave profiles, including the correct shock strength
and position. A most stringent test case is generated in Figs. 9.17c, d using a depth
ratio hd/hu = 0.01, hu = 1 m and identical values of computational parameters.
Critical flow establishes at the dam axis x = 0 in a point of the rarefaction wave. As
depicted in the comparison of the numerical solution with the exact results, the
MUSCL-Hancock scheme produces a highly accurate resolution of transcritical
flow near the dam axis. Therefore, the scheme is entropy preserving and unphysical
jumps are avoided. Finally, a dry-bed dam break wave is considered in Figs. 9.17e, f.
Dry beds were handled in the numerical model using Eqs. (9.76) and (9.77), and
the computational results show good agreement with Ritter’s analytical solution
(Ritter 1892). Note that the dry front in Fig. 9.16e is faster than the shock fronts in
Figs. 9.16a, c. A more detailed study of numerical schemes applied to the dam break
problem is available (Zoppou and Roberts 2003).
9.9.2 Dam Break Waves with Friction
The MUSCL-Hancock scheme implemented in the file DamBreakWave.xls permits

simulation of real fluid flows based on Manning’s equation for the friction slope
U2
Sf ¼ n2 4=3
; ð9:161Þ
Rh
where n is Manning’s roughness coefficient and Rh the hydraulic radius. For an

explicit treatment of friction, the updating after the advection step is conducted
using
!adv
adv U2
qki þ 1 ¼ qadv
i ghSf i
Dt ¼ qadv
i gn 2
h 4=3
Dt: ð9:162Þ
Rh i
For an implicit treatment, the updating equation is

!k þ 1
k þ 1 U2
qki þ 1 ¼ qadv
i ghSf i
Dt ¼ qadv
i gn 2
h 4=3
Dt; ð9:163Þ
Rh i
Fig. 9.17 Ideal dam break wave simulations: comparison of numerical simulations using the
MUSCL-Hancock scheme with exact solutions for a, b subcritical wet-bed test, c, d transcritical
wet-bed test, e, f dry-bed test
or
k þ 1 2
qadv 2 Ui
Uik þ 1 ¼ i
gn k þ 1 4=3 Dt: ð9:164Þ
hki þ 1 R hi
The experimental data of Ozmen-Cagatay and Kocaman (2010) at various

normalized times T = t(g/ho)1/2 starting at abrupt gate removal are considered in
Fig. 9.18 for a subcritical dam break wave test (hd/ho > 0.138) in a horizontal
channel. Their upstream water depth was ho = 0.25, the flume width 0.3 m, and the
downstream water depth for this series hd = 0.1 m. Simulations were conducted
with the MUSCL-Hancock scheme using CFL ¼ 0:9 and Dx = 0.05 m. Manning’s
roughness coefficient was set to n = 0.005 m−1/3s, with an implicit treatment of the

numerical simulations using
the MUSCL-Hancock scheme
with experimental data
(Ozmen-Cagatay and
Kocaman 2010) for a
subcritical dam break wave
friction source term and the minmod slope limiter. During initiation of motion, the
wave profile is affected by non-hydrostatic flow effects (Castro-Orgaz and Chanson
2017), so that precise agreement of experiments with simulations based on the SWE
is not expected. The overall position of the surge front is reasonably predicted, but
the detailed wave shape is affected by turbulence, a feature not accounted for in the
mathematical model. Experiments (Ozmen-Cagatay and Kocaman 2010) and
numerical simulations using the MUSCL-Hancock scheme for a transcritical dam
break wave test (hd/ho < 0.138) are considered in Fig. 9.19, with hd = 0.025 m. For
this test, the position of the surge is reasonably well predicted, but the effects of
turbulence are strong in the supercritical portion of the wave profile. A dry-bed dam
break wave is examined in Fig. 9.20, showing a reasonable agreement of numerical
predictions with experimental data by Ozmen-Cagatay and Kocaman (2010). The
shape of the experimental curves of Fig. 9.20 is similar to those previously mea-
sured by Dressler (1954) or Lauber (1997).
The experiments by Schoklitsch (1917) for a dam break wave in a dry, rect-
angular, horizontal flume are considered as an additional test case. The flume was
0.093 m wide, 0.08 m high, and 20 m long. The dam was located at x = 10 m, and
the removal was considered instantaneous. The tailwater flume portion was initially
dry, and the water depth at the dam 0.074 m. Experimental data recorded by
Schoklitsch (1917) for two times after dam removal, namely t = 3.75 s and
t = 9.4 s, are plotted in Fig. 9.21a, b, respectively. The predictions using the
MUSCL-Hancock scheme were conducted using CFL ¼ 0:9, Dx = 0.1 m, and
n = 0.008 m−1/3s. A comparison of simulations and data shows good agreement.
The effect of flow resistance in this dataset is significant for the wave front, as
observed from the comparison of the numerical simulation with Ritter’s solution
plotted in the same figure.
9.9.3 Positive Surge with Friction
Favre (1935) conducted measurements of a subcritical surge propagating in a

horizontal flume 0.42 m wide with initially static water of depth 0.205 m. The surge
was generated by setting a discharge increase at the inlet section of 0.028 m3/s.
Manning’s coefficient for this flume was n = 0.01 m−1/3s (Favre 1935; Terzidis and
Strelkoff 1970). The time needed to switch on the pumps and generate the discharge
pulse into the flume was computationally accounted for as a time lag of 1 s, given that
the generation of the discharge pulse is instantaneous in the mathematical model, as
explained in Chap. 7. The predictions using the MUSCL-Hancock scheme were
conducted using CFL ¼ 0:9, Dx = 0.25 m and an explicit treatment of the friction
source term. This test is available in the file FavreWave.xls in Chap. 12.
A comparison of simulations and test data in Fig. 9.22 shows good agreement.
Comparing wave profiles at two different instants of time, the absolute velocity of
the surge predicted by the MUSCL-Hancock scheme is in excellent agreement with
that indicated by the data.

(Ozmen-Cagatay and
Kocaman 2010) for a

(Ozmen-Cagatay and
Kocaman 2010) for a dry-bed
dam break wave

with experimental data of
Schoklitsch (1917) for a
dry-bed dam break wave and
a t = 3.75 s, b t = 9.4 s
9.9.4 Hydraulic Jump Beyond a Sluice Gate
Gharangik and Chaudhry (1991) conducted measurements of steady hydraulic

jumps in a horizontal flume 0.46 m wide and 14 m long. A supercritical profile was
initially generated in the flume using a gate at the inlet, and the hydraulic jump was
thereafter generated raising the tailwater level with another gate (Fig. 9.16c).
Consider a test conducted for an inflow Froude number F1 ¼ 7: The upstream
boundary section was located at x = 0.305 m from the gate, and the depth and unit
discharge there resulting from experimentation are h = 0.031 m and q = 0.118 m2/s.
The tailwater level was set to h = 0.265 m. This test is available in the file
Hydraulicjump_gate.xls of Chap. 12. The predictions using the MUSCL-Hancock
scheme were conducted using CFL ¼ 0:5; Dx = 0.05 m and an explicit treatment of
the friction source term with n = 0.008 m−1/3s. The initial condition was a static
layer of 0.1 m along the flume. Upstream boundary conditions were then settled,
and, after 15 s of routing, the tailwater level was raised instantaneously to 0.265 m.
The model was run up to t = 1500 s to ensure steady flow conditions.
A comparison of simulations and data in Fig. 9.23a shows good agreement of the
steady wave profile with the exact result obtained by solving the steady-state flow
equations of Chap. 4. The numerical model confines the jump to three computa-
tional cells only, but ideally this jump length should be zero. The numerical dis-
charge of Fig. 9.23b attains a constant value, except just at the jump, where a peak

with experimental data (Favre
1935) for a positive surge
is detected, given the use of Eq. (9.12). This is purely a numerical issue common to
many numerical schemes (Khan and Lai 2014). A comparison of the water surface
profile with experimental data by Gharangik and Chaudhry (1991) in Fig. 9.23c
shows good agreement, although the finite length of the jump produced by tur-
bulence is not a feature tractable with the numerical model of the SWE.

numerical simulation using
with exact and experimental
results (Gharangik and
Chaudhry 1991) for a steady
hydraulic jump
9.9.5 Flow over Round-Crested Weirs
An important transcritical open channel flow feature is the passage from sub-
(F\1Þ to supercritical (F [ 1) flow over a round-crested weir (Fig. 9.16b). Here,
this steady transcritical flow is simulated using the MUSCL-Hancock scheme
assuming ideal fluid flow. An initial steady free surface profile over the weir, for
which qo = const. and ho = ho(x) are known, must be prescribed to initiate unsteady
computations [Eq. (9.160)]. Assuming Sf = 0, Eq. (9.1) reduces for steady flow to
(Henderson 1966)
dh @zb
Sf @zb
¼ @x 2 ¼ @x
2 : ð9:165Þ
dx 1F 1 gh3
q
In this test, Eq. (9.165) was used to produce an initial free surface profile over the
weir. The profile was numerically computed using the fourth-order Runge–Kutta
method (Chaudhry 2008), as described in Chaps. 3 and 4. Computations started at the
crest section where the flow is critical (F ¼ 1Þ; that is, h = hc = (q2/g)1/3. If F ¼ 1;
then Eq. (9.165) must equal the indeterminate identity dh/dx = 0/0. This singularity
is removed by applying L’Hospital’s rule to Eq. (9.165), resulting in (see Chap. 4)
1=2
dh hc @ 2 z b
¼ : ð9:166Þ
dx c 3 @x2
At the crest section, Eq. (9.166) was implemented in the Runge–Kutta solver, and
the corresponding sub- and supercritical branches of the water surface profile were
computed in the up- and downstream directions, respectively.
For unsteady transcritical flow over a weir, one boundary condition must be
prescribed at the subcritical section on the upstream weir side, whereas no boundary
conditions need to be prescribed at the supercritical outlet section. The inlet
boundary condition is given by an instantaneous rise in the discharge, which is kept
constant during all transient flow. Unknown values of conserved variables at
boundary sections are then computed using ghost cells by extrapolation of values at
adjacent interior cells (LeVeque 2002), as previously explained. The steady water
surface profile over a weir of bed shape zb = 0.2 − 0.01x2 (m) was computed using
the MUSCL-Hancock method with the HLL approximate Riemann solver and the
DGM for reconstruction, for a target discharge of q = 0.2 m2/s. It was settled
instantaneously at the inlet, once transient computations were initiated, and kept
constant during all the routing. The initial free surface profile was computed for the
steady discharge qo = 0.05 m2/s.
This particular weir is widely used to test unsteady numerical models (i.e., Zhou
et al. 2001; Ying et al. 2004; Castro-Orgaz and Chanson 2016). In this test case,
Δx = 0.05 m and CFL ¼ 0:9 were used. The code is available in the file
Weir_DGM_parabolic.xls. In the code Weir_DGM_parabolic_MC.xls, the
upstream water depth is determined by using a backward characteristic, showing
variations at initiation of motion only (t < 0.05 s). Some snapshots of the unsteady
flow motion are presented in Fig. 9.24, from where it is observed how a shock wave
passes over the weir (Fig. 9.24a–c). Once it disappears, e.g., at t = 3 s (Fig. 9.24d),
the unsteady free surface flow profile is smooth and gradually adjusts to the dise-
quilibrium in discharge along the weir profile. The results of Fig. 9.24e involve a
simulation time of t = 350 s to ensure that steady-state conditions are reached. The
steady water surface profiles computed using Eqs. (9.165) and (9.166) are presented
in the same figure (labeled as exact solution), showing the excellent agreement of
the finite volume computations with these results. Note further that the discharge is
well conserved by the unsteady flow model. The same computations conducted
Fig. 9.24 Unsteady transcritical flow using the MUSCL-Hancock scheme: evolution of bore
passage over a weir profile and comparison of steady-state solution with exact results
using the one-sided upwind finite volume method yield results almost identical
(Castro-Orgaz and Chanson 2016) (the two profiles deviate only in the third digit).
This numerical model is implemented in the file Weir_upwind.xls of Chap. 12.
The code Weir_dry_SGM-DGM_MC.xls permits to simulate this steady flow and
allows for flooding over dry terrain.
Figure 9.25 contains the experimental data of Sivakumaran et al. (1983) for a
Gaussian hump of profile zb = 20exp[− 0.5(x/24)2] (cm) for two test cases. This
weir test is implemented in the file Weir_DGM_Sivakumaran.xls in Chap. 12. The
computed solution using the MUSCL-Hancock scheme is presented for both cases
and compared in Fig. 9.25a, c with the exact steady solution obtainable using
Eqs. (9.165) and (9.166). Steady solutions obtained with the MUSCL-Hancock
Fig. 9.25 Transcritical flow over Gaussian weir profile using the MUSCL-Hancock scheme.
Comparison of steady-state solutions with exact results and experimental data (Sivakumaran et al.
1983) for q = a, b 0.111 m2/s, c, d 0.0359 m2/s
scheme are in excellent agreement with exact results (Fig. 9.25a, c). The departure
between simulations and experiments for the test case of Fig. 9.25b (q = 0.111 m2/s)
indicates that the effect of the vertical acceleration as the flow passes from sub- to
supercritical is notable, so that the solution of the SWE does not fully agree with
physical experiments due to non-hydrostatic effects. For the test case of Fig. 9.25d
(q = 0.0359 m2/s), deviations between numerical results and experiments are small.
The SWE produce realistic free surface profile solutions across the critical depth
using shock-capturing numerical methods. The computation of a steady flow profile
using an unsteady flow computation produces a solution that automatically crosses
the critical depth. This unsteady flow computation is performed without any further
special treatment at the critical point, as the unsteady computation does not suffer
from any mathematical indetermination. However, the steady gradually-varied flow
equation has a mathematical indetermination at critical flow conditions resolved
with L’Hopital’s rule. The unsteady computation produces a singular point at the
weir crest automatically as the steady state is asymptotically reached (Castro-Orgaz
and Chanson 2016).
9.9.6 Dam Break Wave Evolution over a Bottom Sill
Consider the experimental tests of Ozmen-Cagatay and Kocaman (2011) for a dam
break wave propagated over a trapezoidal sill. The upstream water depth was
ho = 0.25 m, the flume width was 0.3 m, and the downstream channel was dry.
A trapezoidal bottom sill 0.075 m high, of 0.3 m crest length, with up- and
downstream slopes of 7.5/35, was inserted 1.53 m downstream of the gate used to
simulate the dam section. The upstream flume portion was used to store the water
and simulate a reservoir, over a length of 4.65 m. This test was modeled with the
one-sided first-order finite volume method (Ying et al. 2004). Simulations were
conducted using CFL ¼ 0:9 and Dx = 0.01 m assuming ideal fluid flow. The
downstream boundary conditions were modeled as transmissive boundary condi-
tions (Toro 2001), e.g.,
ð9:167Þ
hkNþþ11 ¼ hkNþ 1 :
These are also the conditions used to model supercritical flow at an outlet. The
upstream portion of the computational domain is a solid wall, which must be
modeled using reflective boundary conditions (Toro 2001)
qko þ 1 ¼ qk1 þ 1 ;
ð9:168Þ
hko þ 1 ¼ hk1 þ 1 :
The physical effect of this condition is to represent what actually happens at a solid
wall: There is no mass flow, and there is a hydrostatic thrust acting on the wall.
Instead of using Eq. (9.168) it is possible to fix the numerical flux directly. The
computational model is available in the code of the file “Dambreak_sill.xls”,
Chap. 12. The evolution of this flow is observed in Fig. 9.26, showing the initial
run-up of a supercritical current over the trapezoidal sill, and its subsequent
drowning with wave breaking and bore propagation in the upstream direction. The
experimental data of Ozmen-Cagatay and Kocaman (2011) at different normalized
times T = t(g/ho)1/2 are considered in Fig. 9.27, where computational results at the
same times are also included. Note that the SWE produce an excellent represen-
tation of the supercritical flow passage over the trapezoidal sill and adequately
represent wave breaking and formation of the bore propagating in the upstream
direction at later times. The precise shape of the upstream bore is not well predicted,
given that the SWE are unable to model the finite bore length due to turbulence, but
the overall position of the bore is in good agreement with computations. The
propagation of a supercritical flow over a sill and subsequent wave breaking and
propagation of an upstream bore is a beautiful hydraulic feature (Fig. 9.28) to be
handled with satisfactory results using the SWE.
Fig. 9.26 Evolution of a dam

break wave over a trapezoidal
bottom sill computed with the
one-sided first-order finite
volume scheme (Ying et al.
2004), for the experimental
setup of Ozmen-Cagatay and
Kocaman (2011)
Fig. 9.27 Evolution of a dam

break wave over a trapezoidal
bottom sill computed with the
one-sided first-order finite
volume scheme (Ying et al.
2004) for the experimental
setup of Ozmen-Cagatay and
Kocaman (2011): comparison
of experiments and
simulations
Fig. 9.28 Supercritical flow over a sill resulting in wave breaking and a surge propagating in the
upstream direction (flow is from left to right) [Taken from movie Fluid motion in a gravitational
field, by Rouse (1961), IIHR-Hydroscience & Engineering, The University of Iowa]
9.9.7 Solitary Wave Run-Up on a Plane Beach
Synolakis (1986) conducted a series of laboratory experiments for incident solitary

waves on a plane beach of slope 1:19.85. The crest of the initial solitary wave
solution is placed at half a wavelength from the toe of the plane beach (located at
x = 0 m), which is approximately given by
1=2 rffiffiffiffiffiffiffiffiffi!
4ho 1
L¼ arcosh ; ð9:169Þ
3H 0:05
where ho = 1 m is the still water depth and H the solitary wave amplitude. A test for
non-breaking solitary wave run-up with H/ho = 0.04 is presented in Fig. 9.29 to
check the predictions of the SWE. The initial condition is given by the solitary
wave solution
" 1=2 #
3H
gðx; 0Þ ¼ Hsech2 ð x þ LÞ ; ð9:170Þ
4h3o
Fig. 9.29 Non-breaking solitary wave run-up on a plane beach for H/ho = 0.04 at various
normalized times T = t(g/ho)1/2
1=2
g
Uðx; 0Þ ¼ gðx; 0Þ ; ð9:171Þ
ho
where η = h – ho is the free surface displacement. Solutions were obtained with the
MUSCL-Hancock scheme using an implicit treatment of the friction effects.
Computations were conducted using CFL ¼ 0:4 and Dx = 0.1 m. The depth gra-
dient method was used for reconstruction of the solution, along with a
wetting-drying procedure to track the position of the shoreline. To simulate the bed
roughness effects during both the run-up and the drawdown, a Manning’s coeffi-
cient n = 0.01 m1/3/s was used, assuming wide channel conditions, the scheme is
implemented in the file solitarywaverunup.xls in Chap. 12. Simulations at different
normalized times T = t(g/ho)1/2 are considered in Fig. 9.29, where the experimental
data of Synolakis (1986) at identical times are also included. The SWE produce an
excellent representation of the wave run-up and drawdown features, including the
formation of the hydraulic jump.
A test for the breaking solitary wave run-up with H/ho = 0.3 is presented in
Fig. 9.30, where the predictions of the SWE are compared with experimental data
of Synolakis (1986). The numerical solution disagrees with the experimental data at
T = 10 and 15, given that non-hydrostatic effects are significant, not accounted for
by the SWE. Wave breaking is observed at T = 20 and 25. Once the wave is
broken, the SWE satisfactorily reproduce the run-up characteristics. At T = 30, 35,
and 40, the SWE predictions are in good agreement with the experimental data
during the run-up process. From T = 45 to 60, a moving hydraulic jump is pro-
gressively formed, linked to the drawdown process. Overall, the bore propagation is
satisfactorily predicted by the SWE model. Further research on this topic was
presented by Hafsteinsson et al. (2017). A sequence of photographs during a model
test is shown in Fig. 9.31, where the run-up flow front and the formation of a
moving hydraulic jump during the drawdown are observed.
As highlighted in this chapter, the finite volume solutions of the 1D SWE
produce results in conformity with exact and experimental results for a variety of
flows, including cases with subcritical, transcritical, and supercritical conditions,
the existence of wet–dry interfaces, uneven topography, steady conditions, and
propagation of bores.
Fig. 9.30 Breaking solitary wave run-up on a plane beach for H/ho = 0.3 at various normalized
times T = t(g/ho)1/2
Fig. 9.31 Model tests of non-breaking solitary wave run-up on slope (VAW photographs)
References
Aureli, F., Maranzoni, A., Mignosa, P., & Ziveri, C. (2008). A weighted surface-depth gradient
method for the numerical integration of the 2D shallow water equations with topography.
Advances in Water Resources, 31(7), 962–974.
Bermudez, A., & Vazquez-Cendón, M. E. (1994). Upwind methods for hyperbolic conservation
laws with source terms. Computers & Fluids, 23(8), 1049–1071.
Bradford, S. F., & Sanders, B. F. (2002). Finite-volume model for shallow-water flooding of
arbitrary topography. Journal of Hydraulic Engineering, 128(3), 289–298.
Bradford, S. F., & Sanders, B. F. (2005). Performance of high-resolution, nonlevel bed,
shallow-water models. Journal of Engineering Mechanics, 131(10), 1073–1081.
Brocchini, M., & Dodd, N. (2008). Nonlinear shallow water equation modeling for coastal
engineering. Journal of Waterway, Port, Coastal, and Ocean Engineering, 134(2), 104–120.
References 413
Brufau, P., Vázquez‐Cendón, M. E., & García‐Navarro, P. (2002). A numerical model for the
flooding and drying of irregular domains. International Journal for Numerical Methods in
Fluids, 39(3), 247–275.
unsteady open channel flow. Journal of Irrigation and Drainage Engineering, 142(1),
04015030.
Castro-Orgaz, O., & Chanson, H. (2017). Ritter’s dry-bed dam-break flows: Positive and negative
wave dynamics. Environmental Fluid Mechanics, 17(4), 665–694.
Colella, P., & Woodward, P. (1984). The piecewise-parabolic method (PPM) for gas dynamical
simulations. Journal of Computational Physics, 54(1), 174–201.
Cunge, J. A. (1975). Rapidly varying flow in power and pumping canals. In K. Mahmood & V.
Yevjevich (Eds.), Unsteady flow in open channels, 14, 539–586. Fort Collins, CO, USA: Water
Resources Publications.
Dressler, R. (1954). Comparison of theories and experiments for the hydraulic dam-break wave. In
Proceedings of International of Association of Scientific Hydrology, 3(38), 319–328. Rome,
Italy: Assemblée Générale.
découverts [Theoretical and experimental study of travelling surges in open channels]. Dunod,
Paris, France (in French).
Glaister, P. (1987). Difference schemes for the shallow water equations (Numerical Analysis
Report 9/87). UK: Department of Mathematics, University of Reading.
Godunov, S. K. (1959). A difference method for numerical calculation of discontinuous solutions
of the equations of hydrodynamics. Matematicheskii Sbornik, 47(3), 271–306 (in Russian).
Gottlieb, S., & Shu, C. W. (1998). Total variation diminishing Runge-Kutta schemes. Mathematics
of Computation, 67(221), 73–85.
Hafsteinsson, H. J., Evers, F. M., & Hager, W. H. (2017). Solitary wave run-up: wave breaking
and bore propagation. Journal of Hydraulic Research, 55(6), 787–798.
Hancock, S. L. (1980). PISCES industrial simulation code manual. Physics International.
Harten, A. (1983). High resolution schemes for hyperbolic conservation laws. Journal of
Computational Physics, 49(3), 357–393.
Harten, A., Lax, P., & van Leer, B. (1983). On upstream differencing and Godunov-type scheme
for hyperbolic conservation laws. Society for Industrial and Applied Mathematics, 25(1), 35–
61.
Hirsch, C. (1988). Numerical computation of internal and external flows, 1: Fundamentals of
numerical discretization. Chichester, England: Wiley.
Hirsch, C. (1990). Numerical computation of internal and external flows, 2: Computational
methods for inviscid and viscous flows. Chichester, England: Wiley.
Marcel Dekker.
Karni, S. (2011). Nonlinear hyperbolic conservation laws-a brief informal introduction. Lecture
notes on numerical methods for hyperbolic equations: short book course. London: Taylor and
Francis.
Khan, A. A., & Lai, W. (2014). Modeling shallow water flows using the discontinuous Galerkin
method. Taylor and Francis, New York: CRC Press.
Lauber, G. (1997). Experimente zur Talsperrenbruchwelle im glatten geneigten Rechteckkanal
[Experiments to the dam break wave in smooth sloping rectangular channel]. Ph.D. thesis, ETH
Zurich, Zürich, Switzerland (in German).
Lax, P. (1954). Weak solutions of nonlinear hyperbolic equations and their numerical
computation. Communications on Pure and Applied Mathematics, 7, 159–193.
Lax, P. D., & Wendroff, B. (1960). Systems of conservation laws. Communications in Pure and
Applied Mathematics, 13(2), 217–237.
LeVeque, R. J. (1992). Numerical methods for conservation laws. Basel, Switzerland: Birkhäuser.
University Press.
Mingham, C. G., & Causon, D. M. (1998). High-resolution finite-volume method for shallow
water flows. Journal of Hydraulic Engineering, 124(6), 605–614.
Nujic, M. (1995). Efficient implementation of non-oscillatory schemes for the computation of
free-surface flows. Journal of Hydraulic Research, 33(1), 100–111.
Ozmen-Cagatay, H., & Kocaman, S. (2011). Dam-break flow in the presence of obstacle:
Experiment and CFD simulation. Engineering Applications of Computational Fluid
Mechanics, 5(4), 541–552.
Ritter, A. (1892). Die Fortpflanzung von Wasserwellen [Propagation of water waves]. Zeitschrift
Verein Deutscher Ingenieure, 36(2), 947–954 (in German).
Roache, P. J. (1972). Computational fluid dynamics. Albuquerque NM: Hermosa Publishers.
Roe, P. L. (1981). Approximate Riemann solvers, parameter vectors, and difference schemes.
Journal of Computational Physics, 43(2), 357–372.
Rouse, H. (1961). Fluid motion in a gravitational field. IIHR film, The University of Iowa, Iowa.
Sanders, B. F. (2001). High-resolution and non-oscillatory solution of the St. Venant equations in
non-rectangular and non-prismatic channels. Journal of Hydraulic Research, 39(3), 321–330.
Schoklitsch, A. (1917). Über Dammbruchwellen [On dam break waves]. Kaiserliche Akademie
der Wissenschaften, Wien, Mathematisch-Naturwissenschaftliche Klasse. Sitzungberichte IIa,
126, 1489–1514 (in German).
Interscience Publishers.
Sweby, P. K. (1984). High resolution schemes using flux limiters for hyperbolic conservation
laws. Journal of the Society for Industrial and Applied Mathematics, Series B Numerical
Analysis, 21(5), 995–1011.
Sweby, P. K. (1999). Godunov methods (Numerical Analysis Report 7/99), UK: Department of
Mathematics, University of Reading.
Synolakis, C. E. (1986). The runup of long waves (Ph.D. thesis). California Institute of
Technology, Califormia.
Terzidis, G., & Strelkoff, T. (1970). Computation of open channel surges and shocks. Journal of
the Hydraulics Division, ASCE, 96(HY12), 2581–2610.
Toro, E. F. (1992). Riemann problems and the WAF method for solving the two-dimensional
shallow water equations. Philosophical Transactions of the Royal Society of London, Series A,
338, 43–68.
References 415
Toro, E. F. (2009). Riemann solvers and numerical methods for fluid dynamics: A practical
introduction. Berlin, Germany: Springer.
van Albada, G. D., van Leer, B., & Roberts, W. W. (1982). A comparative study of computational
methods in cosmic gas dynamics. Astronomy & Astrophysics, 108(1), 76–84.
van Leer, B. (1979). Towards the ultimate conservative difference scheme, V: A second-order
sequel to Godunov´s method. Journal of Computational Physics, 32, 101–136.
van Leer, B. (2006). Upwind and high-resolution methods for compressible flow: From Donor cell
to residual distribution schemes. Computer Physics Communications, 1(2), 192–206.
Vazquez-Cendón, M. E. (2015). Solving hyperbolic equations with finite volume methods. New
York: Springer.
Ying, X., Khan, A., & Wang, S. (2004). Upwind conservative scheme for the Saint Venant
Ying, X., & Wang, S. (2008). Improved implementation of the HLL approximate Riemann solver
for one-dimensional open channel flows. Journal of Hydraulic Research, 46(1), 21–34.
Zhou, J. G., Causon, D. M., Mingham, C. G., & Ingram, D. M. (2001). The surface gradient
method for the treatment of source terms in the shallow water equations. Journal of
Chapter 10
Sediment Transport and Movable Beds
10.1 Introduction
Transportation of sediment is an important and frequent phenomenon in rivers.

