ARR 190514 Book2

A GUIDE TO
FLOOD ESTIMATION
BOOK 2 - RAINFALL ESTIMATION

The Australian Rainfall and Runoff: A guide to flood estimation (ARR) is licensed under the
Creative Commons Attribution 4.0 International Licence, unless otherwise indicated or
marked.
Please give attribution to: © Commonwealth of Australia (Geoscience Australia) 2019.
Third-Party Material
The Commonwealth of Australia and the ARR’s contributing authors (through Engineers
Australia) have taken steps to both identify third-party material and secure permission for its
reproduction and reuse. However, please note that where these materials are not licensed
under a Creative Commons licence or similar terms of use, you should obtain permission
from the relevant third-party to reuse their material beyond the ways you are legally
permitted to use them under the fair dealing provisions of the Copyright Act 1968.
If you have any questions about the copyright of the ARR, please contact:
arr_admin@arr.org.au
c/o 11 National Circuit,
Barton, ACT
ISBN 978-1-925848-36-6
How to reference this book:

Ball J, Babister M, Nathan R, Weeks W, Weinmann E, Retallick M, Testoni I, (Editors)
Australian Rainfall and Runoff: A Guide to Flood Estimation, © Commonwealth of Australia
(Geoscience Australia), 2019.
How to reference Book 9: Runoff in Urban Areas:

Coombes, P., and Roso, S. (Editors), 2019 Runoff in Urban Areas, Book 9 in Australian
Rainfall and Runoff - A Guide to Flood Estimation, Commonwealth of Australia, ©
Commonwealth of Australia (Geoscience Australia), 2019.
PREFACE
Since its first publication in 1958, Australian Rainfall and Runoff (ARR) has remained one of
the most influential and widely used guidelines published by Engineers Australia (EA). The
3rd edition, published in 1987, retained the same level of national and international acclaim as
its predecessors.
With nationwide applicability, balancing the varied climates of Australia, the information and
the approaches presented in Australian Rainfall and Runoff are essential for policy decisions
and projects involving:
• infrastructure such as roads, rail, airports, bridges, dams, stormwater and sewer
systems;
• town planning;
• mining;
• developing flood management plans for urban and rural communities;
• flood warnings and flood emergency management;
• operation of regulated river systems; and
• prediction of extreme flood levels.
However, many of the practices recommended in the 1987 edition of ARR have become
outdated, and no longer represent industry best practice. This fact, coupled with the greater
understanding of climate and flood hydrology derived from the larger data sets now available
to us, has provided the primary impetus for revising these guidelines. It is hoped that this
revision will lead to improved design practice, which will allow better management, policy
and planning decisions to be made.
One of the major responsibilities of the National Committee on Water Engineering of

Engineers Australia is the periodic revision of ARR. While the NCWE had long identified the
need to update ARR it had become apparent by 2002 that even with a piecemeal approach the
task could not be carried out without significant financial support. In 2008 the revision of
ARR was identified as a priority in the National Adaptation Framework for Climate Change
which was endorsed by the Council of Australian Governments.
In addition to the update, 21 projects were identified with the aim of filling knowledge gaps.
Funding for Stages 1 and 2 of the ARR revision projects were provided by the now
Department of the Environment. Stage 3 was funded by Geoscience Australia. Funding for
Stages 2 and 3 of Project 1 (Development of Intensity-Frequency-Duration information
across Australia) has been provided by the Bureau of Meteorology. The outcomes of the
projects assisted the ARR Editorial Team with the compiling and writing of chapters in the
revised ARR. Steering and Technical Committees were established to assist the ARR
Editorial Team in guiding the projects to achieve desired outcomes.
Assoc Prof James Ball Mark Babister
ARR Editor Chair Technical Committee for
ARR Revision Projects
ARR Technical Committee:
Chair: Mark Babister

Members:
Associate Professor James Ball
Professor George Kuczera
Professor Martin Lambert
Associate Professor Rory Nathan
Dr Bill Weeks
Associate Professor Ashish Sharma
Dr Bryson Bates
Steve Finlay
Related Appointments:
ARR Project Engineer: Monique Retallick
ARR Admin Support: Isabelle Testoni
Assisting TC on Technical Matters: Erwin Weinmann, Dr Michael Leonard
ARR Editorial Team:
Editors:James Ball
Mark Babister
Rory Nathan
Bill Weeks
Erwin Weinmann
Monique Retallick
Isabelle Testoni
Associate Editors for Book 9 - Runoff in Urban Areas
Peter Coombes
Steve Roso
Editorial assistance: Mikayla Ward

Status of this document
This document is a living document and will be regularly updated in the future.
In development of this guidance, and discussed in Book 1 of ARR 1987, it was recognised
that knowledge and information availability is not fixed and that future research and
applications will develop new techniques and information. This is particularly relevant in
applications where techniques have been extrapolated from the region of their development
to other regions and where efforts should be made to reduce large uncertainties in current
estimates of design flood characteristics.
Therefore, where circumstances warrant, designers have a duty to use other procedures and
design information more appropriate for their design flood problem. The Editorial team of
this edition of Australian Rainfall and Runoff believe that the use of new or improved
procedures should be encouraged, especially where these are more appropriate than the
methods described in this publication.
Care should be taken when combining inputs derived using ARR 1987 and methods
described in this document.
What is new in ARR 2019?
Geoscience Australia, on behalf of the Australian Government, asked the National

Committee on Water Engineers (NCWE) - a specialist committee of Engineers Australia - to
continue overseeing the technical direction of ARR. ARR's success comes from practitioners
and researchers driving its development; and the NCWE is the appropriate organisation to
oversee this work. The NCWE has formed a sub-committee to lead the ongoing management
and development of ARR for the benefit of the Australian community and the profession. The
current membership of the ARR management subcommittee includes Mark Babister, Robin
Connolly, Rory Nathan and Bill Weeks.
The ARR team have been working hard on finalising ARR since it was released in 2016. The
team has received a lot of feedback from industry and practitioners, ranging from substantial
feedback to minor typographical errors. Much of this feedback has now been addressed.
Where a decision has been made not to address the feedback, advice has been provided as to
why this was the case.
A new version of ARR is now available. ARR 2019 is a result of extensive consultation and
feedback from practitioners. Noteworthy updates include the completion of Book 9,
reflection of current climate change practice and improvements to user experience, including
the availability of the document as a PDF.
Key updates in ARR 2019
Update ARR 2016 ARR 2019
Book 9 Available as “rough” draft Peer reviewed and completed
Guideline Epub version Following practitioner feedback, a pdf version of ARR

formats Web-based version 2019 is now available
User Limited functionality in web-based version Additional pdf format available

experience
Climate Reflected best practice as of 2016 Climate Updated to reflect current practice
change Change policies
PMF chapter Updated from the guidance provided in 1998 Minor edits and reflects differences required for use in
to include current best practice dam studies and floodplain management
Examples Examples included for Book 9
Figures Updated reflecting practitioner feedback
As of May 2019, this version is considered to be final.

BOOK 2
Rainfall Estimation
Rainfall Estimation
Table of Contents
1. Introduction ........................................................................................................................ 1
1.1. Scope and Intent .................................................................................................... 1
1.2. Application of these Guidelines .............................................................................. 1
1.3. Climate Change ...................................................................................................... 1
1.4. Terminology ............................................................................................................ 2
1.5. References ............................................................................................................. 2
2. Rainfall Models .................................................................................................................. 3
2.1. Introduction ............................................................................................................. 3
2.2. Space-Time Representation of Rainfall Events ...................................................... 4
2.3. Orographic Enhancement and Rain Shadow Effects on Space-Time Patterns ...... 7
2.4. Conceptualisation of Design Rainfall Events .......................................................... 8
2.4.1. Event Definitions .......................................................................................... 8
2.4.2. Rainfall Event Duration ................................................................................ 9
2.4.3. Event Rainfall Depth (or Average Intensity) ................................................. 9
2.4.4. Temporal Patterns of Rainfall ....................................................................... 9
2.4.5. Spatial Patterns of Rainfall .......................................................................... 9
2.5. Spatial and Temporal Resolution of Design Rainfall Models .................................. 9
2.6. Applications Where Flood Estimates are Required at Multiple Locations ............ 11
2.7. Climate Change Impacts ...................................................................................... 11
2.8. References ........................................................................................................... 12
3. Design Rainfall ................................................................................................................ 13
3.1. Introduction ........................................................................................................... 13
3.2. Design Rainfall Concepts ..................................................................................... 13
3.4. Frequent and Infrequent Design Rainfalls ............................................................ 15
3.4.1. Overview .................................................................................................... 15
3.4.2. Rainfall Database ...................................................................................... 17
3.4.3. Extraction of Extreme Value Series ........................................................... 28
3.4.4. Regionalisation .......................................................................................... 33
3.4.5. Gridding ..................................................................................................... 35
3.4.6. Outputs ...................................................................................................... 37
3.5. Very Frequent Design Rainfalls ............................................................................ 37
3.5.1. Overview .................................................................................................... 37
3.5.4. Ratio Method ............................................................................................. 41
3.5.5. Gridding ..................................................................................................... 41
3.5.6. Outputs ...................................................................................................... 42
3.6. Rare Design Rainfalls ........................................................................................... 42
3.6.1. Overview .................................................................................................... 42
3.6.4. Regionalisation .......................................................................................... 45
3.6.5. Gridding ..................................................................................................... 45
3.6.6. Outputs ...................................................................................................... 45
3.7. Probable Maximum Precipitation Estimates ......................................................... 46
3.7.1. Overview .................................................................................................... 46
3.7.2. Estimation of PMPs ................................................................................... 46
3.7.3. Generalised Methods for Probable Maximum Precipitation Estimation ..... 46
3.7.4. Generalised Method of Probable Maximum Precipitation Estimation ........ 48
clxxv
Rainfall Estimation
3.8. Uncertainty in Design Rainfalls ............................................................................. 49

3.9. Application ............................................................................................................ 50
3.9.1. Design Rainfalls ......................................................................................... 50
3.9.2. Frequent and Infrequent Design Rainfalls (IFDs) ...................................... 52
3.9.3. Very Frequent Design Rainfalls ................................................................. 53
3.9.4. Rare Design Rainfalls ................................................................................ 54
3.9.5. Probable Maximum Precipitation Estimates .............................................. 56
3.10. Acknowledgements ............................................................................................ 57
3.11. References .......................................................................................................... 57
4. Areal Reduction Factors .................................................................................................. 61
4.1. Introduction ........................................................................................................... 61
4.2. Derivation of Areal Reduction Factors .................................................................. 61
4.3. Areal Reduction Factor Recommendations .......................................................... 62
4.3.1. Areal Reduction Factors for Catchments up to 30 000 km2, Durations
up to 7 days and Events More Frequent than 0.05% AEP .................................. 62
4.3.2. Events That are Rarer than 0.05% Annual Exceedance Probability ......... 66
4.3.3. Catchments with Areas Greater than 30 000 km2 ..................................... 66
4.4. Worked Example .................................................................................................. 66
4.5. Limitations and Recommended Further Research ............................................... 66
4.6. Recommended Further Research ........................................................................ 67
4.6.1. Areal Reduction Factors ............................................................................ 67
4.7. References ........................................................................................................... 68
5. Temporal Patterns ........................................................................................................... 70
5.1. Introduction ........................................................................................................... 70
5.2. Temporal Pattern Concepts .................................................................................. 70
5.2.1. Storm Components .................................................................................... 70
5.2.2. Pattern Variability ....................................................................................... 72
5.2.3. History of Design Temporal Pattern Development ..................................... 74
5.3. Storm Database .................................................................................................... 77
5.3.1. Data Quality ............................................................................................... 79
5.3.2. Event Selection and Analysis .................................................................... 79
5.3.3. Regional Characteristics ............................................................................ 80
5.4. Pre Burst Rainfall and Antecedent Conditions ..................................................... 83
5.5. Design Point Temporal Patterns ........................................................................... 85
5.6. Temporal Patterns for Areal Rainfall Bursts .......................................................... 88
5.6.1. Areal Rainfall Time Series Grid for All Australia ........................................ 88
5.6.2. Average Areal Rainfall Calculation ............................................................ 88
5.6.3. Areal Temporal Pattern Selection .............................................................. 89
5.6.4. Design Areal Temporal Patterns ................................................................ 90
5.7. Other Temporal Pattern Options ........................................................................... 91
5.7.1. Use of Historical Temporal Patterns .......................................................... 91
5.7.2. Complete Storm Patterns .......................................................................... 91
5.7.3. Continuous Data ........................................................................................ 92
5.9. Temporal Pattern Application and Pre-burst ......................................................... 93
5.9.1. General ...................................................................................................... 93
5.9.2. Ensemble Considerations .......................................................................... 93
5.9.3. Upscaling of Patterns ................................................................................. 94
5.9.4. Dealing with Inconsistencies and Smoothing of Results ........................... 94
5.9.5. Practical Issues .......................................................................................... 94
5.9.6. Point and Areal Temporal Pattern Meta-Data ............................................ 96
5.9.7. Very Rare Point Temporal Patterns ........................................................... 96
clxxvi
Rainfall Estimation
5.9.8. Region Considerations .............................................................................. 96

5.9.9. Pre-burst .................................................................................................... 97
5.10. Example .............................................................................................................. 97
5.11. References .......................................................................................................... 99
6. Spatial Patterns of Rainfall ............................................................................................ 103
6.1. Introduction ......................................................................................................... 103
6.2. Methods for Deriving Spatial Patterns of Rainfall for Events .............................. 103
6.2.1. Precipitation Observation Methods and Uncertainties Associated with
Reconstructing Space-Time Rainfall Patterns ................................................... 103
6.2.2. Data Availability ....................................................................................... 104
6.2.3. Construction of Space-Time Patterns from Rainfall Gauge Networks ..... 105
6.2.4. Space-Time Patterns for Calibration ........................................................ 108
6.3. Spatial and Space-Time Patterns for Design Flood Estimation .......................... 108
6.3.1. Guidance for Catchments up to and Including 20 km2: Single Uniform
Spatial Pattern ................................................................................................... 109
6.3.2. Guidance for Catchments Greater than 20 km2: Single Non-Uniform
Spatial Pattern ................................................................................................... 109
6.3.3. Alternative Approach: Monte Carlo Sampling from Separate
Populations of Spatial and Temporal Patterns ................................................... 110
6.3.4. Alternative Approach: Monte Carlo Sampling from Single Population of
Space-Time Patterns .......................................................................................... 111
6.3.5. Spatial Patterns for Pre-Burst and Post-Burst Rainfall ............................. 111
6.3.6. Spatial Patterns for Continuous Rainfall Series ........................................ 111
6.4. Potential Influences of Climate Change on Areal Reduction Factors, Spatial
and Space-Time Patterns .......................................................................................... 113
6.5. Worked Examples ............................................................................................... 113
6.5.1. Catchment Used for Worked Examples ................................................... 113
6.5.2. Worked Example 1: Interpolation of Spatial Patterns for an Event Using
Various Methods ................................................................................................ 114
6.5.3. Worked Example 2: Calculation of Catchment Average Design Rainfall
Depths and Areal Reduction Factors ................................................................. 119
6.5.4. Worked Example 3: Calculation of Spatial Pattern for Design Flood
Estimation .......................................................................................................... 123
6.5.5. Worked Example 4: Application to Design Flood Estimation ................... 127
6.6. Recommended Further Research ...................................................................... 128
6.6.1. Deriving Spatial and Space-Time Patterns of Rainfall for Events ............ 128
6.6.2. Space-Time Patterns for Calibration of Rainfall Runoff Models to
Historical Floods ................................................................................................ 129
6.6.3. Spatial and Space-Time Patterns for Design Flood Estimation ............... 129
6.6.4. Potential Influences of Climate Variability and Climate Change .............. 130
6.7. References ......................................................................................................... 130
7. Continuous Rainfall Simulation ..................................................................................... 132
7.1. Use of Continuous Simulation for Design Flood Estimation ............................... 132
7.2. Rainfall Data Preparation ................................................................................... 134
7.2.1. Errors in Rainfall Measurements ............................................................. 134
7.2.2. Options for Catchments with no Rainfall Records ................................... 137
7.2.3. Missing Rainfall Observations ................................................................. 138
7.3. Stochastic Rainfall Generation Philosophy ......................................................... 140
7.4. Rainfall Generation Models ................................................................................ 140
7.4.1. Daily Rainfall Generation ......................................................................... 140
7.4.2. Sub-daily Rainfall Generation .................................................................. 153
7.4.3. Identifying ‘Nearby’ Stations - Application to Sydney Airport ................... 162
clxxvii
Rainfall Estimation
7.4.4. Modification of Generated Design Rainfall Attributes .............................. 163

7.4.5. Example of Daily and Sub-Daily Rainfall Generation .............................. 166
7.5. Implications of Climate Change .......................................................................... 170
7.6. References ......................................................................................................... 170
clxxviii
List of Figures
2.2.1. Conceptual Diagram of Space-Time Pattern of Rainfall .............................................. 5
2.2.2. Conceptual Diagram of the Spatial Pattern and Temporal Pattern Temporal and
Spatial Averages Derived from the Space-Time Rainfall Field ...................................... 6
2.2.3. Conceptual Diagram Showing the Temporal Pattern over a Catchment and the
Spatial Pattern Derived over Model Subareas of the Catchment .................................. 7
2.3.1. Classes of Design Rainfalls ....................................................................................... 14
2.3.2. Frequent and Infrequent (Intensity Frequency Duration) Design Rainfall Method .... 16
2.3.3. Daily Read Rainfall Stations and Period of Record ................................................... 20
2.3.4. Continuous Rainfall Stations and Period of Record .................................................. 21
2.3.5. Daily Read Rainfall Stations Used for ARR 1987 and ARR 2016 Intensity
Frequency Duration Data ...................................................................................... 22
2.3.6. Continuous Rainfall Stations Used for ARR 1987 and ARR
2016 Intensity Frequency Duration Data ............................................................... 23
2.3.7. Length of Available Daily Read Rainfall Data ............................................................ 24
2.3.8. Length of Available Continuous Rainfall Data ........................................................... 24
2.3.9. Number of Long-term Daily Read Stations Used for ARR 1987 and ARR 2016
Intensity Frequency Duration Data ....................................................................... 25
2.3.10. Length of Record of Continuous Rainfall Stations Used for ARR 1987 and ARR
2016 Intensity Frequency Duration Data ............................................................... 26
2.3.11. Analysis Areas Adopted for the BGLSR .................................................................. 32
2.3.12. Daily Read Rainfall Stations and Continuous Rainfall Stations Used for Very
Frequent Design Rainfalls ......................................................................................... 39
2.3.13. Procedure to Derive Very Frequent Design Rainfall Depth Grids From Ratios ....... 42
2.3.14. Daily Read Rainfall Stations with 60 or More Years of Record ............................... 44
2.3.15. Generalised Probable Maximum Precipitation Method Zones ................................ 47
2.3.16. Design Rainfall Point Location Map Preview ........................................................... 51
2.3.17. IFD Outputs ............................................................................................................. 52
2.3.18. Very frequent Design Rainfall Outputs .................................................................... 53
2.3.19. Rare design rainfall outputs ..................................................................................... 54
2.3.20. Design Rainfall Output Shown as Table .................................................................. 55
2.3.21. Design Rainfall Output Shown as Chart .................................................................. 56
2.4.1. Area Reduction Factors Regions for Durations 24 to 168 Hours ............................. 65
2.5.1. Typical Storm Components ....................................................................................... 71
2.5.2. Two Different Storm Events with Similar Intensity Frequency Duration Characteristics
(Sydney Observatory Hill) – Two Hyetographs plus Burst Probability Graph ............ 73
2.5.3. Ten 2 hr Dimensionless Mass Curves ....................................................................... 74
2.5.4. Decay Curve of Ten Dimensionless Patterns and AVM Patterns .............................. 75
2.5.5. Pluviograph Stations Record Lengths ....................................................................... 77
2.5.6. Pluviograph Stations used Throughout South-Eastern Australia with Record Lengths 78
clxxix
Rainfall Estimation
2.5.7. Temporal Pattern Regions ......................................................................................... 81

2.5.8. Example of Front, Middle and Back Loaded Events ................................................ 82
2.5.9. Pre-burst Rainfall ....................................................................................................... 84
2.5.10. Pre-burst to burst ratio ............................................................................................. 84
2.5.11. Standardised pre-burst to burst ratio distributions ................................................... 85
2.5.12. Temporal Pattern Ranges ........................................................................................ 86
2.5.13. Combinations of Aspect Ratio and Rotation for Hypothetical Catchments ............. 89
2.5.14. Comparison of Point Temporal Patterns and Areal Temporal Patterns - East Coast South
Region - 1 Day ......................................................................................................... 90
2.5.15. Comparison of Areal Temporal Patterns and the Temporal Pattern of the Closest
Pluviograph for the Same Event ................................................................................ 91
2.5.16. Tennant Creek Catchment ....................................................................................... 98
2.5.17. Duration Box plot for the 1% AEP ........................................................................... 99
2.6.1. Stanley River Catchment, Showing Runoff-routing Model Subcatchments and the Locations
of Daily rainfall and pluviograph gauges ................................................................. 114
2.6.2. Rainfall totals (mm) Recorded at Rainfall Gauges for the January 2013 Event in the
Vicinity of the Stanley River Catchment .................................................................. 115
2.6.3. Application of Thiessen Polygons- Rainfall Totals for the January 2013 Event -
Stanley River Catchment ......................................................................................... 116
2.6.4. Application of Inverse Distance Weighting - Rainfall Totals for the January 2013
Event - Stanley River Catchment ............................................................................ 117
2.6.5. Application of Ordinary Kriging - Rainfall Totals for the January 2013 Event -
Stanley River Catchment ......................................................................................... 118
2.6.6. Observed Semi-variogram and Fitted Linear Semi-variogram for the January 2013
Rainfall Event for Stanley River catchment, Applied in the Ordinary Kriging
Algorithm ................................................................................................................. 119
2.6.7. Non-dimensional Spatial Pattern (percentage of catchment average design rainfall
depths) for Events with AEP of 1% and more Frequent for Stanley River to Woodford (top
panel) and Stanley River to Somerset Dam (bottom panel) .................................... 125
2.6.8. Design Spatial Pattern of Design Rainfall Depths 1% AEP 24 hour Event for Stanley
River to Woodford (top panel) and Stanley River to Somerset Dam (bottom panel) 126
2.6.9. Flood Frequency Curves for Stanley River at Somerset Dam Inflow Derived from Analysis
of Estimated Annual Maxima and from RORB Model Simulations .......................... 128
2.7.1. Flood Events for a Typical Australian Catchment - Scott Creek, South Australia ... 133
2.7.2. Double Mass Curve Analysis for Rainfall at Station A (from
World Meteorological Organisation (1994)) ........................................................... 136
2.7.3. Total Rainfall Amounts for Rainfall Station 009557 over the Period 1956-1962 (from
Viney and Bates (2004)) ........................................................................................ 137
2.7.4. Rainfall Stations used in Table 2.7.1 for the Transition
Probability Model (Srikanthan et al., 2003) ........................................................... 146
2.7.5. Identification of “Similar” Locations for Daily Rainfall Generation using
RMMM ................................................................................................................. 151
clxxx
Rainfall Estimation
2.7.6. Generation of Daily Rainfall Sequences using the Regionalised Modified Markov
Model Approach .................................................................................................. 152
2.7.7. Disaggregated Rectangular Intensity Pulse Model (extracted from
Heneker et al. (2001)) ......................................................................................... 155
2.7.8. Schematic of Non-dimensional Random Walk used in DRIP disaggregate pulses . 156
2.7.9. Heneker et al. (2001) Model Fitted to Monthly Inter-event Time Data for
Melbourne in January ............................................................................................ 157
2.7.10. Heneker et al. (2001) Model Fitted to Monthly Storm Duration Data for
Melbourne in May .................................................................................................. 158
2.7.11. State-based Method of Fragments Algorithm used in the Regionalised Method of
Fragments Sub-daily Rainfall Generation Procedure ............................................ 159
2.7.12. Sydney Airport and nearby pluviograph stations .................................................. 163
2.7.13. Main Steps Involved in the Adjustment of Raw Continuous Rainfall Sequences to
Preserve the Intensity Frequency Duration relationships ...................................... 165
2.7.14. Annual Rainfall Simulations for Alice Springs using 100 Replicates ..................... 167
2.7.15. Intensity Frequency relationship for 24 hour Duration. ......................................... 167
2.7.16. 6 minute (left column) and 6 hour (right column) Annual Maximum Rainfall against
Exceedance Probability for Alice Springs. ............................................................ 169
2.7.17. Intensity Duration Frequency Relationships for Target and Simulated Rainfall before
and after Bias Correction at Alice Springs ............................................................. 170
clxxxi
List of Tables
2.3.1. Classes of Design Rainfalls ....................................................................................... 14
2.3.2. Frequent and Infrequent (Intensity Frequency Duration) Design Rainfall
Method .................................................................................................................... 17
2.3.3. Rainfall Reporting Methods ....................................................................................... 18
2.3.4. Restricted to Unrestricted Conversion Factors ......................................................... 29
2.3.5. Intensity Frequency Duration Outputs ....................................................................... 37
2.3.6. Very Frequent Design Rainfall Method ...................................................................... 38
2.3.7. Very Frequent Design Rainfall Outputs ..................................................................... 42
2.3.8. Rare Design Rainfall Method .................................................................................... 43
2.3.9. Rare Design Rainfall Outputs .................................................................................... 45
2.4.1. ARF Procedure for Catchments Less than 30 000 km2 and
Durations up to and Including 7 Days ....................................................................... 62
2.4.2. ARF Equation (2.4.2) Coefficients by Region for Durations 24 to 168
hours Inclusive .......................................................................................................... 65
2.5.1. Number of Pluviographs by Decade .......................................................................... 78
2.5.2. Regions- Number of Gauges and
Events ....................................................................................................................... 80
2.5.3. Burst Loading by Region and
Duration ..................................................................................................................... 82
2.5.4. Regional Temporal Pattern Bins ................................................................................ 86
2.5.5. Temporal Pattern Durations ....................................................................................... 86
2.5.6. Temporal Pattern Selection Criteria ........................................................................... 87
2.5.7. Areal Rainfall Temporal Patterns - Catchment Areas and Durations ........................ 88
2.5.8. Minimum Number of Pluviographs Required for Event
Selection for Each Catchment Area .......................................................................... 89
2.5.9. Areal Temporal Pattern Sets for Ranges of Catchment Areas .................................. 93
2.5.10. Alternate Regions Used for Data ............................................................................. 95
2.5.11. Flows for the 1% Annual Exceedance Probability for Ten Burst Events .................. 99
2.6.1. Calculation of Weighted Average of Point Rainfall Depths for the 1% AEP 24 hour
Design Rainfall Event for the Stanley River at Woodford ...................................... 120
2.6.2. Stanley River Catchment to Woodford: Calculation of Catchment Average Design
Rainfall Depths (bottom panel) from Weighted Average of Point Rainfall Depths (top panel)
and Areal Reduction Factors (middle panel) ........................................................... 121
2.6.3. Stanley River Catchment to Somerset Dam: Calculation of Catchment Average Design
Rainfall Depths (bottom panel) from Weighted Average of Point Rainfall Depths (top panel)
and Areal Reduction Factors (middle panel) ........................................................... 122
2.6.4. Calculation of Design Spatial Pattern for Stanley River at Woodford ...................... 123
2.6.5. RORB Model Scenarios Run for Worked Example on Stanley River Catchment to Somerset
Dam ......................................................................................................................... 127
2.7.1. Alternative Methods for Stochastic Generation of Daily Rainfall ............................. 141
clxxxii
Rainfall Estimation
2.7.2. Number of States used for Different Rainfall Stations in the Transition
Probability Model (Srikanthan et al., 2003) ......................................................... 145
2.7.3. State Boundaries for Rainfall Amounts in the Transition Probability
Model ................................................................................................................... 146
2.7.4. Daily Scale Attributes used to Define Similarity between Locations ....................... 149
2.7.5. Commonly used Sub-daily Rainfall Generation Models .......................................... 153
2.7.6. Sub-daily Attributes used to Define Similarity between Locations .......................... 160
2.7.7. Logistic Regression Coefﬁcients for the Regionalised Method of Fragments
Sub-daily Generation Model ................................................................................ 161
2.7.8. Statistical Assessment of Daily Rainfall from RMMM for Alice Springs using 100
Replicates 67 years Long .................................................................................... 166
2.7.9. Performance of extremes and representation of zeroes (for 6 minute time-steps)
from the sub-daily rainfall generation using RMOF for at-site generation using
observed sub-daily data (option 1), at-site disaggregation using observed daily data
(option 2), and the purely regionalised case (option 3). ...................................... 168
clxxxiii
List of Equations
2.4.1. Short duration ARF Equation .................................................................................... 64
clxxxiv
Chapter 1. Introduction
Mark Babister, Monique Retallick
Chapter Status Final
Date last updated 14/5/2019
1.1. Scope and Intent