Sediment is mobilized as bed-load with particles sliding, saltating and rolling over
the river bed, or as a suspended-load, where particles move with the turbulent water
flow away from the bed. The dynamic processes of sediment erosion and deposition
modify the average density of the river fluid and reshape the river contours, which
are the physical boundaries conveying the flow. The study of sediment transport in
free surface flow is subject matter for a complete book. The works of Wu (2008)
and Dey (2014) are recommended for a comprehensive study. The purpose of this
chapter is to introduce how sediment transport is accounted for in shallow water
models by introducing the fluid–sediment mixture density and the river bed
deformation. The two-layer model of Wu and Wang (2007, 2008) is used for this
task and solved with the first-order upwind method. The problems of dam break
waves over an erodible bed and dike erosion due to overtopping are used as test
cases.
10.2 Flow Equations
10.2.1 Shallow Water Equations
Consider 1D unsteady free surface flow over an erodible bed in a vertical plane
(Fig. 10.1). The elevation of the static erodible sediment bed is zb(x, t), and the fluid
flow above is composed of a mixture of water and sediment. The dynamic flow
above the bed is composed of a bed-load layer, whose flux per unit width is qb, and
a suspended-load layer. The latter flow depth is h(x, t), the discharge per unit width

https://doi.org/10.1007/978-3-030-13073-2_10
418 10 Sediment Transport and Movable Beds
Fig. 10.1 One-dimensional two-layer flow over erodible bed
is q(x, t) and the depth-averaged velocity in x-direction is U(x, t) = q/h. The

depth-averaged hydrostatic mass and momentum conservation equations for the
suspended-load layer are, when integrating the RANS equations [see Eqs. (1.161)
and (1.181)],
@ @ @Ai
ðqAÞ þ ðqQÞ þ qi ¼ 0; ð10:1Þ
@t @x @t
2
@ @ Q @zs 1 @q
ðqQÞ þ q ¼ qgA qgASf g hp A: ð10:2Þ
@t @x A @x 2 @x
Here, Q(x, t) = discharge, A(x, t) = flow area, Sf = friction slope, Ai = transverse

bed sediment area taking the interface of suspended and bed-load layers as refer-
ence, q(x, t) = mass density of the water–sediment mixture in the suspended-load
layer, t = time, and x = streamwise coordinate.
Consider the thickness of the bed-load layer negligible as compared to h, which
is thereby approximated as the thickness of the suspended plus the bed-load layers.
The flow density is thus computed in terms of the total sediment concentration as
q = qw(1 − Ct) + qsCt, where qw = clear water mass density, qs = sediment mass
density, Ct(x, t) = depth-averaged total sediment concentration in the fluid layer of
thickness h (suspended plus bed-load layers), qi qb = mass density of static bed
layer, defined as qb = qwpm+ qs(1 − pm) and pm = sediment porosity. For a rect-
angular section, Eqs. (10.1)–(10.2) reduce with zs zb+ h to
@ @ @zi @zb
ðqhÞ þ ðqhUÞ ¼ qi qb ; ð10:3Þ
@t @x @t @t
@ @ @zs 1 2 @q
ðqhUÞ þ qhU 2 ¼ qgh gh sb : ð10:4Þ
@t @x @x 2 @x
10.2 Flow Equations 419
The bed shear stress is computed with n = Manning’s roughness coefficient and
Rh = hydraulic radius from
n2 U jU j
sb ¼ qg 1=3
: ð10:5Þ
Rh
10.2.2 Sediment Transport Layers and Bed Deformation
The sediment transport phenomenon is included in the vertically integrated flow

equations [Eqs. (10.3)–(10.4)] as highlighted by the presence of the functions q(x, t)
and zb(x, t). These flow variables are related to the bed- and suspended-load fluxes,
thereby implying that additional transport equations are needed. Here, the sediment
transport model of Wu and Wang (2007) is used to complete the system of governing
equations. The 1D depth-averaged mass conservation equation for the suspended sed-
iment is (Wu and Wang 2007; Wu 2008)
@ @
ðhCs Þ þ ðhUCs Þ ¼ E D: ð10:6Þ
@t @x
Here, Cs(x, t) = depth-averaged suspended sediment concentration; and E(x, t) and

D(x, t) = entrainment and depositional rates of sediment across the interface of
suspended- and bed-load layers. The mass balance in the bed-load layer is (Wu and
Wang 2007; Wu 2008)
@ @qb @zb
ðhb Cb Þ þ þ ð1 pm Þ ¼ D E; ð10:7Þ
@t @x @t
where hb(x, t) = thickness of bed-load layer, Cb(x, t) = depth-averaged bed-load

sediment concentration, and qb(x, t) = bed-load flux. The evolution of the static
sediment bed can be determined from the mass balance equation (Wu 2008)
@zb 1 qb qb
¼ DEþ ; ð10:8Þ
@t 1 pm L
where L = non-equilibrium adaptation length of total-load transport and qb*(x,

t) = equilibrium bed-load flux. In unsteady flows, a certain distance is required to
reach the equilibrium condition of sediment transport. This length L is defined as
the non-equilibrium adaptation length. Inserting Eq. (10.8) into Eq. (10.7), the
bed-load mass balance equation is rewritten as
@ @qb qb qb
ðhb Cb Þ þ ¼ : ð10:9Þ
@t @x L
Note that qb = hbUbCb, where Ub = bed-load velocity, approximated here as U (Wu

and Wang 2007).
10.2.3 Modified Flow Equations
For numerical implementation, it is recommended to remove the density q from the

temporal and flux terms of the shallow water equations, forming an additional
source term (Cao et al. 2004). Inserting the definitions q = qw(1 − Ct) + qsCt and
Ct = Cs+ qb/(hU) into Eqs. (10.3)–(10.4), the system of conservation laws is
rewritten as
@h @ 1 qb qb
þ ðhUÞ ¼ E Dþ ; ð10:10Þ
@t @x 1 pm L
@ @ 2 @zs sb 1 g 2 @q
ðhUÞ þ hU ¼ gh h
@t @x @x q 2 q @x
ð10:11Þ
q q U qb qb
b E Dþ ;
q 1 pm L
where the mixture density at the bed is qb = qs(1 − pm) + qwpm.
10.2.4 Total-Load Equation
Rather than using Eqs. (10.6)–(10.7), it is possible to define a transport equation for
the total load if separate computations for each load are of no interest. Summing
Eqs. (10.6)–(10.7) produces
@ @ @zb
ðhCs þ hb Cb Þ þ ðhUCs þ qb Þ þ ð1 pm Þ ¼ 0: ð10:12Þ
@t @x @t
Using the definition of the total-load concentration Ct = Cs + qb/(hU), and resorting

to Eq. (10.8), the total sediment transport mass conservation equation is (Wu and
Wang 2007, 2008)
@ @ qb qb
ðhCt Þ þ ðhUCt Þ ¼ D E þ : ð10:13Þ
@t @x L
We now define the source term of Eq. (10.13) as function of the total load as (Wu
2008)
qb qb qt qt
DEþ w¼ ; ð10:14Þ
L L
where qt* = total-load sediment transport capacity rate, qt = qCt = actual total-load
sediment transport rate, and L = non-capacity adaptation length of the total-load
sediment transport.
10.2.5 Sediment Transport Closure
According to Wu et al. (2004), L is given as the larger adaptation lengths of

bed-load (Lb) and suspended-load (Ls). It is with a = non-capacity adaptation
coefficient of suspended-load

U
L ¼ max½Lb ; Ls ¼ max Lb ; ; ð10:15Þ
axs
calculated with a0 = 2 as (Wu 2008)
a ¼ min½a0 ; ð1 pm Þ=Ct : ð10:16Þ
The settling velocity of sediment particles in turbid water xs is determined from the
Richardson–Zaki formula with m = 4 (Wu 2008)
xs ¼ xo ð1 Ct Þm ; ð10:17Þ
where the settling velocity of a single particle in clear water is with d as the particle
diameter and m the kinematic viscosity of water (Wu 2008)
h i0:5
xo ¼ ð13:95m=d Þ2 þ 1:09½ðqs =qÞ 1gd 13:95m=d: ð10:18Þ
The actual total load is qt= hUCt, while the total-load capacity is determined as
the contribution of both the suspended-load and the bed-load as
qt ¼ qs þ qb : ð10:19Þ
Here, qs* and qb* = suspended- and bed-load sediment transport capacity rates,
respectively, computed using Wu et al.’s (2000) formulas. To compute the equi-
librium bed-load flux qb*, alternative empirical formulations could be considered,
including the classical equations by Meyer-Peter and Müller, Yalin and van Rijn
(Dey 2014).
Eugen Meyer-Peter was born on February 25, 1883, at Herisau, Switzerland,

and passed away aged 86 years on June 18, 1969, at Zürich, Switzerland. He
received the ETH civil engineering diploma in 1905, joining then the con-
sulting firm of Conradin Zschokke (1842–1918), an expert in hydraulic
engineering, with Meyer-Peter eventually becoming an expert in caisson
foundation. The latter was involved in harbor works at Dieppe (France) and
Venice (Italy), as also in the Rhine power plant Augst-Wyhlen. In 1920, he
took over as ETH professor of hydraulic engineering and in parallel aimed to
set up a hydraulic laboratory. Financial constraints postponed its inauguration
until 1930.
He set out with excellent collaborators, including Henry Favre (1901–
1966), Charles Jaeger (1901–1989), Hans-Albert Einstein (1904–1973), and
Robert Müller (1908–1987). Next to contract work, a number of important
research projects were initiated, such as on surges in tailrace tunnels,
unsteady free surface flows, scour due to plunging jets, and spatially varied
flows. His name remains related to his researches pertaining to sediment
transport, with a first formula for the bed-load transport of uniform sediment
in 1934, and the Meyer-Peter-Müller (MPM) formula in 1948, currently
mainly applied for relatively steep rivers, such as in the Alpine Rhine envi-
ronment. He was also involved in the prediction of future river degradation.
He was awarded honorary doctorates from Zürich University in 1933 and the
University of Grenoble in 1950. He is considered a great organizer of
research campaigns and was the founder or co-founder of various institutes at
ETH Zurich.
According to Wu et al. (2000), both capacity rates can be determined by using

1=2 1:74
qs s U
qs ¼ 0:0000262 1 gd 3 1 ; ð10:20Þ
qw sc xs
1=2 " 0 3=2 #2:2
qs n sb
qb ¼ 0:0053 1 gd 3
1 ; ð10:21Þ
qw n sc
where d = single sediment size diameter, s = shear stress on the wetted perimeter
of the cross-section, sb = bed shear stress tangent to sediment–fluid interface,
sc = critical bed shear stress determining the threshold for incipient sediment
motion, n′ = Manning’s coefficient corresponding to the grain roughness for a
movable bed (= d1/6/20), and n = Manning’s roughness coefficient for the channel
bed, with d in [m] and n in [sm−1/3]. Assuming a channel width larger than the flow
depth, s sb. By numerical experimentation in shallow water flows, it was found
that the last assumption does not reduce the accuracy of the solution. The critical
shear stress is calculated with H = Shields parameter = 0.03 as (Wu 2008)
sc ¼ Hðqs qÞgd: ð10:22Þ
To apply Eqs. (10.20)–(10.21), the following ratio needs to be determined
sb qu2
¼ 2 : ð10:23Þ
s c q u c
The bed shear stress yields the shear velocity as
g1=2 U
u ¼ n 1=6
; ð10:24Þ
Rh
whereas the critical shear velocity is from Eq. (10.22)

1=2 h i1=2
qs s
ð u Þ c ¼ H 1 gd ¼ H 1 gd : ð10:25Þ
q K
Here, s = qs/qw is the particle submerged density ratio and

q
K¼ ¼ ð1 Ct Þ þ sCt : ð10:26Þ
qw
Note that K is based on the depth-averaged mixture density. Fraccarollo and Capart
(2002) observed extremely high sediment concentration near the bed in geomorphic
dam break flows. Thus, for dam break waves, the empirically enhanced version of
K used is
qb
K ¼ ½ð1 Ct Þ þ sCt b : ð10:27Þ
qw
The near-bed value for dam break waves over erodible beds considered below is
Ct = 0.6.
10.3 Numerical Scheme
10.3.1 Conservation Laws
Equations (10.10), (10.11), (10.13), and (10.8) can be written as (Cao et al. 2004;
Wu and Wang 2008)
@U @F
þ ¼ S; ð10:28Þ
@t @x
where
0 w
1
1p
0 1 0 1 B m
@zs sb 1 g 2 @q C
B C
h hU B gh h C
B hU C B hU 2 C B @x q 2 q @x C
B C
U¼B C
@ hCt A; F¼B C
@ hUCt A; S¼B q q U C:
B þ b w C
B q 1 pm C
zb 0 B C
@ w A
w
1pm
ð10:29Þ
Here, U is the dependent variable vector, F the flux in the x-direction, and S the
source term. Note that the thickness of the bed-load layer is not resolved by this
model, as originally assumed, that is, zs = zb+ h.
10.3.2 First-Order Upwind Finite Volume Method
Consider a rectangular finite volume mesh in the x-t plane. Cell i is limited by both
cell interfaces i − 1/2 and i + 1/2. In the t-axis, the control volumes are delimited
by the time levels k and k + 1. Integration of Eq. (10.28) over a control volume
yields exactly (Toro 2001) [see Eq. (9.7)]
10.3 Numerical Scheme 425
Dt
Uki þ 1 ¼ Uki Fi þ 1=2 Fi1=2 þ DtSi ; ð10:30Þ
Dx
where k = time level, Δt = time step, and Δx = grid step. According to Ying et al.
(2004), the one-sided upwind method is used to evaluate the intercell fluxes in
Eq. (10.30) as
0 1k
h
B hU 2 C
Fi þ 1=2 ¼B C
@ hUCt A : ð10:31Þ
0 iþl
Here, l = operator determining the upwind direction as function of the discharge

q = Uh sign as: l = 0 if qi > 0 and qi+1 > 0, l = 1 if qi < 0 and qi+1 < 0, and l = 1/2
in other cases, where l = 1/2 refers to an average of the flow variable between its
value at i and i + 1.
10.3.3 Source Terms
To increase model robustness, the flow depth in the momentum topographic source
term is evaluated at the time level k + 1, given that it is available by previously
solving the continuity equation. To ensure numerical stability of the one-sided
first-order upwind method, the water surface gradient in Eq. (10.30) is evaluated
following Ying et al. (2004) by using a weighted average of the downwind and the
upwind gradients as
!
@zs zks;iþþ11 zks;iþ 1 zks;iþ 1 zks;i1
þ1
gh ghki þ 1 w1 þ w2 : ð10:32Þ
@x i Dx Dx
The weighting factors are
Dt Uikþ 1l + Uil

k
Dt Uikþ l + Ui1
k
þl
w1 ¼ 1 and w2 ¼ : ð10:33Þ
Dx 2 Dx 2
The sediment transport source term in the momentum equation is discretized in

the numerical model as

1 g 2 @q qb q U 1 g 2 qi þ 1 qi1 qb qi Ui
h þ w h þ w;
2 q @x q 1 pm i 2 qi i 2Dx qi 1 pm i
ð10:34Þ
and sb/q is computed from Eq. (10.5).

10.3.4 Stability Condition
For stability of an explicit scheme, the Courant–Friedrichs–Lewy number must be

CFL\1 (Toro 2001). Given the value of CFL, Dt is determined at time level
k from [Eq. (9.29)]
2 3
6 Dx
Dt ¼ CFL4

7
5: ð10:35Þ

k k 1=2
max
Ui þ ðghi Þ
The maximum value for CFL was determined by trial-and-error numerical exper-
iments. The value was typically less than 0.5.
10.3.5 Computational Sequence
The process of the numerical solution is as follows:

1. A longitudinal finite volume mesh of cell width Dx is defined.
2. The initial conditions are set.
3. The value of Dt is computed by applying Eq. (10.35).
4. The first-order upwind method is applied to compute Fi+1/2.
5. The sediment transport computations are conducted to evaluate w at each cell.
6. The water depth at time hk+1 is computed from the first component of
Eq. (10.30).
7. The source terms in the momentum equation due to the friction slope, free
surface slope, and sediment transport are evaluated.
8. The discharge qk+1 is computed from the second component of Eq. (10.30).
9. The total sediment load (hCt)k+1 is computed from the third component of
Eq. (10.30).
10. The bed profile is updated using the fourth component of Eq. (10.30).
11. If the actual time equals the final, then stop.
12. Boundary conditions are set.
13. Go back to step 3 for a new time loop.
10.4 Test Cases
10.4.1 Dam Break Wave over Movable Bed
Dam break flow is a topic of continued research interest given its highly detrimental
effects. The usual engineering approach to predict these flows relies on the use of
10.4 Test Cases 427
depth-averaged shallow water wave models based on a hydrostatic pressure dis-

tribution (Wu 2008). In recent years, hydrostatic shallow water flow models were
extensively studied. Wu and Wang (1999, 2007, 2008) developed
Saint-Venant-type mixture flow equations for dam break flows over erodible beds.
Fraccarollo and Capart (2002) proposed a two-layer movable bed dam break flow
model with separate simulations of clear water and mixture sediment–water layers.
Capart and Young (1998, 2002) and Cao et al. (2004) developed models for dam
break waves over erodible bed considering non-equilibrium sediment transport
conditions. Geomorphic shallow water flows, as dam break waves over an erodible
bed, require consideration of fluid motion and erosion/deposition of particles. It is
accepted that the shallow water flow approximation with hydrostatic pressure
produces a reasonable balance between computational efforts and accuracy of
results (Pontillo et al. 2010; Greco et al. 2012).
In this section, dam break waves over an erodible bed generated under laboratory
conditions (Fig. 10.2) are considered, where the movable bed is made of uniform
particles, the tailwater portion of the flume is dry, and the dam is simulated with a
gate. The experimental data included in Figs. 10.3 and 10.4 are taken from the
Taipei and Louvain test cases (Capart and Young 1998; Fraccarollo and Capart
2002). Both sets correspond to laboratory observations, where the flow depth
upstream of the dam was ho = 0.1 m and the tailwater portion of the flume was dry.
Values of Δx = 0.01 m and CFL ¼ 0:3 are set in the numerical model implemented
in the file Movablebed_DamBreak.xls, available in Chap. 12. According to Wu and
Wang (2007), Ctb = 0.6, Lb = 0.25 m, a0 = 2, and n = 0.025 s/m1/3 are used as
simulation data for both test cases. The bed sediment layer porosities are pm = 0.28
and 0.30 for the Taipei and Louvain tests, respectively. The dry downstream bed is
simulated by adopting a minute flow depth of 0.0005 m. This technique is accepted
for dam break flow simulations (Wu 2008), although this is conceptually incorrect
(Toro 2001). Alternatively, a zero tailwater depth can be prescribed if an algorithm
to preserve positivity in the computed water depths near the wet–dry front is
implemented (Wu 2008). The computational domain was selected so that waves do
not reach the boundaries.
Fig. 10.2 Laboratory setup

to generate a dam break flow
over erodible bed
Fig. 10.3 Water and bed surfaces at different times for the Taipei test case
In Fig. 10.3, the Taipei experiment (Capart and Young 1998) is simulated by
solving Eqs. (10.28)–(10.29) at times t = 3to, 4to, and 5to after the dam failure,
where the time scaling is to = (ho/g)1/2 0.1 s. The sediment particles used were
artificial pearls covered with a shiny white coating, having d = 6.1 mm, qs =
1048 kg/m3, and xo = 7.6 cm/s. The flume was sufficiently long and deep with
b = 0.2 m. The model produces a bed profile eroded by the dam break flow in
overall agreement with the experiments. The experimental free surface profile is
composed of a negative smooth wave, followed by a train of undulations above the
10.4 Test Cases 429
scour hole, ending in a positive wave where its edge is a wet–dry front propagating
in the positive x-direction. Note that the experimental data show free surface
undulations at the vicinity of the positive wave portion, not predicted by hydrostatic
computations. However, the position of the wet–dry front predicted by the
hydrostatic model is in agreement with the experimental data. The overall wave
features are thus well predicted by the SWE.
The study of the Louvain test experiment at times t = 5to, 7.5to, and 10to after
the dam break is presented in Fig. 10.4. In this experiment, Fraccarollo and Capart
Fig. 10.4 Water and bed surfaces at different times for Louvain test case
(2002) used cylindrical PVC pellets as sediment particles, having d = 3.5 mm,
qs = 1540 kg/m3, and xo = 18 cm/s. The flume had a width b = 0.10 m. The
hydrostatic model produces simulations of the eroded bed profiles with some
divergence from the experimental results, but in overall agreement. The free surface
profile predicted is in fair agreement with observations, but the undulations due to
non-hydrostatic pressure are not reproduced (Cantero-Chinchilla et al. 2016).
10.4.2 Dike Erosion due to Overtopping
Dike breach experiments of Schmocker (2011) were selected for comparison purposes
with numerical simulations using Eq. (10.30). Figure 10.5 shows a definition sketch of
his experimental setup. Tests 42 and 55 were selected, involving variations in sediment
diameter (non-cohesive material) and discharge. The tests features are summarized in
Table 10.1. The dike erosion process produces a round-crested weir-like profile
(Fig. 10.6), with features different from the dam break wave problem.
The experimental data of Test 42 are compared in Fig. 10.7 with the numerical
solution resulting from Eq. (10.30). Values of Dx = 0.025 m and CFL ¼ 0:25
were set in the numerical model, which is implemented in the file
Movablebed_Dikebreaching.xls, available in Chap. 12. For the simulations K = 1
[see Eq. (10.26)], L = 0.01 m, xs = xo, and n = 0.018 s/m1/3 were used. The bed
sediment layer porosity pm is calculated from (Wu and Wang 2008)
Fig. 10.5 Definition sketch of experimental setup (Schmocker 2011), subscript “0” defines the
inflow conditions
Table 10.1 Dike test Test d (mm) Lk b (m) xD Sd Q0

characteristics (m) (m) (−) (m3s−1)
42 1.0 0.1 0.2 1.0 1:2 0.006
55 2.0 0.1 0.2 1.0 1:2 0.016
10.4 Test Cases 431
Fig. 10.6 Dike erosion test (photograph O. Castro-Orgaz)
Fig. 10.7 Water and bed surfaces at different times for dike erosion Test 42
0:21
pm ¼ 0:13 þ : ð10:36Þ
½d ðmmÞ þ 0:0020:21
The dry bed was simulated by adopting the minute flow depth 0.0001 m. The
upstream discharge was implemented in the numerical model as an instantaneous
pulse in a ghost cell, whereas the upstream water depth at the ghost cell was
determined by extrapolating the interior solution at the previous instant. At the
downstream section, transmissive boundary conditions were implemented in a
ghost cell (Toro 2001). For Test 42, the discharge is low, and non-hydrostatic
effects are expected to be small. This is in turn verified by comparing simulations
with measurements in Fig. 10.7, from where it is observed that the water level
upstream of the dike crest is in conformity with the theory. Beyond the dike crest,
the numerical model predicts excessive erosion on the downstream slope, thereby
diverging theory from experimental results.
Fig. 10.8 Water and bed surfaces at different times for dike erosion Test 55
10.4 Test Cases 433
The experimental and computational results for Test 55 are presented in

Fig. 10.8, whose particle size and inflow discharge are larger than for Test 42. Note
that the upstream flow depth is systematically overestimated resulting from the
hydrostatic pressure assumption. The model produces in this test again an over-
estimation of erosion in the supercritical flow portion. The entire process is highly
complex and still not fully understood. Recent results suggest that non-hydrostatic
pressures shall be introduced in shallow water models to produce a more accurate
flow profile up to the dike crest, whereas for simulating the tailwater portion the
displacement of the granular dike has a non-negligible effect (Cantero-Chinchilla
et al. 2018, 2019).
In this chapter, the extended shallow water equations for geomorphic flows are
presented and solved by using an upwind finite volume scheme. Applications to
dam break waves over erodible beds and dike failures due to overtopping highlight
that the physical processes involved are extremely complex and still not fully
accounted for in these families of computational models.
References
Cantero-Chinchilla, F., Castro-Orgaz, O., Dey, S., & Ayuso, J. L. (2016). Nonhydrostatic dam
break flows II: One-dimensional depth-averaged modelling for movable bed flows. Journal of
Cantero-Chinchilla, F. N., Castro-Orgaz, O., Schmocker, L., Hager, W. H., & Dey, S. (2018).
Depth-averaged modelling of granular dike overtopping. Journal of Hydraulic Research,
56(4), 537–550.
Cantero-Chinchilla, F. N., Castro-Orgaz, O., Dey, S. (2019). Prediction of overtopping dike
failure: A sediment transport and dynamic granular bed deformation model. Journal of
Cao, Z., Pender, G., Wallis, S., & Carling, P. (2004). Computational dam-break hydraulics over
erodible sediment bed. Journal of Hydraulic Engineering, 130(7), 689–703.
Capart, H., & Young, D. L. (1998). Formation of a jump by the dam-break wave over a granular
bed. Journal of Fluid Mechanics, 372, 165–187.
Capart, H., Young, D. L. (2002). Two-layer shallow water computations of torrential flows.
In: Proceedings of River Flow, vol. 2, (pp. 1003–1012). Lisse, The Netherlands: Balkema.
Dey, S. (2014). Fluvial hydrodynamics: Hydrodynamic and sediment transport phenomena.
Berlin: Springer.
Fraccarollo, L., & Capart, H. (2002). Riemann wave description of erosional dam break flows.
Journal of Fluid Mechanics, 461, 183–228.
Greco, M., Iervolino, M., Leopardi, A., & Vacca, A. (2012). A two-phase model for fast
geomorphic shallow flows. International Journal of Sediment Research, 27(4), 409–425.
Pontillo, M., Schmocker, L., Greco, M., & Hager, W. H. (2010). 1D numerical evaluation of dike
erosion due to overtopping. Journal of Hydraulic Research, 48(5), 573–582.
Schmocker, L. (2011). Hydraulics of dike breaching. Ph.D. thesis. Zürich, Switzerland: Swiss
Federal Institute of Technology.
Wu, W. (2008). Computational river dynamics. London, U.K.: Taylor and Francis.
Wu, W., & Wang, S. S. Y. (1999). Movable bed roughness in alluvial rivers. Journal of Hydraulic
Engineering, 125(12), 1309–1312.
Wu, W., & Wang, S. S. Y. (2007). One-dimensional modeling of dam-break flow over movable
beds. Journal of Hydraulic Engineering, 133(1), 48–58.
Wu, W., & Wang, S. S. Y. (2008). One-dimensional explicit finite-volume model for sediment
transport. Journal of Hydraulic Research, 46(1), 87–98.
Wu, W., Wang, S. S. Y., & Jia, Y. (2000). Nonuniform sediment transport in alluvial rivers.
Journal of Hydraulic Research, 38(6), 427–434.
Wu, W., Vieira, D. A., & Wang, S. S. Y. (2004). One-dimensional numerical model for
nonuniform sediment transport under unsteady flows in channel networks. Journal of
Ying, X., Khan, A. A., & Wang, S. S. Y. (2004). Upwind conservative scheme for the Saint
Venant equations. Journal of Hydraulic Engineering, 130(10), 977–987.
Chapter 11
Numerical Modeling of Non-hydrostatic
Free Surface Flows
11.1 Introduction
The shallow water equations (SWE) are a dispersionless system of hyperbolic PDEs
obtained by assuming that the vertical flow acceleration is negligible. Under this
assumption, the vertical momentum balance is reduced to the hydrostatic pressure
law (Toro 2001; Castro-Orgaz and Hager 2017). This approach produces good
solutions for shallow flows if the vertical length scale [H] is negligible as compared
to the horizontal length scale [L]. In turn, this scenario is realistic in many open
channel and river hydraulics problems. However, there are as well a large number
of practical questions where this level of mathematical approximation is not well
suited. It includes flows across hydraulic structures like weirs, intakes, overfalls,
energy dissipators, and water wave motion in maritime and fluvial hydraulics, as the
solitary and cnoidal waves, and the undular bore propagation in a river. In this large
portfolio of problems, the vertical length scale [H], though it can be small, is not
negligible as compared to [L] (Steffler and Jin 1993; Castro-Orgaz and Hager 2017).
A non-hydrostatic vertical pressure distribution must therefore be accounted for in
the depth-averaged equations. A higher-order mathematical model based on the
so-called Boussinesq equations is a possible choice to simulate these flows.
Depending on the technique and terms retained while introducing the vertical
acceleration and/or turbulent stresses on the depth-averaged equations, different
types of Boussinesq equations are obtained. In this chapter, we only consider
inviscid flow solutions following Castro-Orgaz and Cantero-Chinchilla (2019). The
non-hydrostatic flows considered are shallow, and waves are thus long and dis-
persion effects weak. The modeling of highly dispersive waves from deep to
intermediate depths is beyond the scope of this chapter. First, steady potential open
channel flow problems in a vertical plane are considered, then generalizations for
inviscid unsteady flow over a 3D terrain. Illustrative problems relating to hydraulic
structures and wave motion are solved using second- and higher-order numerical
schemes.

https://doi.org/10.1007/978-3-030-13073-2_11
436 11 Numerical Modeling of Non-hydrostatic Free Surface Flows
11.2 Two-Dimensional Steady Potential Flow
11.2.1 Basic Equations
Consider steady two-dimensional (2D) potential free surface flows over a curved
bed (Fig. 11.1). The governing equations are (Rouse 1938; Vallentine 1969)
@u @w
þ ¼ 0; ð11:1Þ
@x @z
@u @u 1 @p
u þw ¼ ; ð11:2Þ
@x @z q @x
@w @w 1 @p
u þw ¼ g; ð11:3Þ
@x @z q @z
@u @w
¼ 0: ð11:4Þ
@z @x
Here, u is the velocity component in the x-direction, w that in the vertical

z-direction, p is pressure, g the gravity acceleration, and q the fluid density.
Equation (11.1) is the continuity equation of an incompressible fluid, Eqs. (11.2)
and (11.3) are the momentum equations in the x- and z-directions, and Eq. (11.4) is
the irrotational flow condition. Equations (11.1)–(11.3) are the Euler equations for
inviscid fluid flow.
The potential function / is defined by Thom and Apelt (1961), Vallentine (1969)
@/ @/
u¼ ; w¼ : ð11:5Þ
@x @z
Fig. 11.1 Potential flow over a curved bed

11.2 Two-Dimensional Steady Potential Flow 437
This flow field automatically satisfies Eq. (11.4). Inserting Eq. (11.5) in Eq. (11.1),
the flow field obeys the Laplace equation for /,
@2/ @2/
þ 2 ¼ 0: ð11:6Þ
@x2 @z
For an incompressible fluid, a stream function w is defined by Thom and Apelt

(1961), Vallentine (1969) as
@w @w
u¼ ; w¼ : ð11:7Þ
@z @x
Using these relations in Eq. (11.1), the continuity equation is automatically satis-
fied. Using Eq. (11.7) in Eq. (11.4), the stream function satisfies the Laplace
equation
@2w @2w
þ 2 ¼ 0: ð11:8Þ
@x2 @z
The lines w = const. are streamlines, and the velocity vector, of modulus
V = (u2 + w2)1/2 and inclined with respect to the x-axis by the angle h = tan−1(w/u),
is tangent to them. The equipotential lines / = const. are normal to the streamlines
at intersection points, forming the so-called flow net of a potential flow (Fig. 11.1).
The Cauchy–Riemann equations are derived by equaling the velocity components
in Eqs. (11.5) and (11.7) as
@/ @w @/ @w
u¼ ¼ ; w¼ ¼ : ð11:9Þ
@x @z @z @x
A final result of relevance to solve potential free surface flows is the Bernoulli
equation, stating conservation of the total energy head H within the fluid domain
(Vallentine 1969)
p u2 þ w2
H ¼ zþ þ ¼ const: ð11:10Þ
qg 2g
Two-dimensional potential flow solutions are generally based on Eqs. (11.6),

(11.8), (11.9), and (11.10), or variations using other mappings. The velocity (u, w) and
pressure p of the 2D potential flow can be inserted into the vertically integrated form of
the momentum equations to obtain additional mathematical statements. One of great
utility is the vertically integrated x-momentum balance, obtained by integration of
Eq. (11.2) from the channel bottom (subscript b) to the free surface (subscript s), using
Leibniz’s rule, and imposing kinematic boundary conditions, expressing tangency of
the velocity at the basal and free surfaces, namely [see Eqs. (1.13) and (1.15)]
@zs @zb
ws ¼ us ; w b ¼ ub : ð11:11Þ
@x @x
The result is [see Eq. (1.126)]
dS pb dzb
¼ ; ð11:12Þ
dx qg dx
where
b þh
zZ
u2 p
S¼ þ dz ð11:13Þ
g qg
zb
is the specific momentum (Jaeger 1956; Matthew 1991; Montes 1998).
11.2.2 Picard Iteration
An approximate 1D model to solve potential flow problems is considered here. In turn,

this is a topic where a large research activity was conducted since the 1980s following
Boussinesq’s (1877) theory, see, e.g., Hager and Hutter (1984a, b), Hager (1985),
Marchi (1992, 1993), and Matthew (1991). There is a vast amount of Boussinesq-type
approximations to construct 1D models in civil, environmental, and coastal engineering
[see Castro-Orgaz and Hager (2017) for a detailed review]. Here, an elegant and
rigorous method developed by Matthew (1991) is presented. It is based on the Picard
iteration of the full 2D potential flow equations. The idea of all 1D methods is to
eliminate the z-dependence of the equations and to produce a model where the position
of the free surface as function of the x-coordinate is part of the solution itself. Van Dyke
(1975) discussed methods to solve systems of partial differential equations in fluid flow
problems by approximations. One of the possible techniques is to iterate the solution of
the equations ∇2/ = 0 and ∇2w = 0 starting with an initial solution. This is the idea
pursued by Matthew (1991), who iterated the Cauchy–Riemann equations [Eq. (11.9)],
using uniform flow as initial solution (Matthew 1991; Castro-Orgaz and Hager 2013,
2014). The process is described by Matthew (1991) for steady flow and extensively
discussed by Castro-Orgaz and Hager (2017) in their generalization for unsteady
potential flow, so that only the final results are stated here. The potential velocity
components approximated to second-order accuracy are
2
h h g2
u ¼ U þ ð2Ux gx þ Ugxx Þ g þ Uxx ; ð11:14Þ
2 6 2
w ¼ Ux g Ugx ; ð11:15Þ
with the variable η(x, z) = z − zb(x) as the vertical distance above the channel
bottom, U = q/h the depth-averaged velocity, h the vertical flow depth, and q the
unit discharge. Subscripts indicate differentiation, e.g., Ux =@U=@x, Uxx =@ 2 U=@x2 ,
ηx = −@zb =@x, and ηxx = −@ 2 zb =@x2 . For steady flow, the spatial derivatives of
U and η take the form
qhx qhxx qh2x