Nearly all design flood estimation techniques rely on rainfall inputs to estimate flood
quantiles. These methods use catchment modelling techniques to estimate the flood
quantiles that would be derived from Flood Frequency Analysis if a long-term gauge record
was available. While simple methods just use rainfall intensity frequency duration data more
complex approaches require temporal and spatial rainfall information and continuous
simulation approaches require long-term rainfall sequences. Irrespective of the approach, it
is important to understand how the design rainfall inputs were derived and how they vary
from observed events.
Despite the advances in flood estimation many design inputs are assumed to be much
simpler than real or observed events. The more complex methods continue to make
assumptions including the use storm burst instead of a complete storm and spatial uniform
temporal patterns. For these reasons actual rainfall events tend to show considerably more
variability than design events and often have different probabilities at different locations.
This book describes the different rainfall inputs can be derived and how they can be used.
Book 2, Chapter 2 provides an introduction to rainfall models. Book 2, Chapter 3 details the
development of the design rainfalls (Intensity Frequency Duration data) by the Bureau of
Meteorology. Book 2, Chapter 4 and Book 2, Chapter 5 discuss the spatial and temporal
distributions of rainfall respectively. Book 2, Chapter 7 covers the development of continuous
rainfall time series for use in continuous simulation models.
1.2. Application of these Guidelines

The application of the design inputs discussed in this Book to Very Rare and Extreme floods
is discussed in Book 8.
1.3. Climate Change

These guidelines apply to the current climate. Statistically significant increases in rainfall
intensity have been detected in Australia for short duration rainfall events and are likely to
become more evident towards the end of the 21st century (Westra et al., 2013). Changes in
long duration events are expected to be smaller and harder to detect, but projections
analysed by CSIRO and Australian Bureau of Meteorology (2007) show that an increase in
daily precipitation intensity is likely under climate change. It is worth noting that a warming
climate can lead to decreases in annual rainfall along with increases in flood producing
rainfall.
The IFD’s presented in this chapter can be adjusted for future climates using the method
outlined in Book 1, Chapter 6 Which recommends an approach based on temperature
scaling using temperature projections from the CSIRO future climates tool. Scaling based on
temperature is recommended, as climate models are much more reliable at producing
temperature estimates than individual storm events.
1
Introduction
The impact of climate change on storm frequency, mechanism, spatial and temporal
behaviour is less understood. Work by (Abbs and Rafter, 2009) suggests that increases are
likely to be more pronounced in areas with strong orographic enhancement. There is
insufficient evidence to confirm whether this result is applicable to other parts of Australia.
Work by (Wasco and Sharma, 2015) analysing historical storms found that, regardless of the
climate region or season, temperature increases are associated with patterns becoming less
uniform, with the largest fractions increasing in rainfall intensity and the lower fraction
decreasing.
1.4. Terminology
The terminology for frequency descriptor described in Figure 1.2.1 applies to all chapters of
this book other than Book 2, Chapter 3 Design Rainfall.
1.5. References
Abbs, D. and Rafter, T. (2009), Impact of Climate Variability and Climate Change on Rainfall
Extremes in Western Sydney and Surrounding Areas: Component 4 - dynamical
downscaling, CSIRO.
CSIRO and Australian Bureau of Meteorology (2007), Climate Change in Australia, CSIRO
and Bureau of Meteorology Technical Report, p: 140. www.climatechangeinaustralia.gov.au
Wasko, C. and Sharma, A. (2015), Steeper temporal distribution of rain intensity at higher
temperatures within Australian storms, Nature Geoscience, 8(7), 527-529.
Westra, S., Evans, J., Mehrotra, R., Sharma, A. (2013), A conditional disaggregation
algorithm for generating fine time-scale rainfall data in a warmer climate, Journal of
Hydrology, 479: 86-99
2
Chapter 2. Rainfall Models
James Ball, Phillip Jordan, Alan Seed, Rory Nathan, Michael Leonard,
Erwin Weinmann

2.1. Introduction
The philosophical basis for use of a catchment modelling approach is the generation of data
that would have been recorded if a gauge were present at the location(s) of interest for the
catchment condition(s) of interest. For reliable and robust predictions of design flood
estimates with this philosophical basis, there is a need to ensure that rainfall characteristics
as one of the major influencing factors are considered appropriately.
There are many features of rainfall to consider when developing a rainfall model for design
flood prediction; exploration of these features can be undertaken using historical storm
events as a basis. In using this approach, there is a need to acknowledge that consideration
of historical events is an analysis problem and not a design problem. Nonetheless, insights
into the characteristics of rainfall events for design purposes can be obtained from this
review.
Rainfall exhibits both spatial and temporal variability at all spatial and temporal scales that
are of interest in flood hydrology. High resolution recording instruments have identified
temporal variability in rainfall from time scales of less than one minute to several days
(Marani, 2005). Similarly, observations of rainfall from high resolution weather radar and
satellites have demonstrated spatial variability in rainfall at spatial resolutions from 1 km to
more than 500 km (Lovejoy and Schertzer, 2006).
While it is important to be aware of this large degree of variability, for design flood estimation
based on catchment modelling it is only necessary to reflect rainfall variability at space and
time scales that are influential in the formation of flood events. The main focus is generally
on individual storms or bursts of intense rainfall within storms that cover the catchment
extent. However, it needs to be recognised that, depending on the design problem (e.g. flood
level determination in a system with very large storage and small outflow capacity), the
relevant ‘event’ to be considered may consist of rainfall sequences that include not just one
storm but extend over several months or even years.
Rainfall models are designed to capture in a simplified fashion those aspects of the spatial
and temporal variability of rainfall that a relevant to specific applications. A broad distinction
between different rainfall models can be made on the basis of their scope. Commonly rainfall
models consider only the temporal dimension by neglecting the spatial dimension. Inclusion
of the spatial dimension together with the temporal dimension results in an alternative form
of a rainfall model. This leads to the following categorisation of rainfall models:
• Models that concentrate on significant rainfall events (storms or intense bursts within
storms) at a point or with a typical spatial pattern that have the potential to produce floods;
• Models that attempt to simulate rainfall behaviour over an extended period at a point,
producing essentially a complete (continuous) rainfall time series incorporating flood
3
Rainfall Models
producing bursts of rainfall, low intensity bursts of rainfall and the dry periods between
bursts of rainfall(Book 2, Chapter 7); and
• Models that attempt to replicate rainfall in both the spatial and temporal dimensions.
Currently, models in this category are being researched and are not in general usage.
There are, however, many problems where rainfall models of this form may be applicable.
Rainfall models that concentrate on the flood producing bursts of rainfall have the inherent
advantage of conciseness (from a flood perspective, only the interesting bursts of rainfall are
considered). Hence, there is great potential to consider interactions of rainfall with other
influential flood producing factors but they also need to allow for the impact of varying initial
conditions.
Continuous rainfall models (Book 2, Chapter 7) have the inherent advantage of allowing the
initial catchment conditions (e.g. soil moisture status and initial reservoir content) at the
onset of a storm event to be simulated directly. However, the need to model the rainfall
characteristics of both storm events (intense rainfall) and inter-event periods (no rainfall to
low intensity rainfall) adds significant complexity to continuous rainfall models. The greater
range of events these models cover tends to be achieved at the cost of reduced ability to
represent rarer, higher intensity rainfall events. Additionally, very long sequences of rainfall
observations are required to properly sample rarer events. These issues make continuous
rainfall models more suitable for simulation of frequent events.
Rainfall data are mostly obtained from individual gauges (daily read gauges or pluviographs)
and only provide data on point rainfalls. However, for catchment simulation the interest is on
rainfall characteristics over the whole catchment. Rainfall models thus are needed to allow
extrapolation of rainfall characteristics from the point scale to the catchment scale. In
extrapolating rainfall characteristics from a point to a catchment or subcatchment, there is a
need to ensure that the extrapolation does not introduce bias into the predictions. This
applies to both continuous rainfall models and event rainfall models.
2.2. Space-Time Representation of Rainfall Events

When combined, the spatial and temporal variability of rainfall will be referred to as the
space-time variability of rainfall. The space-time pattern of rainfall over a catchment or study
area is therefore defined in three dimensions: two horizontal dimensions, which are normally
latitude and longitude (or easting and northing in a projected coordinate system) and one
temporal dimension. In practice, the space-time pattern of rainfall will often be described as
a three dimensional matrix, with the value in each element of the matrix representing either
the accumulated rainfall or the mean rainfall intensity for a grid cell over the catchment and a
specified period of time within the event, as shown in Figure 2.2.1.
4
Rainfall Models
Figure 2.2.1. Conceptual Diagram of Space-Time Pattern of Rainfall
If the space-time pattern of rainfall is considered as a field defined in three dimensions, then
the temporal and spatial patterns of rainfall that have conventionally been used in hydrology
can be considered as convenient statistical means of summarising that field. The temporal
pattern of rainfall over a catchment area is derived by taking an average in space (over one
or more grid elements) of the rainfall depth (or mean intensity) over each time increment of
the storm. The spatial pattern of rainfall for an event is defined by taking an average in time
(over one or more time periods) of the rainfall depth (or mean intensity) over each grid cell of
the catchment. Derivation of spatial and temporal patterns is demonstrated with the
conceptual diagram in Figure 2.2.2. Commonly, the spatial pattern is defined by averaging
over each subarea to be used in a model of the catchment or study area as shown in
Figure 2.2.3. The application of some catchment modelling systems (for example, rainfall-on-
grid models commonly used to simulate floods in urban areas), however, require grid based
spatial patterns of rainfall. In these situations, each grid element can be considered as a
subarea or subcatchment.
The space-time pattern of rainfall varies in a random manner between events and within
events influenced by spatial and temporal correlation structures that are an inherent
observed property of rainfall. The random space-time variability may make it difficult to
specify typical or representative spatial patterns for some catchments. Umakhanthan and
Ball (2005) in a study of the Upper Parramatta River Catchment in NSW showed the
variation in the temporal and spatial correlation between storm events on that catchment.
However, there are often hydrometeorological drivers, as discussed in Book 2, Chapter 4

that cause some degree of similarity in spatial and space-time patterns of flood producing
rainfall between events for a particular catchment. This similarity increases for the rarer
events and decreases for the more frequent events.
5
Rainfall Models
Figure 2.2.2. Conceptual Diagram of the Spatial Pattern and Temporal Pattern Temporal and
Spatial Averages Derived from the Space-Time Rainfall Field
6
Rainfall Models
Figure 2.2.3. Conceptual Diagram Showing the Temporal Pattern over a Catchment and the
Spatial Pattern Derived over Model Subareas of the Catchment
2.3. Orographic Enhancement and Rain Shadow Effects on

Space-Time Patterns
Orographic precipitation, also known as relief precipitation, is precipitation generated by a
forced upward movement of air upon encountering a physiographic upland. This lifting can
be caused by two mechanisms:
• Upward deflection of large scale horizontal flow by the topography; or
• Anabatic or upward vertical propagation of moist air up an orographic slope caused by

daytime heating of the mountain barrier surface.
Upon ascent, the air that is being lifted will expand and cool. This adiabatic cooling of a
rising moist air parcel may lower its temperature to its dew point, thus allowing for
condensation of the water vapour contained within it, and hence the formation of a cloud.
Rainfall can be generated from the cloud through a number of physical processes (Gray and
Seed, 2000). The cloud liquid droplets grow through collisions with other droplets to the size
where they fall as rain. Rain drops from clouds at high altitude may fall through the clouds
near the surface that have formed because of the uplift due to topography and grow as a
result of collisions with the cloud droplets. Air may also become unstable as it is lifted over
higher areas of terrain and convective storms may be triggered by this instability. These
influences combine to typically produce a greater incidence of rainfall on the upwind side of
hills and mountains and also typically larger rainfall intensities on the upwind side than would
otherwise occur in flat terrain.
7
Rainfall Models
The space-time pattern will vary between every individual rainfall event that occurs in a
catchment. In catchments that are subject to orographic influences, there will commonly be
similarity in the space-time pattern of rainfall between many of the different events that are
observed over the catchment. This will typically be the case for catchments that are subject
to flood producing rainfall events that have similar hydrometeorological influences. For
example, the spatial patterns of rainfall for different events may often demonstrate similar
ratios of total rainfall depth in the higher elevations of the catchment to total rainfall depth at
lower elevations.
The spatial patterns of rainfall in catchments that are influenced by orographic effects
represent a systematic bias away from a completely uniform spatial pattern. The influence of
this systematic bias in spatial pattern of rainfall should be explicitly considered in design
flood estimation. Other hydrometeorological influences, such as the distance from a
significant moisture source like the ocean may also give rise to systematic bias in the spatial
pattern of rainfall.
2.4. Conceptualisation of Design Rainfall Events

Ideally, the space-time variation of rainfall over a catchment would be represented as a
moving space-time field of rainfall at the appropriate spatial and temporal resolution.
However, while current developments are progressing in that direction (for example,
stochastic-space-time rainfall models developed by (Seed et al., 2002; Leonard et al., 2008),
the rainfall models widely used in practice are based on a more reductionist approach,
dealing separately with the spatial and temporal variability of rainfall. A result of this
reductionist approach is that rainfall bursts are assumed to be stationary; in other words, the
storm does not move during the period of rainfall.
For ease of modelling, storm events can be conceptualised and represented by four main
event characteristics that are analysed and modelled separately:
• Duration of storm or burst event;
• Total rainfall depth (or average intensity) over the event duration, at a point or over a
catchment;
• Spatial distribution (or pattern) of rainfall over the catchment during the event; and
• Temporal distribution (or pattern) of rainfall during the event.
These rainfall event characteristics are discussed in Book 2, Chapter 2, Section 4. to Book 2,
Chapter 2, Section 4
2.4.1. Event Definitions

The modelling of rainfall events first requires a clear definition of what constitutes an event
(Hoang et al., 1999). Given the variation of rainfall in time and space, it is not immediately
apparent when an event starts and ends. Start and end points of rainfall events need to be
defined by rainfall thresholds or separation in time from preceding/subsequent rainfall. For
an event to be significant, it may also need to exceed a total event rainfall threshold.
Two different types of rainfall events are relevant for design flood estimation: complete storm
events and internal bursts of intense rainfall. While complete storm events are the
theoretically more appropriate form of event for flood simulation, the internal rainfall bursts of
given duration, regardless of where they occur within a storm event, lend themselves more
8
Rainfall Models
readily for statistical analysis. The Intensity Frequency Duration (IFD) data covered in Book
2, Chapter 3 are thus for rainfall burst events.
2.4.2. Rainfall Event Duration

Actual storm events vary in their duration, from local thunderstorms lasting minutes to
extended rainfall events lasting several days. This variation occurs in a random fashion, and
rainfall event duration for a particular region can be characterised by a probability
distribution. However, for practical design flood estimation, the occurrence of rainfalls of
different durations within an appropriate range is generally assumed to have equal
probability, and the ‘critical rainfall duration’ is then determined as the one that maximises
the value of the design flood characteristic of direct interest.
The design rainfall data provided in ARR covers the range of rainfall burst durations from 1
minute to 7 days.
2.4.3. Event Rainfall Depth (or Average Intensity)

The basic methods for estimating design rainfall depths (or average intensities) for different
durations are discussed in Book 2, Chapter 3 for both point rainfalls . The principal modelling
approach used is to fit a probability distribution to series of rainfall depth observations
(annual maximum or peak over threshold) for the selected event duration at sites with long,
reliable rainfall records. The results of these at-site analyses are then generalised over
regions with similar rainfall characteristics and mapped over the whole of Australia. The
conversion of point design rainfalls to average catchment design rainfalls is modelled
through rainfall areal reduction factors is discussed in Book 2, Chapter 4.
2.4.4. Temporal Patterns of Rainfall

There are two distinct model representations of the temporal variability of rainfall within
events (for complete storms or internal bursts), depending on whether the model only
reflects the central tendency of different observed patterns or the variability of patterns for
different events is also modelled. These differences in modelling approach are further
discussed in Book 2, Chapter 2, Section 5 and Book 2, Chapter 5.
2.4.5. Spatial Patterns of Rainfall

In larger catchments and where there is a consistent spatial trend in observed rainfall
depths, (see Book 2, Chapter 2, Section 4) this needs to be represented by a non-uniform
spatial rainfall pattern. . The models for representing the typical spatial variability of rainfall
are based either on the analysis and generalisation of historical storms or on spatial trends
derived from analysis of design rainfall depths (IFD maps). The application of these models
is explained in Book 2, Chapter 6, Section 4
In the following, a number of different approaches to model the space-time characteristics of

event rainfall for design flood estimation are introduced briefly.
2.5. Spatial and Temporal Resolution of Design Rainfall

Models
The space-time pattern of rainfall for an individual flood event will often have an appreciable
influence on the flow hydrograph generated at the outlet of a catchment. Two rainfall events
may have identical total volumes over a defined catchment area and duration but differences
9
Rainfall Models
in their space-time patterns may produce very different hydrographs at the outlet of the
catchment. Both the runoff generation and runoff-routing processes in catchments are
typically non-linear, so a space-time pattern that exhibits more variability will normally
generate a higher volume of runoff and larger peak flow at the catchment outlet than a
space-time pattern that is more uniform.
Variability in hydrographs introduced by space-time variability in rainfall will be accentuated

in catchments that have spatial and temporal variability in runoff generation and routing
processes. For example, in a partly urbanised catchment, a rainfall event with a spatial
pattern that has larger depths on the urban part of the catchment than the rural part would
normally produce both a larger volume of runoff and flood peak at the catchment outlet than
a storm of the same depth and duration that has a spatially uniform rainfall pattern. Other
factors in catchments that may accentuate the influence of the space-time rainfall pattern on
the variability in hydrographs produced at the catchment outlet include:
• the presence of reservoirs and lakes, for which all rainfall on the water surface is
converted to runoff;
• the presence of dams, weirs, drains and other flow regulating structures;
• significant variations in soil type;
• significant variations in vegetation type, such as forested and cleared areas;
• the arrangement of the drainage network of the catchment the dependency of alternative
flow paths on event magnitude and differences in contributing area with length of network;
• significant variations in stream channel and floodplain roughness;
• significant variations in slope of stream channels and floodplains;
• significant variations in antecedent climatic conditions across the catchment prior to the
events; and
• variations in elevation, snowpack depth, density and temperature in those catchments

subject to rain-on-snow flood events.
The required resolution of rainfall models to adequately reflect the variability of rainfall in
historical rainfall events has been investigated by (Umakhanthan and Ball, 2005) for the
Upper Parramatta River catchment. (Umakhanthan and Ball, 2005) categorised the
variability of recorded storm events in the spatial and temporal domains and confirmed that
the degree of spatial and temporal resolution of rainfall inputs to flood estimation models can
have a significant impact on resulting flood estimates. A range of other studies have come to
similar conclusions but have found it difficult to give more than qualitative guidance on the
required degree of spatial and temporal resolution of rainfall for different modelling
applications. The conclusions can be summarised in qualitative terms as:
• “Spatial rainfall patterns are understood to be a dominant source of variability for very
large catchments and for urban catchments but for other hydrological contexts, results
vary. Much of this knowledge is either site specific or is expressed qualitatively” (Woods
and Sivapalan, 1999).
• Where short response times are involved in urban catchments, inadequate representation
of temporal variability of rainfall can lead to significant underestimation of design flood
peaks (Ball, 1994). More generally, the importance of temporal variability of rainfall in flood
10
Rainfall Models
modelling depends on the degree of ‘filtering’ of shorter term rainfall peaks through
catchment routing processes (ie. the amount of storage in the catchment system) and the
interaction of flood contributions from different parts of a catchment system.
Sensitivity analyses can be applied to determine for a specific application the influence of the
adopted spatial and temporal resolution of design rainfalls on flood estimates and their
uncertainty bounds.
2.6. Applications Where Flood Estimates are Required at

Multiple Locations
Design flood estimates are often required at multiple locations within a catchment or study
area. Ideally, flood simulation (e.g. using Monte Carlo approaches) should consider a large
number of complete storm events that cover the whole AEP spectrum of interest and have
internal characteristics which automatically reproduce the critical rainfall bursts over a range
of temporal and spatial scales. Unfortunately, such comprehensive ensembles of synthetic
storm events are not currently available, and combined system wide analysis is thus not yet
feasible. Instead, separate analysis at the different locations (subcatchments) of interest is
required, using design rainfall events for the relevant space and time scales. To this end, it is
necessary to derive design rainfall inputs for the catchment upstream of each required
location. This involves:
• Deriving average values of the point design rainfalls for the total catchment upstream of
each location;
• Conversion of average point design rainfall values to areal estimates by multiplying by the
ARF applicable to the total catchment area upstream of each location; and
• Adoption of space-time patterns of rainfall relevant to the total catchment area upstream of
each location.
It is commonly found that design flood estimates are required at one or more locations in a
catchment where flow gauges are not located. If so, it will be necessary to use the above
procedure to derive design rainfalls for the catchment upstream of each gauge location so
that the rainfall-based design floods estimates can be verified against estimates derived from
Flood Frequency Analysis at each flow gauge. Different sets of design rainfall intensities,
ARF and space-time patterns should be calculated for the each of the catchments draining
to the other locations of interest, which are not at flow gauges.
2.7. Climate Change Impacts

Statistically significant increases in rainfall intensity have been detected in Australia for short
duration rainfall events and are likely to become more evident towards the end of the 21st
century (Westra et al., 2013). Changes in long duration events are expected to be smaller
and harder to detect, but projections analysed by CSIRO and Australian Bureau of
Meteorology (2007) show that an increase in daily precipitation intensity is likely under
climate change. It is worth noting that a warming climate can lead to decreases in annual
rainfall along with increases in flood producing rainfall.
The impact of climate change on storm frequency, mechanism, spatial and temporal
behaviour is less understood.
Work by Abbs and Rafter (2009) suggests that increases are likely to be more pronounced in
areas with strong orographic enhancement. There is insufficient evidence to confirm whether
11
Rainfall Models
this result is applicable to other parts of Australia. Work by Wasco and Sharma (2015)
analysing historical storms found that, regardless of the climate region or season,
temperature increases are associated with patterns becoming less uniform, with the largest
fractions increasing in rainfall intensity and the lower fraction decreasing.
The implications of these expected climate change impacts on the different design rainfall
inputs to catchment modelling are discussed further in the relevant sub-sections of the
following chapters.
2.8. References
downscaling, CSIRO.
Ball, J.E. (1994), 'The influence of storm temporal patterns on catchment response', Journal
of Hydrology, 158(3-4), 285-303.
CSIRO and Australian Bureau of Meteorology (2007), Climate Change in Australia, CSIRO
and Bureau of Meteorology Technical Report, p: 140. www.climatechangeinaustralia.gov.au
Gray, W. and Seed, A.W. (2000), The characterisation of orographic rainfall, Meteorological
Applications, 7: 105-119.
Hoang, T.M.T., Rahman, A., Weinmann, P.E., Laurenson, E.M. and Nathan, R.J. (1999),
Joint Probability Description of Design Rainfalls. 25th Hydrology and Water Resources
Symposium, Brisbane, 2009, Brisbane.
Leonard, M., Lambert, M.F., Metcalfe, A.V. and Cowpertwait, P.S.P. (2008), A space-time
Neyman-Scott rainfall model with defined storm extent. Water Resources Research, 44(9),
1-10, http://doi.org/10.1029/2007WR006110
Lovejoy, S. and D. Schertzer, 2006: Multifractals, cloud radiances and rain, J. Hydrology,
322: 59-88.
Marani, M., 2005: Non-power-law-scale properties of rainfall in space and time, Water
Resour. Res., 41, W08413, doi:10.1029/2004WR003822.
Seed, A.W., Srikanthan, R., and Menabde, M. (2002), Stochastic space-time rainfall for
designing urban drainage systems. Proc. International Conference on Urban Hydrology for
the 21st Century, pp: 109-123, Kuala Lumpur.
Umakhanthan, K. and Ball, J.E. (2005), Rainfall models for catchment simulation. Australian
Journal of Water Resources, 9(1), 55-67.
Woods and Sivapalan (1999), A synthesis of space-time variability in storm response:

rainfall, runoff generation and routing. Water Resources Research, 35(8), 2469-2485.
12
Chapter 3. Design Rainfall
Janice Green, Fiona Johnson, Catherine Beesley, Cynthia The
3.1. Introduction
Obtaining an estimated rainfall depth for a specified probability is an essential component of
the design of infrastructure including gutters, roofs, culverts, stormwater drains, flood
mitigation levees, retarding basins and dams.
If sufficient rainfall records are available, at-site frequency analysis can be undertaken to
estimate the rainfall depth corresponding to the specified design probability in some cases.
However, limitations associated with the spatial and temporal distribution of recorded rainfall
data necessitates the estimation of design rainfalls for most projects.
The purpose of this chapter is to outline the processes used to derive temporally and
spatially consistent design rainfalls for Australia by the Bureau of Meteorology. The classes
of design rainfall values for which estimates have been developed are described in Book 2,
Chapter 3, Section 2. The practitioner is advised that this chapter uses different frequency
descriptors (Table 2.3.1) used to describe events to other the rest of this Guideline (which
use Figure 1.2.1).
Book 1, Chapter 6 summarises the current recommendations on how climate change should
be incorporated into design rainfalls for those situations where the design life of the structure
means that it could be affected by climate change.
Book 2, Chapter 3, Section 4 summarises the steps involved in deriving the frequent and
infrequent design rainfalls (also known as the Intensity Frequency Duration (IFD) design
rainfalls) for Australia. Book 2, Chapter 3, Section 5 and Book 2, Chapter 3, Section 6
describe how the very frequent and rare design rainfalls were estimated. The methods
adopted are only briefly outlined in these sections, with additional references provided to
facilitate access to further technical information for interested readers. More detail on each of
the methods is provided in Bureau of Meteorology (2016).
In Book 2, Chapter 3, Section 7 a summary of the methods adopted for the estimation of
Probable Maximum Precipitation is provided. Book 2, Chapter 3, Section 8 provides
information on the uncertainties associated with the design rainfalls and Book 2, Chapter 3,
Section 9 explains how to access estimates of each of the design rainfall classes.
3.2. Design Rainfall Concepts

Design rainfalls are a probabilistic or statistically-based estimate of the likelihood of a
specific rainfall depth being recorded at a particular location within a defined duration. This is
generally classified by an Annual Exceedance Probability (AEP) or Exceedances per Year
(EY) (as defined in Book 1, Chapter 2, Section 2). Design rainfalls are therefore not real (or
observed) rainfall events; they are values that are probabilistic in nature.
There are five broad classes of design rainfalls that are currently used for design purposes,
generally categorised by frequency of occurrence. These are summarised below and
13
Design Rainfall
presented graphically in Figure 2.3.1. However, it should be noted that there is some overlap
between the classes. Different methods and data sets are required to estimate design
rainfalls for the different classes and these are discussed in the following sections. The
practitioner is advised that this chapter uses different frequency descriptors (Table 2.3.1)
used to describe events to other the rest of this Guideline (which use Figure 1.2.1).
Table 2.3.1. Classes of Design Rainfalls

Design Rainfall Class Frequency of Occurrence Probability Range
Very Frequent Design Very frequent 12 EY to 1 EY
Rainfalls
Intensity Frequency Duration Frequent 1 EY to 10% AEP
(IFD Infrequent 10% to 1% AEP
Rare Design Rainfalls Rare 1 in 100 AEP to 1 in 2000
AEP
Probable Maximum Extreme < 1 in 2000 AEP
Precipitation (PMP)
Figure 2.3.1. Classes of Design Rainfalls

The design rainfalls provided as part of these guidelines are based on observed rainfall data
that represent, primarily, the climate of the 20th century. In order to assess the impact of
future climates an adjustment must be made to the design rainfalls provided in this chapter.
14
Design Rainfall
As part of the ARR revision projects a summary of the scientific understanding of how
projected changes in the climate may alter the behaviour of factors that influence the
estimation of the design floods was undertaken. Climate change research undertaken as
part of the ARR revision projects has lead to an interim recommendation to factor the design
rainfalls based on temperature scaling using temperature projections from the CSIRO future
climates tool. Advice on how to adjust design rainfalls for climate change is detailed in Book
1, Chapter 6.
3.4. Frequent and Infrequent Design Rainfalls

3.4.1. Overview
This section summarises the steps involved in deriving frequent and infrequent designs
rainfalls (Intensity Frequency Duration (IFDs)) for the probabilities from 1EY to 1% AEP.
These classes of design rainfalls constitute the traditional IFD design rainfalls (Table 2.3.1 ).
The main steps involved in the derivation of the frequent and infrequent design rainfalls
include the collation of a quality controlled database, extraction of the extreme values series,
frequency analysis, regionalisation and gridding processes. These steps are discussed in
more detail in Book 2, Chapter 3, Section 4 to Book 2, Chapter 3, Section 4 and summarised
in Figure 2.3.2 and in Table 2.3.2. The Sections in which each of the steps is discussed are
shown in Figure 2.3.2.
15
Design Rainfall
Figure 2.3.2. Frequent and Infrequent (Intensity Frequency Duration) Design Rainfall Method
16
Design Rainfall
Table 2.3.2. Frequent and Infrequent (Intensity Frequency Duration) Design Rainfall Method
Step Method/Data
Number of rainfall stations Daily read - 8074 gauges
Continuous – 2280 gauges
Period of record All available records up to 2012
Length of record used in analyses Daily read ≥ 30 years
Continuous > 8 years
Source of data Organisations collecting rainfall data across
Australia
Series of Extreme values Annual Maximum Series (AMS)
Frequency analysis Generalised Extreme Value (GEV)
distribution fitted using L-moments
Extension of sub-daily rainfall statistics to Bayesian Generalised Least Squares
daily read stations Regression (BGLSR)
Regionalisation Region of Influence (ROI)
Gridding Regionalised at-site distribution parameters
gridded using ANUSPLIN
3.4.2. Rainfall Database

Integral to the estimation of design rainfalls was the creation of a database containing data
from all available rainfall stations across Australia. These rainfall data were collected at
rainfall stations operated by various organisations using a range of types of collecting
methods and instrumentation. Further information on the collection and archiving of rainfall
data can be found in Book 1, Chapter 4, Section 9.
3.4.2.1. Types of Rainfall Data

Rainfall data are collected using a number of different types of instrumentation. These
provide different temporal and spatial resolutions of rainfall data, depending on the
instrument type and reporting method used. A brief summary of each of the main types of
rainfall data are provided below. Table 2.3.3 summarises the types of rainfall reporting
methods used and indicates their use in the estimation of the design rainfalls. Most of the
rainfall data used to derive the design rainfalls were recorded by a daily read, pluviograph or
Tipping Bucket Rain Gauge (TBRG) (Table 2.3.3).
3.4.2.1.1. Daily Read Rainfall Gauges
Daily read rainfall gauges are read at 9:00 am each day and a total rainfall depth for the
previous 24 hours is reported (refer to Book 1, Chapter 4, Section 9).
3.4.2.1.2. Continuous Rainfall Gauges
Continuous rainfall stations measure rainfall depth at much finer time intervals. In Australia
there have been two main types of continuous rainfall stations as discussed below.
i. Dines Tilting Syphon Pluviographs
17
Design Rainfall
Dines Tilting Syphon Pluviographs (DINES) record rainfall on a paper chart which is then
digitised manually. Due to the limitations in the digitisation process, the minimum interval
at which rainfall data could be accurately provided was 5 or 6 minutes.
ii. Tipping Bucket Raingauges (TBRG)
Since the 1990s the majority of Dines pluviographs have been replaced by Tipping Bucket
Raingauges (TBRG) which typically have a 0.2 mm bucket capacity. Each time the bucket
is filled the gauge tips creating an electrical impulse which is logged. Rainfall data from
TBRGs can be accurately provided for intervals of less than one minute (refer to Book 1,
Chapter 4, Section 9).
3.4.2.1.3. Event-Reporting Radio Telemetry Systems

The Event-Reporting Radio Telemetry Systems (ERTS) network consists of over 1000
stations across Australia, operated by the Bureau of Meteorlogy, local government and other
water agencies. As the purpose of these gauges is to provide information for use in flood
forecasting and warning, the location and calibration of these gauges is not necessarily in
accordance with the procedures adopted for the main rainfall station networks. However, the
data from the ERTS stations do provide an additional source of information on large rainfall
events.
3.4.2.1.4. Radio Detection and Ranging

The Bureau of Meteorology's network of Radio Detection and Ranging (RADAR) provides
near-real time estimates of rainfall accumulations which can be used in weather forecasting,
flood modelling and flash flood warning. It also provides information on the spatial extent of
rainfall events and can be used in combination with rainfall station measurements to improve
estimates of rainfall in areas between rainfall stations.
However, because of the relatively sparse spatial distribution of the RADAR network across
Australia and the short period of record, data from RADAR were not able to be used in the
estimation of the design rainfalls.
3.4.2.1.5. Meta-data
Meta-data provides essential information about the rainfall station such as the location of the
rainfall station, the type of instrumentation and data collection method. It therefore provides
context for the rainfall data collected at a station and an indication of its quality. At a
minimum, meta-data relating to location in terms of latitude and longitude were collated for
each rainfall station. However, any additional meta-data that were available including
elevation, details on siting and clearance and photographs were also collated.
Table 2.3.3. Rainfall Reporting Methods
Type Reporting Recording Reporting Reporting Used for

Resolution Interval Method Design
Rainfalls
Daily Daily 24 hour totals Daily to Paper Yes
(9:00 am – monthly
9:00 am)
DINES Continuous 5-6 minutes Daily to Digitised Yes
Pluviograph weekly
18
Design Rainfall
Type Reporting Recording Reporting Reporting Used for

Resolution Interval Method Design
Rainfalls
TBRG Continuous On Hourly to six Logger Yes
occurrence monthly
ERTS Event On On Electronic Some
occurrence occurrence
RADAR Spatial 10 minutes 10 minutes Digital No
3.4.2.2. Sources of Data

The rainfall database used for the estimation of the design rainfalls included rainfall data
collected at rainfall stations operated by organisations around Australia. There were two
main sources of data; the Australian Data Archive for Meteorology and Australian Water
Resources Information System (more detail on these data sources can be found in Book 1,
Chapter 4, Section 9).
Rainfall data collected by the Bureau of Meteorology are stored in the Australian data
Archive for Meteorology (ADAM) which contains approximately 20 000 daily read rainfall
stations (both open and closed) starting in 1800; and nearly 1500 continuous rainfall stations
– using both DINES and TBRG instrumentation.
Under the terms of the Water Regulations 2008, water information (including rainfall data)
collected by organisations across Australia are required to be provided to the Bureau of
Meteorology. The rainfall data collected by organisations including local and state
government water agencies, hydropower generators and urban water utilities are stored in
the Australian Water Resources Information System (AWRIS) together with other water
information. At present, AWRIS contains:
• approximately 350 daily read rainfall stations; and
• approximately 2500 continuous rainfall stations.
Of particular importance to design rainfall estimation are the dense networks of continuous
rainfall stations operated by urban water utilities which provide data in areas of steep rainfall
gradients and urban areas.
3.4.2.3. Spatial Distribution of Rainfall Data
3.4.2.3.1. Daily Read Rainfall Stations
The location and period of record of the daily read rainfall stations operated by the Bureau of
Meteorology are shown in Figure 2.3.3.
Figure 2.3.3 depicts the spatial coverage of the daily read rainfall stations across Australia is
reasonably good, especially over the eastern states and around the coast. Gaps in the
spatial coverage of the daily read rainfall station network occur in the eastern half of Western
Australia; the western and north eastern parts of South Australia; and the parts of the
Northern Territory that are removed from the road and rail networks.
19
Design Rainfall
Figure 2.3.3. Daily Read Rainfall Stations and Period of Record
3.4.2.3.2. Continuous Rainfall Stations

The location and period of record of the continuous rainfall stations operated by the Bureau
of Meteorology and other organisations are shown in Figure 2.3.4. The sparseness of the
network of continuous rainfall stations across Australia can be seen from Figure 2.3.4,
especially when compared to the spatial distribution of the daily read rainfall stations. In spite
of the significant improvement in the spatial coverage of the continuous rainfall stations by
the inclusion of rainfall stations operated by other organsiations, there are still large areas of
Australia with either no or very few continuous rainfall stations.
20
Design Rainfall
Figure 2.3.4. Continuous Rainfall Stations and Period of Record
3.4.2.3.3. Increase in Spatial Coverage

The increase in the spatial coverage of the daily read and continuous rainfall stations used
for the design rainfalls in this edition of ARR compared to the spatial coverage available for
the IFDs provided in ARR 1987 (Pilgrim, 1987) are shown in Figure 2.3.5 and Figure 2.3.6.
21
Design Rainfall
Figure 2.3.5. Daily Read Rainfall Stations Used for ARR 1987 and ARR 2016 Intensity
Frequency Duration Data
The increase in daily read rainfall stations is due to the increased number of stations that
met the minimum period of record criterion.
Figure 2.3.6 shows the inclusion of data from continuous rainfall stations operated by other
organisations has resulted in a significant increase in the spatial coverage of these data. In
particular, the spatial coverage along the east coast of Australia; the west coast of Tasmania;
and large areas in Western Australia has been improved.
22
Design Rainfall
Figure 2.3.6. Continuous Rainfall Stations Used for ARR 1987 and ARR 2016 Intensity
Frequency Duration Data
3.4.2.4. Temporal Distribution of Rainfall Data

Official daily read rainfall data are available from the early 1800s with the longest rainfall
records in Australia being approximately 170 years. Some of these early rainfall stations are
still open. The Bureau of Meteorology’s ADAM database contains approximately 3000 daily
read rainfall stations with more than 100 years of record.
In Figure 2.3.7 the distribution of record lengths for daily read rainfall stations is shown. It
can be seen that, although there are a reasonable number of long term stations,
approximately half of the daily read rainfall stations have less than 10 years of record.
23
Design Rainfall
Figure 2.3.7. Length of Available Daily Read Rainfall Data
While there are a small number of continuous rainfall stations with more than 70 years of
record, the majority of stations have less than 40 years of record and a high proportion have
less than 10 years of record. Figure 2.3.8 shows the distribution of available length of record
for the continuous stations.
Figure 2.3.8. Length of Available Continuous Rainfall Data
24
Design Rainfall
3.4.2.4.3. Increase in Length of Available Record Compared to ARR 1987

The inclusion of nearly 30 years additional daily read rainfall data since the estimation of the
ARR 1987 IFDs has increased both the amount of data available for the frequent and
infrequent design rainfalls (IFDs) as well as the number of daily read rainfall stations which
now met the minimum length of record criterion as shown in Figure 2.3.9.
Figure 2.3.9. Number of Long-term Daily Read Stations Used for ARR 1987 and ARR 2016
Intensity Frequency Duration Data
For the continuous rainfall data, the inclusion of stations operated by other organisations and
the nearly 30 years of additional data resulted in a significant increase in both the length of
record available and the number of rainfall stations that met the minimum record length
criterion (Figure 2.3.10).
25
Design Rainfall
Figure 2.3.10. Length of Record of Continuous Rainfall Stations Used for ARR 1987 and
ARR 2016 Intensity Frequency Duration Data
3.4.2.5. Quality Controlling Data

In addition to ensuring that as much data as possible was used in estimating the design
rainfalls, it was also necessary that the rainfall data be quality controlled. In light of the
volume of data that needed to be quality controlled, automated procedures were developed
for the identification of suspect data and, as far as possible, the correction of these data.
However, the quality controlling of the data could only be automated so far and a significant
amount of data was required to be manually checked.
The quality controlling undertaken of both the daily read and continuous rainfall data is
summarised below (refer to Green et al. (2011) for more information). The quality controlled
database prepared for the estimation of the design rainfalls will be archived in AWRIS and
made available from the Bureau of Meteorology’s website via the Water Data Online product.

For the daily read rainfall data automated quality controlling procedures were developed in
order to:
• infill missing data;
• disaggregate flagged accumulated daily rainfall totals;
• detect, identify and correct suspect data;
• unflagged accumulated totals; and
• time shifts.
• identify gross errors - data inconsistent with neighbouring records but not captured by
either of the above two categories.
26
Design Rainfall
Manual correction of gross errors identified during the automated quality controlling
procedures was facilitated through the use of the Bureau of Meteorology’s Quality Monitoring
System. The Bureau of Meteorology’s Quality Monitoring System is a suite of programs that
has functionalities to map the suspect value in relation to nearby stations and to link to
Geographic Information System (GIS) data from other systems including RADAR, Satellite
Imagery and Mean Sea Level Pressure Analysis.

The automated quality control procedures for continuous rainfall data used comparisons with
other data sources including the Australian Water Availability Project (AWAP) gridded data,
daily read rainfall stations, automatic weather stations, and synoptic stations to identify
spurious and missing data.
In order to reduce the amount of continuous rainfall data that needed to be quality controlled
to a manageable volume, only a subset of the largest rainfall events was quality controlled.
The subset was created by extracting the number of highest rainfall records equal to three
times the number of years of record at each site for each duration being considered.
Each continuous rainfall value in the data subset that was flagged as being spurious by the
automated quality controlling procedures was subjected to manual quality controlling. The
manual quality controlling of the data was undertaken in order to determine whether the
flagged value was correct or not. The manual quality controlling procedure adopted involved
comparing 9:00 am to 9:00 am continuous rainfalls with daily (also 9:00 am to 9:00 am)
rainfalls at the co-located daily read rainfall station. For continuous rainfall sites with no co-
located daily site, the continuous rainfall record was compared with the daily rainfall record
of the nearest site. The continuous rainfall value was not modified in any way - the
comparison with daily values was made in order to assess whether it was valid or not.
Where it was assessed that the flagged value was definitely incorrect it was excluded from
the analyses, otherwise values were retained in the continuous rainfall database.
3.4.2.5.3. Meta-data
The meta-data associated with each of the rainfall stations were also checked. For the
Bureau of Meteorology operated rainfall stations, the Bureau of Meteorology’s meta-data
database, SitesDB, includes details of the station’s location in latitude and longitude, and
elevation. For rainfall stations operated by other organisations, meta-data were provided with
the rainfall data and stored in AWRIS. Gross error checks on station locations and elevation
were performed by comparing elevations derived using a Digital Elevation Model (DEM) to
those recorded in the station’s meta-data. Checks of latitude and location were also carried
out by plotting the latitudes and longitudes in GIS. Revisions to station locations or
elevations were carried out using Google Earth and information on the station provided in
the Bureau of Meteorology’s station meta-data catalogue.
For the limited number of closed stations for which an elevation was not included in the
meta-data, the station elevation was extracted from the Geoscience Australia 9 second
DEM1 based on the latitude and longitude.
3.4.2.6. Stationarity Assessment

The quality controlled database that was established contained rainfall data for the period
extending from the 1800s to the present. However, if climate change has caused non-
1http://www.ga.gov.au/metadata-gateway/metadata/record/gcat_66006
27
Design Rainfall
stationarity in the recorded rainfalls, then possibly only a portion of the observed record
should have been used in deriving the design rainfalls. This is because a key assumption in
the statistical methods adopted for the derivation of the design rainfalls is that the data are
stationary. In order to determine whether the complete period of available rainfall records
could be adopted in estimating the design rainfalls, it was necessary to assess the degree of
non-stationarity present in the historic record at rainfall stations across Australia (Green and
Johnson, 2011).
Two methods were used to establish if there are trends in the Annual Maximum Series of
rainfalls for Australia. The first examined the records at individual stations which were tested
to assess trends in the time series of the annual maximum rainfalls and changes in the
probability distributions fitted to the annual maxima to estimate design rainfall quantiles. The
second method used an area averaged approach to check for regional trends in the number
of exceedances of pre-determined thresholds. The approach was based on that carried out
by Bonnin et al. (2010) to assess trends in large rainfall events in the USA as part of the
revisions by the National Oceanic and Atmospheric Administration to design rainfalls.
It was concluded that although some stations showed strong trends in the annual maximum
time series, particularly for short durations and more frequent events, the magnitude of these
changes was within the expected accuracy of the fitted design rainfall relationships. It was
therefore considered appropriate to assume stationarity and use the complete period of
record at all stations in the estimation of the design rainfalls.
3.4.3. Extraction of Extreme Value Series

Rainfall frequency analysis was an integral part of the estimation of the design rainfalls as it
enabled rainfall depths corresponding to a probability quantile to be ascertained. In
estimating the frequent and infrequent design rainfalls it is large rainfalls that were being
considered and therefore it was the extreme value series that was of interest.
The extreme value series can be defined using the Annual Maximum Series (AMS) or the
Partial Duration Series (PDS) (also known as Peak over Threshold) (more information can
be found in Book 3, Chapter 2). For the frequent and infrequent design rainfalls, the AMS
was used to define the extreme value series because of its lack of ambiguity in defining the
series; its relatively simple application and the problem of bias associated with the PDS for
less frequent AEPs.
It should be noted that in extracting the AMS, the focus was on obtaining the largest rainfall
depth in each year for each of the durations considered. Therefore the extracted depths
comprised both total storm depths and bursts within storms.
In order to reduce the uncertainty in the design rainfall estimates, minimum station record
lengths were adopted. The criteria were:
• 30 or more years of record for daily read rainfall stations; and
• More than 8 years of record for continuous rainfall stations.
These criteria were selected on the basis of optimising the spatial coverage of the rainfall
stations while ensuring that there were sufficient AMS values at each site to undertake
frequency analysis.
The daily read rainfall data are for the restricted period from 9:00 am to 9:00 am rather than
for the actual duration of the event. As this may not lead to the largest rainfall total, it was
28
Design Rainfall
necessary to convert these ‘restricted’ daily read rainfall depths to unrestricted rainfall
depths. In order to do this, ‘restricted’ to unrestricted conversions factors were estimated
using co-located daily read and continuous rainfall gauges at a number of locations around
Australia of differing climatic conditions. The resultant factors are shown in Table 2.3.4.
Table 2.3.4. Restricted to Unrestricted Conversion Factors

Duration 1 Day 2 Days 3 Days 4 Days 5 Days 6 Days 7 Days
Factor 1.15 1.11 1.07 1.05 1.04 1.03 1.02
3.4.3.1. At-Site Frequency Analysis

In order to assess the most appropriate distribution to adopt across Australia for the AMS, a
range of distributions was trialled using single site analysis. Five distributions – Generalised
Extreme Value (GEV), Generalised Logistic (GLO), Generalised Normal (GNO), Log
Pearson III (LP III) and Generalised Pareto (GPA) – were fitted to the AMS extracted from
the available long-term continuous rainfall stations for a range of durations. The goodness of
fit of each distribution was assessed using the approach recommended by Hosking and
Wallis (1997). It was found that the Generalised Extreme Value (GEV) distribution produced
the best fit to the AMS on an at-site analysis. The comparison of distributions was
subsequently repeated for regional estimates with the same results (Green et al., 2012b).
3.4.3.2. Estimation of L-moments