Ux ¼ ; U xx ¼ þ 2 ;
h2 h2 h3 ð11:16Þ
gx ¼ zbx ; gxx ¼ zbxx :
Inserting Eq. (11.16) into Eqs. (11.14) and (11.15) results in (Castro-Orgaz and
Hager 2017)
2
q 2hx zbx 2g h hxx h2x 3g h2
u ¼ 1 þ zbxx þ ; ð11:17Þ
h h 2 2h h2 3
qh g i
w ¼ zbx þ hx : ð11:18Þ
h h
Inserting Eqs. (11.17)–(11.18) in the Bernoulli Eq. (11.10), neglecting

second-order terms, and particularizing the result for the free surface (p = 0, η = h),
results in the extended Bernoulli-type equation (Matthew 1991; Castro-Orgaz and
Hager 2017)

q2 2hhxx h2x
H ¼ zb þ h þ 1 þ þ hz bxx þ z bx :
2
ð11:19Þ
2gh2 3
This theory considers the curvature effect of both the free surface and the bottom,
accounted for by the inclusion of hxx = d2h/dx2 and zbxx = d2zb/dx2; it also considers
the slope effect of the free surface and the bottom by inclusion of hx = dh/dx and
zbx = dzb/dx. The values of these slopes and curvatures are not necessarily small,
given that no restriction on non-linearity was imposed. Therefore, Matthew’s
(1991) theory, used by Castro-Orgaz and Hager (2009), is an approximate potential
flow model that includes the effects of finite curvatures and slopes. This equation
was presented by Naghdi and Vongsarnpigoon (1986) based on the theory of a
Cosserat surface (Green and Naghdi 1976a, b; Naghdi 1979). It is remarkable to
state that this equation is the steady-state form of the well-known Serre-Green-
Naghdi equations, which are widely applied in ocean research. This issue will be
further exploited below. Equation (11.19) was also obtained by Marchi (1992,
1993) by expanding the stream function in power series, and by Zhu and Lawrence
(1998) by using a perturbation method. It is a second-order differential equation,
from which the free surface profile h = h(x) is determined. For given q and
H = const., and prescribed flow depths at two boundary sections, Eq. (11.19) is
solved numerically as a two-point boundary value problem (Zhu 1996; Castro-

Orgaz and Hager 2009).
Using Bernoulli’s equation, the pressure distribution to this order of accuracy is
(Matthew 1991; Castro-Orgaz and Hager 2017)

p q2 g g2
¼ h gþ ð2hzbxx 2hx zbx Þ 1 þ hhxx hx 1 2 : ð11:20Þ
2
c 2gh2 h h
The value at the channel bottom pb = p(η = 0) is [see Eq. (1.132)]
pb q2
¼ hþ 2hzbxx þ hhxx h2x 2zbx hx : ð11:21Þ
c 2gh 2
After having established the energy equation, the momentum principle is con-
sidered. Inserting Eqs. (11.17) and (11.20) for the distributions of u and p into
Eq. (11.13), the specific momentum S is given by (Castro-Orgaz and Hager 2017)
[see Eq. (1.133)]

h2 q2 hhxx h2x hzbxx hx zbx
S¼ þ 1þ þ : ð11:22Þ
2 gh 3 2 2
Using Eqs. (11.21) and (11.22), the vertically integrated momentum balance
[Eq. (11.12)] can be equally used to compute the free surface profile h = h(x). It
remains to delineate the connection between the energy [Eq. (11.19)] and
momentum [Eqs. (11.12), (11.21) and (11.22)] models for steady potential flow
over curved beds. Consider first Eq. (11.19), given that the energy head is a con-
stant in potential flow, its differentiation with respect to x yields

dH q2 2hhxx h2x
¼ zbx þ hx 3 hx 1 þ þ hzbxx þ z2bx
dx gh 3
ð11:23Þ
q2 2hhxxx
þ þ hz bxxx þ h z
x bxx þ 2z z
bx bxx ¼ 0:
2gh2 3
The result is the following third-order ODE after elementary re-arrangement
q2 2 q2 1 q2 3 1 q2
zbx þ hx h x h x h xx þ h þ hxxx
gh3 3 gh2 3 gh3 x 3 gh
ð11:24Þ
1 q2 q2 1 q2 q2
h x z bxx h x z2
þ z bxxx þ zbx zbxx ¼ 0:
2 gh2 gh3 bx 2 gh gh2
Now, Eqs. (11.21)–(11.22) are inserted into Eq. (11.12), producing


d h2 q2 hhxx h2x hzbxx hx zbx
þ 1þ þ
dx 2 gh 3 2 2
q2 dzb
¼ hþ 2hzbxx þ hh xx h 2
x 2z h
bx x : ð11:25Þ
2gh2 dx
Expanding the various terms in Eq. (11.25) yields Eq. (11.24). This implies that the
energy and momentum balances produce the identical governing equation to this
order of accuracy. Therefore, it is in principle easier to adopt Eq. (11.19), which is
an integral form of Eq. (11.25). This is interesting from a numerical standpoint,
given that the higher order of differentiation is reduced to second order, avoiding
instability problems linked to the discretization of third-order derivatives in
Boussinesq-type equations (Bonneton et al. 2011). Equation (11.25) is the
steady-state version of the Serre–Green–Naghdi equations; thus, Eq. (11.19) is an
integral form of these equations under steady flow. Equation (11.19) admits ana-
lytical solutions for selected problems like the flow profile upstream of a free
overfall, the solitary wave, or the supercritical jet beyond a sluice gate
(Castro-Orgaz and Hager 2017). However, in general, a numerical solution is
required for flows over curved beds.
11.2.3 Spillway Crest
Reconsider Eq. (11.23). Imposing dzb/dx = 0 yields, after re-arrangement,

q2 2hhxx h2x h2 hxxx hzbxx h2 zbxxx
1þ þ ¼ 1: ð11:26Þ
gh3 3 3hx 2 2hx
This relation is valid at the spillway crest (Fig. 11.2); accepting a physical solution
with finite free surface slope (hx < 0), it is the critical flow condition for curvilinear
flows. This condition serves as the mathematical equation describing the critical
depth in non-hydrostatic flows (Castro-Orgaz and Hager 2017).Equation (11.26)
indicates that the non-hydrostatic critical flow condition does not depend on the
section conditions alone, e.g., the actual value of h, but in addition on the con-
figuration of the flow profile h = h(x) in the vicinity of the crest section, and on the
local bottom geometry variation zb = zb(x). The words by Rouse (1938) on p. 326
of his book state:
… Since the ratio of depth to specific energy at the true critical section is so definitely a
function of the curvature imposed by the fixed boundaries, it is almost futile to expect that a
simple relationship may be found expressing this ratio in terms of boundary geometry. It is
to be hoped, nevertheless, that a broader understanding of true critical discharge may soon
lead to definite progress in this essential field.
Fig. 11.2 Flow over spillway

crest (VAW photo)
Equation (11.26) is an advance in this line; the non-hydrostatic effects imposed by the
boundary conditions are included in the critical flow statement. Equation (11.26) is
an approximate solution at the weir crest to the elliptic problem posed by Laplace’s
equation for potential flow. Therefore, the approximation to this elliptic problem
depends not only on the crest conditions, but rather on the complete flow solution,
which, in turn, determines the spatial derivatives hx, hxx, and hxxx.
It is possible, however, to find approximations to these derivatives at the weir
crest on the basis of the lower-order energy head equation for hydrostatic flows
(Hager 1985; Matthew 1991) [Eq. (2.12)]
q2
H ¼ zb þ h þ ¼ const: ð11:27Þ
2gh2
This theory for critical curvilinear flows is presented here following Castro-Orgaz
and Hager (2017). Differentiation of Eq. (11.27) produces for critical flow (F ¼ 1)
the following approximations for h2x and hhxx at the weir crest, with R as the crest
curvature radius (Castro-Orgaz and Hager 2017),
hzbxx h
h2x ¼ ¼ ; ð11:28Þ
3 3R
4zbxx 4
hxx ¼ ¼ : ð11:29Þ
9 9R
The specific energy at the weir crest (dzb/dx = 0) is from Eq. (11.19)

q2 2hhxx h2x
E ¼ hþ 1þ þ hzbxx ; ð11:30Þ
2gh2 3
and the discharge coefficient of the spillway is defined by Poleni’s equation as

(Montes 1998)
1=2
q ¼ Cd gE 3 : ð11:31Þ
Inserting Eqs. (11.28)–(11.29) into Eq. (11.30), and using Eq. (11.31) produce,
retaining first-order terms (Castro-Orgaz and Hager 2017)
3=2
2 22 E
Cd ¼ 1þ : ð11:32Þ
3 81 R
The critical depth in curvilinear motion is determined from Eq. (11.26). The
third-order flow depth derivative term needed in this equation is approximated by
differentiation of Eq. (11.27) as (Castro-Orgaz and Hager 2017)
h2 hxxx 5 5h
¼ hzbxx ¼ : ð11:33Þ
hx 9 9R
Inserting Eqs. (11.28), (11.29), and (11.33) into Eq. (11.26) produces when
retaining first-order terms (Castro-Orgaz and Hager 2017)
hcrest E
¼1 : ð11:34Þ
hc 9R
The bottom pressure head at the weir crest is from Eq. (11.21)
pb q2
¼ hþ 2hzbxx þ hhxx h2x ; ð11:35Þ
c 2gh 2
where c is the specific weight of fluid. Using Eqs. (11.28), (11.29), and (11.34) into
Eq. (11.35) yields to first-order accuracy
ðpb Þcrest 20E

¼1 : ð11:36Þ
chc 27R
Equations (11.32), (11.34), and (11.36) comprise the first-order solutions for the
problem of critical flow in curvilinear motion. They are expected to be valid for
weakly curved flows, e.g., up to E/R = 1 (Castro-Orgaz and Hager 2017). This
Fig. 11.3 Critical flow at crest of a parabolic weir
theory is compared in Fig. 11.3 with the experimental data of Blau (1963) for a
parabolic weir of crest radius R = 0.919 m. Note that the approximate theory
produces a solution in fair agreement with observations. The advantage of this type
of model is that the full solution of Eq. (11.19) is avoided. The full solution h(x) of
flow over curved beds can only be obtained numerically, given the lack of known
exact analytical expressions.
11.2.4 Steep Slope
Consider flows away from a slope break, where streamline curvature effects can be
neglected. The flow is gradually varied on a steep slope (Fig. 11.4), given that the
variation of h with x is small, as confirmed by experiments (Castro-Orgaz and
Hager 2009, 2017).
For these flows, it can be assumed that h2x hxx 0. Further, on the slope, the
bed is flat, resulting in zbxx = 0. On this slope, however, the term zbx is finite.
Therefore, despite hx will be small, the product (hx ⋅ zbx) remains finite. Therefore,
Eqs. (11.19) and (11.21) for gradually varied, 1D potential flow on a finite slope
read (Castro-Orgaz and Chanson 2016)
q2
H ¼ zb þ h þ 2
1 þ z2bx ; ð11:37Þ
2gh
pb q2
¼h ð2zbx hx Þ: ð11:38Þ
c 2gh2
Differentiation of Eq. (11.37) produces for supercritical flow on the steep slope
(large F) the ODE (Castro-Orgaz and Hager 2009)
dh zbx
¼ q2
: ð11:39Þ
dx
gh3 1 þ z2bx
Inserting Eq. (11.39) into Eq. (11.38), the bottom pressure head is [see Eq. (1.137)]

pb q2 z2bx h
¼h ð 2z h
bx x Þ ¼ h 1 ¼ : ð11:40Þ
c 2gh 2 1 þ zbx2 1 þ z2bx
Fig. 11.4 Transition from a

mild to a steep slope (VAW
photo)
Equation (11.40) is the bottom pressure head in gradually varied flows on a steep
slope, implying non-hydrostatic conditions. This is the value to which 1D com-
putations tend on a slope using Eq. (11.19) (Castro-Orgaz and Hager 2009).
11.2.5 Flow over Round-Crested Weir: Numerical Method
On inspecting Eqs. (11.36) and (11.40), it can be appreciated that Matthew’s (1991)
theory accounts for curvature (weir crest) and slope (chute) effects in non-
hydrostatic fluid pressure. The theory has the advantage that it is mathematically
elaborated. Thus, the salient results are rigorous statements, without the need of
artificial tuning of the final equations, based on ad hoc assumptions. A numerical
method to solve Eq. (11.19) is considered in this section for the flow over a
round-crested weir. The method will be presented taking as test case transcritical
flow over a bell-shaped hump prescribed by

x 2
zb ¼ a exp b : ð11:41Þ
L
Here, a is the maximum weir elevation, b a coefficient, and L a horizontal length

scale. The hump experimentally tested by Sivakumaran et al. (1983) was designed
for a = 0.2 m, b = 0.5, and L = 0.24 m. The computational domain is discretized
using a finite-difference mesh divided in N nodes separated by Δx (Fig. 11.5). For a
given flow rate q and known boundary flow depths h(i = 1) and h(i = N),
Eq. (11.19) is a second-order ODE to be solved for the computational nodes i = 2,
3,…, N − 1. Further, the constant H must be evaluated prior to any numerical
Fig. 11.5 Finite-difference model for the solution of the steady Serre–Green–Naghdi equations
[adapted from Castro-Orgaz and Cantero-Chinchilla (2019)]
solution. The nodes i = 1 and i = N are located away from the hump in zones where the
streamlines are parallel and the pressure is hydrostatic. Therefore, H is estimated from
q2
H h1 þ : ð11:42Þ
2gh21
However, neither h(i = 1) nor h(i = N) is known in advance for weir flow. For a
transition from upstream subcritical to downstream supercritical flow, these flow
depths are the alternate depths corresponding to the total head H. For a numerical
solution, an iterative strategy is necessary. Equation (11.19) can be solved numeri-
cally using a shooting method transforming it into a pair of ODEs. Starting with a
guessed value of h1 and setting dh/dx = 0 there, given that the streamlines are parallel
at the inflow section (Castro-Orgaz and Hager 2017), the flow profile is explicitly
computed based on the results at the previous computational section. Once h1 is
assumed, the corresponding value of H is determined by Eq. (11.42). Using a Runge–
Kutta solver, the system of two ODEs is integrated along the computational domain
from i = 1 up to i = N, with the flow depth there determined as the final result of the
numerical integration. This value of the flow depth is compared to the supercritical
alternate depth of h1, and, if they do not match within a prescribed tolerance, then h1
and thus also H are both in error; the value of h1 must be therefore corrected. This
method of attack for transcritical flow over a round-crested weir was first discussed
by Naghdi and Vongsarnpigoon (1986), who numerically solved Eq. (11.19) for the
weir experiments of Sivakumaran et al. (1983). They found that if the value of h1 is
too low, the flow profile intersects the bed profile elsewhere (h = 0 at this point), and
computations are thus aborted before reaching the end section. If the head is too large,
an undular jump is formed at the tailwater portion of the weir. The value of h1 can be
iterated, but a high precision is needed to obtain the correct physical solution. This
was later discussed by Fenton (1996) and Castro-Orgaz and Hager (2017).
Castro-Orgaz and Hager (2013) demonstrated that a physically correct solution for
transcritical weir flow follows with this shooting technique.
An alternative method of solution is to treat Eq. (11.19) as a boundary value
problem directly, discretizing the equations using finite-differences and solving the
implicit system of equations iteratively. The first solution of a steady
Boussinesq-type equation using an implicit finite-difference method was elaborated
by Hosoda and Tada (1994) to compute the undular jump profile. For the specific
case of transcritical flow over a weir, Zhu (1996) and Zhu and Lawrence (1998)
solved Eq. (11.19) as a boundary value problem using a collocation method,
finding excellent agreement between the numerical solution and the data of
Sivakumaran et al. (1983). Onda and Hosoda (2004) solved Eq. (11.15) for flow
over curved bed forms, but added terms to account for bed friction and turbulence.
They used a second-order accurate central finite-difference scheme and solved the
resulting equations iteratively with a technique similar to a Newton–Raphson
method. Zerihun and Fenton (2006) computed transcritical flows with an implicit
finite-difference scheme solved iteratively with a Newton–Raphson iteration
assisted by a lower-upper (LU) decomposition of an analytical Jacobian matrix.
Castro-Orgaz and Hager (2009) applied Eq. (11.19) for the transition from a
mild to a steep slope using a centered finite-difference scheme fourth-order accurate
and solved the resulting implicit system of equations as an optimization problem.
Here, a simpler version of the finite-difference method of Castro-Orgaz and Hager
(2009) is elaborated. Following Onda and Hosoda (2004), the free surface
derivatives of steady Boussinesq-type equations can be discretized using second-
order central finite-differences as
dh hi þ 1 hi1 d2 h hi þ 1 2hi þ hi1

¼ ; 2¼ : ð11:43Þ
dx 2Dx dx ðDxÞ2
Inserting Eq. (11.43) into Eq. (11.19) yields the algebraic equation at node i as
"
q2 2hi ðhi þ 1 hi1 Þ2
H i ¼ ð z b Þ i þ hi þ 2
1þ 2
ðhi þ 1 2hi þ hi1 Þ
2ghi 3ðDxÞ 12ðDxÞ2
#

þ hi ðzbxx Þi þ z2bx i :
ð11:44Þ
If the discrete flow depths hi−1, hi, and hi+1 are an accurate representation of the true
flow solution at these nodes, then the estimated head at node Hi can be considered
to equal H, within a prescribed tolerance. Thus
H Hi ¼ fi ðhi þ 1 ; hi ; hi1 Þ eH: ð11:45Þ
Here, e is the tolerance, typically set to 10−6. In this event, the problem is finished
and there is no need to iterate the flow depths at the finite-difference nodes.
However, the initially assumed distribution of flow depths at the nodes of the mesh
is in general not accurate enough, and there is a need of iterative solutions. Thus, at
each node i, the error ei generated by the assumed solution is
ei ¼ H Hi [ eH: ð11:46Þ
Now the correction strategy is elaborated. Let k be the iteration index, then the
variation of the error at node i between two consecutive iterations can be expressed
using a truncated Taylor series as

@ei @ei @ei
dei ¼ eki þ 1 eki ¼ dhi1 þ dhi þ dhi þ 1 : ð11:47Þ
@hi1 @hi @hi þ 1
We seek setting to zero the error after the current iteration, that is, eki þ 1 0. Thus,
Eq. (11.47) for the N − 2 computational nodes is written in matrix form with J as
the Jacobian matrix as
ek ¼ Jk dh: ð11:48Þ
Thanks to the second-order differences used [Eq. (11.43)] to discretize the steady
Boussinesq-type equation [Eq. (11.19)], the structure of the linear system (11.48) is
tridiagonal, and the vector of corrections dh can be efficiently determined using the
Thomas algorithm (Hoffman 2001). Once this task is accomplished, the new esti-
mation of the flow depths is given by
hk þ 1 ¼ hk þ dh: ð11:49Þ
The non-zero elements of the Jacobian matrix J are given analytically as

( " #
@ei q2 2hi ðhi þ 1 hi1 Þ2 2
¼ 1 3 1þ ðhi þ 1 2hi þ hi1 Þ þ hi ðzbxx Þi þ zbx i
@hi ghi 3ðDxÞ2 12ðDxÞ2
" #)
q2 2 4hi
þ ðhi þ 1 2hi þ hi1 Þ þ ðzbxx Þi ;
2gh2i 3ðDxÞ2 3ðDxÞ2
ð11:50Þ
" #
@ei q2 2hi ðhi þ 1 hi1 Þ
¼ þ ; ð11:51Þ
@hi1 2gh2i 3ðDxÞ2 6ðDxÞ2
" #
@ei q2 2hi ðhi þ 1 hi1 Þ
¼ : ð11:52Þ
@hi þ 1 2gh2i 3ðDxÞ2 6ðDxÞ2
A reasonable initial solution must be implemented in the Newton–Raphson scheme

to obtain a convergent solution. Computation of the transcritical flow profile over
the weir using the hydrostatic version of Eq. (11.19) for the actual discharge q,
namely
q2
H ¼ zb þ h þ ; ð11:53Þ
2gh2
yields an estimation of flow depths in the subcritical flow portion, which are too far
away from the Boussinesq solution for some tests, however. This produces diver-
gent numerical computations.
A different strategy was therefore adopted based on Eq. (11.53). The assumed
flow depth h1 was used to estimate the head as H = h1; the specific energy at the
weir crest is then Emin = h1 − a. The hydrostatic discharge compatible with this
minimum specific energy is (Chanson 2004)
3=2
2 3 1=2
qo ¼ gEmin : ð11:54Þ
3
The two positive solutions of Eq. (11.53) are (Chanson 2004)

1 2 C
h ¼ ðH zb Þ þ cos ð11:55Þ
3 3 3

1 2 C 4p
h ¼ ðH zb Þ þ cos þ ð11:56Þ
3 3 3 3

27 H zb 3
cos C ¼ 1 ; ð11:57Þ
4 hc
with hc = (q2o/g)1/3. The transcritical hydrostatic flow profile computed for qo was
found to produce a reasonable initial estimate of the flow depths at the
finite-difference nodes. Once the free surface profile is accurately predicted with
Eq. (11.19), the result must be analyzed. If a train of standing waves is formed
upstream from the weir, this points at an incorrect estimate of H. It basically
indicates that the corresponding value of the discharge coefficient Cd is incorrect,
with
q
Cd ¼ h i1=2 : ð11:58Þ
gð H aÞ 3
For transcritical weir flow, the flow passes from upstream subcritical to down-
stream supercritical flow without appreciable waves on the flow profile. The rela-
tion between q and H is unique, and the value of Cd is therefore fixed. However, for
a given q, other flow profiles with upstream waves are mathematically possible
from Eq. (11.19) for different values of H. These are not physical solutions to the
weir flow problem, however. This phenomenon is described by Zhu (1996), who
also found upstream trains of standing waves while solving Eq. (11.19) as a
boundary value problem using a collocation method. The head H (or h1) must be
iterated until the upstream waves are suppressed.
The process of numerical solution is as follows:
1. Given the discharge q, a value of h1 is assumed, and H is then computed using
Eq. (11.42).
2. The value of hN is the supercritical alternate depth of h1.
3. The flow depths h1 and hN are set as boundary conditions.

4. An initial free surface profile is determined for the computational nodes h2, h3,
…hN−1 using Eqs. (11.54)–(11.57).
5. The error ei at each computational node is determined from Eqs. (11.44) and
(11.46).
6. The Jacobian matrix is formed using Eqs. (11.50)–(11.52).
7. The linear system of Eq. (11.48) is solved using the Thomas algorithm, resulting
in the vector of corrections at each node, with
0 1 0 kþ1 1
dh2 h2 hk2
B dh3 C B hk þ 1 hk C
B C B 3 3 C
dh ¼ B
B C¼B
C B C:
C ð11:59Þ
@ A @ A
kþ1
dhN1 hN1 hN1
k
8. The updated values of h are computed using Eq. (11.49).

9. The error ei at each computational node is determined from Eqs. (11.44) and
(11.46). If the maximum error is not within a prescribe tolerance eH, then go
back to step 6 for an improved estimation of h.
10. Once the free surface profile converges, if a train of standing waves is formed at
the upstream weir side, the flow depth h1 must be iterated to produce the correct
discharge coefficient Cd of weir flow, which is reached when these waves are
suppressed (Zhu 1996).
The numerical model described is implemented in a code available on the file
“Matthew_SVK_Implicit.xls,” in Chap.12. The selection of a reasonable value of h1
to start the cycle can be done by resorting to equations available in the literature,
e.g., using the result of a third-order Boussinesq equation (Matthew 1991)
3=2 " 2 #
2 22 E E
Cd ¼ 1þ 0:045 ; ð11:60Þ
3 81 R R
where E = H − a. For the Gaussian hump, d2zb/dx2 = –2ab/L2 at the crest. If the
assumed head is too far from its correct value, the numerical solution may diverge
while updating h. This was avoided using a relaxation factor of 0.25 in the Newton–
Raphson algorithm.
Figure 11.6 displays the experimental data of Sivakumaran et al. (1983) for a
symmetrical hump of shape zb = 0.2 exp[−0.5(x/0.24)2] (m). The unit discharge is
0.11197 m2/s. The up- and downstream boundary sections are located at
Fig. 11.6 Iteration of upstream flow depth in transcritical non-hydrostatic flow over a weir a RUN
1, b RUN 2 (see Table 11.1) [adapted from Castro-Orgaz and Cantero-Chinchilla (2019)]
x = ±1.5 m. Equation (11.19) was numerically solved with the implicit finite-
difference method described above using Δx = 0.05 m. A first computation was
conducted assuming h1 = 0.35 m, and the corresponding free surface profile and
discharge coefficient are reported in Fig. 11.6a and Table 11.1 as RUN1, respec-
tively. Note that this arbitrary h1 value, though close to experiments, produces an
inacceptable variation of Cd, with the corresponding train of upstream waves
(Fig. 11.6a). For RUN 2, the value of h1 after some iterations is given in
Table 11.1, resulting in a smooth free surface profile (Fig. 11.6b). The predicted Cd
value is now very close to the experimental value.
Once the final free surface profile is determined, the bottom pressure head is
estimated from the finite-difference version of Eq. (11.21) as
"
pb q2 hi
¼ hi þ 2hi ðzbxx Þi þ ðhi þ 1 2hi þ hi1 Þ
qg i 2gh2i ðDxÞ2
# ð11:61Þ
ðhi þ 1 hi1 Þ2 ðhi þ 1 hi1 Þ
ðzbx Þi :
4ðDxÞ2 Dx
The computed water surface and bottom pressure profiles are compared in Fig. 11.7
with the corresponding test data (Sivakumaran 1981; Sivakumaran et al. 1983),
resulting in excellent agreement.
Table 11.1 Iteration of Cd in weir flow (q = 0.11197 m2/s)

h1 (m) H (m) Cd Variation Cd (%)
RUN 1 0.35 0.35521 0.584602 –5.8
RUN 2 0.343447 0.348865 0.622411 +0.29
Experiments 0.3437 0.349154 0.620603 –
Fig. 11.7 Comparison of computed and measured (Sivakumaran 1981; Sivakumaran et al. 1983)
free surface and piezometric bottom pressure head in transcritical flow over a weir; a scaled photo
from experiments (Sivakumaran 1981) is also shown
Paul Mansour Naghdi was born on March 29, 1924, in Tehran, Iran, and
passed away on July 9, 1994, aged 70 years at Berkeley CA, USA, having
been naturalized there in 1948. He made studies at the University of
Michigan, Ann Arbor MI, receiving the PhD degree in 1951. He was there
assistant until 1958, then associate and full professor. He moved to the
University of California, Berkeley CA, as a professor of engineering science,
chairing from 1964 to 1969 its Department of Applied Mechanics. He was a
member of the National Academy of Engineering NAE, the American
Society of Mechanical Engineers ASME, and ASME Fellow from 1969. He
was the 1958 Guggenheim Fellow, recipient of the Timoshenko Medal in
1980, and was awarded ASME Honorary Membership in 1983.
Naghdi’s work on continuum mechanics extended over forty years
including most aspects of the mechanical behavior of solids and fluids. He
was strongly attracted by fundamental questions and always sought to treat
these at the highest level of generality. He is known for his works in the areas
of shell theory and plasticity, but also in viscoelasticity, fluid sheets and jets,
continuum thermodynamics, and the mixture theory. His talents as a teacher
were legendary. At Berkeley, he continually refined a magnificent series of
courses on theoretical mechanics. These were well prepared, clearly deliv-
ered, original, and intellectually provocative. They also reflected his deep
understanding of the history of mechanics and his encyclopedic knowledge of
literature.
11.2.6 Transition from Mild to Steep Slope
Transitional flows from mild to steep bottom slopes (Fig. 11.4) involve a continuous
free surface profile and a significant departure of the bottom piezometric pressure
profile from the free surface (Montes 1994; Castro-Orgaz and Hager 2009).
The numerical model developed in the previous section was applied to the experi-
mental data of Hasumi (1931) for a slope transition composed of a horizontal reach
followed by a circular-shaped transition profile of R = 0.1 m that finishes in a steep
slope reach of 45° inclination. The unit discharge is 0.0995 m2/s. The up- and
downstream boundary sections were located away from the slope break using
Dx = 0.005 m and 180 nodes to model the flow. The upstream boundary flow depth
was taken as the critical depth for hydrostatic flow hc = (q2/g)1/3, resulting in the total
head of the potential flow problem of H = 3hc/2. At the downstream boundary
section, the supercritical non-hydrostatic flow depth for streamlines nearly parallel to
the bed was estimated to hd = [(1 + S2o) q2/(2gE)]1/2, where E is the specific energy
at the downstream section and So = 1 is the chute slope for this test (Castro-Orgaz and
Hager 2009). The success in the iteration process depends upon the selection of a
plausible initial flow profile. In this test case, the initial flow profile was taken as
a rough linear interpolation between the critical depth and the brink flow depth
hb = 0.7hc along the horizontal reach, and between hb and hd along the chute portion.
The code for this test is available on the file Matthew_slopebreak.xls, in Chap. 12.
The bed profile derivatives were obtained using second-order central
finite-differences. The computed water surface and bed pressure profiles are com-
pared in Fig. 11.8 with the experimental data, resulting in a good agreement. Note the
computed pressure peak near the end of the transition circle, which is in disagreement
with the experiments, resulting from the downstream bed curvature discontinuity.
This computational effect can be removed by substituting the real bed profile by an
approximate transition curve with a smooth curvature variation (Castro-Orgaz and
Hager 2009). However, this task is not accomplished here, given that the emphasis is
on the computational process rather than on improving the simulation results.