The linear combinations of the data (L-moments) (Hosking and Wallis, 1997) of mean,
variation (L-CV) and skewness (L-Skewness) were used to summarise the statistical
properties of the extreme value series data at each station location. L-moments are
commonly used in rainfall and Flood Frequency Analysis (Hosking and Wallis, 1997) due to
their efficiency in fitting the data and lack of bias in the sample estimates, particularly in the
higher order moments, when compared to ordinary moments.
While for durations of one day and longer this was a fairly straightforward approach, for sub-
daily durations the scarcity of long-term continuous rainfall records meant that an alternative
approach was needed to supplement the available data. For the IFD revision project, a
Bayesian Generalised Least Squares Regression (BGLSR) approach was adopted, a
summary of which is provided in Book 2, Chapter 3, Section 4; more details can be found in
Johnson et al. (2012a) and Haddad et al. (2015).
3.4.3.2.1. Daily Durations

For daily read rainfall stations with 30 or more years of record, the mean, L-CV and L-
Skewness, were determined from the at-site extreme value series for each duration.
3.4.3.2.2. Sub-Daily Durations – at Continuous Rainfall Stations

For continuous rainfall stations with more than eight years of record, the mean, L-CV and L-
Skewness, were determined from the at-site extreme value series for each duration.
The spatial coverage of sub-daily rainfall stations is considerably less than that of the daily
read stations (refer to Figure 2.3.4 and Figure 2.3.5 ). Therefore, a method was needed to
improve the spatial coverage of the sub-daily data. This is most commonly done using
information from the daily read stations with statistics of sub-daily data being inferred from
those of the daily data. Previously, adopted techniques for predicting rainfall depths at
durations below 24 hours from those for the 24, 48 and 72 hour durations have been
29
Design Rainfall
factoring of the 24 hour IFDs; principal component analysis followed by regression; and
Partial Least Squares Regression. However, a major weakness of these previously adopted
approaches is their inability to account for variation in record lengths from site to site and
inter-station correlation.
The approach adopted for the frequent and infrequent design rainfalls was Bayesian
Generalised Least Squares Regression (BGLSR) as it accounts for possible cross-validation
and unequal variance between stations by constructing an error co-variance matrix and can
explicitly account for sampling uncertainty and inter-site dependence. Details of the BGLSR
approach can be found in Reis et al. (2005), Madsen et al. (2002) and Madsen et al. (2009).
In Australia, Haddad and Rahman (2012a) used BGLSR to obtain regional relationships to
estimate peak streamflow in ungauged catchments and for pilot studies for the design rainfall
project (Haddad et al., 2009; Haddad et al., 2011; Haddad et al., 2015).
3.4.3.2.3. Bayesian Generalised Least Squares Regression – Overview

BGLSR is an extension of Ordinary Least Squares (OLS) regression such that the predictor
and (dependent variable) is calculated from a linear combination of a number of predictor
variables (independent variables) with a suitable error model. In general the predictions for
the rainfall statistic, y, of interest for site i, are made according to Equation (2.3.1).
�
�� = �0 + ∑ � ��ij + �� + �� (2.3.1)
j−1
Where Xij (j = 1,…,k) are the k predictor variables, βjare the parameters of the model that
must be estimated, ε is the sampling error and δ is the model error.
A further advantage of the BGLSR is that the Bayesian formulation allows for the separation
of sampling and statistical modelling errors. This is important because it was found that the
sampling errors dominate the total error in the statistical model. The BGLSR produces
estimates of the standard error in:
• the regression coefficients;
• the predicted values at-site used in establishing the regression equations; and
• the predicted values at new sites (that is, sites not used in deriving the regression). In the
application of the BGLSR these are the daily rainfall stations where the predictions of sub-
daily rainfalls statistics are required.
The error variances for the predictions are comprised of the regional model error and the
sampling variance.
The errors in the BGLSR model are assumed to have zero mean and the co-variance
structure described in Equation (2.3.2).
�2� ,
�=� � 2, �=�
�
�� , � � = ; �� , � � = � (2.3.2)
�� ,� ≠ � 0, �≠�
� � ��
Where �2� is the sampling error variance at site i, �� is the correlation coefficient between
� ��
�2��
sites i and j, is the model error variance. For the Bayesian framework introduced by Reis
et al. (2005), the parameters of the model (β) are modelled with a multivariate normal
30
Design Rainfall
distribution using a non informative prior. A quasi analytic approximation to the Bayesian
formulation of the GLSR has been developed by Reis et al. (2005) to solve for the posterior
distributions of the mean and variance for β.
3.4.3.2.4. Application of Bayesian Generalised Least Squares Regression

The aim of the BGLSR is to predict sub-daily rainfall statistics at the location of daily rainfall
stations. As discussed previously, L-moments have been used to summarise the statistical
properties of the AMS data (Hosking and Wallis, 1997) because L-moments are relatively
robust against outliers in the datasets. The statistics that are required for the project are:
• Mean of the AMS (also called the index rainfall);
• L-coefficient of variation (L-CV); and
• L-Skewness.
These three statistics can then be used to define the parameters of any appropriate
probability distribution which in the case of the design rainfalls had been shown to be the
GEV distribution.
The initial work required to apply the BGLSR was to determine the appropriate predictors
(i.e. X from Equation (2.3.1)) to estimate the three rainfall statistics listed above. A review of
literature and meteorological causative mechanisms selected a number of site and rainfall
characteristics for use as possible predictors as reported in Johnson et al. (2012a). These
predictors were:
• Latitude and longitude;
• Elevation;
• Slope;
• Aspect;
• Distance from the coast;
• Mean annual rainfall; and
• Rainfall statistics (mean, L-CV and L-Skewness) for the 24, 48 and 72 hour duration
events.
Haddad and Rahman (2012b) provide extensive details of the cross-validated predictor
selection process for each of the study areas. It was found that the most important predictor
is the 24 hour rainfall statistic. However performance of the model was not changed
significantly by including all predictors so this approach was adopted.
As well as determining the optimum combination of predictor variables, the testing for the
BGLSR needed to determine the number of stations to contribute to each regression
equation. Ideally, the number of stations in each analysis area would be maximised to
improve the accuracy of the regression equations. However the number of stations is limited
to approximately 100 by the requirement for the error co-variance matrices to be invertible.
The delineation of the analysis areas thus needed to balance these two competing
requirements.
31
Design Rainfall
It was also important that stations were grouped into analysis areas where the causative
mechanisms for large rainfall events are similar. The rainfall stations were grouped primarily
according to climatic zones by considering the seasonality of rainfall events and mean
annual rainfalls. Australian drainage divisions were also used to guide the division of larger
climatic zones into smaller areas over which the BGLSR calculations are tractable, such as
in the northern tropics where three analysis areas have been adopted (NT, GULF and
NORTH_QLD). The final analysis areas are shown in Figure 2.3.11. The South East Coast
and South Western WA areas are considered Regions of Interest. A 0.2 degree buffer has
been used in assigning stations to each analysis area to provide a smooth transition
between adjacent areas.
For each analysis area, a regression relationship was developed which could be applied to
all stations within the analysis area. Where the density of stations was high, a Region Of
Influence (ROI) approach (Burn, 1990) was adopted such that each station has its own ROI.
This allowed the regression equations to smoothly vary across the data dense analysis
areas. For sparser analysis areas, a clustering, or fixed region, approach was adopted such
that stations were grouped by spatial proximity into analysis areas with rigid boundaries. All
stations in each analysis area were used to derive one regression equation that was then
adopted for the predictions at those stations.
Figure 2.3.11. Analysis Areas Adopted for the BGLSR
To improve the predictions from the BGLSR it was desirable that the distribution of each
predictor variable was relatively symmetric and preferably approximately normally
distributed. For each analysis area the distribution of the predictor variables from all sites in
the area were examined using histograms and quantile-quantile plots. For predictors that
appeared to be strongly skewed, a range of transformations were trialled to attempt to
reduce the skewness of the variable. The transformations included a natural logarithm,
32
Design Rainfall
square root transformation and Box-Cox (i.e. power) transformation. In general the log
transformation and the Box-Cox transformation were successful in reducing the skewness of
the predictors.
After determining the regression coefficients for the analysis areas, these coefficients were
combined with the set of predictors for the daily station locations to produce the estimates of
the sub-daily rainfall statistics. There are no observations of the sub-daily rainfall statistics to
which these predictions at daily sites can be compared. However “sanity” checks on the
values were carried out by comparing the estimates to the 24 hour rainfall statistics and to
the possible range of values for L-CV and L-Skewness (both limited to -1 to 1).
The result of using estimated sub-daily rainfall statistics was that the number of locations
with sub-daily information was increased from approximately 2300 to approximately 9700
when both the daily and continuous rainfall stations locations are used. This substantially
increased density of sub-daily rainfall data assisted in the subsequent gridding of the rainfall
quantiles across Australia described in Book 2, Chapter 3, Section 4.
3.4.4. Regionalisation
Regionalisation recognises that for stations with short records, there is considerable
uncertainty when estimating the parameters of probability distributions and short records can
bias estimates of rainfall statistics. To overcome this, it is assumed that information can be
combined from multiple stations to give more accurate estimates of the parameters of the
extreme value probability distributions. One approach that is widely used to reduce the
uncertainty and overcome bias in estimating rainfall quantiles is regional frequency analysis,
also known as regionalisation.
For the design rainfalls, regionalisation was used to estimate the L-CV and L-Skewness with
more confidence. The regionalisation approach adopted is generally called the “index flood
procedure” (Hosking and Wallis, 1997). This approach assumes that sites can be grouped
into homogenous regions, such that all sites in the region have the same probability
distribution, other than a scaling factor. The scaling factor is termed the index flood or in this
case, since the regionalisation is of rainfall data, the “index rainfall”. The index rainfall is the
mean (that is, first L-moment) of the extreme value series data at the station location.
The homogenous regions for the frequency analysis can be defined in a number of ways.
Cluster and partitioning methods divide the set of all stations into a fixed number of
homogenous groups (Hosking and Wallis, 1997) where generally every site is assigned to
one group. Alternatively, a ROI approach (Burn, 1990) can be adopted, such that for each
station an individual homogenous region is defined. Each ROI will contain a potentially
unique set of sites, with each site possibly contributing to multiple ROIs.
For the design rainfalls, the station point estimates were regionalised using a ROI as the
advantage of this approach is that the region sizes can be easily varied according to station
density and the available record lengths. The assumptions of the approach are, firstly, that
the specified probability distribution (GEV in the case of the AMS) is appropriate; that the
region is truly homogenous; and, finally, that sites are independent or that their dependence
is quantified.
In the application of the ROI method, it was first necessary to establish how big the ROIs
should be. The size of the ROI can be defined in two ways; either using the number of
stations included in the region or alternatively by calculating the total number of station-years
in the region as the sum of the record lengths of the individual stations included in the ROI.
33
Design Rainfall
Region sizes from 1 to 50 stations and from 50 to 5000 station-years were investigated to
establish the optimum ROIs for estimating rainfall quantiles across Australia using a simple
circular ROI and the Pooled Uncertainty Measure (PUM) (Kjeldsen and Jones, 2009). The
minimum PUM values occur where there is an optimum size in the trade-off between bias
and variance of generally lead to the minimum PUM value. When considering the region
defined using the number of stations it was found that a region of 8 stations performed best.
Given that the average record length for stations used in the analysis was 66 years, a region
of 8 stations will have on average 528 years of data which is consistent with the region size
using the station-year criteria. The findings were generally independent of rainfall event
duration and frequency.
Defining regions in terms of station-years is attractive as this approach can adapt to different
station densities and station record lengths. Given the similar results from both methods, the
station-years definition for the region size was adopted.
After finalising the optimum region size, a number of geographic and non-geographic
similarity measures were investigated as methods to define membership of each ROI. Three
different alternatives for defining the ROIs using geographical similarity were investigated:
• Distance between sites (in kilometres) defined using latitude and longitude;
• Euclidean distance between sites where distance was defined using latitude, longitude
and scaled elevation (Hutchinson, 1998); and
• Nearest neighbours defined using distance in kilometres inside an elliptical ROI.
Non-geographic characteristics were selected based on their potential influence on the

properties of large rainfall events at a site (Johnson et al., 2012a). The site characteristics
that were trialled were:
• Location (latitude and longitude);
• Elevation;
• Mean Annual Rainfall;
• Aspect;
• Slope;
• Distance from the coast;
• Mean date of AMS (seasonality); and
• Variability of AMS occurrence (seasonality).
The results of the trialling showed that the best results were provided using:
• a circular ROI; and
• distances defined in three dimensions using latitude; longitude and elevation.
3.4.4.1. Regionalisation - Application

To undertake the regionalisation the following procedure was followed initially using the 24
hour rainfall data:
34
Design Rainfall
• For each station location, a circular ROI was expanded until 500 stations years of record
was achieved. The resultant region was tested for homogeneity using the H measure of
(Hosking and Wallis, 1997);
• If the region was not homogenous the stations in the regions were checked according to
the discordancy measures of Hosking and Wallis (1997) and the region membership
revised where appropriate;
• The average L-CV for each region was calculated using a weighted average of the L-CV
at all stations in the region, with the weights proportional to the station lengths. This was
repeated for the L-Skewness; and
• The regionalised L-CV and L-Skewness were used to estimate the scale (α) and shape (κ)
parameters of the growth curve (scaled GEV distribution) at each location.
The regions defined for the 24 hour duration rainfall data were used for all daily and sub-
daily durations. More details on the regionalisation can be found in Johnson et al. (2012b).
3.4.5. Gridding
The regionalisation process resulted in estimates of the GEV parameters at all station
locations, which were combined with the mean of the extreme value series at that site to
estimate rainfall quantiles for any required exceedance probability. However frequent and
infrequent design rainfall estimates are required across Australia, not just at station locations
and therefore the results of the analyses needed to be extended in some way to ungauged
locations.
3.4.5.1. Selection of Approach to be Adopted for Gridding

For the design rainfalls, the software package ANUSPLIN (Hutchinson, 2007) was chosen to
grid the GEV parameters so that frequent and infrequent design rainfall estimates are
available for any point in Australia. ANUSPLIN applies thin plate smoothing splines to
interpolate and smooth multi-variate data. The degree of smoothing of the fitted functions
was determined through generalised cross-validation. The splines are fitted using three
independent variables; latitude, longitude and elevation. The elevation scale was
exaggerated by a factor of 100 to represent the importance that elevation has on
precipitation patterns (Hutchinson, 1998).
3.4.5.2. Selection of Parameters to be Gridded

The GEV parameters were gridded in ANUSPLIN rather than the rainfall depths, as testing
showed little difference in the resulting quantile estimates irrespective of whether point GEV
parameters or point rainfall depths were gridded. Gridding the GEV parameters provided
more flexibility in the choice of exceedance probabilities that could be extracted and enables
the provision of a greater number of AEPs.
3.4.5.3. Optimisation of Gridding

In undertaking the gridding, a considerable number of iterations was required to achieve an
optimum outcome that represented the observed rainfalls but which did not place too much
significance on short rainfall records or from poorly located rainfall stations. The appropriate
degree of smoothing of the fitted functions was determined through generalised cross-
validation with the number of knots and transformation adopted varied to achieve optimal
results. In addition to the statistical tests to determine the appropriate degree of smoothing,
35
Design Rainfall
qualitative assessments were also conducted by preparing maps which compared the index
rainfall derived from at-site frequency analysis of rainfall records, the length of record
available at each station, and the spatial density of the rainfall gauge network to the gridded
index rainfalls produced by ANUSPLIN for daily durations.
The final IFD grids were produced by the application of ANUSPLIN using a 0.025 degree
DEM resolution and adopting 3570 knots with no transformation of the data. More details on
the gridding approach adopted can be found in The et al. (2012), The et al. (2014) and
Johnson et al. (2015).
3.4.5.4. Calculation of Growth Factors and Rainfall Depths

The outputs of the ANUSPLIN analysis were grids across Australia of index rainfall and the
GEV scale (�) and shape (�) parameters for each duration. These were then processed to
firstly estimate the growth factors for each grid location and then the rainfall depths for each
exceedance probability, according to the following equations:
� = 1 − � 1 − �(1 + �) /� (2.3.3)
where � is the location parameter for the regionalised growth curve and � represents the
Gamma function.
�(�) = � + � 1 − ( − log(�))� /� (2.3.4)
where �(�) is the quantile function of the growth curve for the cumulative probability �.
�(�) = ��(�) (2.3.5)
where �(�) is the quantile function of the scaled growth curve, which is multiplied by the
index rainfall �.
3.4.5.5. Derivation of Sub-Hourly Rainfall Depths

To derive frequent and infrequent design rainfalls for durations of less than one hour to one
minute the ‘simple scaling’ model developed by Menabde et al. (1999) was adopted. The
model was calibrated using the AMS from the Bureau of Meteorology’s continuous rainfall
stations with more than eight years of data. For each continuous rainfall station the scaling
factor, η, was determined and the at-site η values gridded to provide estimates for all grid
locations. The model was then applied to the one hour duration rainfall depth grids to
estimate the rainfall depths for the 1 minute to 30 minute rainfall events according to the
following equation:
� �
�� = � (2.3.6)
� �
Where �� is the sub-hourly rainfall intensity for duration � , �� is the 60 minute rainfall
intensity (ie. duration � is 60 minutes).
3.4.5.6. Consistency Checking and Smoothing

In order to reduce inconsistencies across durations and smooth over discontinuities in the
gridded data (unevenly spaced differences in design rainfall estimates at neighbouring
durations) arising from application of the method, a smoothing process was undertaken. This
36
Design Rainfall
was done by applying a sixth order polynomial to each grid point to all the standard durations
from one minute up to seven days.
Although polynomials up to order 12 were investigated, a sixth order polynomial was

adopted as investigations showed that this order polynomial gives adequate results.
Inconsistencies with respect to duration (rainfall depths at lower durations exceeding those
at higher durations) were also found and were addressed.
Inconsistencies were detected by subtracting each grid from a longer duration grid at the
same probability and checking for negative values. Inconsistencies were addressed by
adjusting the longer duration rainfall upwards so that the ratio of shorter duration rainfall to
the longer duration rainfall equals 0.99 or
Rainfall depth at the shorter duration
Rainfall depth at the longer duration
= 0.99
The smoothing procedure was applied first to the original grids and the smoothed grids
adjusted for inconsistencies. The grids were smoothed once again and a final adjustment for
inconsistencies across durations was performed. The final grids were also checked for
inconsistencies across AEP.
Grids of the polynomial coefficients were prepared in order to enable IFDs for any duration to
be determined.
3.4.6. Outputs
The method described in Book 2, Chapter 3, Section 4 produced frequent and infrequent
design rainfall estimates across Australia. The design rainfall estimates are provided both as
rainfall depths in millimetres (mm) and rainfall intensities in millimetres per hour (mm/hr) for
the standard durations and standard probabilities described in Table 2.3.5.
Table 2.3.5. Intensity Frequency Duration Outputs

Output Values Units
Standard durations 1, 2, 3, 4, 5, 10, 15, 30 Minutes
1, 2, 3, 6, 12 Hours
1, 2, 3, 4, 5, 6, 7 Days
Standard probabilities 1 EY
63.2%, 50%, 20%, 10%, 5%, AEP
2%, 1%
3.5. Very Frequent Design Rainfalls

3.5.1. Overview
The previous section summarised the steps involved in deriving the frequent and infrequent
design rainfall values (IFDs) for probabilities from 1EY to 1% AEP. This range of probabilities
is suitable for most design situations, however, many stormwater quality or Water Sensitive
Urban Design guidelines recommend a flow threshold of Q3month for the design of
stormwater quality treatment devices.
Design rainfalls for the three month Average Recurrence Interval (or 4 EY) have not been
previously available, with agencies giving their own advice on the approach for estimating
37
Design Rainfall
very frequent design rainfalls. To address this need, estimates for probabilities more frequent
than 1 EY have been derived.
To ensure consistency between the very frequent design rainfalls and the frequent and
infrequent design rainfall, the overall approach adopted for the very frequent design rainfalls
was very similar to that adopted for the frequent and infrequent design rainfall. However,
some modifications to the approach were necessary because of the increased frequency of
occurrence that was being considered. A summary of the method is presented in Table 2.3.6
and in Book 2, Chapter 3, Section 5 to Book 2, Chapter 3, Section 5. Further details can be
found in The et al. (2015).
Table 2.3.6. Very Frequent Design Rainfall Method

Step Method
Number of Rainfall Stations Daily read – 15 364
Continuous – 2722
Period of Record All available records up to 2012
Length of Record used in Analyses Daily read > 5 years
Continuous > 5 years

Source of Data Organisations collecting rainfall data across
Australia
Extreme Value Series Partial Duration Series (PDS)
Frequency Analysis Generalised Pareto (GPA) distribution fitted
using L-moments
Ratios Ratio X EY to 50% AEP
Gridding Regionalised at-site distribution parameters
gridded using ANUSPLIN

The data adopted consisted of the stations used for the frequent and infrequent design
rainfalls (Green et al., 2011) and an additional 7290 stations with shorter periods of record,
which have undergone the same rigorous quality controlling as the frequent and infrequent
design rainfall database (Green et al., 2011; Green et al., 2012a). The locations of the
stations used for the estimation of the very frequent design rainfalls are shown in
Figure 2.3.12.
Additional stations could be used as the minimum number of years of record was reduced
from 30 (for the frequent and infrequent design rainfalIs) to five years for the very frequent
design rainfalls. A threshold of five effective years was selected for daily and sub-daily sites
as this was deemed to be statistically acceptable given the high frequency of the estimated
exceedances compared to the previous 1 EY. The shorter record length ensures greater use
of available sites but also ensures that there is sufficient information available to derive the
more frequent probabilities from 12 EY to 2 EY.
38
Design Rainfall
Figure 2.3.12. Daily Read Rainfall Stations and Continuous Rainfall Stations Used for Very
Frequent Design Rainfalls

A Partial Duration Series (PDS) approach was adopted to estimate probabilities for events
occurring more frequently than once a year. The advantage of using the PDS is that it
extracts as much information as possible about large events and produces more accurate
estimates for very frequent probabilities. As a PDS was used for the at-site series the
selection of independent events was based on rank rather than temporal periods, thus
completeness of record was not a consideration.
As a PDS approach was being adopted it was necessary to define the threshold above
which all events will be included. It was important to identify the number of values per year
that are required to accurately estimate the more frequent IFD's. Given that the most
frequent probability is 12 EY, a minimum of 12 events per year was used to adequately
represent the at-site distribution for these higher frequency events.
An assumption of the method, is that the events in the PDS are independent. In order to
ensure that the events in the PDS were independent, a method that provided a consistent
and meteorologically rigorous approach to defining independence of rainfall events across
Australia was developed. The event independence testing criteria used were based on the
Minimum Inter-event Time (MIT) approach (Xuereb and Green, 2012). The analyses
suggested that a MIT that varied from two to six days with latitude across Australia was
appropriate for event durations up to three days while, for durations longer than three days
the MIT adopted was zero. For durations of less than one day, the MIT for the one day
duration was adopted (Green et al., 2015).
39
Design Rainfall
3.5.3.1. At-Site Frequency Analysis

The approach adopted effectively treated the PDS as a Monthly Partial Duration Series or,
more correctly for the current dataset, a Monthly Exceedance Series (MES) (PDS where
number of values = number of effective months: 12 nE). While the extracted PDS is the
same, the averaging duration is changed, representing the time in months rather than years.
The selection of an MES rather than an annual series allowed significantly more records to
be included from each site to establish the at-site rainfall distribution, capturing the more
frequent rainfall patterns.
As discussed in Book 2, Chapter 3, Section 4 and Green et al. (2012b), testing of the most
appropriate distribution to adopt for both the AMS and the PDS was undertaken as part of
the derivation of the IFDs with results identifying the GEV distribution as the most
appropriate for the AMS and the GPA distribution for the PDS. However, as a monthly
exceedance data series was adopted for the very frequent design rainfalls there is some
added uncertainty; to address this, a comparison was conducted of the GEV and GPA
distributions. Twenty-four geographically distributed test sites with medium to long record
lengths were selected for assessing the relative fit of the distributions to the at-site data. The
test sites indicated that the GPA provides a closer fit to the site data in the majority of cases.
On the basis of this, the very frequent design rainfalls used the GPA distribution fitted to the
PDS for all stations which met the required record length.
3.5.3.2. Estimation of L-moments

Regional frequency analysis was undertaken using L-moments extracted from each of the
at-site frequency distributions for sub-daily and daily data. The L-moments were used to
estimate the parameters of the selected GPA distribution.
Extracting 12 independent events per year of record for the MES introduced the issue of
zero values included in the PDS at some sites. This particularly occurred through the arid
areas of central Australia to the west coast, where annual rainfall is highly variable and
strong seasonality can occur. These areas have short wet seasons and can fail to have 12
rain events on average that are independent of one another for every year. However, given
the previously defined minimum number of events being 12, these zero values events are
considered as part of the distribution. To manage the occurrence of the zero values in the
extreme value series, Hosking and Wallis (1997) suggest using a ‘mixed distribution’ or more
correctly a conditional probability adjustment that gives a probability of a zero value, and
cumulative distribution for the non-zero values as seen in Equation (2.3.7) (Guttman et al.,
1993).
�(�) = � + (1 − �)�(�) (2.3.7)
Where � is the probability of a zero rainfall value which is estimated by dividing the numbers
of zeros by the total number of events and �(�) is the cumulative distribution function of the
non-zero rainfall events. Using this approach, if the Non-Exceedance Probability (NEP) of
interest is less than � , then the quantile estimate is zero and if the NEP is greater than �,
the quantile is estimated from �(�) using the adjusted NEP shown in Equation (2.3.8).
�� = (�� − �)/(1 − �) (2.3.8)
For series with a small proportion of zeros, the impact on the distribution and resulting
quantiles was negligible. For records with less than 10% zeros, there is very little difference
and for up to nearly 20% zeros there is less than 10% average difference in the quantile
40
Design Rainfall
depths. However, the differences become much more significant when the proportion of
zeros increases.
3.5.4. Ratio Method