computed and measured free
surface and piezometric
bottom head in transition from
mild to steep slope
11.2.7 Flow over Trapezoidal Profiled Weir
Zerihun (2004) conducted steady non-hydrostatic flow tests for transcritical flow
over trapezoidal profiled weirs. The steady numerical model was applied to a
trapezoidal weir of up- and downstream slopes 2:1 (H:V) and crest width and height
both of 0.15 m. The unit discharge is 0.06128 m2/s. The up- and downstream
boundary sections were located away from the weir using Dx = 0.01 m and 300
nodes. The upstream boundary flow depth h1 was determined by iteration until the
upstream undulations in the approach flow were significantly suppressed. At the
downstream boundary section, the supercritical alternate depth of the upstream
depth was settled in the mathematical model. For the initial free surface profile,
linear variations along the weir involving the inflow depth, critical depth, brink
depth, supercritical depth at the toe of the weir, and the tailwater flow depth were
assumed. The code for this test is available on the file Matthew_embankmentweir.
xls, in Chap. 12. The bed profile derivatives were obtained using second-order
central finite-differences. The computed water surface and bed pressure profiles
after the iteration of h1 are compared in Fig. 11.9a with the experimental data
(Zerihun 2004), resulting in a good agreement. The pressure peaks at the slope
discontinuities can be removed using transition curves (Zerihun 2004), but this task
is not accomplished here.
A simulation for the same weir for q = 0.07102 m2/s using Dx = 0.005 m is
presented in Fig. 11.9b, showing again a good match with experimental observa-
tions. Lesleighter et al. (2008) simulated this weir test with CFD solving the
Reynolds-Averaged Navier–Stokes equations, and their results for the bottom
pressure head are presented in Fig. 11.9c. Note that the experimental data of
Zerihun (2004) did not reveal pressure peaks at the bed-slope breaks, but CDF
simulations highlight those peaks, although they are of smaller magnitude than
predicted by the 1D model. It is impossible to know the experimental magnitude of
these peaks, given that no pressure taps were installed at these positions (Zerihun
2004). However, the physical model experiments conducted by Lesleighter et al.
(2008) on trapezoidally profiled spillways reveal pressure peaks at slope breaks.
11.3 Unsteady Ideal Fluid Flow
11.3.1 Two-Dimensional Serre Equations
Important problems like the propagation of undular, transcritical waves over

topography (Fig. 11.10) require a generalization of the results to unsteady 2D flows
(Nadiga et al. 1996). The mathematical procedure to generate 2D non-hydrostatic
equations consists in vertically integrating the governing equations at an arbitrary
horizontal (x, y)-position from zb to zs, where subscripts b and s, respectively, refer
to the bed and the free surface (Fig. 11.10), apply Leibniz’s rule, and impose the
11.3 Unsteady Ideal Fluid Flow 457

computed and measured
(Zerihun 2004) free surfaces
and bottom pressure heads
under transcritical flow
over a trapezoidal weir for
a q = 0.06128 m2/s, b,
c q = 0.07102 m2/s
Fig. 11.10 Undular waves

over an obstacle
kinematic and dynamic boundary conditions (Castro-Orgaz et al. 2015;

Castro-Orgaz and Hager 2017). For the continuity equation of clear water flow with
vanishing entrainment or deposition from above and below, this process yields the
exact result
@h
þ DivðuhÞ ¼ 0: ð11:62Þ
@t
Here, u = (U, V) is the depth-averaged velocity vector with U and V as

depth-averaged velocities in the x- and y-directions, respectively, Div() =
@ðÞ=@x þ @ðÞ=@y the two-dimensional horizontal divergence operator, and h the
flow depth. The general depth-integrated momentum equations in the (x, y, z) di-
rections are given by Castro-Orgaz et al. (2015) and Castro-Orgaz and Hager
(2017). Following Serre (1953), the (u, v) velocity components in the horizontal (x,
y) directions are approximated by their depth-averaged values, that is,
Zzs Zzs
1 1
uðx; y; z; tÞ U ðx; y; tÞ ¼ udz; vðx; y; z; tÞ V ðx; y; tÞ ¼ vdz: ð11:63Þ
h h
zb zb
The vertical velocity is then approximately computed by integrating the continuity

equation in the vertical direction, resulting in (Castro-Orgaz et al. 2015)
@ @
wðx; y; z; tÞ ¼ ½U ðz zb Þ ½V ðz zb Þ ¼ Div½uðz zb Þ
@x @y
ð11:64Þ
@zb @zb
¼ wb gðDivuÞ; wb ¼ U þV ¼ u Gradðzb Þ;
@x @y
where η = z – zb and wb is the vertical velocity at the bed. Within this level of
mathematical approximation, boundary layers are neglected, and a slip velocity at
the bed is accepted for depth-averaged modeling purposes. If we replace u and
v everywhere by U and V in the general depth-integrated (x, y) momentum equa-
tions of Castro-Orgaz and Hager (2017), and turbulence stresses are neglected, then
the resulting inviscid equations can be expressed as a system of partial differential
equations in general conservative form as (Castro-Orgaz and Hager 2017) [see
Eqs. (1.24), (1.35) and (1.36)]
@U @F @G
þ þ @y0¼ S;
0 1 0
@t
1
@x 1
h Uh Vh 0 1
Rzs 0
B C B 2 B C
U ¼ @ Uh A; F ¼ @U hþ
1
pdz C
A; G¼B
VUh
Rzs C; S¼
@zb
q1 @ pb @x A:
q @ 2 A
zb V h þ q pdz
1
pb @z b
Vh UVh zb @y
ð11:65Þ
In these equations, U is the dependent variable vector, F and G are the fluxes in the
x- and y-directions for non-hydrostatic pressure conditions, respectively, and S is
the source term. The pressure distribution p(z) is needed for model closure. The
vertical pressure distribution for depth-independent horizontal velocity components
and inviscid flow is given by (Castro-Orgaz et al. 2015) [see Eq. (1.60)]
2 z 3 2 z 3
Zzs Zs Zs
@ @ 4 @
pðx; y; z; tÞ ¼ qgðh gÞ þ q wdz þ q U wdz5 þ q 4V wdz5 qw2
@t @x @y
z z z
@I
¼ qgðh gÞ þ q þ qDivðIuÞ qw2 :
@t
ð11:66Þ
The auxiliary variable I is determined using Eq. (11.64) as

0 1
Zzs Zzs Zzs
I¼ wðx; y; zÞdz ¼ Div½uðz zb Þdz ¼ Div@u ðz zb ÞdzA þ uh Gradðh þ zb Þ
z z z

h2 g2
¼ Div u þ uh Gradðh þ zb Þ
2
h2 g2
¼ DivðuÞ þ ðh gÞu Gradðzb Þ;
2
ð11:67Þ
where Grad() = @ðÞ=@x þ @ðÞ=@y is the 2D horizontal gradient operator.

Equations (11.66) and (11.67) determine pressure at the vertical position z when-
ever the horizontal depth-averaged velocity components u = (U, V) are known. The
general system of Eqs. (11.65)–(11.67) must be manipulated to produce a form
suitable for numerical implementation. Using Eq. (11.67), we may deduce
(Castro-Orgaz and Cantero-Chinchilla 2019)
"( ) #
h2 g2
DivðIuÞ ¼ Div DivðuÞ u þ uðh gÞu Gradðzb Þ
2
"( ) #
h2 g2
¼ Div DivðuÞ u þ Div½u fðh gÞu Gradðzb Þg :
2 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} B
A
ð11:68Þ
The terms A and B are determined after some elementary operations by

"( ) #
h2 g2 h2 g2 h2 g2
A ¼ Div DivðuÞ u ¼ u Grad½DivðuÞ ½DivðuÞ2
2 2 2
hDivðuÞu GradðhÞ gDivðuÞu Gradðzb Þ;
ð11:69Þ
and
B ¼ Div½u fðh gÞu Gradðzb Þg ¼ ½DivðuÞu Gradðzb Þðh gÞ

ð11:70Þ
þ ½u Gradðzb Þu Gradðh þ zb Þ þ fu Grad½u Gradðzb Þgðh gÞ:
Further, one gets from Eq. (11.64)
w2 ¼ ½wb gðDivuÞ2 ¼½u Gradðzb Þ2 þ ½DivðuÞ2 g2 2½u Gradðzb ÞDivðuÞg;

ð11:71Þ
and resorting again to Eq. (11.67)

@I @ h2 g2 @
¼ DivðuÞ þ ½ðh gÞu Gradðzb Þ
@t @t 2 @t
2
@u h g2 ð11:72Þ
¼ Div þ h2 ½DivðuÞ2 þ hDivðuÞu GradðhÞ
@t 2
@u
DivðhuÞu Gradðzb Þ þ ðh gÞ Gradðzb Þ;
@t
in which @h=@t has been replaced by −Div(uh) via Eq. (11.62). The vertical
pressure distribution can now be evaluated from Eq. (11.66) collecting
Eqs. (11.68)–(11.72). The result after some manipulation is
2
p 2 @u h g2
¼ gðh gÞ þ ½DivðuÞ Div u Grad½DivðuÞ
q @t 2
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
f1
ð11:73Þ
@u
þ Gradðzb Þ þ u Grad½u Gradðzb Þ ðh gÞ:
@t
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
f2
For 2D waves propagating over horizontal terrain f2 = 0 results (Hutter and

Castro-Orgaz 2016)
2
p @u h g2
¼ gðh gÞ þ ½DivðuÞ2 Div u Grad½DivðuÞ : ð11:74Þ
q @t 2
For 1D waves, it reduces to the original result of Serre (1953)

( )
p @U 2 @ 2 U @ 2 U h2 g2
¼ gð h gÞ þ U 2 : ð11:75Þ
q @x @x@t @x 2
Using Eq. (11.73), the pressure integral is
Zh
p h2 h3 h2
dg ¼ g þ f1 þ f2 ; ð11:76Þ
q 2 3 2
0
and the bottom pressure becomes

pb
¼ gh þ f1 h2 þ f2 h: ð11:77Þ
q
Inserting Eqs. (11.76)–(11.77) into Eqs. (11.65), the general 2D inviscid Serre or
fully non-linear Boussinesq equations are obtained. These equations are used to model
fully non-linear and weakly dispersive water waves. These results presented here are in
agreement with those obtained by Nadiga et al. (1996) using a perturbation method.
11.3.2 One-Dimensional Serre Equations
For 1D water waves propagating over arbitrary topography, Eqs. (11.65) and
(11.73) read
@U @F
þ ¼ S;
@t @x
h Uh 0
U¼ ; F¼ 2 ; S ¼ @z b
;
Uh U h þ 2 gh þ 3 f1 h þ 2 f2 h
2 1 2 1 3 1 @x gh þ f1 h2 þ f2 h
ð11:78Þ
where the 1D non-hydrostatic coefficients are

@U 2 @ 2 U @2U
f1 ¼ U 2 ;
@x @x@t @x
ð11:79Þ
@U @zb @ zb
2
@U @zb
f2 ¼ þ U2 2 þ U :
@t @x @x @x @x
Serre (1953) derived 1D equations for turbulent, weakly dispersive, fully

non-linear water waves over nearly horizontal and flat topography. The same
system was obtained by Su and Gardner (1969) for a horizontal bed applying the
irrotational flow assumption and by Green and Naghdi (1976a, b) for 2D flows over
arbitrary terrain using the theory of directed fluid sheets. The system is called the
Serre-Green-Naghdi (SGN) equations in coastal engineering, but in civil and
environmental engineering these are referred to as the Serre equations or simply
Boussinesq equations (Soares-Frazão and Zech 2002; Soares-Frazão and Guinot
2008; Mohapatra and Chaudhry 2004; Cantero-Chinchilla et al. 2016). They apply
to model a variety of fully non-linear and weakly dispersive wave phenomena, like
Fig. 11.11 Undular bore in

Severn River (courtesy of
Mark Humpage; a collection
of undular bore photos is
available at https://www.
markhumpage.com/Mother-
Nature/Severn-Bore/)
the undular bore (Peregrine 1966, 1967, 1972) (Fig. 11.11). The equations are
discussed by Barthelemy (2004), Cienfuegos et al. (2006), Dias and Milewski
(2010), and Bonneton et al. (2011) for coastal engineering problems.
Equations (11.78)–(11.79), the 1D inviscid Serre equations for weakly dispersive
and fully non-linear water waves over topography, were derived and numerically
solved by Seabra-Santos et al. (1987). The x-momentum equation in Eq. (11.78)
can be rewritten as
@ ðUhÞ @M pb @zb
þ ¼ ; ð11:80Þ
@t @x q @x
where M is the momentum function, given by

" 2 #
1 @U @2U @2U 1 3 @U @zb @ 2 zb @U @zb 1 2
M ¼ U h þ gh2 þ
2
U 2 h þ þ U2 2 þ U h;
2 @x @x@t @x 3 @t @x @x @x @x 2
ð11:81Þ
and the bottom pressure is

" 2 #
pb @U @2U @2U 2 @U @zb @ 2 zb @U @zb
¼ gh þ U 2 h þ þ U2 2 þ U h:
q @x @x@t @x @t @x @x @x @x
ð11:82Þ
Note that M = gS. If unsteadiness is dropped, e.g., @ðÞ=@t and U = q/h is used to
transform the U-derivatives into h-derivatives, Eqs. (11.81)–(11.82) reduce to
Eqs. (11.22) and (11.21), respectively. This means that the steady version of the
SGN equations derived here equals the steady Picard iteration model. The imme-
diate implication is that Eq. (11.19), an extended Bernoulli-type equation, is the
integral form of Eq. (11.78) in steady state. It has important applications for water
wave flow problems. It is common practice while developing unsteady flow solvers
of the SWE to test their ability to converge to steady flow solutions. The conver-
gence to steady transcritical flow over a weir is a common and widely accepted test
(see Sect. 9.9.5). This issue was exported to non-hydrostatic Boussinesq-type sol-
vers forcing a version of these models with the dispersive terms deactivated to fight
with this test and to demonstrate their convergence to the steady flow solution.
However, it is more logic and stringent to test the ability of a non-hydrostatic
unsteady flow solver to converge to the correct steady non-hydrostatic flow solu-
tion. This opens the path of using Eq. (11.19) as a means to generate “exact” steady
non-hydrostatic flow solutions and to test the ability of non-hydrostatic solvers of
Eq. (11.78) to converge to these solutions. We remark then that the relevance of the
steady flow solver of Eq. (11.19) is therefore not only its ability to reveal the role
on non-hydrostaticity in weir flow, but also their utility to generate steady flow
solutions of the more general unsteady flow system.
For numerical computations, it is desirable to collect in a single vector all the
terms with temporal derivatives. This task is accomplished here. Consider the
following identity

@ h3 @U h3 @ 2 U @h @U
¼ þ h2 ; ð11:83Þ
@t 3 @x 3 @t@x @t @x
that permits to write

@ h3 @ 2 U @ @ h3 @U @h @U
¼ h2
@x 3 @t@x @x @t 3 @x @t @x
3 ð11:84Þ
@ 2
h @U @ 2 @ ðUhÞ @U
¼ þ h :
@x@t 3 @x @x @x @x
Using this identity, it is possible to split terms with temporal derivatives. Following
the same strategy for the remaining terms with time derivatives in M and pb, the
following alternative form of Eq. (11.78) is obtained
@W @F
þ ¼ S þ Sd ;
@t @x
h Uh
W¼ ; F¼ ; ð11:85Þ
r U 2 h þ 12 gh2

@zb 0 @zb 0 @D 0
S¼ ; Sd ¼ ;
@x gh @x p1 @x 1
where

1 3 @2U 2 @U @h @zb @zb @h 1 2 @ 2 zb
r ¼ Uh h h þ Uh þ þ Uh ; ð11:86Þ
3 @x2 @x @x @x @x @x 2 @x2
" #
@U 2 @2U 1 3 @U @zb 2 @ zb 1 2
2
D¼ U 2 h þ U þU h
@x @x 3 @x @x @x2 2
ð11:87Þ
@ ðUhÞ @zb @U
þ U h h;
@x @x @x
" #
@U 2 @2U 2 @U @zb 2 @ zb
2
p1 ¼ U 2 h þ U þU h
@x @x @x @x @x2
ð11:88Þ
@ ðUhÞ @zb @U
þ U h :
@x @x @x
Equation (11.86) is a Helmholtz equation for the non-hydrostatic velocity field. In

Eq. (11.85), all the time derivatives are collected in a single time-stepping term, and
the remaining contributions originating from non-hydrostatic pressure with only
spatial derivatives are treated as source term Sd. The SWE are obtained from the
SGN equations by setting Sd = 0 and r = Uh. An alternative to the SGN model is
the VAM (vertically averaged and moment equations) model (Khan and Steffler
1996a, b). The accuracy of this system for simulating water waves involving both
large amplitudes and high dispersion effects was demonstrated (Cantero-Chinchilla
et al. 2018). The SGN equations can be improved to model short waves using the
method by Nwogu (1993). This technique is, however, beyond the scope of this
chapter.
11.3.3 Solitary Wave
Consider waves of permanent form propagating over still water with celerity
c. Using the Galilei transformation X = x − ct for a wave displacement in the
positive x-direction over a horizontal bed, the Serre–Green–Naghdi equations,
Eq. (11.78), take the form (Castro-Orgaz and Hager 2017)
@h @ðUhÞ
c þ ¼ 0; ð11:89Þ
@X @X

@ðUhÞ @ 1 2 1
c þ gh þ U 2 h þ UX2 UUXX þ cUXX h3 ¼ 0: ð11:90Þ
@X @X 2 3
Integration of Eq. (11.89) yields the progressive discharge q

q ¼ ðU cÞh ¼ const: ð11:91Þ
Using Eq. (11.91), (11.90) is written in the simplified form

@h @ 1 2 1
c2 þ gh þ U 2 h þ UX2 ðU cÞUXX h3 ¼ 0: ð11:92Þ
@X @X 2 3
The solitary wave is a solution of Eq. (11.92) subject to the boundary conditions
h ! ho and U ! 0 for x ! ±∞, with ho as still water depth. Integrating twice, the
1=2
final result is with Fo ¼ q= gh3o (Serre 1953; Benjamin and Lighthill 1954;
Castro-Orgaz and Hager 2017)
" 1=2 #
hðx; tÞ 3F2o 3 ðx ctÞ
¼ 1 þ F2o 1 sech2 : ð11:93Þ
ho Fo 2ho
Using the conditions h = ho and U = 0 at x ! ±∞, the progressive discharge

q verifies the identity
q ¼ ½U ð X Þ chð X Þ ¼ cho ; ð11:94Þ
from which the depth-averaged velocity is

ho
U ðx; tÞ ¼ c 1 : ð11:95Þ
hðx; tÞ
The maximum flow depth at the solitary wave crest is obtained from Eq. (11.93) at
x − ct = 0 as
c2 h2o c2
hmax ¼ ho F2o ¼ ho ¼ : ð11:96Þ
gh3o g
This then yields

1=2 1=2 hmax 1=2 1=2 A 1=2
c ¼ ðghmax Þ ¼ ðgho Þ ¼ ðgho Þ 1þ ; ð11:97Þ
ho ho
where A = is the solitary wave amplitude. Another useful relation is thus

A
F2o ¼ 1þ : ð11:98Þ
ho
For given values of ho and A, the degree of non-linearity of the solitary wave is
fixed. The celerity c and Fo are thus deduced from Eqs. (11.97)–(11.98), and,
resorting to Eqs. (11.93) and (11.95), the functions h(x, t) and U(x, t) are fully
determined. This analytical solution is of great utility to check the quality of

numerical models to solve Eq. (11.78).
Joseph Valentin Boussinesq was born on March 15, 1842, at St. André-
de-Sangonis, France, and passed away on February 19, 1929, in Paris. He was
self-taught, starting his scientific writing in 1865. He thereby took into
consideration during his long career all branches of mathematical physics
except for electro-magnetism. After having served as teacher at various
colleges of France, he was appointed in 1873 Lecturer at the University of
Lille. In 1886, Boussinesq was appointed to the chair of mechanics at the
famous Sorbonne University, Paris, taking over in 1896 as professor of
mathematical physics at Collège de France.
Boussinesq’s life work in hydraulics is outstanding but extremely hard to

follow, given his complicated writing style. His colleagues Alfred A. Flamant
(1839–1915) or Auguste Boulanger (1866–1923) were able to present in their
books a more popular approach of Boussinesq’s ideas. His 1872 paper
explains the observations of solitary waves of John Scott Russell (1808–
1882) from a physical perspective thereby overcoming the many attempts
offered in the past decades. It was noted that many hydraulic phenomena
could only be explained by inclusion of the streamline curvature effects. This
paper particularly attracted the interest of Adhémar Barré de Saint-Venant
(1797–1886), who in 1871 had published his famous paper on the shallow
water equations, yet by assuming hydrostatic pressure and uniform velocity
distributions. The monumental 1877 Essay of Boussinesq made his name
definitely known to the hydraulics community given the large number of
relevant problems discussed. In the 1880s, he started in addition a close
collaboration with Henry Bazin (1829–1917) on the weir flow features, for
which streamline curvature effects again are significant. In hydraulics, this
collaboration between the then best experimenter and scientist marked the
start of engineering hydraulics, leading to the close relationship between
scientists in mathematical physics and hydraulics in the twentieth century.
The outstanding merits of Boussinesq were awarded by his nomination to the
mechanics chair at Sorbonne, taking over the chair of his intimate colleague
de Saint-Venant.
11.3.4 MUSCL-Hancock Scheme
The system of Eq. (11.85) is solved here using a finite volume-finite difference
method based on the MUSCL-Hancock scheme, a second-order accurate model in
space and time (see Sect. 9.7). Boussinesq-type water wave propagation models are
solved for coastal engineering applications using fourth-order accurate schemes in
space, and third-order accuracy or fourth-order accuracy in time. The reason for
imposing such accuracy is that truncation errors originating from the discretization
to second-order accuracy of the Saint-Venant type leading terms can induce
numerical dispersion. This problem is serious for large-scale simulations using
sparse time-space meshes (Wei et al. 1995). However, using a fine mesh in
second-order accurate schemes, this effect disappears or is significantly limited.
There are good reasons for testing a second-order non-hydrostatic scheme, how-
ever: The current knowledge of the finite volume technology for the solution of the
SWE is in a state of maturity permitting the development of a large variety of 1D
and 2D solvers worldwide. Research groups developed their own codes and verified
them extensively. If the inclusion of non-hydrostatic effects into a solver is offered
in a simple way, implying only an improvement of an existing SWE solver, there is

a great probability that this development will reach a large echo in the hydraulics
community. Driven by this line of thinking, it is then logic to investigate how
accurate can be a second-order non-hydrostatic module introduced in a
second-order SWE solver. A numerical experiment is conducted here solving the
SGN equations with the MUSCL-Hancock scheme originally developed for the
SWE. The ability of the model to replicate steady flow over a weir will be inves-
tigated, and the aptness of the second-order accuracy of the scheme will be assessed
against a third-order in time, fourth-order in space solver with an architecture
similar to that used in Boussinesq-type models for ocean research. To solve
Eq. (11.85), a splitting approach is used. First, an advection step is tackled dis-
carding Sd. For a rectangular control volume in the x-t plane, the integral form of
Eq. (11.85) then reads (Toro 2001)
Dt
Wadv ¼ Wki Fi þ 1=2 Fi1=2 þ DtSi : ð11:99Þ
i
Dx
Here, Δt and Δx are the step sizes in the x and t axes, respectively, k refers to the
time level, i is the cell index in the x-direction, and Fi+1/2 is the numerical flux
crossing the interface i + 1/2 between cells i and i + 1. Note that the advection step
as defined here includes the bed-slope source term directly. The topography source
term is discretized for a well-balanced scheme as [see Eq. (9.99)]

@zb 0 hi þ 1=2 þ hi1=2 zbi þ 1=2 zbi1=2 0
Si ¼ g h ¼ g: : ð11:100Þ
@x i 1 2 Dx 1
The solution process starts with the cell-averaged values of conserved variables at
time level k, Uki . For second-order space accuracy, a piecewise linear reconstruction
is conducted within each cell (Toro 2001). Linear slopes resulting from the
reconstructed solution must be limited to avoid spurious oscillations near discon-
tinuities. Let letters L and R denote the reconstructed variables at the left and right
sides of a cell interface; the resulting values of U at each of its sides are with DUi
and DUi þ 1 as the limited jumps (Toro 2001) [see Eqs. (9.122)–(9.125)]
1 1
ULi þ 1=2 ¼ Uki þ DUi ; URi þ 1=2 ¼ Ukiþ 1 Uki DUi þ 1 : ð11:101Þ
2 2
The surface gradient method (SGM) (Zhou et al. 2001) is adopted, consisting of
the reconstruction of the free surface elevation, instead of the water depth h. The
minmod limiter is used in the tests presented here. In the MUSCL-Hancock method,
an evolution of boundary-extrapolated values ULi þ 1=2 and URi þ 1=2 at interface i + 1/2
over half the time step is conducted to regain second-order accuracy in time. Based
on a Taylor series expansion in space and time, interface values are then given by
(Toro 2001) [see Eq. (9.134)]
L Dt h L i Dt
Ui þ 1=2 ¼ ULi þ 1=2 F Ui þ 1=2 F URi1=2 þ Si ;
2Dx 2 ð11:102Þ
R Dt h L i Dt
Ui þ 1=2 ¼ Ui þ 1=2
R
F Ui þ 3=2 F URi þ 1=2 þ Si þ 1 :
2Dx 2
L R
With these evolved boundary-extrapolated variables Ui þ 1=2 and Ui þ 1=2 defining states
L and R, the numerical flux is computed using the HLL approximate Riemann
solver as (Toro 2001) [see Eq. (9.44)]
8
> FL if SL 0
>
<
Fi þ 1=2 ¼ ; if SL 0 SR : ð11:103Þ
>
> SR SL
:
FR if SR 0
Note that the Riemann problem corresponding to the homogeneous version of

Eq. (11.85), @W=@t þ @F=@x, implies the determination of a numerical flux Fi+1/2
as function of two states WL and WR (Montecinos et al. 2016). However, we rather
assume that this numerical flux is determined by the left and right states of vector
U. While this is incorrect, it is a common practice in phase resolving
Boussinesq-type models. Keeping this limitation in mind, the computational results
will be critically looked at. Here, FL and FR are the fluxes computed at states L and
R. Robust wave speed estimates SL and SR for a wet bed are given by (Toro 2001)
[see Eqs. (9.45)–(9.46)]
SL ¼ UL cL kL ; SR ¼ UR þ cR kR ; ð11:104Þ
where c = (gh)1/2, and kK(K = L, R) is

8 1=2
>
< 1 h ðh þ hK Þ
h [ hK
kK ¼ 2
h2K : ð11:105Þ
>
:
1 h hK
The flow depth at the star region of the Riemann problem at each interface h* is
(Toro 2001) [see Eq. (9.47)]
2
1 1 1
h ¼ ð cL þ cR Þ þ ð U L U R Þ : ð11:106Þ
g 2 4
For the dry-bed problem, the celerity of the signals is given by [see Eqs. (9.48)–
(9.49)]
SL ¼ UR 2cR ðif hL ¼ 0Þ; SR ¼ UL þ 2cL ðif hR ¼ 0Þ: ð11:107Þ
For stability in time of the explicit scheme, the Courant–Friedrichs–Lewy number

CFL must be less than unity (Toro 2001). Given the value of CFL, Dt is determined
at time level k from [see Eq. (9.29)]
2 3
6 Dx
Dt ¼ CFL4 7
5: ð11:108Þ
k 1=2
max Ui þ ðghki Þ
Once the solution of Eq. (11.99) is available, the value obtained for the flow depth
is hk+1; the auxiliary variable r, however, must be updated to include the effect of
Sd. Here, we will use a predictor–corrector finite-difference scheme to incorporate
Sd in the solution. First, a predictor step is conducted as

@zb @D adv
rpi ¼ radv þ Dt p 1 : ð11:109Þ
i
@x @x i
All the spatial derivatives in the non-hydrostatic source term (linked to p1, D, and
its gradient) are approximated using second-order central finite-differences. In
general, a derivative is discretized in this work as

@mf 1 Xþk
¼ x k fk ; ð11:110Þ
@xm i Dxm k
where the weights are given in Table 11.2. The only exception is the first derivative
of U, which is discretized in the predictor step using a forward finite-difference as
(Mohapatra and Chaudhry 2004)
Ui þ 1 Ui
Ux ¼ : ð11:111Þ
Dx
Once rpi is determined, the predicted velocity field must be resolved. This is
accomplished by solving an elliptic problem posed by Eq. (11.86). The process is
explained below, given that it must be also applied after the correction step.
The corrector step is given by

@zb @D p
rki þ 1 ¼ radv þ Dt p 1 ; ð11:112Þ
i
@x @x i
Table 11.2 Weighting factors for discretization of spatial derivatives in non-hydrostatic terms
based on Abramowitz and Stegun (1972) and Fornberg (1988)
Type of Order of Order of Weighting factor xk at nodes
derivative derivative accuracy k = −2 k = −1 k = 0 k =+1 k = +2
Centered 1 2 0 −1/2 0 +1/2 0
Centered 1 4 1/12 −2/3 0 2/3 −1/12
Centered 2 2 0 1 –2 1 0
Centered 2 4 −1/12 4/3 –5/2 4/3 −1/12
Upwind 1 1 – −1 1 – –
Upwind 1 2 1/2 −2 3/2 – –
which is adopted as the final step. The first derivative of U is discretized in the
corrector step with a backward finite-difference as (Mohapatra and Chaudhry 2004)
Ui Ui1
Ux ¼ : ð11:113Þ
Dx
Once the values of r are determined at each finite volume for the new time level,
the following elliptic problem is stated at each cell of the computational domain
rki þ 1 ¼ d1 Ui1
kþ1
þ d2 Uik þ 1 þ d3 Uikþþ11 ; ð11:114Þ
which is obtained by discretizing Eq. (11.86) using second-order central

finite-differences. Here
h3i h2i
d1 ¼ 2
þ ðhi þ 1 hi1 Þ;
3ðDxÞ 4ðDxÞ2
2h3i
d2 ¼ hi þ þ a;
3ðDxÞ2
h3i h2i
d3 ¼ 2
ðhi þ 1 hi1 Þ;
3ðDxÞ 4ðDxÞ2