The parameters of the GPA distribution derived from the L-moments were used to calculate
at-site quantiles. The parameters or quantiles could be gridded in a similar method to the
frequent and infrequent design rainfalls and smoothed or rescaled to integrate them with the
design rainfalls less frequent than 1 EY. Alternatively, as adopted, a ratio method was
applied to derive the very frequent design rainfall estimates. A general ratio approach is
currently used by various councils and authorities in Australia and internationally (Huff and
Angel, 1992). It involves using the at-site data to determine the ratio of the various very
frequent design rainfall values to either the 1 EY or 50% AEP gridded design rainfalls.
The ratio method adopted involves estimating at-site quantiles, using the at-site 50% AEP as
the reference values for the ratios and gridding the calculated ratios. The advantage of this
approach and using the at-site 50% AEP, was that it allows for the spatial variability in the
ratios. In addition, the ratio was generally a more accurate representation of the X EY to
50% AEP ratio since it was calculated from the same dataset and resulted in a smooth
spatial pattern. Consistency was also inherent since the ratios would always decrease with
increasing probability. Since the ratios were spatially consistent, the final very frequent
design rainfall depths follow the frequent and infrequent 50% AEP depths closely. These
depth estimates were calculated using the gridded ratios, and multiplying by the 50% AEP
design rainfall.
3.5.5. Gridding
As with the frequent and infrequent design rainfalls the ratios for all durations and EYs were
gridded using the splining software ANUSPLIN (Hutchinson and Xu, 2013). To determine the
most appropriate method to adopt for the gridding of the ratios a range of tests was
undertaken of combinations of variates and different knot sets. The final case adopted was a
spline that incorporated latitude, longitude and elevation using 4000 knots for the daily
dataset and 1000 knots for the sub-daily dataset. The 0.025 degree Digital Elevation Model
of Australia was used to provide the elevation data which were the same as that used in the
derivation of the frequent and infrequent grids (The et al., 2014).
3.5.5.1. Depth Estimates

Very frequent design rainfall depth estimates for each duration and EY were calculated by
multiplying the ratio grids with the corresponding 50% AEP design rainfall grids
(Figure 2.3.13). As the frequent and infrequent design rainfall grids were based on AMS
estimates, an AMS/PDS conversion factor was applied to account for the lower estimates
(Green et al., 2012b).
The grids were smoothed to reduce any inconsistencies across durations and to smooth
over discontinuities in the gridded data. A sixth order polynomial was applied to each grid
point for all the standard durations from 1 minute up to 7 days. Grids were also checked for
inconsistencies across EY.
41
Design Rainfall
Figure 2.3.13. Procedure to Derive Very Frequent Design Rainfall Depth Grids From Ratios
3.5.6. Outputs
The method described in Book 2, Chapter 3, Section 5 to Book 2, Chapter 3, Section 5
produced very frequent design rainfall estimates across Australia. The very frequent design
rainfall estimates are provided both as rainfall depths in millimetres (mm) and rainfall
intensities in millimetres per hour (mm/hr) for the standard durations and standard
probabilities in Table 2.3.7:
Table 2.3.7. Very Frequent Design Rainfall Outputs
Output Values Units

Standard Durations 1, 2, 3, 4, 5, 10, 15, 30 Minutes
1, 2, 3, 6, 12 Hours
1, 2, 3, 4, 5, 6, 7 Days
Standard Probabilities 12, 6, 4, 3, 2, 1, 0.5, 0.2 EY
3.6. Rare Design Rainfalls

3.6.1. Overview
Rare design rainfalls (for 1 in 100 to 1 in 2000 AEP) are used by engineers, hydrologists,
and planners for a range of purposes including:
• the design of dams that fall into the Significant and Low Flood Capacity Category where
the Acceptable Flood Capacity is the 1 in 1000 AEP design flood (ANCOLD, 2000);
• the design of bridges, where the ultimate limit state adopted in the Australian bridge
design code is defined as ‘the capability of a bridge to withstand, without collapse, the
design flood associated with a 2000 year return interval’ (Austroads, 1992);
• the incorporation of climate change into IFDs in accordance with Book 1, Chapter 6 which
recommends that if the design probability for a structure is 1% AEP, then the possible
impacts of climate change should be assessed using 0.5% and 0.2% AEP (Bates et al.,
2015); and
• the undertaking of spillway adequacy assessments of existing dams as the Dam Crest
Flood (DCF) of many dams lies between the 1% AEP flood and the Probable Maximum
Flood (as defined by the Probable Maximum Precipitation, PMP). Rare design rainfalls
enable more accurate definition of the design rainfall and flood frequency curves between
the 1% AEP and Probable Maximum Events.
Unlike the derivation of very frequent, frequent and infrequent design rainfalls which are
based on observed rainfall events that lie within the range of probabilities being estimated,
rare design rainfalls are an extrapolation beyond observed events. The longest period for
which daily read rainfall records are available is around 170 years (Figure 2.3.3 and
42
Design Rainfall
Figure 2.3.7) however rare design rainfalls are required for probabilities much rarer than this.
As a consequence it is difficult to validate the resultant rare design rainfalls and therefore the
method adopted needs to be based on a qualitative assessment that the assumptions made
in the method are reasonable and that the adopted approach is consistent with methods
used to derive more frequent design rainfalls where the results can be validated.
The method adopted for deriving the rare design rainfalls was based on the data and
method adopted for the more frequent design rainfalls but places more weight on the largest
observed rainfall events which are of most relevance to rare design rainfalls. The adopted
regional LH-moments approach is summarised in Table 2.3.8 and Book 2, Chapter 3,
Section 6 to Book 2, Chapter 3, Section 6. More detail can be found in Green et al. (2015)
and Bureau of Meteorology (2016).
Table 2.3.8. Rare Design Rainfall Method

Step Method
Number of rainfall stations Daily read – 8074 for index value
Daily read – 3955 for LCV and LSK

Period of record All available records up to 2012
Length of Record Used in Analyses Daily read ≥ 30 years for index value
Daily read ≥ 60 years for LCV and LSK

Source of Data Bureau of Meteorology
Extreme Value Series Annual Maximum Series (AMS)
Frequency Analysis Generalised Extreme Value (GEV)
distribution fitted using LH(2)-moments
Regionalisation Region of Influence
Gridding Regionalised GEV parameters gridded using
ANUSPLIN

The quality controlled rainfall database established for the derivation of the more frequent
design rainfalls was used as the basis for the database used for the rare design rainfalls.
However, as the estimation of rare design rainfalls relies on long-term records, only those
stations with more than 60 years of record were selected. This reduced to data set to
approximately 4000 stations, the locations of which are shown in Figure 2.3.14. In order to
ensure consistency with the more frequent design rainfalls, the index values were derived
using the same dataset as the more frequent design rainfalls.
43
Design Rainfall
Figure 2.3.14. Daily Read Rainfall Stations with 60 or More Years of Record

The AMS was extracted from all daily read rainfall stations with 60 or more years of record.
The AMS was used to define the extreme value series for the rare design rainfalls as the
focus is on the largest recorded events.
As discussed in Book 2, Chapter 3, Section 4, the GEV distribution was adopted for AMS for
the frequent and infrequent design rainfalls following extensive testing of a range of
candidate distributions. On the basis of these trials and similar results found by Nandakumar
et al. (1997) and Schaefer (1990), the GEV distribution was adopted for the rare design
rainfall analyses.
In keeping with the approach adopted for the more frequent design rainfalls, the statistical
properties of the at-site data were estimated and then translated into the relevant GEV
distribution parameters. However, whereas L-moments were used for the more frequent
design rainfalls, for the rare design rainfalls LH-moments were adopted (Wang, 1997). LH-
moments were adopted as they more accurately fit the upper tail (rarer probabilities) of the
distribution.
LH-moments are a generalisation of L-moments and allow the distribution to be increasingly

focused on the larger data values depending on the value of ƞ, where ƞ=0 is equivalent to L-
moments. The equations for deriving LH-moments and the associated GEV parameters are
given in Wang (1998). LH-moments with a ƞ=2 were selected as a compromise between
providing a better fit to the tail of the at-site distribution without giving too much influence to
the high outliers. LH-moments (ƞ=2) were derived for all stations with greater than 60 years
AMS.
44
Design Rainfall
3.6.4. Regionalisation
For the rare design rainfalls, the ROI approach adopted for the IFDs was used to reduce the
uncertainty in the estimated LH-moments by regionalising the station point estimates. While
500 station years was found to be an optimum pool size for the IFDs, because the rare
design rainfalls are provided for probabilities up to 1 in 2000 AEP, the ROI needed to be
increased. The tradeoff between gaining improved accuracy from a larger pool of data was
that the assumption of homogeneity may not be satisfied. Testing was conducted to find the
pool size that reduced uncertainty without introducing significant homogeneity, with a
minimum of 2000 station years adopted. However, where necessary, the number of pooled
station years was increased above this number to maximize the available record used, while
ensuring homogeneity.
The average LH-CV for each region was calculated using a weighted average of the LH-CV
at all stations in the region, with the weights proportional to the station lengths. This was
repeated for the LH-Skewness.
3.6.5. Gridding
The Index, regionalised LH-CV and LH-SK values for all durations and AEP's were gridded
using the splining software ANUSPLIN (Hutchinson and Xu, 2013) that was adopted for the
more frequent design rainfalls. To determine the most appropriate method to adopt for the
gridding of the moments a range of tests was undertaken of combinations of different knot
sets. The final case adopted was a spline that incorporated latitude, longitude and elevation
using 3750 knots for the Index (as was adopted for the more frequent design rainfalls) and
2200 knots for the regionalised LH-CV and LH-SK values. The 0.025 degree Digital
Elevation Model (DEM) of Australia was used to provide the elevation data which was the
same as that used in the derivation of the frequent and infrequent grids (The et al., 2014) .
In order to provide consistent design rainfall estimates across all durations and probabilities,
a suitable method was required to integrate the rare design rainfalls with the more frequent
design rainfalls. After testing of various ‘anchor’ points, the rare design rainfalls were
anchored to the more frequent design rainfalls at the 5% AEP as it was considered that the
rare design rainfalls provide a better estimate of the upper tail of the distribution down to the
5% AEP.
3.6.6. Outputs
The method described in Book 2, Chapter 3, Section 6 to Book 2, Chapter 3, Section 6
produced rare design rainfall estimates across Australia for the standard durations and
standard probabilities in Table 2.3.9.
Table 2.3.9. Rare Design Rainfall Outputs

Output Values Units
Standard Durations 1, 2, 3, 4, 5, 6, 7 Days
Standard Probabilities 1 in 100; 1 in 200; 1 in 500; 1 AEP
in 1000; 1 in 2000
45
Design Rainfall
3.7. Probable Maximum Precipitation Estimates

3.7.1. Overview
The design rainfalls classes described in Book 2, Chapter 3, Section 4 and Book 2, Chapter
3, Section 5 were derived using frequency analysis. However, extreme rainfalls events such
as the Probable Maximum Precipitation (PMP), lie beyond both any directly observed events
and the limit to which observed data can be extrapolated. As a result, estimation of extreme
rainfall events is based on the broadest understanding of extreme events and the
meteorological processes that produce them.
The Probable Maximum Precipitation (PMP) is defined as ‘the theoretical greatest depth of
precipitation that is physically possible over a particular catchment’ (World Meteorological
Organisation, 1986). The PMP assumes the simultaneous occurrence in one storm of
maximum amount of moisture and the maximum conversion rate of moisture to precipitation
(maximum efficiency).
3.7.2. Estimation of PMPs

The Bureau of Meteorology has been providing PMP estimates for over 70 years, however,
the methods adopted have changed with time, as the understanding of extreme storms and
the mechanisms which produce them have developed, and the databases of observed
extreme storms have expanded. These methods include:
• In Situ Maximisation Method: During the 1950’s to 1970’s PMP estimates were based
on the maximisation of the moisture content of storms which had been observed over the
catchment of interest. The limitation of this method was that the differing lengths of rainfall
records and occurrence or non-occurrence of an extreme storm led to inconsistent PMP
estimates for catchments within the same region.
• Storm Transposition Method: During the late 1960’s and early 1970’s the size of the
extreme storm sample for a specific catchment was increased by the transposition to the
catchment of interest of extreme storms which had been observed over nearby
catchments which had similar hydrometeorological and topographic features. Although this
improved the within-region consistency of PMP estimates, the method was limited, as only
storms from a similar topographic region could be transposed, and the selection of storms
introduced a significant level of subjectivity.
• Generalised Methods: From the mid-1970’s generalised methods were introduced into
Australia. Generalised methods make use of all available storm data for a large region by
making adjustments for moisture availability and differing topographic effects. The
generalised methods currently adopted in Australia are described in Book 2, Chapter 3,
Section 7.
3.7.3. Generalised Methods for Probable Maximum

Precipitation Estimation
There are three main generalised methods used for PMP estimation in Australia. These
methods and their area of applicability are shown in Figure 2.3.15 and described in Book 2,
Chapter 3, Section 7 to Book 2, Chapter 3, Section 7. There are two regional methods (Book
2, Chapter 3, Section 7, Book 2, Chapter 3, Section 7).
46
Design Rainfall
Figure 2.3.15. Generalised Probable Maximum Precipitation Method Zones
3.7.3.1. Generalised Short-Duration Method

The Generalised Short Duration Method (GSDM) is applicable across Australia for
catchment areas less than 1000 km2 and for durations up to three hours or six hours
depending on the location within of Australia (Bureau of Meteorology, 2003).
3.7.3.2. Generalised South-east Australia Method

The Generalised South-east Australia Method (GSAM) is applicable to the southern third of
Australia where it is assumed that the causative mechanism of the PMP would not be
tropical. The GSAM method is applicable for durations from 12 hours to 120 hours (Bureau
of Meteorology, 2006; Minty et al., 1996).
3.7.3.3. Revised General Tropical Storm Method

The Revised Generalised Tropical Storm Method (GTSMR) is applicable to the northern two-
thirds of Australia where it is assumed that the causative mechanism of the PMP would be a
tropical storm. The GTSMR method is applicable for durations from 12 hours to 120 hours
(Bureau of Meteorology, 2005; Walland et al., 2003).
3.7.3.4. West Coast Tasmania Method Zone

The West Coast Tasmania Method Zone applies to the west coast region of Tasmania which
is outside the region of applicability of the Tasmanian GSAM coastal zone. It is applicable for
durations between 24 and 72 hours (Xuereb et al., 2001).
47
Design Rainfall
3.7.3.5. GSAM-GTSMR Coastal Transition Zone

The GSAM-GTSMR Coastal Transition Zone method to the coastal area in NSW where is it
considered that PMPs could be caused by either tropical or non-tropical storms.
3.7.4. Generalised Method of Probable Maximum Precipitation

Estimation
Although each of the three generalised methods has specific features, the generalised
method can be summarised as follows:
3.7.4.1. Development of Storm Database

The ten highest one to seven day rainfalls which were common to a number of stations were
selected and the storms prioritised according to the rarity of the event in order to identify the
100 or more largest storms. For each storm the following analyses were undertaken.
• The rainfall totals for the total storm duration were plotted on a topographic map and
isohyets drawn to determine the spatial extent and distribution of each storm.
• To determine the storm temporal distribution, parallelograms were drawn around the storm
centre for standard areas of 100; 500; 1 000; 2 500; 10 000; 40 000 and 60 000 km2. The
average daily rainfall depths within a parallelogram were determined using Thiessen
weights. For each standard area, the percentage of the total storm that fell during each 24
hour period was determined. These daily data were supplemented by pluviograph and 3
hourly synoptic charts.
• The representative dew point temperature for each storm was determined using a number
of sources including the Australian Region Mean Sea Level charts, National Climate
Centre Archives and Observers’ Logbooks.
3.7.4.2. Generalisation of Storm Database

The ‘site specific’ attributes of each storm were removed in order to attain a homogenous
data set.
The effects of storm type were removed from the data set by the dividing of Australia into the
GSAM and GTSMR regions on the basis of the type of storm that produces the largest
observed rainfall depths. The two regions were further divided into Coastal and Inland Zones
on the basis that different mechanisms produce the largest rainfall depths in each of the
zones (refer Figure 2.3.15).
The specification of each storm in terms of depth-area-duration curves as done previously

effectively removed the storm specific spatial distribution.
The removal of the site specific topographic effects was undertaken using 72 hour, 50 year
ARI ‘flat land’ rainfall intensity field in order to produce the convergence component of each
storm.
3.7.4.3. Removal of the Site-Specific and Storm-Specific Moisture

Content
To remove the storm-specific moisture content from each storm and simultaneously to
maximise the moisture content, the convergence depths were multiplied by a moisture
48
Design Rainfall
maximisation factor. The moisture maximisation factor is defined as the ratio of extreme
precipitable water associated with the extreme dew point temperature at the storm location.
The site-specific moisture content of each storm was removed by transposition to a single
location which for the GSAM was chosen as Brisbane and for GTSMR as Broome. For each
location, representative seasonal extreme 24 hour persisting dew point temperatures were
selected and the moisture content for each storm standardised.
3.7.4.4. Determination of ‘Storm’ of Maximum Moisture Content

The moisture maximisation factor and the standardised convergence depths were combined
to estimate the maximised standardised convergence depths. To determine a single
hypothetical storm of maximum moisture content, an envelope curve was drawn to the set of
maximised, standardised convergence depth-area curves.
3.7.4.5. Determination of Catchment Specific Probable Maximum

Precipitation
The envelope curves represent the maximised convergence component of the PMP at the
standardising locations (Brisbane for the GSAM and Broome for GTSMR). To obtain an
estimate of the PMP for a specific catchment it is necessary to build in the moisture content
and topographic influences specific to the catchment of interest. The moisture content of the
standard PMP convergence depth is adjustment using a Moisture Adjustment Factor (MAF)
such that:
MAF=EPWcatchment/EPWstd (2.3.9)
where MAF = Moisture Adjustment Factor; EPWcatchment is the Extreme Precipitable Water
associated with the catchment extreme dew point temperature; EPWstd is the Extreme
Precipitable Water associated with the standard extreme dew point temperature for
appropriate season.
The Topographic Enhancement Factor (TEF) for the catchment PMP is estimated in the
same manner as the topographic component of the storms in the database using the 72
hour 50 year ARI rainfall intensities.
The total PMP for a specific catchment for each of the standard durations is estimated as:
PMPcatchment=MCDstd*MAFcatchment*TEFcatchment (2.3.10)
3.8. Uncertainty in Design Rainfalls

The design rainfalls described in this Chapter are the best estimates currently available and
have been derived using an extensive rainfall database which has been analysed using the
most appropriate techniques.
However, there are uncertainties associated with the design rainfalls which arise from
various sources including:
• errors in the data due to short record length, instrumentation errors, gaps in the data,
unidentified errors in the data;
• sampling errors including network sparsity, poorly placed gauges, non-representativeness

of gauging networks; and
49
Design Rainfall
• limitations in the adopted methods including delineation of regions, lack of homogeneity in

regions, selection of distribution, parameter and quantile estimation, and extrapolation of
data to ungauged locations.
These uncertainties need to be taken into consideration when the design rainfalls are being
used in conjunction with other design flood inputs. Quantification of the uncertainties
associated with the design rainfalls is described in Bureau of Meteorology (2016) as well as
advice on how to incorporate uncertainty when using the design rainfalls.
3.9. Application
3.9.1. Design Rainfalls
The very frequent, frequent, infrequent, and rare design rainfalls are available via the Bureau
of Meteorology’s website2.
3.9.1.1. Input Data Requirements and Options

To estimate a design rainfall for a point location it is necessary to enter the co-ordinate of the
location in one of three co-ordinate format options:
• Decimal degrees;
• Degrees, Minutes, Seconds; and
• Easting, Northing, Zone.
The location of the entered co-ordinate can be seen by using the map preview option and a
location label can also be entered (see Figure 2.3.16).
2http://www.bom.gov.au/water/designRainfalls/ifd/index.shtml
50
Design Rainfall
Figure 2.3.16. Design Rainfall Point Location Map Preview
Determine the design rainfall class for which design rainfalls are required:
• Very frequent
• Frequent and infrequent (IFDs)
• Rare
Determine the durations for which design rainfalls are required:
• Standard
• Non-standard
Determine the units in which the design rainfalls will be provided:
• Depths as millimetres (mm)
• Intensities as millimetres per hour (mm/hr)
51
Design Rainfall
3.9.2. Frequent and Infrequent Design Rainfalls (IFDs)
Figure 2.3.17. IFD Outputs
52
Design Rainfall
3.9.3. Very Frequent Design Rainfalls
Figure 2.3.18. Very frequent Design Rainfall Outputs
53
Design Rainfall
3.9.4. Rare Design Rainfalls
Figure 2.3.19. Rare design rainfall outputs
3.9.4.1. Output Options

The design rainfall information for the selected location can be seen in either tabular form
(Figure 2.3.20) or graphical form (Figure 2.3.21).
54
Design Rainfall
Figure 2.3.20. Design Rainfall Output Shown as Table
55
Design Rainfall
Figure 2.3.21. Design Rainfall Output Shown as Chart
3.9.4.2. Export Options

The table can be exported as a .csv file for ease of incorporation into software and the chart
as a .png to facilitate integration into reports.
3.9.5. Probable Maximum Precipitation Estimates

Probable Maximum Precipitation estimates for the GSDM can be estimated using the
method contained in Bureau of Meteorology (2003) which can be downloaded from the
Bureau of Meteorology’s website.
56
Design Rainfall
Guides for the application of the GSAM and GTSMR methods are available from the Bureau
of Meteorology (Bureau of Meteorology, 2005; Bureau of Meteorology, 2006).
3.10. Acknowledgements
The authors would like to acknowledge the contributions made by the following people
during the IFD Revision Project:
• Dr Louise Minty and Dr Ian Prosser for their direction of the project
• Andrew Frost, Khaled Haddad, Catherine Jolly, Garry Moore, Scott Podger, Ataur
Rahman, Lionel Siriwardena and Karin Xuereb for their contribution during various stages
of the project
• Professor Mike Hutchinson for his ongoing advice on the application of ANUSPLIN
• Erwin Jeremiah, Deacon McKay, Scott Podger and Michael Sugiyanto for their cheerful
quality controlling of the data
• Dr George Kuczera, Dr Nanda Nandakumar; Dr Rory Nathan and Erwin Weinmann for
their sage advice on the rare design rainfalls
• Martin Chan, Damian Chong, Ceredwyn Ealanta Murray Henderson, Chris Lee, Quentin
Leseney, Maria Levtova, Max Monahon, Julian Noye, and William Tall for the webpages.
3.11. References
ANCOLD (Australian National Committee on Large Dams) (2000), Guidelines on Selection
of Acceptable Flood Capacity for Dams. Australian National Committee on Large Dams.
AUSTROADS (1992) '92 AUSTROADS Bridge Design Code.
Bates, B.C., McLuckie, D., Westra, S., Johnson, F., Green, J., Mummery, J. and Abbs, D.
(2015), Revision of Australian Rainfall and Runoff - The Interim Climate Change Guideline.
Presented at National Floodplain Management Authorities Conference, Brisbane, Qld, May.
Bonnin, G., Maitaria, K. and Yekta, M. (2010), Trends in Heavy Rainfalls in the Observed
Record in Selected Areas of the U.S. IN PALMER, R. N. (Ed.) World Environmental and
Water Resources Congress 2010: Challenges of Change. Providence, Rhode Island,
American Society of Civil Engineers.
Bureau of Meteorology (2016), Design rainfalls for Australia: data, methods and analyses.
Bureau of Meteorology, Melbourne, VIC.
Bureau of Meteorology (2003), The Estimation of Probable Maximum Precipitation in

Australia: Generalised Short-Duration Method, [Online] Bureau of Meteorology, Melbourne,
Australia, June 2003, (p: 39). Available at: http://www.bom.gov.au/water/
designRainfalls/pmp/gsdm_document.shtml.
Bureau of Meteorology (2006), Guidebook of the Estimation of Probable Maximum

Precipitation: Generalised Southeast Australia Method, included on the compact disc 'Guide
to the Estimation of Probable Maximum Precipitation: Generalised Southeast Australia
Method', Hydrometeorological Advisory Service, Bureau of Meteorology, October.
Bureau of Meteorology (2005), Guidebook of the Estimation of Probable Maximum

Precipitation: Generalised Tropical Storm Method, Included on compact disc 'Guide to the
57
Design Rainfall
Estimation of Probable Maximum Precipitation: Generalised Tropical Storm Method',

Hydrometeorological Advisory Service, Bureau of Meteorology, September.
Burn, D.H. (1990), An appraisal of the 'region of influence' approach to flood frequency
analysis, Hydrological Sciences - Journal-des Sciences Hydrologiques, 35(2), 149-165.
Green, J. and Johnson, F.J. (2011), Stationarity Assessment of Australian Rainfall, Internal
Bureau of Meteorology Report.
Green, J.H., Xuereb, X. and Siriwardena, L. (2011), Establishment of a Quality Controlled

Rainfall Database for the Revision of the Intensity-Frequency Duration (IFD) Estimates for
Australia. Presented at 34th IAHR Congress, Brisbane, Qld, June.
Green, J.H., Beesley, C., Frost, A., Podger, S. and The, C. (2015), National Estimates of
Rare Design Rainfall. Presented at Hydrology and Water Resources Symposium, Hobart,
Tas, December 2015.
Green, J.H., Johnson, J., McKay, D., Podger, P., Sugiyanto, M. and Siriwardena, L. (2012a),
Quality Controlling Daily Read Rainfall Data for the Intensity-Frequency Duration (IFD)
Revision Rainfall Project. Presented at Hydrology and Water Resources Symposium,
Sydney, NSW, November.
Green, J.H., Xuereb, K., Johnson, J., Moore, G. and The, C. (2012b), The Revised Intensity-
Frequency Duration (IFD) Design Rainfall Estimates for Australia - An Overview. Presented
at Hydrology and Water Resources Symposium, Sydney, NSW, November.
Guttman, N.B., Hosking, J.R.M. and Wallis, J.R. (1993), Regional precipitation quantile
values for the continental U. S. computed from L-moments. Journal of Climate, 6:
2326-2340.
Haddad, K. and Rahman, A. (2012a), Regional flood frequency analysis in eastern Australia:
Bayesian GLS regression-based methods within fixed region and ROI framework: quantile
regression vs parameter regression technique. Journal of Hydrology, 430: 142-161.
Haddad, K. and Rahman, A. (2012b), A Pilot Study on Design Rainfall Estimation in

Australia using L-moments and Bayesian Generalised Least Squares Regression:
Comparison of Fixed Region, Facets and Region of Influence Approach, EnviroWater
Sydney.
Haddad, K., Johnson, F., Rahman, A., Green, J. and Kuczera, G. (2015), Comparing three
methods to form regions for design rainfall statistics: Two case studies in Australia. Journal
of Hydrology, 527: 62-76.
Haddad, K., Pirozzi, J., McPherson, G., Rahman, A. and Kuczera, G. (2009),'Regional flood
estimation technique for NSW: application of generalised least squares quantile regression
technique', Proc. 32nd Hydrology and Water Resources Symp., Newcastle, pp: 829-840.
Haddad, K., Rahman, A. and Green, J. (2011), Design Rainfall Estimation in Australia: A
Case Study using L moments and Generalized Least Squares Regression, Stochastic
Environmental Research & Risk Assessment, 25(6), 815-825.
Hosking, J.R.M. and Wallis, J.R. (1997), Regional Frequency Analysis: An Approach Based
on L-Moments. CambridgeUniversity Press, Cambridge, UK. p: 224.
Huff, F.A. and Angel, J.R. (1992), Rainfall Frequency Atlas of the Midwest. Illinois State
Water Survey, Champaign, Bulletin 71, 1992.
58
Design Rainfall
Hutchinson, M.F. (1998), Interpolation of rainfall data with thin plate smoothing splines - part
2: Analysis of topographic dependence, Journal of Geographic Information and Decision
Analysis, 2(2), 152-167.
Hutchinson, M.F. (2007), ANUSPLIN version 4.37 User Guide, The Australian National
University, Centre for Resources and Environmental Studies, Canberra.
Hutchinson, M.F. and Xu, T. (2013), ANUSPLIN Version 4.4 User Guide, The
AustralianNationalUniversity. FennerSchool of Environment and Society, Canberra,
Australia.
Johnson, F., Hutchinson, M.F., The, C., Beesley, C. and Green, J. (2015), Topographic
relationships for design rainfalls over Australia. Accepted for publication Journal of
Hydrology.
Johnson, F., Xuereb, K., Jeremiah, E. and Green, J. (2012a), Regionalisation of Rainfall
Statistics for the IFD Revision Project. Presented at Hydrology and Water Resources
Symposium, Sydney, NSW, November 2012.
Johnson, F., Haddad, K., Rahman, A., and Green, J. (2012b), Application of Bayesian GLSR
to Estimate Sub Daily Rainfall Parameters for the IFD Revision Project. Presented at
Hydrology and Water Resources Symposium, Sydney, NSW, November 2012.
Kjeldsen, T.R. and Jones, D.A. (2009), A formal statistical model for pooled analysis of
extreme floods, Hydrology Research, 40(5), 465-480.
Madsen, H., Mikkelsen, P.S., Rosbjerg, D. and Harremoes, P. (2002), Regional estimation of
rainfall intensity duration curves using generalised least squares regression of partial
duration series statistics. Water Resources Research, 38(11), 1239.
Madsen, H., Arnbjerg-Neilsen, K. and Mikkelsen, P.S. (2009), Update of regional intensity-
duration-frequency curves in Denmark: Tendency towards increased storm intensities.
Atmospheric Research, 92: 343-349.
Menabde, M., Seed, S. and Pegram, G. (1999), A simple scaling model for extreme rainfall.
Water Resources Research, 35: 335-339
Minty, L.J., Meighen, J. and Kennedy, M.R. (1996), Development of the Generalised
Southeast Australia Method for Estimating Probable Maximum Precipitation, [Online] HRS
Report No. 4, Hydrology Report Series, Bureau of Meteorology, Melbourne, Australia,
August 1996, p: 48. Available at: http://www.bom.gov.au/water/designRainfalls/pmp/
gsam.shtml
Nandakumar, N., Weinmann, P.E., Mein, R.G. and Nathan, R.J. (1997), Estimation of
Extreme Rainfalls for Victoria using the CRCFORGE Method, CRC for Catchment
Hydrology.
Pilgrim, DH (ed) (1987) Australian Rainfall and Runoff - A Guide to Flood Estimation,
Institution of Engineers, Australia, Barton, ACT, 1987.
Reis Jr., D.S., Stedinger, J.R. and Martins, E.S. (2005), Bayesian GLS regression with
application to LP3 regional skew estimation. Water Resour Res. 41, W10419, (1) 1029.
Schaefer, M.G. (1990), Regional analysis of precipitation annual maxima in Washington

State, Water Resoures Research, 26(1), 119-131.
59
Design Rainfall
The, C., Johnson, F., Hutchinson, M., and Green, J. (2012), Gridding of Design Rainfall
Parameters for the IFD Revision Project for Australia. Presented at Hydrology and Water
Resources Symposium, Sydney, NSW, November.
The, C., Hutchinson, M., Johnson, F., Beesley, C. and Green, J. (2014), Application of
ANUSPLIN to produce new Intensity-Frequency-Duration (IFD) design rainfalls across
Australia. Presentated at Hydrology and Water Resources Symposium, Perth, WA, February
2014.
The, C., Beesley, C., Podger, S., Green, J., Hutchinson, M. and Jolly, C. (2015), Very
frequent design rainfalls - an enhancement to the new IFDs. Accepted for presentation at
Hydrology and Water Resources Symposium, Hobart, TAS, December 2015.
Walland, D.J., Meighen, J., Xuereb, K.C., Beesley, C.A. and Hoang, T.M.T. (2003), Revision
of the Generalised Tropical Storm Method for Estimating Probable Maximum Precipitation
[Online], HRS Report No.8, Hydrology Report Series, Bureau of Meteorology Melbourne,
Australia, available at: http://www.bom.gov.au/water/designRainfalls/hrs8.shtml.
Wang, Q.J. (1997), LH moments for statistical analysis of extreme events. Water Resources
Research, 33(12), 2841-2848.
Wang, Q.J. (1998), Approximate goodness-of-fit tests of fitted generalised extreme value
distributions using LH moments. Water Resources Research, 34(12), 3497-3502.
World Meteorological Organisation (1986), 'Manual for Estimation of Probable Maximum

Precipitation'. Second Edition. Operational Hydrology Report No.1, WMO - No.332, Geneva.
Xuereb, K. and Green, J. (2012), Defining Independence of Rainfall Events with a Partial
Duration Series Approach. Presented at Hydrology and Water Resources Symposium,
Sydney, NSW, November 2012.
Xuereb, K.C., Moore, G.J., and Taylor, B.F. (2001), Development of the Method of Storm
Transposition and Maximisation for the West Coast of Tasmania. HRS Report No.7,
Hydrology Report Series, Bureau of Meteorology, Melbourne, Australia, January 2001.
60
Chapter 4. Areal Reduction Factors
Phillip Jordan, Rory Nathan, Scott Podger, Mark Babister, Peter
Stensmyr, Janice Green
4.1. Introduction
Design rainfall information for flood estimation generally is made available in the form of
rainfall Intensity Frequency Duration (IFD) date (Book 2, Chapter 3) that relates to specific
points in a catchment rather than to the whole catchment area. However, most flood
estimates are required for catchments that are sufficiently large that design rainfall intensities
at a point are not representative of the areal average rainfall intensity across the catchment.
The ratio between the design values of areal average rainfall and point rainfall, computed for
the same duration and Annual Exceedance Probability (AEP), is called the Areal Reduction
Factor (ARF). This allows for the fact that larger catchments are less likely than smaller
catchments to experience high intensity storms simultaneously over the whole of the
catchment area.
It should be noted that the ARF provides a correction factor between the catchment rainfall
depth (for a given combination of AEP and duration) and the mean of the point rainfall
depths across a catchment (for the same AEP and duration combination). Applying an ARF
is a necessary input to computation of design flood estimates from a catchment model that
preserves a probability neutral transition between the design rainfall and the design flood
characteristics. The ARF merely influences the average depth of rainfall across the
catchment, it does not account for variability in the spatial and/or space-time patterns of its
occurrence over the catchment.
Recommendations for ARF to be adopted are provided in Book 2, Chapter 4, Section 3.
4.2. Derivation of Areal Reduction Factors

The method adopted for the derivation of areal reduction factors is a modified version of
Bell’s method (Bell, 1976; Siriwardena and Weinmann, 1996). ARFs derived by this method
have been widely used for some time and have been shown to provide effective estimates of
areal design rainfall for deriving rainfall-based flood frequency curves. The method allows
the dependence between ARF and catchment area to be determined, as well as the
variation with AEP. The method has been applied with data collected and processed for the
estimation of IFDs; they thus provide for a more comprehensive and consistent set of
estimates than were prepared by Jordan et al. (2013).
The modified Bell’s method involves defining hypothetical circular catchments in areas with
sufficient data and creating an areal rainfall time series for each catchment by weighting
point rainfall values based on Thiessen polygon areas (or an equivalent weighting method).
The frequency quantiles calculated from the areal rainfall time series are divided by the
weighted point frequency quantiles for the sites within the catchment, yielding an ARF
estimates for the given catchment area and a range of AEPs. Once ARFs have been
calculated for the required catchment areas, durations and AEPs for as many locations as
possible, they are averaged across these attributes and an equation is fitted to provide a
prediction model for the selected region.
61
Areal Reduction Factors
The adopted methodology is described in more detail in Podger et al. (2015a), Podger et al.
(2015b) and Stensmyr et al (2014).
4.3. Areal Reduction Factor Recommendations

4.3.1. Areal Reduction Factors for Catchments up to 30 000
km2, Durations up to 7 days and Events More Frequent than
0.05% AEP
Areal Reduction Factors (ARF) for catchments with areas up to 30 000 km2, for durations up
to and including 168 hours (7 days) and for AEP more frequent than 0.05% (1 in 2000)
Annual Exceedance Probability (AEP) are recommended based on the values derived by
Podger et al. (2015a), Podger et al. (2015b) and Stensmyr et al (2014). This guidance
should be adopted unless rigorous subsequent research or catchment-specific investigations
have been conducted to define a more appropriate, locally specific, ARF.
The design areal rainfall to be applied in a design flood simulation is the average rainfall over
the total catchment area to the point of interest. Consequently, the ARF should be computed
for the total catchment upstream of each location of interest where a design flood estimate is
required. The ARF should not be computed independently for each subarea in a runoff-
routing model of the catchment of interest, as this would result in systematic overestimation
of catchment rainfalls and simulated design flood hydrographs.
The ARF to be applied to design rainfall is a function of the total area of the catchment, the
duration of the design rainfall event and it’s AEP. The ARF should be computed using the
relevant procedure described in Table 2.4.1.
If the duration of interest is greater than 12 hours, Equation (2.4.2) will be required as part of
the calculation procedure and the coefficients of Equation (2.4.2) vary regionally across
Australia. The applicable ARF region should be selected by referring to Figure 2.4.1. Where
a catchment overlaps the boundary between regions, the ARF should be selected for the
region that has the largest overlap with the boundary of the catchment. The coefficients to be
applied with Equation (2.4.2) should be selected from the appropriate region from
Table 2.4.2.
Table 2.4.1. ARF Procedure for Catchments Less than 30 000 km2 and Durations up to and
Including 7 Days
Catchment Area Duration ≤ 12 hours Duration Between 12 Duration ≥ 24 Hours
and 24 hours (1 Day) and ≤ 7 Days
(168 hours)
≤ 1 km2 ARF = 1
Between 1 and 10 1. Compute ARF(10 1. Compute ARF(24 1. Compute ARF(10
km2 km2) using hr, 10 km2) using km2) using
Equation (2.4.1) for Equation (2.4.2) for Equation (2.4.2) for
2
area = 10 km and 2
area = 10 km and area = 10 km2
selected duration duration = 1440
min 2. Interpolate ARF for
2. Interpolate ARF for catchment area
catchment area 2. Compute ARF(12 and selected
and selected 2
hr, 10 km ) using duration using
Equation (2.4.1) for Equation (2.4.4)
62
Catchment Area Duration ≤ 12 hours Duration Between 12 Duration ≥ 24 Hours

and 24 hours (1 Day) and ≤ 7 Days
(168 hours)
duration using 2
area = 10 km and
Equation (2.4.4) duration = 720 min
3. Interpolate ARF(10
km2) for selected
duration using
Equation (2.4.4)
4. Interpolate ARF for

catchment area
and selected
duration using
Equation (2.4.3)
Between 10 and • Compute ARF 1. Compute ARF(24 • Compute ARF
1000 km2 using Equation hr) using Equation using Equation
(2.4.1) for (2.4.2) for (2.4.2) for
catchment area and catchment area catchment area and
selected duration and duration = selected duration
Between 1000 and Generalised 1440 min
30 000 km2 equations not
2. Compute ARF(12
applicable
hr) using Equation
(2.4.1) for
catchment area
and duration = 720
min
3. Interpolate ARF for

selected duration
using Equation
(2.4.3)
>30 000 km2 Generalised equations not applicable. It is recommended that the
practitioner should perform a frequency analysis of catchment
rainfall data for the catchment of interest.
Notes on Table 2.4.1:
• Equation (2.4.1), Equation (2.4.2) and Equation (2.4.3) require the selected duration to be
provided in minutes.
• There has been limited research on ARF applicable to catchments that are less than 10
km2. The recommended procedure is to adopt an ARF of unity for catchments that are
less than 1 km2, with an interpolation to the empirically derived equations for catchments
that are between 1 and 10 km2 in area (refer to Equation (2.4.3)).
• The ARF equations derived by Podger et al. (2015a), Podger et al. (2015b) and Stensmyr
et al (2014) were derived for the 50% to 1% AEPs. Although these have been
recommended for use for a wider range of AEP, (out to 0.05% AEP), further verification is
ongoing on the validity of this approach. As a result, the coefficients of Equation (2.4.2)
(from Table 2.4.2) and/or the regional boundaries (refer to Figure 2.4.1) may be revised.
63
Short Duration ARF Equation
Equation (2.4.1)
1 − 0.287(��0.265 − 0.439log10(��)) . ��−0.36
�� = �� 1, +2.26 x 10−3 x ��0.226 . ��0.125(0.3 + log10(��))

�� − 180 2
+0.0141 x ��0.213x 10−0.021 (0.3 + log10(��))
1440
where Area is in km2, Duration is in minutes and AEP is a fraction (between 0.5 and 0.0005).
Long Duration ARF Equation
1 − �(�� − �log10��)��−�
+��(0.3 + log10��)
�� = �� 1, (2.4.2)
��
��
1440
+ℎ10 (0.3 + log10��)]
where Area is in km2, Duration is in minutes and AEP is a fraction (between 0.5 and 0.0005).
Interpolation Equation for Durations between 12 and 24 hours
�� − 720
�� = ��12ℎ�� + ��24ℎ�� − ��12ℎ�� (2.4.3)
720
where Duration is in minutes.
ARF Equation for Areas between 1 and 10 km2
�� = 1 − 0.6614 1 − �� 0.4 − 1 (2.4.4)

10��2
where Area is in km2
64
Figure 2.4.1. Area Reduction Factors Regions for Durations 24 to 168 Hours
Table 2.4.2. ARF Equation (2.4.2) Coefficients by Region for Durations 24 to 168 hours
Inclusive
Region1 a b c d e f g h i
East Coast North 0.327 0.241 0.448 0.36 0.000 0.48 -0.21 0.012 -0.001
96 3
Semi-arid Inland 0.159 0.283 0.25 0.308 7.3E- 1 0.039 0 0
Queensland 07
Tasmania 0.060 0.347 0.2 0.283 0.000 0.347 0.087 0.012 -0.000
5 76 7 33
South-West Western 0.183 0.259 0.271 0.33 3.85E 0.41 0.55 0.008 -0.000
Australia -06 17 45
Central New South 0.265 0.241 0.505 0.321 0.000 0.414 -0.021 0.015 -0.000
Wales 56 33
South-East Coast 0.06 0.361 0 0.317 8.11E 0.651 0 0 0
-05
Southern Semi-arid 0.254 0.247 0.403 0.351 0.001 0.302 0.058 0 0
3
Southern Temperate 0.158 0.276 0.372 0.315 0.000 0.41 0.15 0.01 -0.002
141 7
Northern Coastal 0.326 0.223 0.442 0.323 0.001 0.58 -0.374 0.013 -0.001
3 5
65
Region1 a b c d e f g h i
Inland Arid 0.297 0.234 0.449 0.344 0.001 0.216 0.129 0 0
42
1These values are provided on the ARR Data Hub for the relevant region when queried
(Babister et al (2016), accessible at http://data.arr-software.org/)
4.3.2. Events That are Rarer than 0.05% Annual Exceedance

Probability
The ARF equations are only recommended for use for events more frequent than 0.05%
AEP. For more extreme events, the procedures recommended in Book 8, Chapter 3, Section
5 should be used to determine catchment average design rainfall depths. The interpolation
procedure recommended in Book 8, Chapter 3, Section 5 uses the catchment average
design rainfall depth for 0.05% AEP, which would be calculated using the average of the
point design intensities across the catchment multiplied by the ARF estimates recommended
above and the PMP depth, which is already estimated as a catchment average value.
4.3.3. Catchments with Areas Greater than 30 000 km2

The largest (circular) catchments used by Podger et al. (2015a) to estimate ARF were 30
000 km2 and which set the upper limit of applicability of the ARF equations. As the
catchment area increases beyond 30 000 km2, it becomes increasingly likely that storm
events would only influence part of the overall catchment area, which increases the
uncertainty associated with adjusting point design intensities using an ARF.
Design rainfall depths for catchments larger than 30 000 km2 should be derived from
frequency analysis of catchment average rainfall depths over the specific catchment. The
design rainfall depths from the catchment-specific frequency analysis should be checked by
dividing them by the average of the point rainfall depths from point IFD analysis for the
catchment (Bureau of Meteorology, 2013) to infer the ARF for the catchment for each rainfall
duration and AEP. It would be expected that for a catchment larger than 30 000 km2, the
ARF inferred from this check for each duration and AEP should be less than the ARF
calculated from the regional method Equation (2.4.2) for the corresponding duration and
AEP combination. It would also be expected that the inferred ARF (for a given AEP) should
increase with rainfall duration.
For catchments larger than 30 000 km2, it becomes increasingly likely that rainfall events
that would give rise to flooding would be concentrated in one part of the catchment. For
catchments larger than 30 000 km2 it is strongly recommended that partial area storms are
explicitly modelled (using Monte Carlo or other joint probability approaches). Explicit
modelling of partial area storms should also be considered for catchments in the range
between 5 000 km2 and 30 000 km2.
4.4. Worked Example

A worked example for the calculation of the ARF and the areal design rainfall for a
catchment in Queensland is provided in Book 2, Chapter 6, Section 5.
4.5. Limitations and Recommended Further Research

The ARF equations developed in Australia have been derived using data driven and
empirical methods, with limited theoretical underpinning. ARF values for a particular
66
catchment would derive from a combination of the mixture of storm types causing heavy
rainfall within a region, the direction and speed of movement of those storms and the spatial
and temporal characteristics of those storms. Analysis by a hydrometeorologist of the
prevalence of different storm types within different parts of Australia and the advection,
temporal and spatial characteristics of those storms is likely to provide an understanding of
the causes of variations in ARF. Such understanding is difficult to infer directly, on its own,
from the empirically derived ARF equations that are currently recommended for use in
Australia. It is recommended that hydrometeorologists are engaged to investigate the
causes of variations in ARF.
Once the hydrometeological analysis recommended above has been undertaken, the
outcomes of that work may enable further research and improvements in the following
specific areas:
• Clarification of how well the ARFs derived using an empirical method such as Bell’s
method, compare with those derived from a suitable theoretical method that may better
account for hydrometeorological understanding of the drivers of variability in ARFs.
• There are some areas within each of the regions where the ARF values determined
empirically for the circular catchments demonstrated a trend toward being larger or smaller
than obtained from the ARF equations fitted to the mean ARF values from all circular
catchments within the region for a given area, duration and AEP. Hydrometeorological
understanding may enable definition of smaller sub-regions, combining of existing regions
(with the existing regions largely defined using state and territory boundaries), or definition
of new regions in order to reduce the uncertainty introduced by this variability.
• Seasonality was found to be a significant driver of ARFs in Western Australia but has not
been investigated for other parts of Australia. Hydrometeorological understanding may
guide the regions where seasonal dependence in ARF would be likely, the start and end
dates of seasons and how transition periods between seasons should be handled.
It is recommended that after an appropriate study has been undertaken to determine the
hydrometeorological causes of variations in ARF that further studies are then scoped and
prioritised according to areas where the hydrometeorological causes can be best exploited
to reduce residual uncertainty in ARFs.
4.6. Recommended Further Research

4.6.1. Areal Reduction Factors
The ARF equations developed in Australia have been derived using data driven and
empirical methods, with limited theoretical underpinning. ARF values for a particular
catchment would derive from a combination of the mixture of storm types causing heavy
rainfall within a region, the direction and speed of movement of those storms and the spatial
and temporal characteristics of those storms. Analysis by a hydrometeorologist of the
prevalence of different storm types within different parts of Australia and the advection,
temporal and spatial characteristics of those storms is likely to provide an understanding of
the causes of variations in ARF. Such understanding is difficult to infer directly, on its own,
from the empirically derived ARF equations that are currently recommended for use in
Australia. It is recommended that hydrometeorologists are engaged to investigate the
causes of variations in ARF.
Once the hydrometeological analysis recommended above has been undertaken, the
outcomes of that work may enable further research and improvements in the following areas:
67
• how well the ARFs derived using an empirical method such as Bell’s method, compare
with those derived from a suitable theoretical method that may better account for
hydrometeorological understanding of the drivers of variability in ARFs.
• There are some areas within each of the regions where the ARF values determined
empirically for the circular catchments demonstrated a trend toward being larger or smaller
than the fitted ARF equations, which were fitted to the mean ARF values from all circular
catchments within the region for a given area, duration and AEP. Hydrometeorological
understanding may enable definition of smaller sub-regions, combining of existing regions
(with the existing regions largely defined using state and territory boundaries), or definition
of new regions in order to reduce the uncertainty introduced by this variability.
• Seasonality was found to be a significant driver of ARFs in Western Australia but has not
been investigated for other parts of Australia. Hydrometeorological understanding may
guide the regions where seasonal dependence in ARF would be likely, the start and end
dates of seasons and how transition periods between seasons should be handled.
It is recommended that after an appropriate study has been undertaken to determine the
hydrometeorological causes of variations in ARF that further studies are then scoped and
prioritised according to areas where the hydrometeorological causes are best exploited to
reduce residual uncertainty in ARFs.
4.7. References
Babister, M., Trim, A., Testoni, I. and Retallick, M. 2016. The Australian Rainfall and Runoff
Datahub, 37th Hydrology and Water Resources Symposium Queenstown NZ
Bell, F.C. (1976), The areal reduction factors in rainfall frequency estimation. Natural
Environmental Research Council (NERC), Report No. 35, Institute of Hydrology, Wallingford,
U.K.
Bureau of Meteorology (2013), Revised Design Rainfall Intensity Frequency Duration