ðzbi þ 1 zbi1 Þ ðzbi þ 1 zbi1 Þ ðhi þ 1 hi1 Þ ðzbi þ 1 2zbi þ zbi1 Þ
a ¼ hi þ þ h2i :
2Dx 2Dx 2Dx ðDxÞ2
ð11:115Þ
Equation (11.115) produces an algebraic relation containing the unknown velocity

values at the new time level for the three surrounding cells i − 1, i, and i + 1.
Equation (11.115) for the cells i = 2 to N – 1 is assembled, producing a tridiagonal
system of equations, which is easily solved using the Thomas algorithm (Hoffman
2001). Once the velocity Uik þ 1 is determined, the cell-averaged discharge qki þ 1 is
evaluated. It is necessary to incorporate two additional cells at the left and right
ends of the computational domain, called ghost cells (LeVeque 2002; Toro 2009).
Physical boundary conditions are transmitted to the computational model through
these cells. Here, the simplest way of implementing boundary conditions is exposed
using ghost cells with a zero-order extrapolation from the interior solution. Care
should be taken to ensure that spurious waves are not generated when implementing
boundary conditions (LeVeque 2002). Here, transmissive boundary conditions are
implemented allowing waves to enter and leave the computational domain freely.
Mignot and Cienfuegos (2009) incorporated the boundary conditions in a
Serre-type model neglecting the dispersive effects at those cells by using the
characteristic form of the SWE. Here, we follow the same concept but with a rather
different treatment.
Given that dispersive effects will be neglected at cells i = 1 and i = N, the source
term there is Sd = 0. In this way, there are no dispersive effects at those cells
influencing the solution at cells i = 2 and i = N – 1 through the discretization of

@D=@x while solving Eq. (11.85). At the subcritical inflow section, one boundary
condition must be prescribed, as known from the method of characteristics
(Henderson 1966). The other variable needed at the ghost cell is obtained by
extrapolation from the interior solution (LeVeque 2002). This corresponds to the
inlet discharge for weir flow. Thus, we implemented in the upstream ghost cell
qk1 þ 1 ¼ qinlet ;
ð11:116Þ
hk1 þ 1 ¼ hk2 þ 1 :
At the outflow section i = N, a transmissive boundary condition is implemented to

let the waves leave the computational domain as
qkNþ 1 ¼ qkN1
þ1
;
ð11:117Þ
hkNþ 1 ¼ hkN1
þ1
:

1. A longitudinal finite volume mesh with cell width Dx is defined, and the bed
profile function zb(x) is used to define the bed elevation at the cell faces. The
bed elevation at the cell center is taken as the average of cell faces, consistent
with a linear variation of the bed profile within the cell. The bed derivatives for
the dispersive terms are evaluated using the computed coordinates at cell
centers by second-order central finite-differences.
2. The initial conditions are set. These can be arbitrary provided that they generate
the desired transcritical steady flow profile.
3. Boundary conditions are set using Eqs. (11.116)–(11.117).
4. The value of Dt is computed by Eq. (11.108).
5. The MUSCL-Hancock reconstruction is conducted using the vector U.
6. With boundary-extrapolated values of U, the local Riemann problems formed at
cell interfaces are solved with the HLL Riemann solver, resulting in Fi+1/2.
7. The water depth at time hk+1 is computed from the first component of
Eq. (11.99).
8. The bed-slope source term of the momentum equation is determined from
Eq. (11.100) using the definitive values of h.
9. The advection estimation of the conservative variable r is determined from the
second component of Eq. (11.99).
10. A predictor step is conducted to evaluate Sd. Equation (11.109) is applied with
a second-order estimation of the spatial derivatives.
11. Once rp is available at all cells, the tridiagonal system is solved using the
Thomas algorithm to obtain the predicted velocities [see Eq. (11.114)].
12. A corrector step is conducted to evaluate Sd. Equation (11.112) is applied with
a second-order estimation of the spatial derivatives.
13. Once rk+1 is available at all cells, the tridiagonal system is solved using the
Thomas algorithm to obtain the final velocities [Eq. (11.114)].
14. If the actual time is equal to the final time, then stop.
15. Go back to step 3 for a new time loop.
11.3.5 High-Order Scheme
In this section, a high-resolution scheme in both space and time O(Dt3, Dx4) is
developed to assess the second-order MUSCL-Hancock scheme. For the time
stepping, a third-order strong stability preserving (SSP) Runge–Kutta scheme is
given by (Gottlieb et al. 2001)
h i
ð1Þ ð1Þ ð1Þ ð1Þ
Wi ¼ Wki þ L Uki Dt þ ðSd Þi Dt ) Ui ¼ E 1 Wi ;
h i h i
ð2Þ ð1Þ ð1Þ ð2Þ ð2Þ ð2Þ
Wi ¼ Wi þ L Ui Dt þ ðSd Þi Dt ) Ui ¼ E 1 Wi ;
h i
ð3Þ ð2Þ ð3Þ ð3Þ
Wi ¼ 34 Wki þ 14 Wi ) Ui ¼ E 1 Wi ; ð11:118Þ
h i h i
ð4Þ ð3Þ ð3Þ ð4Þ ð4Þ ð4Þ
Wi ¼ Wi þ L Ui Dt þ ðSd Þi Dt ) Ui ¼ E 1 Wi ;
h i
ð4Þ ðk þ 1Þ
Wki þ 1 ¼ 13 Wki þ 23 Wi ) Uki þ 1 ¼ E1 Wi :
Here, E[ ] is the elliptic operator linked to Eq. (11.114), and L[] is the finite
volume-finite difference spatial operator
1
L½Ui ¼ Fi þ 1=2 Fi1=2 þ Si : ð11:119Þ
Dx
In the space operator L(), U is reconstructed with fourth-order accuracy, but the
elliptic operator E() is maintained with second-order accuracy to preserve the
tridiagonal structure of the linear system of equations determining the
non-hydrostatic velocity field. Sd must be updated with U at the corresponding time
stage, so iteration is needed. A high-order total variation diminishing monotone
upstream centred scheme for conservation laws (MUSCL-TVD-4th) is adopted to
reconstruct the solution (Erduran et al. 2005). The local Riemann problem at each
cell interface is then determined by the vector U at its left (L) and right (R) sides
from

1 1
ULiþ 1=2 ¼ Ui þ uðr1 ÞD Ui1=2 þ 2u D Ui þ 1=2 ; ð11:120Þ
6 r1

1 1
URiþ 1=2 ¼ Ui þ 1 2uðr2 ÞD Ui þ 1=2 þ u D Ui þ 3=2 ; ð11:121Þ
6 r1
where the operators used are defined by
1
D Ui þ 1=2 ¼ DUi þ 1=2 DUi þ 3=2 2DUi þ 1=2 þ DUi1=2 ; ð11:122Þ
6
DUi1=2 ¼ minmod½DUi1=2 ; DUi þ 1=2 ; DUi þ 3=2 ; ð11:123Þ
DUi þ 1=2 ¼ minmod½DUi þ 1=2 ; DUi þ 3=2 ; DUi1=2 ; ð11:124Þ
DUi þ 3=2 ¼ minmod½DUi þ 3=2 ; DUi1=2 ; DUi þ 1=2 ; ð11:125Þ
DUi þ 1=2 ¼ Ui þ 1 Ui : ð11:126Þ
The minmod function of 3 numbers is given by
minmod½a; b; c ¼ signðaÞmax½jaj; 2signðaÞb; 2signðaÞc: ð11:127Þ
The van Leer limiting function is used here
ri þ j ri j D Ui þ 1=2 D Ui þ 3=2
uðri Þ ¼ ; r1 ¼ ; r2 ¼ : ð11:128Þ
1 þ j ri j D Ui1=2 D Ui þ 1=2
The surface gradient method is applied to reconstruct the water surface elevation.
Once the reconstruction step is finished, the numerical flux Fi+1/2 is estimated with the
HLL approximate Riemann solver. The dispersive source term Sd (with the exception
of Ux) is discretized using fourth-order accurate central finite-differences (Table 11.2)
for the cells i = 3 to N – 2. At the cells i = 2 and N – 1, Sd is discretized using
second-order differences, and dispersive effects are deactivated at the ghost cells i = 1
and i = N (Sd = 0) to avoid any non-hydrostatic influence transmitted to the cells
i = 2 and i = N − 1 through the gradient @D=@x = (Di+1−Di−1)/(2Δx). The
derivative Ux is computed using upwind differencing of first order at cells i = 2 and
N – 1, and of second order at cells i = 3 to N – 2 (Table 11.2).
1. A longitudinal finite volume mesh with cell width Dx is defined and the bed
profile function zb(x) is used to define the bed elevation at the cell faces. The
bed elevation at the cell center is taken as the average of cell faces, consistent
with a linear variation of the bed profile within the cell. The bed derivatives for
the dispersive terms are evaluated using the computed coordinates at cell
centers by second-order central finite-differences at cells i = 2 and N – 1, and
with fourth-order central differences at the other cells.
2. The initial conditions are set.

3. Boundary conditions are set using Eqs. (11.116)–(11.117).
4. The value of Dt is computed by Eq. (11.108).
5. The fourth-order MUSCL reconstruction is conducted using the vector U.
6. With boundary-extrapolated values of U, the local Riemann problems formed at
cell interfaces are solved with the HLL Riemann solver, resulting in Fi+1/2.
7. The bed-slope source term of the momentum equation is determined from
Eq. (11.100).
8. The dispersive source term Sd is evaluated by second-order central
finite-differences at cells i = 2 and N – 1, and with fourth-order central dif-
ferences at the other cells.
9. The space operator L(U) is determined from Eq. (11.119).
10. The elliptic operator E(U) is inverted using Eq. (11.114).
11. Repeat steps 5–10 for the third-order SSP Runge–Kutta time stepping using
Eq. (11.118), with iteration of Sd at each time stage.
12. If the actual time is equal to the final time, then stop.
13. Go back to step 3 for a new cycle.
11.4 Unsteady Flow Test Cases
11.4.1 Flow over Round-Crested Weir
Figure 11.12 displays the unsteady wave evolution for the symmetrical hump of
shape zb = 0.2exp[−0.5(x/0.24)2] (m) according to Sivakumaran et al. (1983). The
unit discharge is 0.11197 m2/s. The up- and downstream boundary sections are
located at x = ± 3 m. The mesh used is of Δx = 0.01 m with CFL = 0.4. For
comparative purposes, the same simulation was conducted solving the SWE. For
the simulations shown in Fig. 11.12, a dam break numerical setup is implemented;
a gate is located at coordinate x = 1 m. The water is static, with free surface
elevations of 0.34 and 0.03 m up- and downstream of the gate, respectively. The
gate is removed instantaneously at time t = 0; the steady discharge is introduced at
the inlet section as a pulse and kept constant during the simulation. The numerical
model used is implemented in a code available on the file “MUSCLHancock_weir.
xls,” in Chap. 12.
At time t = 0, the dam break initial conditions generate a downstream traveling
dispersive shock wave and an upstream traveling rarefaction wave (Fig. 11.12a, b).
Compare the SGN and the SWE, where a solitary wave at the shock front is
predicted by the former. At the upstream section, the addition of the discharge pulse
generates another wave. In Fig. 11.12c, an undular wave pattern upstream of the
weir is visible, and a dam break front with undulations according to the SGN
equations. The SWE predict a sharp shock front, as expected from the second-order
MUSCL-Hancock scheme. In Fig. 11.12d, the upstream wave train is still active,
Fig. 11.12 Evolution of water surface profile h(x) at various times t in unsteady weir flow
[adapted from Castro-Orgaz and Cantero-Chinchilla (2019)]
11.4 Unsteady Flow Test Cases 477
Fig. 11.12 (continued)
while supercritical flow develops along the downstream weir face, with a hydraulic
jump near the toe of the weir. The undular waves of the dam break front are
progressively leaving the computational domain. In Fig. 11.12e the upstream water
waves continue their activity looking for equilibrium in the upstream water depths,
while the hydraulic jump phenomena increase their intensity at the downstream side
of the obstacle. Dispersive effects gain weight, and a large amplitude initial wave is
formed at the jump front, as is typical for undular hydraulic jumps. Figure 11.12f
shows that the upstream waves start to diminish in wave amplitude, and that the
undular hydraulic jump is pushed out of the computational domain by the super-
critical flow on the downstream weir face. Figure 11.12g finally shows the
steady-state results obtained from the simulation.
Before discussing the steady flow results, the adequacy of the implemented
transmissive boundary condition is assessed in Fig. 11.13. Here, the computational
Fig. 11.13 Evaluation of transmissive boundary condition
results are shown for t = 2.5 s using a domain of double length downstream of the
weir crest. Note that the waves are freely leaving the computational domain without
significant alterations. A more accurate approach would require the inclusion of
sponge layers at both ends of the domain (Cantero-Chinchilla et al. 2018).
The steady flow results generated by the unsteady flow solver are compared in
Fig. 11.14 with the “exact” steady flow solution previously presented in Fig. 11.7.
Note that the two simulations are in excellent agreement both for the free surface
and the bed pressure profiles. The bed pressure head for the unsteady flow solver
was determined using a finite-difference discretization of Eq. (11.82) based on the
computed vector U at each time t. Therefore, the proposed generalization of the
MUSCL-Hancock scheme of the SWE is able to deal with this problem introducing
the simple implementation of an additional source term and a tridiagonal equation
solver.
It is common practice in water wave simulations for ocean research to use
high-resolution schemes, typically of order O(Dt3, Dx4). This is done to reduce
truncation errors and to avoid numerical dispersion. A question may be well asked
if this is also necessary while producing steady non-hydrostatic flow solutions. It

steady and unsteady flow
solvers [adapted from
Castro-Orgaz and
Cantero-Chinchilla (2019)]

steady and unsteady flow
solvers using coarser meshes
in the unsteady flow model
[adapted from Castro-Orgaz
and Cantero-Chinchilla
(2019)]
is shown in Fig. 11.14 that the O(Dt2, Dx2) scheme produces a highly accurate
steady flow solution. Now, we test if this accuracy is maintained using a coarse
mesh. A new unsteady flow simulation was conducted dividing the numbers of
computational cells in Fig. 11.15, e.g., using a mesh of only Δx = 0.1 m, and
imposing the maximum time step for stability by trial and error, which was found at
CFL = 0.7. The results shown in Fig. 11.15a reveal that even for this coarse mesh,
results are still very good. The conclusion is that second-order accuracy is enough
for the simulation of steady non-hydrostatic flows, and centred discretizations
produce highly accurate results both for steady and unsteady solvers. Finally,
Fig. 11.15b presents the results for Δx = 0.2 m and CFL = 0.25. With this mesh,
the total number of cells is 30, and the flow over the obstacle is represented only by
9 of these cells. There is an obvious loss of accuracy for this extremely coarse mesh,
but the prediction is still in fair agreement with the exact results.
A comparison of the MUSCL-Hancock scheme with the third-order RK/
fourth-order MUSCL scheme at t = 1 s and 2.5 s is displayed in Fig. 11.16, using
the same mesh adopted to produce Fig. 11.12. The two are in good agreement, with
some deviations at the undular jump on the tailwater, indicating that the mesh
adopted for MUSCL-Hancock simulation is adequate, but should be refined to get
more precise results. The high-resolution model is implemented in a code available
on the file “rk3MUSCL4_weir.xls,” in Chap. 12. In general, a possible practical
Fig. 11.16 Comparison of MUSCL-Hancock method with the rk3MUSCL4 solver during
transient flow
strategy for using an O(Dt2, Dx2) SWE scheme improved with dispersive terms is to
run a first simulation, and, if extreme waves are expected in any portion of the
computational domain, to refine locally the mesh there to increase the model res-
olution (Popinet 2015).
11.4.2 Solitary Wave Propagation
The solitary wave [Eq. (11.93)] is a solution of the SGN equations in which
non-linear and dispersive effects are balanced. Non-linear effects are present in the
SWE, but this model is dispersionless. The SGN are fully non-linear and weakly
dispersive Boussinesq equations. Thanks to the full non-linearity in the
non-hydrostatic terms, there is no limitation on the wave amplitude modeled by
SGN equations. However, the equations are weakly dispersive, and, thus, only long
waves are correctly propagated. A solitary wave of large amplitude is therefore a
possible theoretical solution within the domain of validity of the SGN equations.
However, exact balancing of non-linear and dispersive effects is necessary to
produce a wave of permanent form traveling over still water. As the solitary wave is
an analytical solution of Eq. (11.78) resulting from such balancing, a numerical
Fig. 11.17 Propagation of solitary wave of A/ho = 0.2 and ho = 1 m using the MUSCL-Hancock
scheme at t = 20 s
scheme to solve them must be able to preserve this equilibrium while propagating
an analytical solitary wave as input data. The ability of the MUSCL-Hancock
scheme to propagate a solitary wave is investigated in Fig. 11.17. A test case
involving ho = 1 m and A = 0.2 m (Fig. 11.17a) is considered in Fig. 11.17b–e at
t = 20 s after initiation of routing. The numerical model is implemented in a code
available on the file “MUSCLHancock_solitarywave.xls,” in Chap. 12.
A comparison between the analytical and numerical solutions as functions of
Δx and CFL indicates that if progressively reducing both, it is possible to find a
mesh where the numerical prediction perfectly matches the theoretical expression
Fig. 11.18 Comparison of a rk3MUSCL4 solver with b MUSCL-Hancock solver for solitary
wave propagation
(Fig. 11.17e). The MUSCL-Hancock scheme for the SGN equations is thus able to
preserve the balance between non-linear and dispersive effects. This is not the case
for the same scheme applied to the SWE (Fig. 11.17f), given that dispersive effects
are absent. After 20 s of routing, the SWE transform the input solitary wave into a
shock.
A test case involving ho = 1 m and A = 0.5 m is considered in Fig. 11.18 at
t = 15 s after initiation of routing. A mesh with Δx = 0.15 m and CFL ¼ 0:8 was
used in the rk3MUSCL4 numerical model, with the results shown in Fig. 11.18a.
The numerical model is implemented in a code available on the file
“rk3MUSCL4_solitarywave.xls,” in Chap. 12. Note that the high-resolution model
produces an excellent result. The same test with the same mesh was implemented in
the MUSCL-Hancock scheme, with the results shown in Fig. 11.18b. There is an
obvious distortion of the solitary wave profile for this mesh, indicating the need to
refine the mesh in the O(Dt2, Dx2) scheme. This comparison was conducted for
a high-amplitude wave, A/ho = 0.5, which clearly reveals the benefits of an
O(Dt3, Dx4) scheme for this extreme non-linearity.
The balance between non-linearity and dispersion in non-hydrostatic waves can
experimentally be observed in surges. An abrupt gate closure generates a surge
propagating in the upstream direction (Rouse 1961). If the wave amplitude is much
less than the final depth, the surge is undular (Fig. 11.19a, b) (Rouse 1961), from
where the quoted balance between non-linear and dispersion effects becomes
apparent. If the amplitude increases, there will be a state where the first wave
breaks, but there is still an undular wave in the tailwater (Fig. 11.19c, d). If the
wave amplitude is much higher than the final depth, a distinctly turbulent broken
wave is generated (Fig. 11.19e, f) (Rouse 1961). The SWE are able to simulate
broken surges, whereas the SGN equations can simulate undular waves. A criterion
for wave breaking is needed to switch from the SWE to the SGN equations and thus
predict the complete wave evolution.
Fig. 11.19 Generation of

surges after gate closure in a
flume; surge displacement is
from right to left a, b undular
surge, c, d undular surge with
incipient wave breaking, e,
f broken surge (taken from
movie Fluid motion in a
gravitational field, by Rouse
(1961), IIHR-Hydroscience &
Engineering, The University
of Iowa)
11.4.3 Dam Break Wave
The experimental data of Ozmen-Cagatay and Kocaman (2010) at various nor-

malized times T = t(g/ho)1/2 starting at abrupt gate removal are considered in
Fig. 11.20 for a subcritical dam break wave test (hd/ho > 0.138) in a horizontal
channel. Its upstream water depth was ho = 0.25 m, the flume width 0.3 m, and the
downstream water depth for this series hd = 0.10 m. Simulations were conducted
with the MUSCL-Hancock scheme for the SGN equations using CFL ¼ 0:5 and
Dx = 0.01 m. The wave profile is well predicted up to T = 4.01, but the detailed
wave shape is affected by turbulence afterward, a feature not accounted for in the
inviscid SGN model. The first solitary wave like undulations of the dispersive bore
is in fact in excellent agreement with measurements. Experiments and numerical
simulations using the MUSCL-Hancock scheme for a transcritical dam break wave
test (hd/ho < 0.138) are considered in Fig. 11.21, with hd = 0.025 m, using CFL ¼
0:5 and Dx = 0.01 m. For this test, the position of the surge is reasonably well
predicted up to T = 2.38, yet the effects of turbulence are strong in the supercritical
portion of the wave profile.
In this test, the predicted first solitary wave undulation is again in excellent
agreement with measurements. The precise wave shape is not well predicted, given
the irregularities produced by turbulence in the experiments. However, for the low
tailwater depth hd in this test, the amplitude of the undulations is small and the
numerical solution is in general satisfactory.
To verify the accuracy of the results obtained with the MUSCL-Hancock solver,
the SGN equations were solved again for the test of Figs. 11.20 and 11.21 using the
rk3MUSCL4 scheme. The output of both schemes is compared in Figs. 11.22 and
11.23, showing minor discrepancies in the secondary wave train.
In general, for hd =ho [ 0:5 an undular wave front is generated, e.g., see
Fig. 11.24, where an undular wave front is clearly visible. Upon reflection
(Fig. 11.24b) at the end of the flume, the undular front is again clearly visible
(Fig. 11.24c).

MUSCL-Hancock scheme for
the SGN equations with
experimental data
(Ozmen-Cagatay and
Kocaman 2010) for a
subcritical dam break wave

MUSCL-Hancock scheme for
the SGN equations with
experimental data
(Ozmen-Cagatay and
Kocaman 2010) for a

with rk3MUSCL4 scheme for
the subcritical dam break
wave of Fig. 11.19

with rk3MUSCL4 scheme for
the transcritical dam break
wave of Fig. 11.20
References 489
Fig. 11.24 Non-hydrostatic dam break wave for hd =ho [ 0:5 (photos by O. Castro-Orgaz at
hydraulic flume of University of Córdoba) a Initial condition b Undular wave front c Reflected
undular wave front
References
Abramowitz, M., & Stegun, I. A. (1972). Handbook of mathematical functions with formulas,
graphs, and mathematical tables (10th ed.). New York: Wiley.
Barthelemy, E. (2004). Nonlinear shallow water theories for coastal waters. Surveys In
Geophysics, 25(3), 315–337.
Benjamin, T. B., & Lighthill, M. J. (1954). On cnoidal waves and bores. Proceedings of the Royal
Society London, A, 224, 448–460.
Wehrschwelle [Distributions of discharge and energy over a parabolic-shaped weir].
Mitteilungen der Forschungsanstalt für Schiffahrt, Wasser- und Grundbau, Berlin, Heft 7,
5–72 (in German).
Bonneton, P., Barthelemy, E., Chazel, F., Cienfuegos, R., Lannes, D., Marche, F., et al. (2011).
Recent advances in Serre-Green-Naghdi modelling for wave transformation, breaking and
runup processes. European Journal of Mechanics B/Fluids, 30(6), 589–597.
Boussinesq, J. (1877). Essai sur la théorie des eaux courantes [Memoir on the theory of flowing
water]. Mémoires présentés par divers savants à l’Académie des Sciences, Paris 23, 1–660; 24,
1-60 (in French).
Cantero-Chinchilla, F. N., Castro-Orgaz, O., Dey, S., & Ayuso, J. L. (2016). Nonhydrostatic dam
break flows I: Physical equations and numerical schemes. Journal of Hydraulic Engineering,
142(12), 04016068.
Cantero-Chinchilla, F. N., Castro-Orgaz, O., & Khan, A. A. (2018). Depth-integrated nonhydro-
static free-surface flow modelling using weighted-averaged equations. International Journal
for Numerical Methods in Fluids, 87(1), 27–50.
Castro-Orgaz, O., & Cantero-Chinchilla, F. N. (2019). Non-linear shallow water flow over
topography with depth-averaged potential equation. Environmental Fluid Mechanics, in Press.
Castro-Orgaz, O., & Chanson, H. (2016). Closure to Minimum specific energy and transcritical
flow in unsteady open channel flow. Journal of Irrigation and Drainage Engineering, 142(10),
07016015.
Castro-Orgaz, O., & Hager, W. H. (2009). Curved streamline transitional flow from mild to steep
slopes. Journal of Hydraulic Research, 47(5), 574–584.
Castro-Orgaz, O., & Hager, W. H. (2013). Velocity profile approximations for two-dimensional
potential channel flow. Journal of Hydraulic Research, 51(6), 645–655.
Castro-Orgaz, O., & Hager, W. H. (2014). 1D modelling of curvilinear free surface flow:
Generalized Matthew theory. Journal of Hydraulic Research, 52(1), 14–23.
Castro-Orgaz, O., & Hager, W. H. (2017). Non-hydrostatic free surface flows. Advances in
Geophysical and Environmental Mechanics and Mathematics. 696 p. Berlin: Springer. https://
doi.org/10.1007/978-3-319-47971-2
Castro-Orgaz, O., Hutter, K., Giraldez, J. V., & Hager, W. H. (2015). Non-hydrostatic granular
flow over 3D terrain: New Boussinesq-type gravity waves? Journal of Geophysical Research:
Earth Surface, 120(1), 1–28.
Cienfuegos, R., Barthélemy, E., & Bonneton, P. (2006). A fourth-order compact finite volume
scheme for fully nonlinear and weakly dispersive Boussinesq-type equations. Part I: Model
development and analysis. International Journal for Numerical Methods in Fluids, 51(11),
1217–1253.
Dias, F., & Milewski, P. (2010). On the fully non-linear shallow-water generalized Serre
equations. Physics Letters A, 374(8), 1049–1053.
Erduran, K. S., Ilic, S., & Kutija, V. (2005). Hybrid finite-volume finite-difference scheme for the
solution of Boussinesq equations. International Journal for Numerical Methods in Fluids, 49
(11), 1213–1232.
Fenton, J. D. (1996). Channel flow over curved boundaries and a new hydraulic theory.
Proceeding of 10th IAHR APD Congress, Langkawi (pp. 266–273). Malaysia.
Fornberg, B. (1988). Generation of finite difference formulas on arbitrarily spaced grids.
Mathematics of Computation, 51(184), 699–706.
Gottlieb, S., Shu, C.-W., & Tadmor, E. (2001). Strong stability-preserving high-order time
discretization methods. SIAM Review, 43(1), 89–112.
Green, A. E., & Naghdi, P. M. (1976a). Directed fluid sheets. Proceedings of the Royal Society.
London, A, 347, 447–473.
Green, A. E., & Naghdi, P. M. (1976b). A derivation of equations for wave propagation in water of
variable depth. Journal of Fluid Mechanics, 78, 237–246.
Hager, W. H. (1985). Critical flow condition in open channel hydraulics. Acta Mechanica, 54(3–4),
157–179.
Hager, W. H., & Hutter, K. (1984a). Approximate treatment of plane channel flow. Acta
Mechanica, 51(3–4), 31–48.
Hager, W. H., & Hutter, K. (1984b). On pseudo-uniform flow in open channel hydraulics. Acta
Mechanica, 53(3–4), 183–200.
Hasumi, M. (1931). Untersuchungen über die Verteilung der hydrostatischen Drücke an
Wehrkronen und -Rücken von Überfallwehren infolge des abstürzenden Wassers [Studies on
the distribution of hydrostatic pressure distributions at overflows due to water flow]. Journal
Department of Agriculture, Kyushu Imperial University 3(4), 1–97 (in German).
References 491

Marcel Dekker.
Hosoda, T., & Tada, A. (1994). Free surface profile analysis on open channel flow by means of
1-D basic equations with effect of vertical acceleration. JSCE Annual Journal of Hydraulic
Engineering, 38, 457–462.
Hutter, K., & Castro-Orgaz, O. (2016). Non-hydrostatic free surface flows: Saint Venant versus
Boussinesq depth integrated dynamic equations for river and granular flows. Continuous Media
with Microstructure (Vol. 2, Chapter 17). Heidelberg: Springer, Berlin.
Khan, A. A., & Steffler, P. M. (1996a). Vertically averaged and moment equations model for flow
over curved beds. Journal of Hydraulic Engineering, 122(1), 3–9.
Khan, A. A., & Steffler, P. M. (1996b). Modelling overfalls using vertically averaged and moment
Lesleighter, E., McPherson, B., Riddette, K., & Williams, J. (2008). Modelling procedures used
for the spillway upgrade for Lake Manchester Dam. In Proceeding of 2008 ANCOLD
Conference, Australian National Committe on Large Dams, Gold Coast.
Univ. Press.
Marchi, E. (1992). The nappe profile of a free overfall. Rendiconti Lincei, Matematica e
Applicazioni, Serie 9, 3(2), 131–140.
Marchi, E. (1993). On the free overfall. Journal of Hydraulic Research 31(6), 777–790; 32(5),
794–796.
Matthew, G. D. (1991). Higher order one-dimensional equations of potential flow in open
channels. Proc. ICE, 91(3), 187–201.
Mignot, E., & Cienfuegos, R. (2009). On the application of a Boussinesq model to river flows
including shocks. Coastal Engineering, 56(1), 23–31.
Mohapatra, P. K., & Chaudhry, M. H. (2004). Numerical solution of Boussinesq equations to
simulate dam-break flows. Journal of Hydraulic Engineering, 130(2), 156–159.
Montecinos, G. I., López-Rios, J. C., Lecaros, R., Ortega, J. H., & Toro, E. F. (2016). An
ADER-type scheme for a class of equations arising from the water-wave theory. Computers
and Fluids, 132, 76–93.
Montes, J. S. (1994). Potential flow solution to the 2D transition from mild to steep slope. Journal
of Hydraulic Engineering, 120(5), 601–621.
Nadiga, B. T., Margolin, L. G., & Smolarkiewicz, P. K. (1996). Different approximations of
shallow fluid flow over an obstacle. Physics of Fluids, 8(8), 2066–2077.
Naghdi, P. M. (1979). Fluid jets and fluid sheets: A direct formulation. In Proceedings of 12th
Symposium Naval Hydrodynamics (pp. 505–515). Washington D.C.: National Academy of
Sciences.
Naghdi, P. M., & Vongsarnpigoon, L. (1986). The downstream flow beyond an obstacle. Journal
of Fluid Mechanics, 162, 223–236.
Nwogu, O. (1993). Alternative form of Boussinesq equations for nearshore wave propagation.
Journal of Waterway Port Coastal and Ocean Engineering, 119(6), 618–638.
Onda, S., & Hosoda, T. (2004). Numerical simulation of the development process of dunes and
flow resistance. In Proceedings River Flow 2004 (pp. 245–252). London: T&F.
Peregrine, D. H. (1966). Calculations of the development of an undular bore. Journal of Fluid
Mechanics, 25(2), 321–330.
Peregrine, D. H. (1967). Long waves on a beach. Journal of Fluid Mechanics, 27(5), 815–827.
Peregrine, D. H. (1972). Equations for water waves and the approximations behind them.
In R. E. Meyer, (ed.), Waves on beaches and resulting sediment transport (pp. 95–122). San
Diego, CA: Academic Press
Popinet, S. (2015). A quadtree-adaptative multigrid solver for the Serre-Green-Naghdi equations.

Journal of Computational Physics, 302, 336–358.
Rouse, H. (1961). Fluid motion in a gravitational field, IIHR movies. Iowa: Univ.
Seabra-Santos, F. J., Renouard, D. P., & Temperville, A. M. (1987). Numerical and experimental
study of the transformation of a solitary wave over a shelf or isolated obstacle. Journal of Fluid
Mechanics, 176, 117–134.
Serre, F. (1953). Contribution à l’étude des écoulements permanents et variables dans les canaux
[Contribution to the study of steady and unsteady channel flows]. La Houille Blanche 8(6–7),
374–388; 8(12), 830–887 (in French).
Sivakumaran, N. S. (1981). Shallow flow over curved beds (Ph.D. thesis) Asian Institute of
Technology, Bangkok, Thailand.
Soares-Frazão, S., & Guinot, V. (2008). A second-order semi-implicit hybrid scheme for
one-dimensional Boussinesq-type waves in rectangular channels. International Journal for
Numerical Methods in Fluids, 58(3), 237–261.
Soares-Frazão, S., & Zech, Y. (2002). Undular bores and secondary waves: Experiments and
hybrid finite-volume modelling. Journal of Hydraulic Research, 40(1), 33–43.
Steffler, P. M., & Jin, Y. C. (1993). Depth-averaged and moment equations for moderately shallow
free surface flow. Journal of Hydraulic Research, 31(1), 5–17.
Su, C. H., & Gardner, C. S. (1969). KDV equation and generalizations. Part III. Derivation of
Korteweg-de Vries equation and Burgers equation. Journal of Mathematical Physics, 10(3),
536–539.
Thom, A., & Apelt, C. (1961). Field computations in engineering and physics. London: Van
Nostrand.
Toro, E. F. (2009). Riemann solvers and numerical methods for fluid dynamics: A practical
introduction. Berlin: Springer.
Vallentine, H. R. (1969). Applied hydrodynamics. London: Butterworths.
Van Dyke, M. (1975). Perturbation methods in fluid mechanics. Stanford, CA: The Parabolic
Press.
Wei, G., Kirby, J. T., Grilli, S. T., & Subramanya, R. (1995). A fully nonlinear Boussinesq model
for surface waves 1: Highly nonlinear unsteady waves. Journal of Fluid Mechanics, 294, 71–
92.
Zerihun, Y. T. (2004). A one-dimensional Boussinesq-type momentum model for steady rapidly
varied open channel flows (Ph.D. thesis). Melbourne University, Australia.
Zerihun, Y. T., & Fenton, J. D. (2006). One-dimensional simulation model for steady transcritical
free surface flows at short length transitions. Advances in Water Resources, 29(11), 1598–
1607.
Zhou, J. G., Causon, D. M., Mingham, C. G., & Ingram, D. M. (2001). The surface gradient
method for the treatment of source terms in the shallow-water equations. Journal of
Zhu, D. Z., & Lawrence, G. A. (1998). Non-hydrostatic effects in layered shallow water flows.
Journal of Fluid Mechanics, 355, 1–16.
Zhu, D. Z. (1996). Exchange flow through a channel with an underwater sill (Ph.D thesis). The
University of British Columbia, Vancouver, Canada.
Chapter 12
Numerical Library of Shallow Water
Equations
12.1 The Library
12.1.1 Introduction: Aims and Scope
Supporting material to the chapters of this book is a collection of source codes

explaining the theory and used to produce the solutions presented. The material was
originally prepared for students attending the course Operation of Rivers and
Reservoirs of the Environmental Hydraulics master program held at the University
of Córdoba, Spain. Most of the codes were used interactively during the lectures,
and others are the homework assignment requested to students. Its purpose is to
teach students how to make the important step from translating the theoretical
material received during the lectures to a code where all the logic sequence is
automatically implemented.
Originally, most of the codes were written in Fortran, but the average student
taking the lectures lacked from previous knowledge of this language. Thus,
MATLAB was tried as an alternative, and programs were therefore adapted, but the
problem of having licenses to let any student use the material on any computer
make this method unpractical. Finally, the approach suggested by Fenton (2010)
was found to be practical and useful; it consists in presenting the codes in Visual
Basic (VBA), which is a high-level programming language, such as Fortran, C++,
or Pascal. Codes can be entered easily via the use of Excel Macros, which is the
approach pursued in the material presented here. Excel is on most personal com-
puters of students; thus, VBA is available to them, whereas separate compilers are
necessary for other high-level languages (Fenton 2010).
Electronic supplementary material The online version of this chapter (https://doi.org/10.1007/

978-3-030-13073-2_12) contains supplementary material, which is available to authorized users.

https://doi.org/10.1007/978-3-030-13073-2_12
494 12 Numerical Library of Shallow Water Equations
The Visual Basic language is used to write scripts, and these scripts are run as
“Macros” in Excel. Prior knowledge of VBA is necessary, however. Given that the
language is fairly simple, this is not a real limitation. Excel itself is used as a means to
organize the input data and receive the output from the code; graphics are immedi-
ately displayed with the computational results. Codes are written as clear as possible
showing all steps. The use of subroutines is done to structure the computations into
separate operations. Programs are by no means optimized; they are simply made to
avoid losing any intermediate step. Further, most codes are tailored for a specific
application, and, thus, are not to be used for general purposes. However, the reader
may take pieces of the codes to develop his/her own general-purpose code. Despite
the codes are written in VBA, their translation to FORTRAN is fairly simple, and the
material can be used by readers as the base to develop codes for scientific purposes.
12.1.2 Summary of Available Codes
The codes available as supporting material are listed in Table 12.1. Below, examples
of the codes for steady and unsteady flows are given. The unsteady flow VBA codes
are adapted from the FORTRAN code HW_MUSH.F by Toro (2000, 2001).
12.2 Examples of Codes
12.2.1 Gradually Varied Flow Computation Using Newton–

Raphson Method
Public b, Q, z, numero, sections, A, f, BSUP, error, direc

Public hsal, Froude, R, Sf, E, n, So, p
Public dEdh, dSfdh, dfdh, dx, h2, h1, ho, xo
Public h, Fo, hnew
Public i As Integer, j As Integer
Sub main()
' Program for the computation of steady gradually varied flow profiles in trapezoidal
channels using
' the standard step method; implicit equation is solved applying Newton–Raphson
iteration
' by Oscar Castro-Orgaz
' read data
12.2 Examples of Codes 495
Table 12.1 Codes as supporting material of the book

File name Book Features of the code
chapter
criticaldepth_NR 2 Computation of critical depth in
trapezoidal section using
criticaldepth_NR_pipe 2 Computation of critical depth in
circular section using Newton–
Raphson method
hump_NR 2 Computation of ideal flow
profile over a hump using
Analytical_hump_subsuper 2 Computation of analytical flow
profile over a hump
conjugatedepth_trapezoidal 2 Computation of conjugate
supercritical depth in
trapezoidal section using
EM_sluicegate 2 Computation of the discharge
and supercritical depth at a
sluice gate with a rejected jump
using the Newton–Raphson
method
normaldepth_NR_Manning 3 Computation of normal depth
using Manning’s equation in
trapezoidal section applying
Lake_discharge 3 Computation of the discharge
released from a reservoir
Bresse_analytical 3 Analytical solution of GVF
equation for wide rectangular
channel
Bresse_RungeKutta4orden 3 Numerical integration of GVF
channel using the fourth-order
Runge–Kutta method
watersurfaceprofiles_RungeKutta4orden 3 Numerical integration of GVF
equation for trapezoidal channel
using the fourth-order Runge–
Kutta method
watersurfaceprofiles_Euler 3 Numerical integration of GVF
using the Euler method
watersurfaceprofiles_trapezoidalrule 3 Numerical integration of GVF
using the trapezoidal rule
(iterative method)
(continued)
Table 12.1 (continued)

chapter
Flowprofiles_NR 3 Computation of water surface
profiles using the standard step
method with Newton–Raphson
iteration
steepslope_RungeKutta4orden 3 Numerical integration of GVF
channel using the fourth-order
Runge–Kutta method
Parabolicweir_transcritical 4 Application of GVF equation to
transcritical flow over a
parabolic weir using the
method
Gaussianhump_transcritical 4 Application of GVF equation to
transcritical flow over a
Gaussian hump using the
method
sluicegate_hydraulicjump 4 Computation of transcritical
flow profile with a hydraulic
jump beyond a sluice gate
rk4 5 Numerical integration of
reservoir ODE using the
method
Euler 5 Numerical integration of
reservoir ODE using the Euler
method
Puls 5 Solution of reservoir routing by
Puls modified method
DamBreakAnalytical 6 Analytical solution of ideal dam
break waves over horizontal
bottom for dry- and wet-bed
conditions
DamBreak_Diffusive 7 Application of Lax diffusive
scheme to ideal dam break
waves
DamBreak_MacCormack_Jameson 7 Application of MacCormack
scheme to ideal dam break
waves with calibrated artificial
viscosity
DamBreak_MacCormack_TVD 7 Application of
MacCormack TVD scheme to
ideal dam break waves
(continued)

chapter
DamBreak_Unstable 7 Application of unstable scheme
to ideal dam break waves with
calibrated artificial viscosity
DamBreak_Upwind 7 Application of upwind scheme
to ideal dam break waves
Favre_FDM_TVD 7 Application of
MacCormack TVD scheme to
Favre waves
Favre_FDM_Jameson 7 Application of MacCormack
scheme with calibrated artificial
viscosity to Favre waves
Favre_FDM_TVD_BoundaryDiscont_Subcrit 7 Variant of MacCormack TVD
scheme implementing boundary
conditions for discontinuous
flow in subcritical conditions
Favre_FDM_TVD_BoundaryDiscont_Supercrit 7 Variant of MacCormack TVD
flow in supercritical conditions
Favre_FDM_TVD_BoundaryDiscont_Supercrit_Cont 7 Variant of MacCormack TVD
flow in supercritical conditions
with naive use of discretized
continuity equation
SurgeAnalytical 7 Analytical solution to positive
surge
ExactRiemannsolver_drybed 8 Solution of the Riemann
problem for dry-bed conditions
ExactRiemannsolver_wetbed 8 Solution of the Riemann
problem for wet-bed conditions
Partialgateopening 8 Solution of the partial gate
opening of a sluice gate for ideal
fluid flow and horizontal bed
DamBreakWave 9 Application of
MUSCL-Hancock scheme to
dam break waves over
horizontal beds
FavreWave 9 Application of
Favre waves over horizontal
beds
Hydraulicjump_gate 9 Application of
computation of the steady
transcritical flow profile beyond
a sluice gate
(continued)

chapter
Weir_DGM_parabolic 9 Application of
transcritical flow profile over a
parabolic weir
Weir_DGM_parabolic_MC 9 Same but with upstream water
depth computed with backward
characteristic
Weir_dry_terrain_SGM-DGM_MC 9 Same but allowing wetting and
drying processes
Weir_upwind 9 Application of first-order
upwind method to computation
of the steady transcritical flow
profile over a parabolic weir
Weir_DGM_Sivakumaran 9 Application of
transcritical flow profile over a
Gaussian hump
Dambreak_sill 9 Computation of the dam break
wave propagation over a
trapezoidal bottom sill using the
first-order upwind method
solitarywaverunup 9 Application of
computation of solitary wave
run-up on a slope
Movablebed_DamBreak 10 Computation of dam break
waves over erodible beds with
Movablebed_Dikebreaching 10 Computation of dike
overtopping erosion with
Matthew_SVK_implicit 11 Application of implicit FDM to
solve Matthew’s equation for
flow over a round-crested weir
Matthew_slopebreak 11 Application of implicit FDM to
flow over a slope break
Matthew_embankmentweir 11 Application of implicit FDM to
flow over a trapezoidal profile
weir
MUSCLHancock_weir 11 Application of
solve the SGN equations for
round-crested weir flow
(continued)

chapter
MUSCLHancock_solitarywave 11 Application of
compute the propagation of a
solitary wave
rk3MUSCL4_weir 11 Application of a high resolution
scheme (3rd-order Runge-Kutta
time stepping, 4th-order
MUSCL reconstruction) to
solve the SGN equations for
round-crested weir flow
rk3MUSCL4_solitarywave 11 Application of a high resolution
scheme (3rd-order Runge-Kutta
time stepping, 4th-order
MUSCL reconstruction) to
compute the propagation of a
solitary wave
backwater_MUSCLHancock 12 Computations of steady
gradually-varied flow profiles
with MUSCL-Hancock scheme
weir_dry_terrain_complex 12 Wetting and drying over
complex topography
steady rk4 12 rk4 solver for wide channel
conditions
frictioncomparison 12 Comparison of explicit, implicit
and semi-implicit discretizations
of friction source team.
Q = Worksheets(''hoja1'').Cells(3, 4).Value
b = Worksheets(''hoja1'').Cells(4, 4).Value
z = Worksheets(''hoja1'').Cells(5, 4).Value
numero = Worksheets(''hoja1'').Cells(6, 4).Value
sections = Worksheets(''hoja1'').Cells(7, 4).Value
So = Worksheets(''hoja1'').Cells(8, 4).Value
n = Worksheets(''hoja1'').Cells(9, 4).Value
dx = Worksheets(''hoja1'').Cells(10, 4).Value
ho = Worksheets(''hoja1'').Cells(11, 4).Value
xo = Worksheets(''hoja1'').Cells(12, 4).Value
direc = Worksheets(''hoja1'').Cells(13, 4).Value
' boundary point
Worksheets(''hoja1'').Cells(7, 7).Value = xo
Worksheets(''hoja1'').Cells(7, 8).Value = ho
' select direction of computation
If direc = 0 Then dx = −dx
h2 = ho
x = xo
' loop to go across all the channel sections
For j = 1 To sections
Newton
x = x − dx
h = h1
h2 = h1
Worksheets(''hoja1'').Cells(7 + j, 7).Value = x
Worksheets(''hoja1'').Cells(7 + j, 8).Value = h
Next j
End Sub
Sub Newton()
'Program to compute the unknown depth using Newton–Rapshon method
' Initiate variables
hsal = h2
section
Fo = So * dx – E − 0.5 * dx * Sf
h1 = h2
' Iterative computation of depth
For i = 1 To numero
hsal = h1
section
functions
hnew = h1 − f/dfdh
If Abs((hnew − h1) /h1) < 10 ^ −8 Then GoTo 10
h1 = hnew
Next i
10
End Sub
Sub section()
' variables of trapezoidal cross section
A = b * hsal + z * hsal ^ 2
BSUP = b + 2 * z * hsal
Froude = (Q ^ 2 /9.81 /A ^ 3 * BSUP) ^ 0.5
p = b + 2 * hsal * (1 + z ^ 2) ^ 0.5
R = A /p
E = hsal + Q ^ 2 /2 /9.81 /A ^ 2
Sf = n ^ 2 * Q ^ 2 /A ^ 2 /R ^ (4 /3)
End Sub
Sub functions()
' functions to apply the NR iteration method
f = E − 0.5 * dx * Sf + Fo
dEdh = 1 − Froude ^ 2
dSfdh = Sf * (−10 /3 * BSUP /A + 4 /3 /p * 2 * (1 + z ^ 2) ^ 0.5)
dfdh = dEdh − 0.5 * dx * dSfdh
End Sub
12.2.2 Dam Break Wave Code

' Define variables
Public x(5000), L, deltax
Public U(2, 5000), PIL(2), PIR(2), V
Public delta, t, dt
Public a, b, sig, d1, d2, factor
Public beta, Sf, nM, a1, a2, a3, Rh, width
Public dtdx, DU, DD, BEXT(2, 2, 5000), CDL(2), CDR(2), DL, VL, CL, DR, VR,
CR
Public FDL(2), FDR(2), Flux(2, 5000), DS, SL, SR, FHLL, U1, U2, c(5000),
Smax, SPELOC, CFL, Dup, Ddown
Public FIL(2), FIR(2), deltaflux, lambda, fric
Public TOLDEP
Public i As Integer, k As Integer, count As Integer
Public N As Integer, Tmax, xdam
Sub Main()
' Program for the computation of Dam Break waves using the Finite Volume method
with: MUSCL linear reconstruction,
' minmod/superbee slope limiter, HLL approximate Riemann solver, and Euler time
stepping
' Second order accuracy is achieved with Hancock's method
' Simplified test conditions are:
' * Horizontal channel
' * Rectangular cross section
' * Dry and wet fronts allowed
' * Friction computed using Manning's equation
' * Higher water level is on the left, so flow is from left to right
' * Waves can not reach the boundary sections of the domain
' by Oscar Castro-Orgaz,
' This code is based on the FORTRAN code HW_MUSH.F by Prof. E. Toro (Numerica
library, 2000)
-15
TOLDEP = 10
lambda = Worksheets("Input").Cells(11, 4).Value
fric = Worksheets("Input").Cells(13, 4).Value
' Read data for simulation

data
' Read initial conditions U(k,i) for conserved variables
initial
' Loop for transient computations
t=0
count = 0
10
' Compute new dt to satisfy the CFL condition
CFLcon
' Reconstruction of the solution within the cells U(x) using the cell-averaged values
U(k,i) obtained in previous time step
MUSCL
' Solve Riemann problem using the boundary extrapolated values of conserved
variables, after application
' of a limiter to avoid spurious oscillations
Riemann
' Compute cell-averaged conserved variables at new time level
evolution
' Check dry bed conditions
dry
' Include friction effects from source terms
' Implicit computation
If fric = 1 Then
friction
End If
' Explicit computation
If fric = 2 Then
friction2
End If
t = t + dt
count = count + 1
' Check final of transient computations
If t < Tmax Then GoTo 10
' Print water depth and discharge data
For i = 1 To N
Worksheets(''Input'').Cells(4 + i, 10).Value = x(i) − 0.5 * deltax
For k = 1 To 2
Worksheets(''Input'').Cells(4 + i, 10 + k).Value = U(k, i)
Next k
If U(1, i) > 0 Then
Worksheets(''Input'').Cells(4 + i, 13).Value = U(2, i) /U(1, i)
Else
Worksheets(''Input'').Cells(4 + i, 13).Value = 0
End If
Next i
End Sub
Sub evolution()
' Program to compute the cell-averaged vector U(k,i) at the new time level using the
computed vector F at interfaces
' Flux(k,i) is the vector of fluxes at interface i + 1/2 of cell i; k make reference to
conservation laws (1 = mass, 2 = momentum)
' The domain is composed of N+2 cells; only N-2 cells are computational
' The conserved variables at cells i = 1 and N are given by the boundary conditions
dtdx = dt /deltax
For k = 1 To 2
For i = 2 To N − 1
U(k, i) = U(k, i) − dtdx * (Flux(k, i) − Flux(k, i − 1))
Next i
Next k
End Sub
Sub data()
' Program to generate the longitudinal mesh of cells
' Cell i is defined in terms of a right (index i) and left hand side nodes (index i − 1)
L = Worksheets(''Input'').Cells(4, 4).Value
deltax = Worksheets(''Input'').Cells(5, 4).Value
' number of cells
N = L /deltax
x(1) = −L /2
' coordinates of right side of each cell
For i = 2 To N
x(i) = x(i − 1) + deltax
Next i
End Sub
Sub initial()
' Program to read simulation data and set initial values of the dam break problem
given the upstream (Dup) and downstream (Ddown) water depths
xdam = Worksheets(''Input'').Cells(6, 4).Value
Dup = Worksheets(''Input'').Cells(7, 4).Value
Ddown = Worksheets(''Input'').Cells(8, 4).Value
CFL = Worksheets(''Input'').Cells(9, 4).Value
Tmax = Worksheets(''Input'').Cells(10, 4).Value
nM = Worksheets(''Input'').Cells(12, 4).Value
width = Worksheets(''Input'').Cells(14, 4).Value
' Computational cells are from i = 2 to N − 1; ghost cells are i = 0, 1, N and
N + 1. Cells i = 0 and N + 1 are only
' used for a fictitious reconstruction of cells i = 1 and N
For i = 0 To N + 1
U(2, i) = 0
If x(i) <= xdam Then
U(1, i) = Dup
Else: U(1, i) = Ddown
End If
Next i
End Sub
Sub MUSCL()
' Linear reconstruction of the solution U(x) within each based on the cell-averaged
values,
' with slope limiters minmod or superbee
' U(k,i) at cells i = 1 and N is required to compute Fi + 1/2 at cells i = 1 and N − 1
in the program Riemann
For i = 1 To N
For k = 1 To 2
' Computation of jumps of conserved variables upstream (U) and downstream (D)
DU = U(k, i) − U(k, i − 1)
DD = U(k, i + 1) − U(k, i)
' Limitation of the jumps in conserved variables
limiters
' MUSCL linear reconstruction of conserved variables within a cell
PIL(k) = U(k, i) − 0.5 * delta

PIR(k) = U(k, i) + 0.5 * delta
Next k
For k = 1 To 2
' Temporal evolution of PIL and PIR (for second order time accuracy)
evolve
deltaflux = 0.5 * dtdx * (FIL(k) − FIR(k))
' Assignation of boundary-extrapolated (BEXT) values to temporal arrays
' BEXT(j,k,i) is the vector of boundary extrapolated values at cell i, with k as the
conservation
' law index (1 = mass, 2 = momentum) and j as the face side index (1 = left face,
2 = right face)
BEXT(1, k, i) = PIL(k) + deltaflux
BEXT(2, k, i) = PIR(k) + deltaflux
Next k
'preserve positivity in water depths
If BEXT(1, 1, i) < 0 Then
BEXT(1, 1, i) = 0
BEXT(1, 2, i) = 0
End If
BEXT(2, 1, i) = 0
BEXT(2, 2, i) = 0
End If
' avoid high velocity in data
If BEXT(1, 1, i) < TOLDEP Then
BEXT(1, 2, i) = 0
End If
BEXT(2, 2, i) = 0
End If
Next i
End Sub
Sub Riemann()
' Computation of Flux vector Fi + 1/2 at each interface using the approximate
Riemann solver HLL
' The flux Fi + 1/2 at cells i = 1 and N − 1 is required in the program evolution to
compute U(k,i) at cells i = 2 and N − 1
For i = 1 To N − 1
' Assignation of values for Riemann problem at interface i + 1/2 of cell i. CDR and
CDL are the vectors of variables U at the right (R)
' and left (L) side of the interface i + 1/2 of cell i
For k = 1 To 2
CDL(k) = BEXT(2, k, i)
CDR(k) = BEXT(1, k, i + 1)
Next k
' Transformation to physical variables flow depth (D), velocity (V), and celerity (C).
Check dry bed conditions
DL = CDL(1)
If DL = 0 Then
VL = 0
Else
VL = CDL(2) /DL
End If
CL = (9.81 * DL) ^ 0.5
DR = CDR(1)
If DR = 0 Then
VR = 0
Else
VR = CDR(2) /DR
End If
CR = (9.81 * DR) ^ 0.5
' Computation of vector fluxes FDL and FDR at interface i + 1/2 of cell i
FDL(1) = Flux1(CDL(2))
FDL(2) = Flux2(CDL(1), CDL(2))
FDR(1) = Flux1(CDR(2))
FDR(2) = Flux2(CDR(1), CDR(2))
' Compute wave estimates SL and SR for dry bed case
If DR < TOLDEP And DL > TOLDEP Then
SL = VL − CL
SR = VL + 2 * CL
GoTo 200
End If
If DL < TOLDEP And DR > TOLDEP Then
SL = VR − 2 * CR
SR = VR + CR
GoTo 200
End If
If DR < TOLDEP And DL < TOLDEP Then
SL = −TOLDEP
SR = TOLDEP
GoTo 200
End If
DS = (0.5 * (CL + CR) + 0.25 * (VL − VR)) ^ 2 /9.81
' Compute wave estimates SL and SR for wet bed case
If DS <= DL Then
SL = VL − CL
Else
SL = VL − CL * (0.5 * DS * (DS + DL)) ^ 0.5 /DL
End If
If DS <= DR Then
SR = VR + CR
Else
SR = VR + CR * (0.5 * DS * (DS + DL)) ^ 0.5 /DR
End If
200 ' Compute the Godunov intercell flux with HLL Riemann solver
' case 1
If SL > = 0 Then
For k = 1 To 2
Flux(k, i) = FDL(k)
Next k
End If
' case 2
If SL <= 0 And SR > = 0 Then
For k = 1 To 2
FHLL = SR * FDL(k) − SL * FDR(k) + SL * SR * (CDR(k) − CDL(k))
Flux(k, i) = FHLL /(SR − SL)
Next k
End If
' case 3
If SR <= 0 Then
For k = 1 To 2
Flux(k, i) = FDR(k)
Next k
End If
Next i
End Sub
Sub CFLcon()
' Computation of dt to satisfy the Courant-Friedrichs-Lewy (CFL) condition and get
a stable time stepping
' Computation of the wave celerity c at each cell i using the cell-averaged values U
(1,i) = flow depth(i)
Smax = 0
For i = 1 To N
' shallow water wave celerity
c(i) = (9.81 * U(1, i)) ^ 0.5
' Local speed; where U(2,i)/U(1,i) = water velocity (i). Check zero water depths
If U(1, i) = 0 Then
SPELOC = 0
Else
SPELOC = Abs(U(2, i) /U(1, i)) + c(i)
End If
If SPELOC > = Smax Then
Smax = SPELOC
End If
Next i
' New value of dt using the fixed value of the CFL number; i.e. 0.5 to 1
dt = CFL * deltax /Smax
' Reduce dt for early times given the approximate Smax computation
If count <= 5 Then
dt = 0.2 * dt
End If
' Check that dt+t is not greater than Tmax and recompute dt if required
If t + dt > Tmax Then
dt = Tmax − t
End If
dtdx = dt /deltax
End Sub
Public Function Flux2(U1, U2) As Double
' Computation of the flux vector component of momentum equation given vector U
If U1 = 0 Then
Flux2 = 0
Else
Flux2 = U2 ^ 2 /U1 + 0.5 * 9.81 * U1 ^ 2
End If
End Function
Public Function Flux1(U2) As Double
' Computation of the flux vector component of continuity equation given vector U
Flux1 = U2
End Function
Sub evolve()
' Computation of vector fluxes associated with PIL and PIR
FIL(1) = Flux1(PIL(2))
FIL(2) = Flux2(PIL(1), PIL(2))
FIR(1) = Flux1(PIR(2))
FIR(2) = Flux2(PIR(1), PIR(2))
End Sub
Sub limiters()
' Limitation of jumps in conserved variables
' If lambda = (1, Minmod limiter; 2, Superbee)
If DU * DD < 0 Then
delta = 0
Else
sig = Sgn(DU)
a = Abs(DU)
b = Abs(DD)
If a < b Then
d1 = a
Else
d1 = b
End If
d1 = lambda * d1
If a > b Then
d2 = a
Else
d2 = b
End If
If d1 < d2 Then
factor = d1
Else
factor = d2
End If
factor = factor * sig
delta = factor
End If
End Sub
Sub friction()
' Implicit computation of the friction effects
If nM = 0 Then
GoTo 250
End If
For i = 2 To N − 1
If U(1, i) > TOLDEP Then
Rh = U(1, i) * width /(2 * U(1, i) + width)
a1 = −9.81 * dt * nM ^ 2 /Rh ^ (4 /3)
a2 = −1
a3 = U(2, i) /U(1, i)
V = (−a2 − (a2 ^ 2 – 4 * a1 * a3) ^ 0.5) /(2 * a1)
U(2, i) = V * U(1, i)
Else
U(2, i) = 0
End If
Next i
250
End Sub
Sub friction2()
' Explicit computation of the friction effects
If nM = 0 Then
GoTo 300
End If
For i = 2 To N − 1
Rh = U(1, i) * width /(2 * U(1, i) + width)
Sf = nM ^ 2 * (U(2, i) /U(1, i)) ^ 2 /Rh ^ (4 /3)
U(2, i) = U(2, i) − 9.81 * U(1, i) * dt * Sf
Else
U(2, i) = 0
End If
Next i
300
End Sub
Sub dry()
' check dry bed conditions

' preserve positivity in water depths
For i = 2 To N − 1
If U(1, i) < 0 Then
U(1, i) = 0
U(2, i) = 0
End If
Next i
For i = 2 To N − 1
If U(1, i) < TOLDEP Then
U(2, i) = 0
End If
Next i
End Sub
12.2.3 Flooding over an obstacle code
' Define variables

Public x(5000), L, z(5000), S(2, 5000), deltax
Public U(3, 5000), PIL(3), PIR(3), V
Public delta, t, dt
Public a, b, sig, d1, d2, factor
Public beta, Sf, nM, a1, a2, a3
Public dtdx, DU, DD, BEXT(2, 2, 5000), CDL(2), CDR(2), DL, VL, CL, DR, VR,
CR
Public FDL(2), FDR(2), Flux(2, 5000), DS, SL, SR, FHLL, U1, U2, c(5000),
Smax, SPELOC, CFL, qo
Public FIL(2), FIR(2), deltaflux, lambda, fric, deltasource(2), Flim
Public xR(5000), xL(5000), zL(5000), zR(5000), zjump(5000), Froude(5000),

depth(5000), Uold(2, 5000), xu, xd, c1, c2, Qtarget
Public qadv(5000), hd, sn, aa1, bb1, cc1, Tmax
Public out As Integer, inlet As Integer, count As Integer
Public TOLDEP
Public i As Integer, k As Integer, N As Integer
Sub Main()
' Program for the computation of flow over a hump using the Finite Volume method
with: MUSCL linear reconstruction, minmod/superbee slope limiter, HLL
approximate Riemann solver, and Euler time stepping
' Second order accuracy is achieved with Hancock’s method
' Simplified test conditions are:
' * Parabolic bed profile
' * Rectangular cross section
' * Dry and wet fronts allowed
' * Hybrid reconstructed solution with SGM and DGM to grant
' the C-property and increase stability of wet-dry front propagation
' * Friction computed using Manning’s equation
' * Inlet discharge is constant
' by Oscar Castro-Orgaz
' This code is based on the FORTRAN code HW_MUSH.F by Prof. E. Toro
(Numerica library, 2000)
TOLDEP = 10-6
lambda = Worksheets(“Input”).Cells(11, 4).Value
fric = Worksheets(“Input”).Cells(13, 4).Value
out = Worksheets(“Input”).Cells(7, 4).Value
inlet = Worksheets(“Input”).Cells(5, 4).Value
Flim = Worksheets(“Input”).Cells(14, 4).Value
' Read data for simulation
data
' Define bed profile
bed
' Read initial conditions U(k,i) for conserved variables
' U(1,i)=flow depth; U(2,i)=discharge
initial
' Loop for transient computations
t=0
count = 0
10
' Store previous solution
For k = 1 To 2
For i = 1 To N
Uold(k, i) = U(k, i)
Next i
Next k
' Compute new dt to satisfy the CFL condition
CFLcon
' Reconstruction of the solution within the cells U(x) using the cell-averaged values
U(k,i) obtained in previous time step
MUSCL
' Solve Riemann problem using the boundary extrapolated values of conserved
variables, after application of a limiter to avoid spurious oscillations
Riemann
' Evaluate bed slope source term
bedslope
' Compute cell-averaged conserved variables at new time level
evolution
' Check dry bed conditions
dry
' Include friction effects from source terms
' Implicit computation
If fric = 1 Then
friction
End If
' Explicit computation
If fric = 2 Then
friction2
End If
' Set boundary conditions
boundary
t = t + dt
count = count + 1
' Check final of transient computations
If t <Tmax Then GoTo 10
' Print water and discharge data
For i = 1 To N
Worksheets(“Input”).Cells(4 + i, 10).Value = x(i) - 0.5 * deltax
Worksheets(“Input”).Cells(4 + i, 14).Value = z(i)
Worksheets(“Input”).Cells(4 + i, 15).Value = z(i) + U(1, i)
For k = 1 To 2
Worksheets(“Input”).Cells(4 + i, 10 + k).Value = U(k, i)
Next k
If U(1, i) > 0 Then
Worksheets(“Input”).Cells(4 + i, 13).Value = U(2, i) / U(1, i)
Else
Worksheets(“Input”).Cells(4 + i, 13).Value = 0
End If
Next i
End Sub
Sub bed()
' Definition of the bed profile
' A parabolic weir profile z=a1+a2x^* 2 is defined
a1 = 0.2
a2 = -0.05
' weir edges
xu = -(-a1 / a2) ^ 0.5
xd = (-a1 / a2) ^ 0.5
' weir profile
For i = 0 To N + 1
xL(i) = x(i) - 0.5 * deltax
xR(i) = x(i) + 0.5 * deltax
Next i
For i = 0 To N + 1
If xL(i) <xu Or xL(i) >xd Then
zL(i) = 0
Else
zL(i) = a1 + a2 * (xL(i)) ^ 2
End If
If xR(i) <xu Or xR(i) >xd Then
zR(i) = 0
Else
zR(i) = a1 + a2 * (xR(i)) ^ 2
End If
z(i) = 0.5 * (zL(i) + zR(i))
Next i
End Sub
Sub initial()
' Program to read simulation data
Qtarget = Worksheets(“Input”).Cells(6, 4).Value
hd = Worksheets(“Input”).Cells(8, 4).Value
CFL = Worksheets(“Input”).Cells(9, 4).Value
Tmax = Worksheets(“Input”).Cells(10, 4).Value
nM = Worksheets(“Input”).Cells(12, 4).Value
' Computational cells are from i=2 to N-1; gosh cells are i=0,1,N and N+1. Cells i=0
and N+1 are only used for a fictitious reconstruction of cells i=1 and N
'Initial condition is static water at elevation hd; dry portions of terrain allowed
For i = 0 To N + 1
If hd>= z(i) And x(i) <= 0 Then
U(1, i) = hd - z(i)
Else
U(1, i) = 0
End If
U(2, i) = 0
Next i
End Sub
Sub boundary()
' Use ghost cells to set boundary conditions
'INLET CONDITIONS
'———————————————————
' Subcritical flow at the inlet section with constant discharge