Estimates for Australia. Available at: http://www.bom.gov.au/water/designRainfalls/revised-
ifd/, Accessed [July 2015].
Jordan, P., Weinmann, P.E., Hill, P. and Wiesenfeld, C. (2013), Collation and Review of Areal
Reduction Factors from Applications of the CRC-FORGE Method in Australia, Final Report,
Australian Rainfall and Runoff Revision Project 2: Spatial Patterns of Rainfall, Engineers
Australia, Barton, ACT.
Podger, S., Green, J., Jolly, C., The, C. and Beesley, C. (2015a),Creating long duration areal
reduction factors for the new Intensity-Frequency-Duration (IFD) design rainfalls, Proc.
Engineers Australia Hydrology and Water Resources Symposium, Hobart, Tasmania,
Australia.
Podger, S., Green, J., Stensmyr, P. and Babister, M. (2015b), Combining long and short
duration areal reduction factors, Proc. Engineers Australia Hydrology and Water Resources
Symposium, Hobart, Tasmania, Australia.
Siriwardena, L. and P.E. Weinmann, (1996), Derivation of areal reduction factors for design
rainfalls in Victoria, 96(4), 60.
68
Stensmyr, P., Babister, M. and Retallick, M. (2014), Australian Rainfall and Runoff Revision
Project 2: Spatial Patterns of Rainfall: Stage 2 Report, Short Duration Areal Reduction
Factors, ARR Report Number P2/S2/019, ISBN 978-085825-9614.
69
Chapter 5. Temporal Patterns
Mark Babister, Monique Retallick, Melanie Loveridge, Isabelle Testoni,
Scott Podger
5.1. Introduction
The majority of hydrograph estimation methods used for flood estimation require a temporal
pattern that describes how rainfall falls over time as a design input. Traditionally a single
burst temporal pattern has been used for each rainfall event duration. The use of a single
pattern has been questioned for some time (Nathan and Weinmann, 1995) as the analysis of
observed rainfall events from even a single pluviograph shows that a wide variety of
temporal patterns is possible.
The importance of temporal patterns has increased as the practice of flood estimation has
evolved from peak flow estimation to full hydrograph estimation. There has been a strong
move toward storage-based mitigation solutions in urban catchments which require realistic
temporal patterns that reproduce total storm volumes as well as the temporal distribution of
rainfall within the event.
This chapter discusses use of temporal patterns for design flood estimation where a fixed
temporal pattern is applied over the entire catchment. Book 2, Chapter 6 discusses the more
complex case of space-time patterns.
This chapter is structured as follows:
• Book 2, Chapter 5, Section 2 – discusses fundamental temporal pattern concepts and how
the concept of using ensembles of temporal pattern developed;
• Book 2, Chapter 5, Section 3 – discusses the storm database that was used to develop
temporal patterns;
• Book 2, Chapter 5, Section 4 – describes the estimation of pre-burst rainfall;
• Book 2, Chapter 5, Section 5 – describes the development of ensembles of design

temporal patterns;
• Book 2, Chapter 5, Section 6 – discusses areal temporal patterns;
• Book 2, Chapter 5, Section 8 – discusses possible effects of climate change on temporal

patterns; and
• Book 2, Chapter 5, Section 9 – discusses the application of design temporal patterns;
5.2. Temporal Pattern Concepts

5.2.1. Storm Components
In order to properly understand the temporal patterns it is necessary to understand the
components of a storm event and how they relate to Intensity Frequency Duration Data (IFD)
70
Temporal Patterns
and catchment response. Figure 2.5.1 depicts a typical storm pattern and how components
of the storm can be characterised. It is important to note the components can be
characterised either by IFD relationships or by catchment response and are highly
dependent on the definitions used. The components of a storm include:
• antecedent rainfall - is rainfall that has fallen before the storm event and is not considered
part of the storm but can affect catchment response. This is not considered in this chapter
but is introduced for completeness.
• pre-burst rainfall - is storm rainfall that occurs before the main burst. With the exception of
relatively frequent events, it generally does not have a significant influence on catchment
response but is very important for understanding catchment and storage conditions before
the main rainfall burst.
• the burst - represents the main part of the storm but is very dependent on the definition
used. Bursts have typically been characterised by duration. The burst could be defined as
the critical rainfall burst , the rainfall period within the storm that has the lowest probability,
or the critical response burst that corresponds to the duration which produces the largest
catchment response for a given rainfall Annual Exceedance Probability (AEP).
• post-burst rainfall - is rainfall that occurs after the main burst and is generally only
considered when aspects of hydrograph recession are important. This could be for
drawing down a dam after a flood event or understanding how inundation times affect
flood recovery, road closures or agricultural land.
Figure 2.5.1. Typical Storm Components
If the critical response burst is not the same as the critical rainfall burst then the critical
response burst is either:
• part of a longer critical rainfall burst; or
• a storm that contains rarer shorter duration bursts.
Rarer shorter duration bursts within a burst are typically called embedded bursts and can
cause problems in modelling as, while the intention may be to assess the catchment
71
Temporal Patterns
response to a burst of a defined duration and probability, the response to a rarer shorter
duration burst is also being assessed.
The distinction between a burst and a complete storm is important as complete storms are
used for calibration and bursts are typically used for design. Though this difference is less
important for catchments with long duration responses, as the bursts typically represent
nearly the entire storm event.
5.2.2. Pattern Variability

It has been well recognised that temporal patterns exhibit significant variability between
rainfall events of similar magnitude, and the adopted pattern can have a significant effect on
the estimated peak flow. Askew (1975), Milston (1979), Brown (1982), Wood and Alvarez
(1982), Cordery et al. (1984) give examples of differences of up to 50% in flood peaks
resulting from different assumed temporal patterns. Ball (1994) discusses how catchment
peak flow tends to increase as the variability of the pattern increases.
Pilgrim et al. (1969) describe the variability of patterns:
“In nature, a wide range of patterns is possible. Some storms have their period of peak
intensity occur early, while other storms have the peak rainfall intensity occur towards the
end of the storm period and a large number have a tendency for the peak to occur more or
less centrally.”
Figure 2.5.2 depicts two very different storms from Sydney Observatory Hill that have similar
IFD characteristics from 15 min to 12 hrs and a 2 hr critical rainfall burst, however on other
criteria they are very different. The first pattern has 182 mm before the 2 hr burst, while the
second pattern has 16 mm. For the 2nd event the commencement of runoff will be closely
aligned with the main burst while the former event is likely to have significant runoff prior to
the main burst. The rainfall burst also exhibits significant variability, Figure 2.5.3 presents the
temporal patterns (as dimensionless mass curves) of ten 2 hr bursts of similar probability
that show the variability described by Pilgrim et al. (1969).
72
Temporal Patterns
Figure 2.5.2. Two Different Storm Events with Similar Intensity Frequency Duration
Characteristics (Sydney Observatory Hill) – Two Hyetographs plus Burst Probability Graph
73
Temporal Patterns
Figure 2.5.3. Ten 2 hr Dimensionless Mass Curves
Most of the historical research on temporal patterns has assumed that the central tendency
of the pattern is more important than the variability, with the aim of producing a typical,
representative or median pattern. French (1985) describes how a pattern can be considered
a two dimensional quantity with most methods breaking the pattern into two manageable one
dimensional quantities that describe the magnitude of the element and the order of the
elements. This can be described as a rank order vector and a decay curve that describes
how the magnitude decreases between ranks.
Monte Carlo and ensemble modelling techniques try to overcome the problems associated
with this simplification by using an ensemble of temporal patterns. A short history of temporal
patterns helps explain that, while the complexity of temporal patterns has been well
understood for a long time, it has been difficult to produce a simple set of design patterns.
5.2.3. History of Design Temporal Pattern Development

Australian Rainfall and Runoff 1977 (Pattison, 1977; Institution of Engineers, Australia.,
1977) notes that there has historically been a slow change in the application of temporal
distributions of rainfall in hydrologic studies from more arbitrary approaches to approaches
that reproduce the IFD characteristics (Keifer and Chu (1957) and Huff (1967)). The modern
interpretation is the duration independent storm (Varga et al., 2009). These approaches
reproduce the IFD loading and at best can reproduce aspects of a single storm. Such
approaches have some applicability for peak flow estimation but cannot reproduce realistic
hydrographs.
Other early approaches included in the development of patterns based on observed

pluviograph records were for complete storms rather than bursts. Such approaches require
the scaling of complete storms. Other methods have allowed the arbitrary rearrangement of
patterns to maximise peak discharge. Such approaches are very far removed from concept
of probability neutrality.
The development of the Average Variability Method by Pilgrim et al. (1969) and Pilgrim and
Cordery (1975) analysed the variability of patterns by separating the analysis of the
74
Temporal Patterns
magnitude or the ranks from the analysis of rank order. This approach is only applicable to
bursts and was applied in ARR 1977 and in a very detailed way in ARR 1987. A variation of
this approach was developed by Hall and Kneen (1973). During the finalisation of ARR 1987
some problems where found with the use of the patterns and extensive testing was carried
out which resulted in some changes to remove some embedded bursts, including some
arbitrary rank order changes. It has been assumed that an AVM pattern would preserve
probability neutrality. Single burst per duration AVM patterns have been extensively used
since ARR 1987 and appear to have performed reasonable well for peak flow estimation.
By their very nature AVM patterns do preserve the average rank magnitudes. Figure 2.5.4
plots the magnitude or decay curve for the ten dimensionless curves in Figure 2.5.3 and the
AVM curve from these events and all 10 events that would be considered in a AVM analysis.
Figure 2.5.4. Decay Curve of Ten Dimensionless Patterns and AVM Patterns
The issues that have arisen have been discussed in Retallick et al. (2009):
• Specific durations tend to dominate;
• Big changes in pattern and corresponding peak flow occur at some region boundaries;
• The burst only approach has made conversion of observed losses to burst losses difficult;
• The design patterns often contain significant embedded bursts; and
• Filtering to remove embedded bursts that was recommended in ARR 1987 (Pilgrim, 1987)
has had a mixed uptake and sometimes produces unrealistically long critical durations.
The AVM approach also works best when there is a dominant typical pattern shape. As part
of the AVM approach the average and standard deviation of the rank of each period is
75
Temporal Patterns
calculated, when there is no dominant pattern these averages can be very similar and the
resultant AVM pattern need not bare any resemblance to any of the observed patterns.
The complexity of producing a representative or median pattern has led many practitioners
to question the concept and ask whether it is better to specifically account for this variability
by modelling an ensemble of temporal patterns.
The problems with the AVM method and other median or representative patterns is that it
assumes the variability of actual patterns is much less important than their central tendency.
Such an approach does not account for how temporal patterns interact with catchments to
produce peak flows and hydrographs. The response can be very catchment-specific, and
there is no guarantee that a representative pattern will produce the medium response from
an ensemble of patterns that properly captures the variability of observed patterns. These
problems can become more pronounced when changes are made to the catchment
response or storage characteristics. Phillips and Yu (2015) examined the impact of storage
on an ensemble of events for the lot, neighbourhood and subcatchment scale.
It is unclear who first proposed the concept of running an ensemble of temporal patterns to
account for the variability of patterns. Practitioners have a long history of comparing the peak
flow estimates from design patterns to estimates from patterns extracted from real storms
that occurred on that catchment of interest. The development of Monte Carlo methods
(Hoang et al., 1999; Rahman et al., 2002; Weinmann et al., 2000) that stochastically sample
from observed events based on complete storms and embedded bursts of maximum
intensity (“storm cores”) is well documented and has helped highlight the value of using
ensembles. Others, such as Nathan et al. (2002) have used ensembles of storm bursts
based on observed point and areal storms to facilitate their use directly with design IFD
information. The development of the Bureau of Meteorology significant storm database
(Meighen and Kennedy, 1995) enabled practitioners to easily test ensembles of areal
patterns and how they affected other flood characteristics; examples include Webb
McKeown and Associates (2003). While there is a long history of testing ensembles of
temporal patterns, Sih et al. (2008) is the first example of detailed testing of ensemble
patterns as a design input outside a Monte Carlo environment. This long development
history of ensemble simulation can probably be explained by the fact that the concept could
not be confirmed until rigorous Monte Carlo techniques became available for validation.
Parallel to the development of ensemble and Monte Carlo approaches, practitioners have
become concerned with using burst patterns where complete storm volume is important.
Rigby and Bannigan (1996) suggest the entire burst approach needed to be reviewed and
design storms needed to replace design bursts. For the Wollongong area Rigby and
Bannigan (1996) demonstrated that historically most short duration events were embedded
in longer duration events. They recommended that short duration events could be
embedded in a 24 hour event of the same probability. They particularly cautioned against
using bursts on catchments with significant natural or man-made storages. Phillips et al.
(1994) had found similar problems in the upper Parramatta River and suggested embedded
storms were more realistic, and that basin storages would be underestimated with a burst
approach unless the embedded nature of events was factored into the starting volumes.
Rigby et al. (2003) extended the earlier work to include guidance on using the embedded
design storms. Roso and Rigby (2006) recommended a storm based approach be used
when there are significant storages or diversions present in the catchment. Kuczera et al.
(2003), inspired by Rigby and Bannigan (1996), explored basin performance using a
theoretical catchment at Observatory Hill, Sydney in a continuous simulation approach and
found similar problems with peak flow being underestimated by a similar amount when
76
Temporal Patterns
storages were present. All of these studies were based on catchments less than 110 km2
that are close to Sydney.
5.3. Storm Database

As part of the IFD revision (Book 2, Chapter 3) the Bureau of Meteorology produced a
quality controlled pluviograph database, containing 2280 stations with more than 8 station
years (refer toBook 2, Chapter 3 and Book 1, Chapter 4). The average station record length
is 25 years, with a combined record length of 57 000 years. A total of 754 of the stations are
owned by the Bureau of Meteorology and the other 1526 are owned by other data agencies
throughout Australia. This data was provided by the Bureau of Meteorology, who undertook
quality controlling to identify suspect data, such as missing data, accumulated totals and
time shifts (Green et al., 2011). All station data was provided as time series data sampled at
a 5 minute time step. In addition to the rainfall depths, each data point had a quality code
indicating the quality of the data. Figure 2.5.5 below depicts the geographical distribution of
the pluviographs across Australia. Each station is represented as a circle, with the size of the
circle indicating the record length of the station.
Figure 2.5.5. Pluviograph Stations Record Lengths
Table 2.5.1 presents the maximum number stations available in any year for decadal
periods. Table 2.5.1 demonstrates that most of the pluviograph record is from 1960 onwards.
Though overall the data comes from a period that starts before 1900, over half of the data is
from the period from 1993 onwards.
77
Temporal Patterns
Table 2.5.1. Number of Pluviographs by Decade

Decade Maximum Number of Pluviographs in any
Year
<1900 2
1900-1940 24
1940-1960 154
1960-1970 392
1970-1980 923
1980-1990 1273
1990 -2000 1926
2000-2010 1837a
aNot all data was available from 2009 onwards
As shown in Figure 2.5.5 the highest density of pluviograph stations is typically found along
coastal areas of Australia around key population centres. In Figure 2.5.6, the pluviograph
stations are seen to be clustered around urban areas, such as Sydney, Wollongong,
Melbourne and Canberra. Less data is available in central Australia, with the exception of
the Alice Springs area. Large areas of central Western Australia contain no data.
Figure 2.5.6. Pluviograph Stations used Throughout South-Eastern Australia with Record
Lengths
78
Temporal Patterns
5.3.1. Data Quality

Where a single station is being analysed it is important that it has a reasonably long record
(preferably longer than 30 years) and only a small percentage of missing data. However, for
studies that pool data the quality of the data is a bigger issue than record length. It is
however, important to remember that most of the pluviograph data is from the last few
decades.
The pluviograph database contains a significant number of events with long periods of
apparently uniform rainfall. While some of this could be from events with relatively uniform
rainfall, most appears to be the result of disaggregating accumulated rainfall totals uniformly.
Even if rainfall is uniform, a tipping bucket rain gauge recording at 5 minutes will only record
uniform rainfall over an extended period if the depth in each five minute period is equal or
slightly larger than an integer multiple of the tipping bucket capacity. It was concluded that
most of the periods of uniform rainfall are probably caused by digitisation and resampling at
different time steps. Events with large periods of uniform rainfall were disregarded, while
events were kept if the uniform period was only a small portion of the entire event.
In addition, several other issues with the data quality were found. Some records contained
significant periods of missing data. There were also periods of interpolated data, where
several data points were indicated as interpolated from a later point, presumably at the end
of an event. Events where a significant part of the rainfall was interpolated were excluded
from further analysis. There were also sections where the rainfall was uniform over many
intervals within an event, indicating that the data points were interpolated, even though the
quality control value did not specify that to be the case. Since storms generally do not have
uniform rainfall at the local scale, events that had a significant part of their total rainfall depth
occurring in consecutive identical intervals were also excluded.
5.3.2. Event Selection and Analysis

Once the data quality checking was complete, all events with a burst greater than 1
Exceedance per Year (EY) using the 2013 IFD and which were not flagged with problems
were extracted for further analysis. For each event the start and finish of the event was
defined using the methodology described in ARR Revision Project 3 - Temporal Patterns of
Rainfall Report Part 1 (WMAwater, 2015a), which was consistent with the storm event
definition used in the ARR Revision Project 6 - Losses (Hill et al., 2014). Once all events with
a burst greater than 1EY were defined for every duration, the following properties where
calculated:
• burst rainfall depth probabilities for each duration;
• start and finish time for each burst;
• time when 50% of the burst depth occurred (ie. burst loading);
• pre-burst rainfall depth for each burst duration;
• pre-burst to burst rainfall ratio for each duration; and
• Post-burst depth.
For each event the rarest or critical rainfall burst (of any duration and location within the
overall event) was also calculated.
79
Temporal Patterns
5.3.3. Regional Characteristics

To assess the regional characteristics, 12 temporal pattern regions were defined based on
the 54 Natural Resource Management (NRM) sub-regions used for investigating the impacts
of climate change (CSIRO and Australian Bureau of Meteorology, 2015). The adopted
regions follow drainage basin boundaries. Figure 2.5.7 depicts the adopted temporal pattern
regions. Table 2.5.2 summarises the number of rainfall gauges and storm events that exist in
each region.
Table 2.5.2. Regions- Number of Gauges and Events

Region Number of Number of Number of Average
Gauges Station Years Events Number of
Events per
Station Year
Southern Slopes 110 2954 3477 1.18
(Tasmania)
Southern Slopes 356 8536 20581 2.41
(mainland)
Murray Basin 233 6316 18399 2.91
Central Slopes 118 2767 7167 2.59
East Coast South 331 8067 19856 2.46
East Coast North 210 5187 12123 2.34
Wet Tropics 99 2474 5437 2.20
Monsoonal North 211 5054 12287 2.43
Rangelands 93 2334 5391 2.31
West
Rangelands 226 5561 12618 2.27
Flatlands West 349 9113 26402 2.90
Flatlands East 56 1401 3450 2.46
80
Temporal Patterns
Figure 2.5.7. Temporal Pattern Regions
The term ‘burst loading’ refers to the distribution of rainfall within a burst and is a defining
characteristic of a rainfall event. For each event the burst loading was calculated as the
percentage of the time taken for 50% of the burst depth to occur. The burst loading can be
used as a simple measure of when the heaviest part of the burst occurs and can be used to
categorise events as ‘front’, ‘middle’ or ‘back’ loaded. Events where categorised into three
groups, depending on where 50% of the burst rainfall occurs:
• front loaded – 0 to 40% of the time;
• middle loaded – 40 to 60% of the time; and
• back loaded – 60 to 100% of the time.
Figure 2.5.8 depicts a typical event from each category.
81
Temporal Patterns
Figure 2.5.8. Example of Front, Middle and Back Loaded Events
This simple categorisation provides a pragmatic means of capturing when the peak loading
occurs, as described by Pilgrim et al. (1969), though it is worth recognising that for a double
peaked event, when most of the rainfall is in the early and later part of the burst, the loading
can somewhat illogically fall into the middle category.
Each region was characterised by its burst loading distribution, which describes the
percentage of front, middle and back loaded events for different durations. The proportion of
front/middle/back loading for each region was determined for less than and greater than 6
hours duration (Table 2.5.3). The proportion was assumed to be constant across all AEPs.
Table 2.5.3. Burst Loading by Region and Duration
Region Duration Front Loaded Middle Loaded Back Loaded

(%) (%) (%)
Southern Slopes ≤ 6hr 21.5 64.1 14.4
(Tasmania) > 6hr 20.5 60.0 19.5
Southern Slopes ≤ 6hr 30.1 53.0 16.9
(mainland) > 6hr 22.7 53.7 23.6
Murray Basin ≤ 6hr 28.3 53.8 17.9
> 6hr 24.7 52.5 22.7
Central Slopes ≤ 6hr 31.0 53.3 15.7
> 6hr 27.0 46.9 26.1
East Coast South ≤ 6hr 26.5 57.1 16.4
82
Temporal Patterns
Region Duration Front Loaded Middle Loaded Back Loaded

(%) (%) (%)
> 6hr 17.1 58.6 24.3
East Coast North ≤ 6hr 28.9 56.5 14.6
> 6hr 23.4 48.5 28.1
Wet Tropics ≤ 6hr 16.0 71.8 12.2
> 6hr 18.7 58.1 23.2
Monsoonal North ≤ 6hr 27.6 63.7 8.8
> 6hr 27.5 41.4 31.2
Rangelands ≤ 6hr 23.7 62.5 13.8
West > 6hr 23.6 49.2 27.2
Rangelands ≤ 6hr 29.0 56.6 14.3
> 6hr 24.4 49.2 26.4
Flatlands West ≤ 6hr 30.8 49.3 19.9
> 6hr 31.4 48.9 19.7
Flatlands East ≤ 6hr 27.6 52.4 20.0
> 6hr 17.4 54.4 28.2
5.4. Pre Burst Rainfall and Antecedent Conditions

The events database allowed the pre-burst behaviour of rainfall events to be characterised,
regionalised and mapped. Temporal Patterns report 2 WMAwater (2015a) presents the
detailed analysis, regionalisation and mapping of pre-burst behaviour for Australia. Due to
the relatively short pluviograph records, the approach assumes that this behaviour can be
pooled to develop reasonably sized samples by transferring storms from nearby locations
with similar IFD characteristics. This was done on a Region of Influence basis and not a
fixed region basis, as the ratio of pre-burst rainfall has some correlation with Intensity
Frequency Duration characteristics. In many parts of Australia the pre-burst rainfall generally
represents a very small amount of the event and generally does not contribute to the runoff
response, so it can be treated in a relatively simply manner. However, in some parts of the
country pre-burst rainfall can represent a significant part of the rainfall event and runoff
response. Pre-burst can also be important in urban catchments with large directly connected
impervious areas (Book 5). Storage strategies need to account for this additional runoff
when sizing storage tanks and basins (Book 9). The pre-burst was characterised based on
the rarest rainfall duration burst within the storm using the 2013 IFDs. Hill et al. (2015) found
this approach gave a biased estimate of the average pre-burst, systematically
underestimating the depth of the pre-burst. Following from this work and the expected
update of the IFDs from the Bureau of Meteorology in 2016 this work will be updated and
Figure 2.5.9, Figure 2.5.10 and Figure 2.5.11 will be updated accordingly.
Figure 2.5.9 and Figure 2.5.10 depict the median ratio of the pre-burst to burst and the depth
of pre-burst in mm for the 6hr duration and probabilities mapped across Australia. The full
data set of maps is available online at the ARR data repository (http://data.arr-software.org/).
Figure 2.5.11 shows the probability distribution of the pre-burst rainfall for each region.
83
Temporal Patterns
Figure 2.5.9. Pre-burst Rainfall
Figure 2.5.10. Pre-burst to burst ratio
84
Temporal Patterns
Figure 2.5.11. Standardised pre-burst to burst ratio distributions
The manner in which pre-burst rainfall is treated depends on the magnitude of the pre-burst
rainfall, how this compares to losses and whether pre-burst runoff is likely and will have a
significant effect on hydrograph volumes. As longer duration bursts tend to represent most of
the storm events, it is generally only an issue for smaller catchments. If pre-burst rainfall is
unlikely to affect the runoff responses, it is best treated in a simple manner with losses. For
simple urban cases pre-burst rainfall can be used to condition storage starting conditions.
Where pre-burst is influential for flood response, it can be sampled from its distribution and
applied with a typical pre-burst temporal pattern.
5.5. Design Point Temporal Patterns

As part of the ARR Revision Project 3 - Temporal Patterns of Rainfall (WMAwater, 2015b) a
series of temporal pattern techniques were trialled. A total of 35 test catchments were
adopted across Australia, ranging in area from 30 to 80 km2. The aims of the testing were to
assess the performance of the design method using best estimates of the revised inputs and
to determine what influence the temporal patterns have on design estimates. Temporal
pattern ensembles were tested within a Monte Carlo framework and a simpler quantile
ensemble framework. Two methods were trialled, the regional temporal patterns (as
discussed in Book 2, Chapter 5, Section 3) along with a Region of Influence (ROI) approach.
Three event types were trialled being burst only, burst plus pre-burst and complete storm
events (for ROI only). A range of initial loss approaches were trailed, including burst and
storm losses, combined with the median value or a distribution of values. The nature of
regional temporal pattern ensembles requires AEP bins to be derived. The ensembles for
each AEP bin were trialled across all AEPs to assess the sensitivity of the AEP bins.
85
Temporal Patterns
Based on the results of this performance testing, the temporal patterns derived by
theregional burst approach are recommended for general use. Temporal patterns for
complete storms derived by the ROI approach are left as an alternative for small, volume-
sensitive systems where pre- and post-burst rainfall is important, though it may be that such
systems are better analysed using continuous simulation approaches.
Other major findings from ARR Revision Project 3 - Temporal Patterns of Rainfall
(WMAwater, 2015b) were:
1. Irrespective of the method used to derive the temporal patterns, when using a
representative ensemble of patterns all methods produce relatively similar results; and
2. Frequent patterns should not be used for rarer events; scaling a temporal pattern
introduces more variability and produces higher design estimates.
Ensembles of 20 temporal patterns were initially envisaged, however a large number of

regions had insufficient data available to warrant a sample of this size. Different ensemble
sizes were tested to determine the sensitivity to ensemble size, with little bias found between
the ensemble sizes. Samples of 10 temporal patterns were therefore adopted as an
appropriate compromise between pattern variability and data availability. Ensembles were
generated for each AEP bin, duration and region. The four AEP bins adopted (see
Table 2.5.4 and Figure 2.5.12) were based on the burst AEP at the source of the event.
Temporal Pattern were extracted for the following durations from 15 minutes to 7 days, as
shown in Table 2.5.5.
Table 2.5.4. Regional Temporal Pattern Bins
AEP Group AEP Range