If inlet = 1 Then
'Inlet discharge
U(2, 1) = Qtarget
U(2, 0) = Qtarget
'Water depth is determined solving the C- backward characteristic
aa1 = 9.81 / (9.81 * Uold(1, 2)) ^ 0.5
bb1 = (Uold(2, 2) / Uold(1, 2)) - (9.81 / (9.81 * Uold(1, 2)) ^ 0.5 * Uold(1, 2))
cc1 = -Qtarget
U(1, 1) = (-bb1 + (bb1 ^ 2 - 4 * aa1 * cc1) ^ 0.5) / (2 * aa1)
U(1, 0) = U(1, 1)
'———————————————————————
' Transmissive condition upstream

ElseIf inlet = 2 Then
For k = 1 To 2
U(k, 1) = U(k, 2)
U(k, 0) = U(k, 3)
Next k
'——————————————————————
'Reflective condition upstream

Else If inlet = 3 Then
U(1, 1) = U(1, 2)
U(2, 1) = -U(2, 2)
U(1, 0) = U(1, 3)
U(2, 0) = -U(2, 3)
End If
' OUTLET CONDITIONS
'————————————————————
' Tailwater gate is fully open; transmissive condition

If out = 1 Then
For k = 1 To 2
U(k, N) = U(k, N - 1)
U(k, N + 1) = U(k, N - 2)
Next k
'——————————————————
' Tailwater gate is closed; reflective condition

Else If out = 2 Then
U(1, N) = U(1, N - 1)
U(2, N) = -U(2, N - 1)
U(1, N + 1) = U(1, N - 2)
U(2, N + 1) = -U(2, N - 2)
End If
End Sub
Sub evolution()
' Program to compute the cell-averaged vector U(k,i) at the new time level using the
computed vector F at interfaces
' Flux(k,i) is the vector of fluxes at interface i+1/2 of cell i; k make reference to
conservation laws (1=mass, 2=momentum)
' The domain is composed of N+2 cells; only N-2 cells are computational
' The conserved variables at cells i=1 and N are given by the boundary conditions
dtdx = dt / deltax
For k = 1 To 2
For i = 2 To N - 1
U(k, i) = U(k, i) - dtdx * (Flux(k, i) - Flux(k, i - 1)) + dt * (S(k, i))

Next i
Next k
End Sub
Sub data()
' Program to generate the longitudinal mesh of cells
' Cell i is defined in terms of a right (index i) and left hand side nodes (index i-1)
L = Worksheets(“Input”).Cells(3, 4).Value
deltax = Worksheets(“Input”).Cells(4, 4).Value
' number of cells
N = L / deltax
x(1) = -0.5 * L
' coordinates of right side of each cell
For i = 2 To N
x(i) = x(i - 1) + deltax
Next i
End Sub
Sub MUSCL()
' Linear reconstruction of the solution U(x) within each based on the cell-averaged
values, with slope limiters minmod or superbee
' U(k,i) at cells i=1 and N is required to compute Fi+1/2 at cells i=1 and N-1 in the
program Riemann
' SGM is needed to grant the C-property (at Froude=0), but the DGM is more robust
during propagation of wet-dry interfaces. Here a limit Froude number Flim is
defined to reconstruct using the SGM subcritical flows and with the DGM the
remaining cases, including the supercritical flow in the tip portion of a wet-dry front
' Surface Gradient method (SGM) is used for reconstruction of water surface ele-
vation for Froude <=Flim
' Depth Gradient method (DGM) is used for reconstruction of flow depths for
Froude >Flim
' Store water depths, surface elevations and Froude number
For i = 1 To N
depth(i) = U(1, i)
'Solution at cell i is vector U(k,i), with k=1 for the water depth, k=2 for the
discharge, and k=3 for the free surface elevation
U(3, i) = U(1, i) + z(i)
If U(1, i) = 0 Then
Froude(i) = 0
Else
Froude(i) = (U(2, i) ^ 2 / 9.81 / U(1, i) ^ 3) ^ 0.5
End If
Next i
For i = 1 To N
For k = 1 To 3
' Computation of jumps of conserved variables upstream (U) and downstream (D)
DU = U(k, i) - U(k, i - 1)
DD = U(k, i + 1) - U(k, i)
' Limitation of the jumps in conserved variables
limiters
' MUSCL linear reconstruction of conserved variables within a cell
PIL(k) = U(k, i) - 0.5 * delta
PIR(k) = U(k, i) + 0.5 * delta
Next k
' Values of the water depth at the interfaces of a cell for Froude<=Flim are com-
puted from SGM
If Froude(i) <= Flim Then
PIL(1) = PIL(3) - zL(i)
PIR(1) = PIR(3) - zR(i)
End If
' In a dry cell the reconstruction is not allowed; thus, water depths at interfaces are
set to zero. All the depths are zero in a cell marked computationally as dry. This is
true in a full dry cell, but incorrect in a partially-filled cell
If depth(i) <= TOLDEP Then
PIL(1) = 0
PIR(1) = 0
End If
' Ensure positivity in water depths at interfaces

If PIL(1) < 0 Then
PIL(1) = 0
End If
If PIR(1) < 0 Then
PIR(1) = 0
End If
If PIL(1) < TOLDEP Then
PIL(2) = 0
End If
If PIR(1) < TOLDEP Then
PIR(2) = 0
End If
For k = 1 To 2
' Temporal evolution of PIL and PIR (for second order time accuracy)
evolve
deltaflux = 0.5 * dtdx * (FIL(k) - FIR(k))
deltasource(k) = 0.5 * dtdx * deltax * deltasource(k)
' Assignation of boundary-extrapolated (BEXT) values to temporal arrays
' BEXT(j,k,i) is the vector of boundary extrapolated values at cell i, with k as the
conservation law index (1=mass, 2=momentum)and j as the face side index (1=left
face, 2=right face)
BEXT(1, k, i) = PIL(k) + deltaflux + deltasource(k)
BEXT(2, k, i) = PIR(k) + deltaflux + deltasource(k)
Next k
'preserve positivity in water depths
BEXT(1, 1, i) = 0
BEXT(1, 2, i) = 0
End If
BEXT(2, 1, i) = 0
BEXT(2, 2, i) = 0
End If
BEXT(1, 2, i) = 0
End If
BEXT(2, 2, i) = 0
End If
Next i
' At the interface of a full wet cell and a dry cell a jump in water depths may be
formed. However, the dry cell is in reality partially filled with water. That is, the dry
front is contained inside the computationally dry cell. Therefore, an unrealistic
dry-bed dam break wave problem at the interface of a full-wet and full-dry cells is
formed. Remember that a cell is marked computationally as dry with zero water
depth (cell-averaged value), but physically there is a portion of water inside this cell
if it is in contact with a full-wet cell
For i = 1 To N
'A reflective condition is imposed in the dry front if the water depth elevation in the
wet cell is below the average bed elevation in the (full) dry cell
If Froude(i) <= Flim Then
' dry front is on the right
If depth(i) < z(i + 1) - z(i) And depth(i + 1) <= TOLDEP Then
BEXT(1, 1, i + 1) = BEXT(2, 1, i)
BEXT(1, 2, i + 1) = -BEXT(2, 2, i)
End If
' dry front is on the left
If depth(i + 1) < z(i) - z(i + 1) And depth(i) <= TOLDEP Then
BEXT(2, 1, i) = BEXT(1, 1, i + 1)
BEXT(2, 2, i) = -BEXT(1, 2, i + 1)
End If
End If
Next i
End Sub
Sub Riemann()
' Computation of Flux vector Fi+1/2 at each interface using the approximate
Riemann solver HLL
' The flux Fi+1/2 at cells i = 1 and N - 1 is required in the program evolution to
compute U(k,i) at cells i = 2 and N - 1
For i = 1 To N - 1
' Assignation of values for Riemann problem at interface i+1/2 of cell i. CDR and
CDL are the vectors of variables U at the right (R) and left (L) side of the interface
i+1/2 of cell i
For k = 1 To 2
CDL(k) = BEXT(2, k, i)
CDR(k) = BEXT(1, k, i + 1)
Next k
' Transformation to physical variables flow depth (D), velocity (V), and celerity (C).
Check dry bed conditions
DL = CDL(1)
If DL = 0 Then
VL = 0
Else
VL = CDL(2) / DL
End If
CL = (9.81 * DL) ^ 0.5
DR = CDR(1)
If DR = 0 Then
VR = 0
Else
VR = CDR(2) / DR
End If
CR = (9.81 * DR) ^ 0.5
' Computation of vector fluxes FDL and FDR at interface i+1/2 of cell i
FDL(1) = Flux1(CDL(2))
FDL(2) = Flux2(CDL(1), CDL(2))
FDR(1) = Flux1(CDR(2))
FDR(2) = Flux2(CDR(1), CDR(2))
' Compute wave estimates SL and SR for dry bed case
If DR < TOLDEP And DL > TOLDEP Then

SL = VL - CL
SR = VL + 2 * CL
GoTo 200
End If
If DL < TOLDEP And DR > TOLDEP Then
SL = VR - 2 * CR
SR = VR + CR
GoTo 200
End If
If DR < TOLDEP And DL < TOLDEP Then
SL = -TOLDEP
SR = TOLDEP
GoTo 200
End If
DS = (0.5 * (CL + CR) + 0.25 * (VL - VR)) ^ 2 / 9.81
' Compute wave estimates SL and SR for wet bed case
If DS <= DL Then
SL = VL - CL
Else
SL = VL - CL * (0.5 * DS * (DS + DL)) ^ 0.5 / DL
End If
If DS <= DR Then
SR = VR + CR
Else
SR = VR + CR * (0.5 * DS * (DS + DL)) ^ 0.5 / DR
End If
200 ' Compute the Godunov intercell flux with HLL Riemann solver
' case 1
If SL >= 0 Then
For k = 1 To 2
Flux(k, i) = FDL(k)
Next k
End If
' case 2
If SL <= 0 And SR >= 0 Then
For k = 1 To 2
FHLL = SR * FDL(k) - SL * FDR(k) + SL * SR * (CDR(k) - CDL(k))
Flux(k, i) = FHLL / (SR - SL)
Next k
End If
' case 3
If SR <= 0 Then
For k = 1 To 2
Flux(k, i) = FDR(k)
Next k
End If
Next i
End Sub
Sub CFLcon()
' Computation of dt to satisfy the Courant-Friedrichs-Lewy(CFL) condition and get
a stable time stepping
' Computation of the wave celerity c at each cell i using the cell-averaged values
U(1,i)=flow depth(i)
Smax = 0
For i = 1 To N
' shallow water wave celerity
c(i) = (9.81 * U(1, i)) ^ 0.5
' Local speed; where U(2,i)/U(1,i)=water velocity (i). Check zero water depths
If U(1, i) = 0 Then
SPELOC = 0
Else
SPELOC = Abs(U(2, i) / U(1, i)) + c(i)
End If
If SPELOC >= Smax Then
Smax = SPELOC
End If
Next i
' New value of dt using the fixed value of the CFL number; i.e. 0.5 to 1
dt = CFL * deltax / Smax
' Reduce dt for early times given the approximate Smax computation
If count <= 5 Then
dt = 0.2 * dt
End If
' Check that dt+t is not greater than Tmax and recompute dt if required
If t + dt>Tmax Then
dt = Tmax - t
End If
dtdx = dt / deltax
End Sub
Public Function Flux2(U1, U2) As Double
' Computation of the flux vector component of momentum equation given vector U
If U1 = 0 Then
Flux2 = 0
Else
Flux2 = U2 ^ 2 / U1 + 0.5 * 9.81 * U1 ^ 2
End If
End Function
Public Function Flux1(U2) As Double
' Computation of the flux vector component of continuity equation given vector U
Flux1 = U2
End Function
Sub evolve()
' Computation of vector fluxes associated with PIL and PIR
FIL(1) = Flux1(PIL(2))
FIL(2) = Flux2(PIL(1), PIL(2))
FIR(1) = Flux1(PIR(2))
FIR(2) = Flux2(PIR(1), PIR(2))
' bed source term effect
' z-jump in the bed profile within a cell
' cell full filled with water
zjump(i) = zR(i) - zL(i)
' cell partially-filled with water, with zero water depth on its right side
If PIR(1) <= TOLDEP Then
zjump(i) = PIL(1)
End If
' cell partially-filled with water, with zero water depth on its left side
If PIL(1) <= TOLDEP Then
zjump(i) = -PIR(1)
End If
deltasource(1) = 0
deltasource(2) = -9.81 * (PIL(1) + PIR(1)) * 0.5 * zjump(i) / deltax
End Sub
Sub limiters()
' Limitation of jumps in conserved variables
' If lambda=(1, Minmod limiter; 2, Superbee)
If DU * DD < 0 Then
delta = 0
Else
sig = Sgn(DU)
a = Abs(DU)
b = Abs(DD)
If a < b Then
d1 = a
Else
d1 = b
End If
d1 = lambda * d1
If a > b Then
d2 = a
Else
d2 = b
End If
If d1 < d2 Then
factor = d1
Else
factor = d2
End If
factor = factor * sig
delta = factor
End If
End Sub
Sub friction()
' Implicit computation of the friction effects
If nM = 0 Then
GoTo 250
End If
For i = 2 To N - 1
If U(2, i) > 0 Then
sn = 1
Else
sn = -1
End If
a1 = -9.81 * dt * nM ^ 2 / U(1, i) ^ (4 / 3) * sn
a2 = -1
a3 = U(2, i) / U(1, i)
V = (-a2 - (a2 ^ 2 - 4 * a1 * a3) ^ 0.5) / (2 * a1)
U(2, i) = V * U(1, i)
Else
U(2, i) = 0
End If
Next i
250
End Sub
Sub friction2()
' Explicit computation of the friction effects
If nM = 0 Then
GoTo 300
End If
For i = 2 To N - 1
' wet bed
qadv(i) = U(2, i)
Sf = nM ^ 2 * (U(2, i) / U(1, i)) * Abs((U(2, i) / U(1, i))) / U(1, i) ^ (4 / 3)
U(2, i) = U(2, i) - 9.81 * U(1, i) * dt * Sf
' Friction can stop the flow, but a flow reversal is not physically feasible
If qadv(i) * U(2, i) < 0 Then U(2, i) = 0
Else
U(2, i) = 0
End If
Next i
300
End Sub
Sub dry()
' check dry bed conditions
' preserve positivity in water depths
For i = 2 To N - 1
' if the water depth is negative, both depth and discharge are set to zero
If U(1, i) < 0 Then
U(1, i) = 0
U(2, i) = 0
End If
' Notethat possible mass errors due to wet-dry updating are not corrected
Next i
' if the water depth is below the tolerance (and positive) only the continuity equation
is solved, that is, discharge is set to zero
For i = 2 To N - 1
If U(1, i) < TOLDEP Then
U(2, i) = 0
End If
Next i
End Sub
Sub bedslope()
' Evaluation of source term from bed profile
For i = 1 To N
S(1, i) = 0
' Computation of the bed z-jump within a cell
' cell full-filled with water
zjump(i) = zR(i) - zL(i)
' cell partially-filled with water, with zero water depth on its right side
If BEXT(2, 1, i) <= TOLDEP Then
zjump(i) = BEXT(1, 1, i)
End If
' cell partially-filled with water, with zero water depth on its left side
If BEXT(1, 1, i) <= TOLDEP Then
zjump(i) = -BEXT(2, 1, i)
End If
S(2, i) = -9.81 * (BEXT(1, 1, i) + BEXT(2, 1, i)) * 0.5 * zjump(i) / deltax
Next i
End Sub
12.3 Using the Library
12.3.1 Short Tutorial: Flooding of an Obstacle
The methods selected for the library are neither the only possible choices, nor uni-
versally valid. However, they are considered illustrative to introduce students to the
computation of Open Channel Flows. In this section we will describe the use of the
Library taking as example a code for flooding of an obstacle, printed in the former
section, and inserted in the file “weir_dry_terrain_SGM-DGM_MC.xls”. This solver
uses the MUSCL-Hancock scheme with a well-balanced discretization of the
bed-slope source term, along with an approximate wetting-drying procedure. The
code permits to simulate the flooding of a parabolic weir of bed profile zb = 0.2 −
0.05x2. The upstream weir side is initially wet with a water surface elevation ho,
whereas the tailwater portion is dry. We will use this code as example to show how to
use the codes available as supporting material of the textbook. Observe in Fig. 12.1
the sheet used to insert the input data, print and plot the computational results. Note
the input data selected for the current simulation.
The following input data must be supplied to the sheet for the computational
simulation:
L = length of computational domain
Δx = cell width
Upstream section:
1 = discharge pulse at inlet section fixed (qo), water depth computed by
using a backward C− characteristic
2 = discharge and water depth computed using a transmissive condition
(open boundary)
3 = discharge and water depth computed using a reflective condition (solid
wall)
Fig. 12.1 Sheet for introducing input data, printing and plotting computational results
12.3 Using the Library 533
Downstream section:
1 = discharge and water depth computed using a transmissive condition
(open boundary)
2 = discharge and water depth computed using a reflective condition (solid
wall)
ho = initial water depth at the upstream weir face. It can be lower than the
maximum weir height (0.2 m), but also higher (to simulate dam-break like flows)
CFL = Courant-Friedrichs-Lewy number
Tmax = simulation time
Limiter:
1 = minmod
2 = superbee
nM = Manning’s roughness coefficient
Friction computation:
1 = implicit
2 = explicit
Froude number for the combined Surface Gradient Method (SGM)-Depth
Gradient Method (DGM). The reconstruction of the solution is done with SGM
below this limiting Froude number, and using the DGM in other cases.
To run the model press “RUN FINITE-VOLUME SOLVER”, and the macros
are called to conduct the simulation. Once finished, numerical results are auto-
matically printed and plotted. Running the model for Tmax = 0, we observe the
initial conditions (Fig. 12.2).
Fig. 12.2 Initial conditions running the solver for Tmax = 0 s

Fig. 12.3 Surge approaching the weir for Tmax = 1 s
In this case we fix a constant discharge pulse at the inlet section and consider the
tailwater section as an open boundary. Running the model for Tmax = 1 s, we
observe a surge approaching the weir crest (Fig. 12.3).
For Tmax = 2 s, the bore is not yet at the weir crest (Fig. 12.4); note the
deformation of the flow profile due to topography effects.
For Tmax = 2.5 s, a dry front is propagated on the lee side of the weir (Fig. 12.5).
For Tmax = 3 s, the dry front is propagated along the tailwater channel
(Fig. 12.6).
For Tmax = 5 s, the tailwater channel is wet (Fig. 12.7).
Fig. 12.4 Wave profile for Tmax = 2 s

Fig. 12.5 Dry front propagation of the lee side of the weir for Tmax = 2.5 s
Fig. 12.6 Wetting of tailwater channel at Tmax = 3 s
Now, let us change the boundary condition at the downstream section. If we

model a solid wall, we must set option “2”. In this case, the dry front is reflected by
the wall and a hydraulic jump propagates in the upstream direction, as seen below
for Tmax = 6 s (Fig. 12.8).
The user can also verify that the scheme is well-balanced; we set qo = 0 m2/s
and ho = 0.15 m, and run the model for Tmax = 50 s. Note the horizontal free
surface (Fig. 12.9).
Fig. 12.7 Wet conditions along the entire computational domain for Tmax = 5 s
Fig. 12.8 Reflection of the wave for Tmax = 6 s
Consider now the flooding if the upstream weir side is partially submerged, e.g.
ho = 0.1 m; see simulation below for Tmax = 1 s with a surge approaching the weir
(Fig. 12.10).
During the run-up, a dry front is formed propagating in the downstream direc-
tion, while a surge propagates in the upstream direction; see simulation below for
Tmax = 2 s (Fig. 12.11).
Fig. 12.9 Testing the C-property for Tmax = 50 s
Fig. 12.10 Surge approaching partially submerged weir for Tmax = 1 s
For Tmax = 3 s, the upstream surge is still propagating, while the dry front
reaches the tailwater channel (Fig. 12.12).
The reader can access and edit the codes at the Vbasic editor (Fig. 12.13).
In the following sections examples on how to use the Library for homework
assignment to students are presented.
Fig. 12.11 Dry front and upstream surge for Tmax = 2 s
Fig. 12.12 Wave profile for Tmax = 3 s
12.3.2 Example of Homework 1: Computation

of Gradually-Varied Steady Flow Profiles
In this example students are requested to modify the former code

“weir_dry_terrain_SGM-DGM_MC.xls” to compute the steady gradually-varied
flow profiles presented in Chap. 3. In a first step, upon editing the code, the bed
profile is modified to a constant slope prismatic channel (Fig. 12.14).
Step 2 is to modify boundary conditions to allow for sub- and supercritical flow
profiles. Following the theory of Chap. 5, for the unsteady computation of a
Fig. 12.13 Access to the code inserted into the sheet
subcritical flow profile, one boundary condition is prescribed at the upstream sec-
tion (typically the discharge), and another at the downstream section (typically the
flow depth). For a supercritical profile, both boundary conditions are prescribed at
the upstream section. The modification to the corresponding routine is seen in
(Fig. 12.15).
Starting with the mesh at x = 0, Fig. 12.16 contains a computation of an S2
steady gradually-varied flow profile (see input data in the figure). The corre-
sponding code is available at “backwater_MUSCLHancock.xls”.
Fig. 12.14 Modification of bed profile for computation of gradually-varied flow profiles
Fig. 12.15 Modification of boundary conditions for computation of gradually-varied flow profiles
Fig. 12.16 Computation of S2 steady gradually-varied flow profile
In Step 3, students are asked to critically look at the correctness of the steady
flow results produced by the unsteady flow solver. They are invited to compare with
a steady flow solver based on the 4th-order Runge-Kutta method, presented in
Chap. 3. The steady flow solver is available at “steadyrk4.xls”, and the results of
both solvers are compared in Fig. 12.17, showing excellent agreement. This
demonstrates the ability of the MUSCL-Hancock solver to converge to steady flow.
Note that the unsteady program correctly approaches steady uniform flow, thus the
discretization used for the friction source term is adequate.
Fig. 12.17 Comparison of steady and unsteady flow solvers for S2 steady gradually-varied flow
profile
12.3.3 Example of Homework 2: Wetting and Drying over

Uneven Topography
Wetting and drying processes over uneven topography count among the most
important problems in Environmental Hydraulics, including dam-break flows,
run-up and overland flows produced by tsunamis, flooding in rivers, and
flooding-draining of wetlands. Students are here requested to modify the code
“weir_dry_terrain_SGM-DGM_MC.xls” to simulate wetting and drying processes
over a bed profile of shape (units in m):
h i
zb ¼ 0:7 expð0:1x2 Þ þ 0:3 exp ðx 5Þ2 ; 7:9 x 7:9: ð12:1Þ
The topography is basically a smooth hump 0.7 m high, with an undulation on

its lee side. For x > 0, the bed profile is a variable topography allowing to produce
interesting flow transitions during wave run-up. Editing the code, the bed profile is
defined as (Fig. 12.18).
The code is available at “weir_dry_terrain_complex.xls”. Now, students are
requested to generate a wave at x = 0, that will produce a wetting of the tailwater
portion of the bed profile (x > 0) while it travels downstream, and then force a
run-up upon reflection on a wall. The purpose is to investigate the flow transitions
during the motion of the wave across this highly variable terrain. For this task,
reflective boundary conditions are prescribed at both ends of the computational
model, and the initial condition to generate a wave is selected as dam-break type
problem with ho = 1 m. Computations are conducted assuming frictionless flow. At
Tmax = 1 s, a dry front is observed travelling down the topography, thereby pro-
ducing wetting of the terrain (Fig. 12.19).
Fig. 12.18 Modification of bed topography for simulation of wetting and drying processes
Fig. 12.19 Wetting of terrain at Tmax = 1 s
At Tmax = 6 s the reflected flow at the wall produces a moving hydraulic jump
travelling in the upstream direction (Fig. 12.20).
At Tmax = 12 s the moving hydraulic jump is in its run-up over the lee-side
undulation, which is covered with a shallow supercritical flow (Fig. 12.21).
At Tmax = 18 s the moving hydraulic jump passes the lee-side undulation crest,
and the resulting wave suffers a modification due to topographic effects
(Fig. 12.22).
At Tmax = 21 s, the front of the moving hydraulic jump is conducting a run-up
over the sloping weir side, but at its tailwater portion a second shock front is formed
past the lee-side undulation crest. The flow pattern is complex but extremely
Fig. 12.20 Moving hydraulic jump at Tmax = 6 s

beautiful: an unsteady transcritical flow profile is formed at the lee-side hump,

thereby forcing the formation of a second shock front at the tailwater portion of the
water body that is in a run-up process over the sloping terrain (Fig. 12.23).
At Tmax = 23 s the back-shock moves in the downstream direction towards the
wall (Fig. 12.24).
Note the free surface depression at the lee-side hump crest, typical of a fully
subcritical flow, and surge approaching the downstream wall, at Tmax = 25 s
(Fig. 12.25).

Fig. 12.23 Continuous flow profile over the lee-side hump and two-shock wave pattern with
run-up over the sloping terrain at Tmax = 21 s
Fig. 12.24 Flow profile at Tmax = 23 s

Fig. 12.25 Flow profile at Tmax = 25 s
12.3.4 Example of Homework 3: Study of Different

Discretizations of the Friction Source Term
In this task, students are asked to go a step further and not only modify the bed
profile, as also the initial and boundary conditions in the code, but actively con-
tribute to add any new numerical component. To produce these skills, it is requested
to investigate different discretizations of the friction source term. In Chap. 9 is
presented the explicit discretization of the friction term as [see Eq. (9.107)]

Uiadv Uiadv
qki þ 1 ¼ qadv
i ðghSf Þadv
i Dt ¼ qadv
i gn Dt
2
; ð12:2Þ
ðhki þ 1 Þ1=3
where all the variables have the meaning as explained in Chap. 9. As previously
stated, this solver works reasonably well for wet-bed conditions, but it may produce
instabilities in propagation over dry terrain if friction is not limited [see Eq. (9.110)],
given that its effect, at the limit, is to stop the flow. Thus, a flow reversal shall not be
numerically permitted.
The implicit version of the model presented in Chap. 9 is [see Eq. (9.109)]
ðUik þ 1 Þ2
qki þ 1 ¼ qadv
i ðghSf Þki þ 1 Dt ¼ qadv
i sgðqadv
i Þgn Dt
2
; ð12:3Þ
ðhki þ 1 Þ1=3
resulting in the quadratic equation

Fig. 12.26 Implementation of semi-implicit friction computation
i Þgn Dt
sgðqadv 2
qadv
ðUik þ 1 Þ2 þ Uik þ 1 i
¼ 0; ð12:4Þ
ðhki þ 1 Þ4=3 hki þ 1
where the sign of the advection estimation of the flow is used to get a physically
correct solution. Implicit and explicit solvers are available in the programs used by
students during the practical sessions. Now, the task is to produce and implement a
new option for the code allowing for a semi-implicit treatment of friction.
A possible semi-implicit treatment is given by the statement
Fig. 12.27 Computational results for explicit estimation of friction effects

Fig. 12.28 Computational results for implicit estimation of friction effects

Uik þ 1 Uiadv
qki þ 1 ¼ qadv
i gn Dt
2
; ð12:5Þ
ðhki þ 1 Þ1=3
from which follows the updating formula
Uiadv
Uik þ 1 ¼ : ð12:6Þ
jUiadv j
1 þ gn2 Dt
ðhik þ 1 Þ4=3
Therefore, the new routine requested to the students based on their analytical
development is as follows (Fig. 12.26).
The resulting code is available in “frictioncomparison.xls”, which was tailored to
simulate dam break waves over horizontal beds. A first run was conducted using
Eq. (12.2) without limiting the maximum value of the friction force (Fig. 12.27),
thereby resulting in instabilities of the wet-dry front due to flow reversal (q < 0).
Therefore, computations are aborted shortly after this numerical artifact. This test
reveals the importance of limiting the friction force in explicit computations.
Implicit (Fig. 12.28) and semi-implicit (Fig. 12.29) simulations were both stable
and similar to each other.
Fig. 12.29 Computational results for semi-implicit estimation of friction effects
12.3.5 Example of Homework 4: Surge Reflection at a Wall
The purpose of this homework is to make the students critically assess the behavior
of a code, without assuming that it works simply because computations are not
crashed. Specifically, they are asked to demonstrate that the implementation of the
reflective boundary condition is producing good physical results. At a solid wall,
U and F are given by

h h Uh 0
U¼ ¼ ; F¼ ¼ 2 : ð12:7Þ
Uh 0 U 2 h þ 12 gh2 1
2 gh
A solid wall is modelled in the MUSCL-Hancock (second-order in space and

time) solver as [see Eq. (9.168)]
qkNþ 1 ¼ qkN1
þ1
;
qkNþþ11 ¼ qkN2
þ1
;
ð12:8Þ
hkNþ 1 ¼ hkN1
þ1
;
hkNþþ11 ¼ hkN2
þ1
;
which produces at the last computational cell a numerical flux of

0
FN1=2 ¼ 1 2 : ð12:9Þ
2 ghN1
The method to check the accuracy of the implementation compares a

numerically-reflected surge with its analytical counterpart. To check the imple-
mentation, we consider a surge travelling in the positive x-direction to be reflected
upon arriving at a solid wall. Its analytical solution is with Vw as the absolute surge
velocity, and subscripts 1 and 2 denoting conditions behind and in front of the surge
[see Eq. (7.102)]

U1 if x\xo þ Vw ðt to Þ
Uðx; tÞ ¼ : ð12:10Þ
U2 if x [ xo þ Vw ðt to Þ
The discontinuity is located at x = xo at start t = to. If a surge is reflected at a

wall, the downstream values consist in a static water layer of depth h2 forced by the
solid barrier, to be determined by using the conservation laws across the surge. The
vector U is thus
8
>
> h1
< U1 ¼ x\xo þ Vw ðt to Þ
q
1
Uðx; tÞ ¼ : ð12:11Þ
>
> h2
: U2 ¼ x [ xo þ Vw ðt to Þ
0
The surge mass conservation states [Eq. (7.103)]
U1 h1 U2 h2
Vw ¼ ; ð12:12Þ
h 1 h2
whereas the momentum equation is [Eq. (7.104)]

1=2
g h1
Vw ¼ U2 þ sg ðh1 þ h2 Þ ; ð12:13Þ
2 h2
where sg = ±1, depending on the direction selected for surge propagation.

Combining Eqs. (12.12) and (12.13) results in
1=2
g h1
q1 ¼ U2 h1 þ ðh1 h2 Þsg ðh1 þ h2 Þ : ð12:14Þ
2 h2
A horizontal and frictionless channel 8 m long is considered in this test; see

Fig. 12.30 for detailed test conditions. For the positive surge stage approaching the
wall (sg = +1), we select the test data q1 = 0.2 m2/s, h2 = 0.1 m and U2 = 0 m/s
(xo = 0 m, to = 0 s), resulting in Vw = +1.796 m/s and h1 = 0.2113 m by solving
Eqs. (12.13)–(12.14). The positive surge arrives at the end of the channel at
t = 4.4532 s, thus xo = 8 m and to = 4.4532 s for the reflected surge. For the
reflection at a wall (sg = −1), q1 = 0.2 m2/s, h1 = 0.2113 m, q2 = 0 m2/s, resulting
in Vw = −1.277 m/s and h2 = 0.3679 m. The analytical solution was easily con-
structed with the solver previously presented in Chap. 7 (Surgeanalytical.xls). The
solution of the MUSCL-Hancock solver is compared with the analytical solution at
the approaching stage to the wall (Tmax = 3.5 s, Fig. 12.30), and after reflection, at
Tmax = 7.453 s (Fig. 12.31). The agreement of the numerical and analytical results
is excellent before and after reflection, thereby confirming that the solid walls are
correctly modeled in the numerical solver.
Fig. 12.30 Numerical prediction of surge approaching the wall at Tmax = 3.5 s
Fig. 12.31 Numerical prediction of reflected surge at Tmax = 7.453 s

12.3.6 Example of Homework 5: Analytical Solution

for Transcritical Weir Flow
The purpose of this task is to construct an analytical solution of transcritical weir

flow for an exponential variation of the primitive variables (h, U). The practical
utility of this development is its use to check the convergence to steady transcritical
flow of the unsteady flow computations. The process consists in setting the desired
analytical solution to the primitive variables, and then resort to the steady
momentum equation to find a bed profile compatible with this solution. Let the
desired exponential variation of the primitive variables be
hðxÞ ¼ ho expðaxÞ;
ð12:15Þ
UðxÞ ¼ Uo expð þ axÞ:
Here a is a coefficient controlling the shape of the profile and “o” refers to the
approach flow conditions. The parameter can be determined by selecting a target
flow depth hd at distance L from the inlet section, thus resulting in

1 hd
a ¼ ln : ð12:16Þ
L ho
Note that the discharge produced by the solution given by Eqs. (12.15) is
qðxÞ ¼ UðxÞhðxÞ ¼ Uo ho ¼ const.; ð12:17Þ
as required in steady flow without sources or sinks of mass. The Froude number is
for Eq. (12.15)

U Uo 3
F¼ ¼ exp ax ; ð12:18Þ
ðghÞ1=2 ðgho Þ1=2 2
which easily reveals the position of the critical point of the free surface profile from
2
F ¼ 1 ) xc ¼ lnðF1
o Þ: ð12:19Þ
3a
The steady momentum equation reads
d 1 dzb
ðU 2 h þ gh2 Þ ¼ gh : ð12:20Þ
dx 2 dx
The momentum function M is thus, using Eq. (12.15),

1 1
M ¼ U 2 h þ gh2 ¼ Uo2 ho ½expðaxÞ þ gh2o ½expð2axÞ: ð12:21Þ
2 2
Inserting this result into Eq. (12.20), the ODE describing the weir bed profile is
dzb U2
¼ o a expð2axÞ þ aho expðaxÞ: ð12:22Þ
dx g
Integration of this equation is simple, resulting in

Uo2 U2
zb ðxÞ ¼ ho þ ho expðaxÞ o expð2axÞ: ð12:23Þ
2g 2g
This is the bed profile compatible with the analytical solution given by
Eq. (12.15) for weir flow. Finally, the free surface position is

Uo2 Uo2
zs ðxÞ ¼ hðxÞ þ zb ðxÞ ¼ ho þ expð2axÞ: ð12:24Þ
2g 2g
It is simple to check analytically an important concept explained in Chap. 4,

namely the critical point of weir flow. At the location where the Froude number is
unity, there must be an extreme in the bottom topography, e.g., the crest of a weir
for the transition from sub- to supercritical flow. Manipulation of Eq. (12.22)
produces

dzb U2
¼ aho expðaxÞ 1 o expð3axÞ ¼ ahð1 F2 Þ; ð12:25Þ
dx gho
from which follows analytically the important result
dzb
F¼1) ¼ 0: ð12:26Þ
dx
The purpose of this task is first to motivate students to do analytical develop-

ments and manipulate the equations of motion. The result, aside from being an
interesting mathematical exercise, is especially illustrative to clarify concepts
relating to steady weir flow, namely the critical point. The analytical solution
enables to check all the aspects with little effort. Second, the final solution is of
utility to check the convergence of unsteady solvers to steady flow. Modification of
the previous unsteady flow solver only entail an update of the bed profile, as
previously explained. The MUSCL-Hancock solver is checked against the analyt-
ical solution in Fig. 12.32, where both the velocity and free surface position are
excellently reproduced. In this test, q = 0.1 m2/s, so roughly hc = 0.1 m. We have
fixed the critical depth about 1 m downstream from the inlet, resulting in
a 1.386. The approach flow depth is 0.4 m.
(a)
(b)
Fig. 12.32 Analytical (red line) and numerical (blue points) solutions for weir flow: a free surface
position b fluid velocity
In the above presentation, it is shown how a numerical solver can be easily

implemented and run as a Vbasic script in Excel, to teach students during their first
contact with the computation of open channel flows. Instructors can easily modify
the material to suit their specific needs while teaching. It is fairly simple to translate
the same logic to Fortran, where the programs would run faster, which is in fact a
desirable exercise to those students interested in taking more advanced courses after
the introduction presented in this book. In turn, the solvers presented in the Library
are slow if a dense mesh is considered, e.g., during the check of mesh-independence
results. However, this is a sacrifice to be accepted, given that debugging is simple in
the Vbasic scripts, which is considered a desirable feature for beginners while
taking the lessons and “getting their hands dirty” with a computational code for the
first time.
References 555
References
Fenton, J.D. (2010). Numerical methods. Lecture notes, Vienna, Austria: Vienna University of
Technology.
Toro, E. F. (2000). HYPER_WAT: Library of 13 source codes for solving the non-linear
time-dependent shallow water equations in one and two space dimensions. Cheadle:
NUMERITEK Ltd.
Author Index
A Cantero-Chinchilla, F. N., 41, 430, 433, 435,

Abbott, M. B., 235 446, 452, 459, 461, 464, 476, 478, 479
Abramowitz, M., 470 Cao, Z., 420, 424, 427
Ackers, P., 59, 119, 121 Capart, H., 423, 427–429
Alcrudo, F., 294, 309 Carling, P., 420, 424, 427
Anderson, J. D., 274, 277, 281, 282, 283 Castro-Orgaz, O., 41, 60, 63, 66, 98, 118, 119,
Apelt, C., 436, 437 121, 122, 124, 125, 130, 132, 184, 186,
Apelt, C. J., 161, 171, 172 187, 189, 196, 241, 268, 402–404, 430,
Aureli, F., 385, 386, 388 433, 435, 438–443, 445–448, 452, 454,
Ayuso, J. L., 41, 63, 66, 245, 246, 430, 461 455, 458–461, 464, 465, 476, 478, 479
Causon, D. M., 386, 468
B Chanson, H., 20, 55, 81, 119, 138, 139, 187,
Babbitt, H. E., 175 196, 205, 232, 241, 253, 268, 396,
Bakhmeteff, B. A., 54, 56, 97, 98, 147 402–404, 445, 449, 450
Barthelemy, E., 441, 462 Chapra, S. C., 171
Baumann, E. R., 175 Chaudhry, M. H., 5, 6, 20, 31, 37, 67,
Bélanger, J. B., 99, 138, 169, 191 141–143, 151, 156, 170–172, 189, 198,
Benjamin, T. B., 465 199, 201–205, 207, 214, 216, 217, 275,
Bermudez, A., 297, 371, 375 279, 281, 283, 286, 287, 292, 293, 300,
Bhallamudi, S. M., 279, 280 365, 399–402, 461, 470, 471
Biot, M. A., 187 Chazel, F., 441, 462
Blau, E., 76, 196, 197, 444 Chen, C. L., 6, 25
Bonneton, P., 441, 462 Chen, J., 187, 188
Bos, M. G., 59, 117 Chow, V. T., 5, 6, 20, 25, 29, 38, 63, 67, 70,
Boussinesq, J., 5, 438 118, 141, 147, 152, 154, 156, 165, 166,
Bradford, S. F., 370, 382 179, 187, 205, 207, 234, 236, 237, 243,
Bresse, J., 163, 165, 166 245, 248, 253
Bretz, N. V., 100 Chu, C. R., 295
Brocchini, M., 369 Cienfuegos, R., 441, 462, 471
Brufau, P., 375 Cimorelli, L., 332
Brunner, G. W., 166, 168 Colella, P., 378
Courant, R., 277, 283
C Covelli, C., 332
Canale, R. P., 171 Cozzolino, L., 332

https://doi.org/10.1007/978-3-030-13073-2
558 Author Index
Cunge, J. A., 29, 30, 201–203, 205, 210, Hancock, S. L., 383
214–216, 238–240, 278, 288, 365, 376 Harbaugh, T. E., 161, 166, 167
Harleman, D. R. F., 111
D Harrison, A. J. M., 59, 119, 121
Dawson, J. H., 116 Harten, A., 355, 361
Della Morte, R., 332 Hasumi, M., 176, 455
Denlinger, R. P., 8, 35, 39 Henderson, F. M., 5, 20, 38, 55, 59, 60, 67, 75,
De Saint-Venant, A. B., 5, 202, 211 77, 94, 105, 117, 118, 130, 140, 147,
Dey, S., 41, 130, 132, 430, 433, 461 184, 186, 187, 205, 235, 237–239, 253,
Dias, F., 462 254, 260, 285, 313, 472
Dodd, N., 369 Henry, H. R., 132
Dracos, T., 187, 188 Hicks, F. E., 216
Dressler, R., 268, 396 Hirsch, C., 290, 292, 293, 296, 367
Dressler, R. F., 7, 257, 260 Hoffman, J. D., 73, 74, 134, 144, 264,
273–276, 290, 292, 305, 319, 367, 372,
E 449, 471
Erduran, K. S., 473 Holly, F. M., 29, 30, 201–203, 205, 216, 278,
Escoffier, F. F., 187 288, 376
Evers, F. M., 410 Hosking, R. J., 196, 198, 403, 404, 446, 447,
451–453, 475
F Hosoda, T., 447, 448
Favre, H., 205, 305, 308, 396, 400 Hseng, M., 295
Fennema, R. J., 293 Hutter, K., 7, 37, 438, 458–460
Fenton, J. D., 141, 142, 244–247, 447, 493
Flannery, B. P., 134 I
Fornberg, B., 470 Iervolino, M., 427
Fraccarollo, L., 423, 427, 429 Ilic, S., 473
Fread, D. L., 161, 166, 167 Ingram, D. M., 385, 386, 388, 402, 468
Friedrichs, K., 277, 283 Ippen, A. T., 109–111, 113, 114, 116
Isaacson, E., 277, 283
G Iverson, R. M., 8–10, 17, 35, 39
Garcia-Navarro, P., 286, 288, 294, 307–309 Iwasa, Y., 187
Gardner, C. S., 461
Gersten, K., 7 J
Gharangik, A. M., 198, 199, 287, 293, Jaeger, C., 6, 30, 51, 52, 54, 59, 66, 94, 142,
399–401 165, 438
Giraldez, J. V., 37, 63, 66, 458, 459 Jain, S. C., 5–7, 25, 30, 51, 53–55, 65, 77, 108,
Glaister, P., 368 110, 117, 141, 147, 151, 156, 161, 179,
Godunov, S. K., 313, 355, 358, 378 180, 192, 202, 205, 209, 215, 217, 218,
Gottlieb, S., 373, 473 231, 232, 237, 253–255, 257, 263, 285,
Greco, M., 427 286, 313, 317, 332, 335, 337, 341, 344,
Greco, V., 334, 336, 340–345 390
Green, A. E., 439, 461 Jameson, A., 293, 300
Grilli, S. T., 467 Jan, C.-D., 165
Guinot, V., 313, 461 Jeppson, R., 55, 73, 81, 97, 102, 105, 132, 144,
313, 344
H Jia, Y., 421, 423
Hafsteinsson, H. J., 410 Jin, Y. C., 8, 19, 33, 34, 39, 435
Hager, W. H., 3–5, 10, 21, 23, 34, 37, 38,
60–64, 76, 93, 109, 116, 118, 119, 121, K
122, 124, 126, 137, 184, 186, 187, 189, Karni, S., 347
241, 247, 268, 410, 427, 433, 435, Katopodes, N. D., 19, 32, 171, 216, 240, 313,
438–443, 445–448, 454, 455, 458, 459, 344
464, 465 Keulegan, G. H., 6, 28
Author Index 559
Khafagi, A., 91 114, 117–119, 121–127, 137, 138, 150,

Khan, A. A., 4, 8, 21, 33, 35, 36, 39, 217, 167, 187, 210, 216, 232, 241, 261, 278,
368–370, 373, 376, 400, 425, 464, 478 438, 443, 454
Kirby, J. T., 467 Montuori, C., 334–336, 340–345
Kirschmer, O., 253, 266
Kocaman, S., 268, 394–398, 405–407, N
484–486 Nadiga, B. T., 456, 461
Kutija, V., 473 Naghdi, P. M., 439, 447, 461
Nujic, M., 388
L Nwogu, O., 464
Lai, C., 6, 216
Lai, W., 217, 368–370, 373, 376, 400 O
Lannes, D., 441, 462 O’Connell, D. R. H., 8, 39
Lauber, G., 268, 396 Onda, S., 447, 448
Lauffer, H., 38 Ortega, J. H., 469
Lawrence, G. A., 439, 447 Ouyang, C., 8, 10
Lax, P., 279, 284, 365 Ozmen-Cagatay, H., 268, 394–398, 405–407,
Lax, P. D., 279, 365 484–486
Lecaros, R., 469
Lee, M. T., 175 P
Le Méhauté, B., 32 Patterson, G. W., 6, 28
Leopardi, A., 427 Pender, G., 420, 424, 427
Lesleighter, E., 456 Percy, A., 121
LeVeque, R. J., 205, 313, 344, 351, 367, 390, Peregrine, D. H., 205, 462
391, 402, 471, 472 Perkins, J. A., 59, 119, 121
Lewy, H., 277, 283 Pianese, D., 332
Liggett, J. A., 5, 6, 19, 28, 52, 67, 93, 106, 112, Poincaré, H., 187
114, 115, 123, 189, 207, 231, 278, 288 Pontillo, M., 427
Lighthill, M. J., 237, 238, 465 Popinet, S., 480
López-Rios, J. C., 469 Posey, C. J., 147, 165
Press, W. H., 134
M Pudasaini, S. P., 7
MacCormack, R. W., 291 Puertas, J., 71–73, 141, 148, 150, 151, 186,
Macdonald, I., 215, 223, 225 192, 193
Maidment, D. R., 29, 207, 234, 236, 237, 243,
245, 248 R
Maranzoni, A., 385, 386, 388 Ré, R., 253
Marche, F., 441, 462 Rees, M., 277, 283
Marchi, E., 438, 439 Regamey, J.-M., 247
Margolin, L. G., 456, 461 Renouard, D. P., 462
Massé, P., 183, 186, 187 Richtmyer, R. D., 284
Mateos, L., 130, 132 Riddette, K., 456
Matthew, G. D., 124, 438–440, 442, 446, 451 Riemann, B., 313
Mays, L. W., 29, 207, 234, 236, 237, 243, 245, Ritter, A., 253, 259, 393
248 Roache, P. J., 215, 276, 277, 351
McPherson, B., 456 Roberts, S., 280, 281, 293, 297, 303, 321–324,
Mignosa, P., 385, 386, 388 393
Mignot, E., 471 Roberts, W. W., 383
Milewski, P., 462 Rodi, W., 17
Mingham, C. G., 386, 468 Roe, P. L., 368
Mohapatra, P. K., 461, 470, 471 Roth, A., 126
Montecinos, G. I., 469 Rouse, H., 7, 52, 55, 79, 80, 85, 87, 88, 117,
Montes, J. S., 5, 6, 20, 28, 30, 33, 51, 52, 54, 118, 122, 124–126, 139, 156, 161, 165,
55, 59, 61, 66, 76, 93, 94, 99, 105, 112, 408, 436, 441, 482, 483
560 Author Index
S 367–369, 371, 372, 378, 424, 426, 427,

Sánchez, M., 71–73, 141, 148, 150, 151, 186, 432, 435, 468, 469, 471, 494
192, 193 Turkel, E., 293, 300
Sanders, B. F., 369, 370, 382
Sargison, J. E., 121 V
Savage, S. B., 7 Vacca, A., 427
Saviron, J. M., 286, 288, 294, 307, 308 Vallentine, H. R., 436, 437
Schlichting, H., 7 van Albada, G. D., 383
Schmidt, W., 293, 300 Van Dyke, M., 438
Schmocker, L., 427, 433 van Leer, B., 378, 383
Schoklitsch, A., 260, 396, 399 Vazquez-Cendón, M. E., 297, 347, 351, 371,
Seabra-Santos, F. J., 462 375
Selby, S. M., 81, 97 Verwey, A., 29, 30, 201–203, 205, 216, 278,
Serre, F., 23, 458, 460, 461, 465 365
Shu, C.-W., 373, 473 Vetterling, W. T., 134
Simmons, A. N., 77 Vieira, D. A., 421
Sinniger, R., 215 Vischer, D. L., 63, 64, 109, 116
Sivakumaran, N. S., 196, 198, 403, 404, 446, Vongsarnpigoon, L., 439, 447
447, 451–453, 475 von Kármán, T., 187
Skaugset, A. E., 77 Vreugdenhil, C. B., 1, 23
Smolarkiewicz, P. K., 456, 461
Soares-Frazão, S., 461 W
Stansby, P. K., 23 Wallis, S., 420, 424, 427
Steffler, P. M., 4, 8, 19, 21, 33–36, 39, 216, Wang, S., 351
435, 464 Wang, S. S. Y., 417, 419, 420, 424, 427, 430
Steger, J. L., 296 Warming, R. F., 296
Stegun, I. A., 470 Wei, G., 467
Stoker, J. J., 223, 225, 232, 239, 253, 255, 257, Wendroff, B., 279, 365
260, 300, 313, 393 White, F. M., 17, 20, 106, 119, 120, 124, 140,
Strelkoff, T., 6, 7, 288, 308, 396 203, 375
Sturm, T. W., 5, 30, 66, 67, 77, 98, 141, 147, White, W. R., 59, 119, 121
179, 180, 189, 201, 202, 237, 275, 278, Whitham, G. B., 237, 238, 260
281 Wilkinson, D. L., 196, 197
Su, C. H., 461 Williams, J., 456
Subramanya, K., 85, 147, 156, 157, 467 Wilson, E. H., 187
Sweby, P. K., 355, 383 Woodward, P., 378
Synolakis, C. E., 408, 410 Woodward, S. M., 147
Wu, W., 8, 29, 41, 42, 46, 47, 242, 417,
T 419–421, 423, 424, 427, 430
Tada, A., 447
Tadmor, E., 473 Y
Temperville, A. M., 462 Yen, B. C., 6–8, 17, 53, 138–141, 143
Terzidis, G., 288, 308, 396 Ying, X., 351, 375–377, 402, 405–407, 425
Teukolsky, S. A., 134 Young, D. L., 427, 428
Thom, A., 436, 437
Tingsanchali, T., 196, 198, 403, 404, 446, 447, Z
451–453, 475 Zech, Y., 461
Toman, E. M., 77 Zerihun, Y. T., 447, 456
Toro, E. F., 1, 23, 203, 205, 213–216, 232, Zhou, J. G., 23, 385, 386, 388, 402, 468
238–240, 253, 274, 279, 283, 296, 313, Zhu, D. Z., 439, 440, 447, 450, 451
316, 317, 319, 320, 323, 327, 329–331, Ziveri, C., 385, 386, 388
347–349, 351, 354, 358, 361, 363–365, Zoppou, C., 280, 281, 293, 297, 303, 321
Subject Index
A 1D shallow water non-hydrostatic equations,

Analytical solution, 163 33
Approximate Riemann solver, 361 2D flows in horizontal plane, 19
3D flow equations, 8
B Dam break wave, 484
Basic numerical concept, 273 Dam break wave evolution, 405
Bed deformation, 419 Dam break wave over movable bed, 426
Boundary condition, 161, 286, 388 Dam break waves with friction, 393
Depth-integrated equations, 5
C Dike erosion due to overtopping, 430
Channel transitions by change of bed elevation, Discontinuous solutions, 203
78 Dry-bed treatment, 369
Channel transitions by change of channel Dry tailwater condition, 254
width, 89
Comparative of numerical methods, 173 E
Complete gate closure, 337 Energy principle, 51
Complete gate opening, 332 Euler’s method, 169
Compound channels, 178 Explicit and implicit scheme, 280
Computation of steady flow profiles, 159 Explicit integral method, 169
Conservative property, 350 Explicit scheme, 288
Continuity equation, 12
Control sections, 117 F
Control volume equation for hydraulic jump, Finite-difference approximation, 274
189 Finite difference method, 273
Critical flow, 67, 118 Finite volume method, 213, 347
Critical flow in arbitrary sections, 73 First-order upwind finite volume method, 424
First-order upwind scheme, 376
D Flooding of an obstacle, 532
1D cross-sectional averaged continuity Flow on mild slope, 175
equation, 41 Flow on steep slope, 37, 175
1D cross-sectional averaged equations, 24 Flow over round-crested weir, 196, 401, 446,
1D cross-sectional averaged x-momentum 475
equation, 43 Flow over trapezoidal profiled weir, 456

https://doi.org/10.1007/978-3-030-13073-2
562 Subject Index
Flow profile for unknown discharge, 157 Momentum principle, 51

Flow profiles in prismatic channels, 147 Movable bed, 41, 417
Flow resistance, 139 MUSCL-Hancock scheme, 467
Fourth-order Runge–Kutta method, 170 MUSCL-Hancock second-order TVD scheme,
Free surface flow 378
fundamental equations, 1 MUSCL reconstruction, 378
FTCS scheme, 288 Muskingum channel routing, 248
Fundamental equations of
free surface flows, 1 N
Negative surge, 232
G Non-conservative form of SWE, 238
Gate flow, 132, 125 Non-hydrostatic free surface flow, 435
Gate maneuver, 332 Non-hydrostatic theory, 33
Godunov first-order scheme, 358 Non-uniform flow
Godunov-type scheme, 348 governing equation, 137
Godunov upwind numerical flux, 351 Numerical library, 493
Governing equation for Numerical scheme, 424
non-uniform flow, 137 Numerical solution of energy-momentum
Gradually varied flow computation, 494 equations, 132
H O
Hancock step, 383 ODE solver, 372
High-order scheme, 473 One-dimensional Serre equations, 461
HLL Riemann solver, 361 Open channel flow, 5
Horizontal plane, 19 Open channel flow classification, 1
Hydraulic jump, 98, 399
Hydraulic jump beyond a sluice gate, 198 P
Hydraulic jump position, 193 Partial gate closure, 338
Hydrologic routing, 242 Partial gate opening, 333
Hydrostatic flow, 3 Picard iteration, 438
Positive surge, 203
I Positive surge moving upstream, 210
Ideal dam break wave, 253, 392 Positive surge with friction, 308, 396
Implicit integral method, 166 Possible wave patterns, 321
Initial and boundary conditions, 221 Predictor–corrector method, 172
Initial condition, 285, 392
R
K Rarefaction wave, 230
Kinematic boundary condition, 10 Reservoir routing, 242
Kinematic wave, 234 Reynolds’ transport theorem, 205
Riemann problem, 313
L Riemann solver, 315
Lax diffusive scheme, 290 Roe Riemann solver, 368
Lax numerical flux, 365
Limitations of SWE, 241 S
Location of critical flow, 59 Saint-Venant theory, 19
Sediment transport, 41, 417
M Sediment transport closure, 421
MacCormack predictor–corrector scheme, 291 Sediment transport layer, 419
MacCormack scheme with calibrated artificial Sequent depths for general cross sections, 101
viscosity, 292 Shallow water equations, 201, 417
Method of characteristics, 217 Shock-capturing method, 214
Momentum equation, 13, 132 Shock-capturing scheme, 278
Subject Index 563
Simple wave, 230 Transition from sub- to supercritical flows, 183

Singular point in free surface flows, 183 Transition from super- to subcritical flows, 189
Sketching of flow profiles, 156 Transitions in supercritical flow, 106
Slope limiting, 380 Treatment of friction, 375
Solitary wave, 464 TVD MacCormack scheme, 293
Solitary wave propagation, 480 Two-dimensional Serre equations, 456
Solitary wave run-up, 408 Two-dimensional steady potential flow, 436
Source term, 425
Specific energy head, 54 U
Specific energy head for general sections, 66 Uniform flow, 117, 138
Specific momentum, 93 Uniform flow depth, 143
Spillway crest, 441 Unsteady flow, 475
Stability, 280 Unsteady ideal fluid flow, 456
Stability condition, 355, 426 Unsteady open channel flow, 201
Steady transcritical open channel flow, 183 Upwind scheme, 295
Steep slope, 445
Subcritical dam break wave, 300 W
Surge reflection at a wall, 549 Water surface slope at critical point, 187
Supercritical surge, 307 Wave celerity, 228
Surface gradient method, 386 Wave motion, 67
SWE in characteristic form, 217 Weir flow, 119
Well-balanced scheme, 373
T Wet tailwater condition, 260
Total-load equation, 420 Wetting and drying over uneven
Transcritical dam break wave, 303 topography, 542
Transition from mild to steep slope, 454

Shallow Water Hydraulics

Uploaded by

Copyright:

Available Formats

Shallow Water Hydraulics

Uploaded by

Document Information

Original Title

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

Shallow Water Hydraulics

Uploaded by

Copyright:

Available Formats

Oscar Castro-Orgaz

Shallow Water Hydraulics

ISBN 978-3-030-13072-5 ISBN 978-3-030-13073-2 (eBook)

Córdoba, Spain Oscar Castro-Orgaz

Oscar Castro-Orgaz would like to deeply thank Jerónimo Puertas Agudo,

1 Fundamental Equations of Free Surface Flows . . . . . . . . . . . . . . . . 1

2.2.5 Computation of Critical Flow in Arbitrary Sections . . . 73

3.6.3 Flow on Mild Slope . . . . . . . . . . . . . . . . . . . . . . . . . . 175

5.10 Hydrologic Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

8.2.3 Complete Wave Proﬁles . . . . . . . . . . . . . . . . . . . . . . . 320

9.8 Boundary and Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . 388

11.3 Unsteady Ideal Fluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 456

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557

1.1.1 Classiﬁcation of Open Channel Flows

© Springer Nature Switzerland AG 2019 1

Uniform and non-uniform flows

Gradually varied and rapidly varied flows

Hydrostatic and non-hydrostatic flows

Fig. 1.4 Rapidly varied flows at a spillway crest, b hydraulic jump

1.1.2 Aims and Scope

1.2 General Depth-Integrated Equations

1.2.1 Previous Open Channel Flow Developments

Garbis Hovannes Keulegan was born at Sebastia-Sivas, Armenia, in today’s

1.2.2 3D Flow Equations

@q @ ðquÞ @ ðqvÞ @ ðqwÞ

Likewise, in the horizontal–vertical Cartesian system of coordinates used, Newton’s

1.2.3 General Kinematic Boundary Conditions

Prior to integrating Eqs. (1.1)–(1.4) in the z-direction, the general kinematic

F ðx; y; z; tÞ ¼ z zb ðx; y; tÞ ¼ 0: ð1:5Þ

The material derivative of F following its motion is, therefore, given by

The derivatives of F are

@F @zb @F @zb @F @zb @F

Inserting Eqs. (1.9) into Eq. (1.8), one ﬁnds

DF @zb @zb @zb

Noting that the modulus of F is given by

Equation (1.10) gives after re-arrangement

@zb @zb @zb

@zs @zs @zs

1.2.4 Continuity Equation

Vertical integration of Eq. (1.1) yields

This integral relation is transformed by computing the following identities by

Summing Eqs. (1.18)–(1.21) produces

Zzs Zzs Zzs

Zzs Zzs Zzs

Zzs Zzs Zzs

This is the general depth-integrated continuity equation.

1.2.5 Momentum Equations

Now, the x-momentum Eq. (1.2) is vertically integrated as

The following transformations are produced using again Leibniz’s rule:

Zzs Zzs Zzs

Collecting terms, one ﬁnds

Zzs Zzs Zzs 

Using Eqs. (1.13) and (1.15) produces

Zzs Zzs Zzs Zzs

Equation (1.35) represents the general x-momentum depth-integrated equation for

Zzs Zzs Zzs Zzs

Finally, the z-momentum Eq. (1.4) is vertically integrated, producing

Employing the transformations:

applied to Eq. (1.37), and subsequently grouping terms, leads to

Zzs Zzs Zzs

Zzs Zzs Zzs