Very Rare Rarest 10 within region
Rare Rarer than 3.2% AEP
Intermediate Between 3.2% and 14.4% AEP
Frequent More frequent than 14.4% AEP
Figure 2.5.12. Temporal Pattern Ranges
Table 2.5.5. Temporal Pattern Durations
Duration
Minutes Hours Days
15 0.25 0.010
30 0.5 0.021
60 1 0.042
120 2 0.083
86
Temporal Patterns
Duration
180 3 0.125
270 4.5 0.188
360 6 0.250
540 9 0.375
720 12 0.500
1080 18 0.750
1440 24 1
2160 36 1.5
2880 48 2
4320 72 3
5760 96 4
7200 120 5
8640 144 6
10080 168 7
When selecting a small ensemble of temporal patterns it is important to capture the typical
variability of the observed events. A methodology was therefore adopted that samples from
observed events with the intent of generating a representative ensemble in terms of the
variability of actual events, with no obvious bias. Whilst it is difficult to quantify or verify the
achievement of this objective, steps have been undertaken to ensure the samples are
broadly representative, including a visual check of all ensembles.
There are a large number of frequent events from which to select, however, the choice is
more limited for rarer events.
For each AEP and duration bin the sampling of an ensemble of 10 patterns used the
preferred criteria in Table 2.5.6. The relaxed criteria were used where less than 10 suitable
patterns could be found using the preferred criteria.
Table 2.5.6. Temporal Pattern Selection Criteria

Preferred Criteria (for Data Rich Regions) Relaxed Criteria (for Data Sparse
Regions)
No rarer embedded bursts n/a
No events overlapping in time n/a
Loading characteristics enforced Loading characteristics ignored
Events from within a region Sampled from within region and
neighbouring region with similar climatic
characteristics
While patterns containing embedded rarer bursts at their recording location were not
selected, events can still have embedded bursts at other locations within the region. The
presence of major embedded bursts may warrant filtering of patterns.
87
Temporal Patterns
5.6. Temporal Patterns for Areal Rainfall Bursts

As part of ARR Revision Project 3 - Temporal Patterns of Rainfall a series of areal average
temporal patterns have been produced for different sized hypothetical catchments. These
patterns average the spatial variability of rainfall in each time step which does remove some
of the variability of actual space-time rainfall fields. The process of calculating areal temporal
patterns involves identifying high rainfall areal events and calculating areal temporal
patterns. A brief summary is provided in this chapter (refer to Podger et al. (2016) for more
detail). Areal rainfall temporal patterns were calculated for the area and duration
combinations listed in Table 2.5.7.
Table 2.5.7. Areal Rainfall Temporal Patterns - Catchment Areas and Durations
Variable
Catchment Area (km2) 100, 200, 500, 1000, 2500, 5000, 10 000, 20
000, 40 000
Durations (Hours) 12, 18, 24, 36, 48, 72, 96, 120, 144, 168
5.6.1. Areal Rainfall Time Series Grid for All Australia

To create sets of areal temporal patterns an areal rainfall grid was derived for 30 minute time
steps using the events database (described in Book 2, Chapter 5, Section 3, ie. 2280
pluviographs). Rainfall values from these pluviographs were further filtered for long
disaggregations and erroneously high rainfall values. A grid resolution of 0.025° identical to
the grid cell size used for the design rainfalls (Book 2, Chapter 3) was chosen for simplicity
of use. A natural neighbours algorithm, which is a variant of Thiessen polygons, was used to
interpolate rainfall values from the pluviograph stations to each grid cell with rainfall values
for each time step from January 1960 to December 2010.
5.6.2. Average Areal Rainfall Calculation

The approach aims to identify the largest set of independent areal rainfall events within the
Temporal Pattern Regions shown in Figure 2.5.7 by not restricting catchments to a particular
shape. For each combination of duration and area, a set of rectangular hypothetical
catchments were used to sample the areal rainfall time series and create catchment average
rainfalls. At each catchment centroid, 12 hypothetical catchments were trialled using the
combinations of 3 aspect ratios (1.5, 2.3 and 3.6) and 4 rotations (0, 45, 90 and 135
degrees) shown in Figure 2.5.13. To best represent Australian catchments, the aspect ratios
were chosen by selecting the 20th, 50th and 80th percentiles of shape factors calculated for
the 798 catchments that were adopted for the RFFE method described in Book 3, Chapter 3.
Average areal rainfall was calculated for a hypothetical catchment’s each time step simply by
averaging the rainfall of all the grid cells with midpoints within the catchment area. To lessen
computational requirements, larger time steps were used for larger catchment areas.
88
Temporal Patterns
Figure 2.5.13. Combinations of Aspect Ratio and Rotation for Hypothetical Catchments
5.6.3. Areal Temporal Pattern Selection

Event rainfall totals were calculated for the sampled areal temporal patterns. For each of the
temporal pattern regions (Figure 2.5.7) a set of events for all combinations of catchment
area and durations (as per Table 2.5.7) with the largest depths and no space-time overlap
were selected. The space-time filter required a minimum of 3 days between the end of one
event and the start of another and for catchment centres to be a minimum of either twice
their longest side or 100 km apart.
To ensure data quality, hypothetical catchments that did not contain the minimum number of
pluviograph stations (Table 2.5.8) within their vicinity that were producing quality data at the
time of the event were disregarded. An additional filter was then applied to remove events
that had too much area assigned to a single gauge or erroneous/unrealistic rainfall values.
Each space-time independent areal temporal pattern had a Pearson correlation coefficient
derived between the areal pattern and all the corresponding temporal patterns of
pluviograph stations within its vicinity. Patterns that had a very high correlation to a single
gauge and no others or did not have enough stations with a reasonable correlation to the
areal pattern were removed.
Table 2.5.8. Minimum Number of Pluviographs Required for Event Selection for Each
Catchment Area
Catchment Areas (km2) Minimum Number of Pluviographs

100 3
200 3
500 3
1000 6
2500 6
5000 6
10 000 9
20 000 20
40 000 40
89
Temporal Patterns
Given that events were chosen simply based on their total rainfall depth, many of the longer
duration patterns selected represented a shorter duration. Using the same procedure
implemented in defining event extents in the events database, events that had extents less
than the duration shorter than the duration of interest were removed.
5.6.4. Design Areal Temporal Patterns

Due to the constraints on data quality, a full set of areal temporal patterns for every region,
duration and area could not be generated. Therefore, to use areal temporal patterns for data
sparse regions, it is necessary to sample patterns from climatically similar or nearby regions.
This is especially a problem for longer durations, with some durations such as the 7 days not
having enough patterns in any region. For these durations patterns must therefore be
sampled from all of Australia, although data dense areas contain the majority of these
events.
Like the point temporal patterns, meta-data on each pattern has been provided that allows
practitioners to track the location and time of the event. Figure 2.5.14 compares the
cumulative mass curves for 24 hour point and area temporal patterns. This demonstrates
how the spatial averaging produced by areal temporal patterns reduces the variability.
Figure 2.5.15 compares areal temporal patterns to the temporal patterns of the pluviograph
closest to the area’s centroid, further highlighting a reduction in variability with the areal
temporal patterns.
Figure 2.5.14. Comparison of Point Temporal Patterns and Areal Temporal Patterns - East
Coast South Region - 1 Day
90
Temporal Patterns
Figure 2.5.15. Comparison of Areal Temporal Patterns and the Temporal Pattern of the
Closest Pluviograph for the Same Event
5.7. Other Temporal Pattern Options

5.7.1. Use of Historical Temporal Patterns
On a large catchment with good quality temporal rainfall data it is not uncommon to have a
reasonable number of historical patterns for events rarer than the 20% AEP event. Where
there is a reasonable amount of local temporal data of appropriate AEPs across the
catchment, it is appropriate to use this local data in place of regional temporal patterns. On
large catchments, local point patterns will exhibit more variability and generally produce
higher flow estimates than locally derived areal temporal patterns and should be used with
caution. Some guidance on the use of local temporal patterns is provided in Book 2, Chapter
6.
Book 2, Chapter 6 provides advice on the assignment of historical temporal patterns when
modelling observed events for calibration or analysis.
5.7.2. Complete Storm Patterns

The methodology developed for the pre-burst estimation pooled storm events from nearby
catchments based on distance and IFD similarity. This approach was extended to develop
ensembles of complete storms based on their critical burst duration by selecting events that
can be put in a site specific duration probability bin with minimal scaling. The approach is
somewhat experimental and has performed as well as burst approaches in limited testing.
While categorising events on the basis of critical storm burst solves a lot of event selection
problems but it is different from more traditional burst approaches. Aspects of the approach
are described in ARR Revision Project 3 – Temporal Patterns of Rainfall Report (WMAwater,
2015c).
91
Temporal Patterns
The major differences are:
• Events are selected and duration binned on the basis of the critical rainfall duration;
• An event can only be used in one duration so a large storm sample is required;
• Duration bins represent a range of storm durations instead of a fixed duration;
• Event filtering is not required; and
• A new sample needs to be created at each location so a regional approach is not

possible.
The process of binning events on the basis of their critical rainfall duration means that events
that produce a catchment response because of rainfall over a certain duration can be placed
in a bin of a very different duration. This means that under a critical duration approach these
event are unlikely to influence the design estimate even if they produce the largest
catchment response. This problem would not occur in a Monte Carlo framework that
samples across durations.
Despite these limitations, a complete storm approach has the advantage of producing
realistic rainfall storm events that have the correct burst IFD and storm volume
characteristics. Coombes et al. (2015) showed considerable difference in basin performance
on a small urban catchment between burst and complete storm approaches.
5.7.3. Continuous Data

While continuous simulation is generally more appropriate for modelling very frequent events
there are some situations where event models might be appropriate. For these situations
temporal patterns could be sampled from a local pluviograph or sampled from a generated
continuous sequence using the techniques described in Book 2, Chapter 7.

Very little information is known about how climate change will affect storm or burst temporal
patterns. The most definitive work is by Wasco and Sharma (2015) which analysed the
relationship between burst patterns from 79 Australian rainfall gauges and temperature
variations. This study found that, regardless of the climate region or season, temperature
increases are associated with patterns becoming less uniform, with the largest fractions
increasing in rainfall intensity and the lower fraction decreasing. The scaling of the largest
fractions was more pronounced in short duration bursts, and rainfall gauges to the north of
the country showed more pronounced changes. While this work is based on the present
climate, it provides an insight into how patterns could change under a future warming. Such
climate changes would also affect IFD characteristics.
Westra et al. (2013) proposed rainfall sequences for future climates could be constructed by
sampling historical rainfall patterns corresponding to warmer days at the same location, or
from locations which have an atmospheric profile more reflective of expected future climate.
Such an approach could conceptually be applied in the selection of design burst patterns but
would require significant testing.
It is currently not possible to provide practical advice to practitioners on this aspect of

temporal patterns. It is recommended that until further studies are completed, for simulations
92
Temporal Patterns
applying projected climate change, the temporal patterns applied should be those derived for
existing climatic conditions but recognising the additional uncertainty in simulation results.
5.9. Temporal Pattern Application and Pre-burst

5.9.1. General
For the majority of problems, the practitioner needs to select a temporal pattern based on
the area and probability of the event they are modelling. For Very Rare and Extreme events
specific advice is provided in Book 8. The derivation of long duration events (both point or
areal temporal patterns) present a number of challenges. Many of the events selected in the
IFD Annual Maxima Series were from shorter duration events and can not be selected as
longer duration temporal patterns. Some long duration events are more like consecutive
events with none or little rainfall in between shorter events. For this reason, in most
locations, it was necessary to borrow temporal patterns (point and areal) from adjoining
regions. Daily rainfall gauges located in the catchment of interest should give a good insight
into the applicability of longer duration temporal patterns approaching 7 days.
Point temporal patterns should be used for catchments less than 75 km2. Areal temporal
patterns have been derived for a number of different catchment areas. Table 2.5.9 provides
a guide to applying the areal patterns. Both point and areal temporal pattern sets can be
downloaded from the ARR Data Hub (Babister et al (2016), accessible at http://data.arr-
software.org/)
Table 2.5.9. Areal Temporal Pattern Sets for Ranges of Catchment Areas
Range of Target Catchment Areas (km2) Catchment Area of Designated Areal
Temporal Pattern Set (km2)
75 – 140 100
140 – 300 200
300 – 700 500
700 – 1600 1000
1600 – 3500 2500
3500 – 7000 5000
7000 – 14,000 10,000
14,000 – 28,000 20,000
28,000 + 40,000
5.9.2. Ensemble Considerations

The use of an ensemble of 10 temporal patterns as discussed in Book 2, Chapter 5, Section
5 is recommended. The temporal patterns have been chosen to represent the variability in
observed patterns. Given the run times of two dimensional hydraulic models it is not practical
to run all 10 patterns (for multiple durations). A more practical approach would be to run a
separate hydrological modelling process of the whole catchment of interest in order to
determine the average pattern in terms of peak flow or volume depending on the problem
(Book 2, Chapter 5, Section 10). It is not recommended that the temporal pattern that
represents the worst (or best) case be used by itself for design. Testing has demonstrated
that on most catchments large number of events in the ensemble patterns are clustered
around the mean and median.
93
Temporal Patterns
When selecting a single representative (average) pattern the practitioner needs to look
at the whole catchment response hydrograph and not local inflow hydrographs.
The ensemble of 10 pattern provides a range of plausible answers. The practitioner should
consider the benefits of investigating multiple temporal patterns or Monte Carlo for sensitive
designs and solutions.
Running an ensemble of ten temporal patterns through a two dimensional model could
be time consuming. One option is to double the grid cell size which will decrease run
time 8 fold.
5.9.3. Upscaling of Patterns

Practitioners are cautioned to avoid or minimise upscaling patterns from frequent events to
rare events. In some locations the dominate rainfall mechanism might change with
probability and frequent events can exhibit more variability than rarer events. While there are
good meteorological reasons for very rare events to exhibit less variability then frequent
events (need reference) the causes are probably more complex than a single reason and
somewhat region specific. Testing carried out as part of the ARR Revision Project 3 –
Temporal Patterns of Rainfall has shown that in some locations that upscaling can be a
significant issue (Loveridge et al., 2015) while in other locations results were insensitive. In
nearly all testing locations results were relatively insensitive to down scaling of patterns.
5.9.4. Dealing with Inconsistencies and Smoothing of Results

Testing has demonstrated that inconsistent results can sometimes occur with AEP when the
bin patterns are drawn from changes particularly between the frequent and intermediate
bins. The problem was more pronounced in drier regions where a large part of the rainfall
was taken up by losses. In some cases, flood magnitude decreased for rarer AEPs while in
others the change with AEP was very stepped. On some catchments, the critical duration
varied inconsistently with AEP. Where problems are mainly caused by the pattern bin
changing two simple options are suggested, either:
1. Draw from both bins at the boundary effectively doubling the ensemble size which
effectively smooths the results; or
2. Replacing the frequent probability bin with the intermediate bin to ensure a smooth
catchment response with rainfall.
Consideration should be given to filtering out (or excluding) embedded bursts of lower AEP
by re-distributing rainfalls of high intensity to other time increments proportionally to their
magnitude (e.g. Herron et al. (2011)). In some situations results will still need to be
smoothed.
5.9.5. Practical Issues

If modelling events greater than 5 day, the practitioner should look at the daily rainfall gauge
totals for gauges within the catchment and confirm what the 7 day events look like in the
catchment compared to the regional patterns.
If 10 temporal patterns are not available for a given region, duration and frequency bin then
patterns were taken from other similar regions (Table 2.5.10). For 7 day events there was not
94
Temporal Patterns
enough patterns and all regions were pooled. Local knowledge can be used to add events to
the ensemble that weren’t selected in the regional sample.
Table 2.5.10. Alternate Regions Used for Data

Region Alternate
Region
Southern Slopes Southern Slopes
(Tasmania) (mainland),
Murray Basin
Southern Slopes Murray Basin,
(mainland) Southern Slopes
(Tasmania),
Central Slopes,
East Coast South
Murray Basin Central Slopes,
Southern Slopes
(mainland),
Flatlands East,
East Coast South
Central Slopes Murray Basin,
East Coast
South, East
Coast North,
Southern Slopes
(mainland)
East Coast South East Coast
North, Central
Slopes, Murray
Basin, Southern
Slopes
(mainland)
East Coast North East Coast
South, Wet
Tropics, Central
Slopes, Murray
Basin
Wet Tropics Monsoonal
North, East
Coast North
Monsoonal North Wet Tropics,
East Coast
North,
Rangelands
West
Rangelands Rangelands,
West Flatlands West
95
Temporal Patterns
Region Alternate
Region
Rangelands Flatlands West,
Flatlands East,
Flatlands West
Flatlands West Flatlands East,
Rangelands
West
Flatlands East Flatlands West,
Flatlands West,
Rangelands
5.9.6. Point and Areal Temporal Pattern Meta-Data

Meta-data is provided (accessed via http://data.arr-software.org/) which describes where the
regional point and areal temporal patterns were originally located and when they occurred.
This includes:
• Event ID - Unique ID given to each pattern;
• Region - The region in which the pattern applies;
• Region (Source) - The region the temporal pattern occurred in;
• Burst Loading - Classified events as front, middle or back (see Book 2, Chapter 5, Section
3);
• Burst Loading (%) - The percentage of the event that falls into the loading category;
• Burst Depth - Source temporal pattern burst depth. This is the total depth the original
pattern had, patterns are given in percentages and not original depths;
• AEP (source) - The source temporal pattern AEP at the location it was recorded. This AEP
is based of the 2016 IFDs and the burst depth of the event;
• Burst Start - The time and date at which the burst began;
• Burst End - The time and date at which the burst ended; and
• Pluviograph number - The ID of the pluviograph as from the Bureau of Meteorology.
There will be events that weren’t picked up in the regional sample however, local experience
will allow them to be used in the ensemble (note when doing this the front, middle and back
loading of the regional needs to be considered).
5.9.7. Very Rare Point Temporal Patterns

For point temporal patterns 4 bins are provided (Figure 2.5.12). The very rare bin contains
the rarest ten patterns within the region. These patterns may not feature in the rare bin for
the region as the rare bin is sampled randomly.
5.9.8. Region Considerations

The regional point temporal patterns have been sourced from various locations throughout
the region and due to IFD gradients within the regions embedded bursts may occur.
96
Temporal Patterns
Depending on the severity of the embedded burst, filtering of rarer bursts is recommended.
When minor embedded bursts occur in only a few of the 10 patterns in the ensemble filtering
can be neglected. Testing has shown with the 2013 Intensity Frequency Duration data that
there are some locations where all observed temporal patterns would have embedded
bursts.
Locations close to region boundaries might experience storm mechanisms represented by

both regions. In these cases it is recommended to run a larger ensemble of all patterns, ie.
All 10 patterns from each neighboring region.
5.9.9. Pre-burst
The treatment is dependent on the approach taken to model losses and the magnitude of
the pre-burst. In most cases practitioners will use just the median pre-burst and median
storm initial loss. However, in some cases the practitioner may sample from distributions of
pre-burst and initial loss. As pre-burst varies with location, duration and probability the
adopted approach will vary, but it is probably sensible to adopt a consistent approach across
durations and probabilities being assessed. Pre-burst for a specific location can be
downloaded from the ARR Data Hub (Babister et al (2016), accessible at http://data.arr-
software.org/).
In locations and for durations that do not have significant pre-burst (Figure 2.5.10), the pre-
burst depth can be ignored when applying a temporal pattern. Therefore the Burst IL (ILb)
can be taken as the Storm IL (ILs). In those locations where the pre-burst is significant a
number of approaches are possible:
• Storm IL is greater than Pre-burst – Pre-burst should be taken out of the storm IL
ie. ILs – Pre-burst = ILb;
• Storm IL is approximately equal to Pre-burst – In the case where storm IL and pre-burst
are close to equal IL is satisfied and no IL needs to be taken from the burst temporal
pattern;
• Pre-burst is greater than storm IL - In the case where pre-burst is larger than the storm IL
there are a number of options:
• Apply a pre-burst temporal pattern after taking out the Storm IL. There is little research
that has investigated pre-burst temporal patterns;
• Test the sensitivity of pre-burst on the resultant flood estimate and determine if it can be
ignored;
• Apply a complete storm approach instead of a burst approach e.g. Coombes et al.
(2015).
5.10. Example
The Tennant Creek Catchment is located in the Northern Territory and was used for the ARR
research Projects 3 and 6. The catchment and catchment location is shown in Figure 2.5.16.
It is located in the Rangeland Region for temporal patterns. The temporal pattern data for the
region can be extracted at http://data.arr-software.org/. In this particular example a RORB
model was set up for the catchment. The model was previously calibrated with an ILb=0 and
a CL=7 mm/hr. As the catchment area is 72.3 km2 it was decided that the critical duration
would be between 60 minutes and 1440 minutes (1 day).
97
Temporal Patterns
Figure 2.5.16. Tennant Creek Catchment
For this example only the 1% AEP patterns were run through the hydrologic model. 10
patterns from the rare AEP bin and the Rangelands region were run for each of the following
durations:
• 60 minute
• 120 minute
• 180 minute
• 270 minute
• 360 minute
• 540 minute
• 720 minute
• 1080 minute
• 1440 minute
Figure 2.5.17 is a presentation of the results in a box plot. The results are also presented in
Table 2.5.11, it should be noted that even though the columns are labelled bursts 1 to 10
they are not the same storms across the durations. The box plot presents clearly that the
180 minute duration is critical. The average peak flow of the 180 minute duration bursts
(277.35 m3/s) should then be taken as the 1% AEP flow. If a practitioner wanted to run a
98
Temporal Patterns
pattern through a hydraulic model and it was not practical to run all 10 patterns, the pattern
closest to the average (shown in Table 2.5.11) is Burst number 1.
Figure 2.5.17. Duration Box plot for the 1% AEP
Table 2.5.11. Flows for the 1% Annual Exceedance Probability for Ten Burst Events
Duration 1% AEP Peak Flow (m3/s)

Burst 1 Burst Burst Burst Burst Burst Burst Burst Burst Burst Average Median
2 3 4 5 6 7 8 9 10
60 146.9 173.8 179.6 145.7 157.2 159.6 169.4 141.6 148.8 150.2 157.28 153.7
120 221.5 230.2 261.9 238.1 260.3 259.5 233 240.2 244.9 262.6 245.22 242.55
180 279.4 264.8 279.6 274.4 263.5 273.5 264.1 272.8 273.1 268.3 277.35 273.95
270 262.5 241.1 284.1 288.5 306.6 264.7 266.8 261.2 287.8 254.4 271.77 265.75
360 255.5 285.5 214.4 243.9 199.7 223.2 246.3 246.3 269.8 283.9 246.85 246.3
540 230.9 163.1 168.7 195.4 254.7 203.2 236.6 235.6 177.5 259.7 212.54 217.05
720 193.7 173.5 188.1 143.8 254.9 243.6 195.2 210.3 172.3 224.2 199.96 194.45
1080 197.7 159.2 103.8 239 179.1 151.3 191.2 160.5 148.1 151.5 168.14 159.85
1440 177.9 113.5 144 108.8 222.5 103.4 173.2 156.9 272 130.5 160.27 150.45
5.11. References
Askew, A.J. (1975), Use of Risk Premium in Chance-Constrained Dynamic Programming.
Water Resources Research, 11(6): 862-866.
99
Temporal Patterns
Babister, M., Trim, A., Testoni, I. and Retallick, M. 2016. The Australian Rainfall and Runoff
Datahub, 37th Hydrology and Water Resources Symposium Queenstown NZ
Ball, J.E. (1994), 'The influence of storm temporal patterns on catchment response', Journal
of Hydrology, 158(3-4), 285-303.
Brown, J.A.H. (1982), A Review of flood estimation procedures, Proceedings of the

workshop on spillway design, Department of National Development and Energy, AWRC
Conference Series No.6.
CSIRO and Bureau of Meteorology (2015), Climate Change in Australia Information for
Australia's Natural Resource Management Regions: Technical Report, CSIRO and Bureau of
Meteorology, Australia.
Coombes, P., Babister, M. and McAlister, T. (2015).Is the Science and Data underpinning the
Rational Method Robust for use in Evolving Urban Catchments, Proceedings of the 36th
Hydrology and Water Resources Symposium Hobart 2015.
Cordery, I., Pilgrim D.H. and Rowbottom I.A. (1984), Time patterns of rainfall for estimating
design floods on a frequency basis. Wat. Sci. Tech., 16(8-9), 155-165.
French, R.H. (1985), Open Channel Hydraulics, McGraw-Hill Book Company, New York.
Green, J.H., Xuereb, K. and Siriwardena, L. (2011), Establishment of a Quality Controlled

Rainfall Database for the Revision of the (IFD) Estimates for Australia, Proceedings of 34th
IAHR World Congress, Brisbane, 26 June-1st July 2011.
Hall, A.J. and Kneen, T.H. (1973), Design Temporal Patterns of Storm Rainfall in Australia,
Proc. Hydrology Symposium, Inst. Eng. Aust., pp: 77-84
Herron, A., Stephens, D., Nathan, R. and Jayatilaka, L. (2011), Monte Carlo Temporal
Patterns for Melbourne. In: Proc 34th World Congress of the International Association for
Hydro- Environment Research and Engineering: 33rd Hydrology and Water Resources
Symposium and 10th Conference on Hydraulics in Water Engineering. Barton, A.C.T.:
Engineers Australia, pp: 186-193.
Hill, P.H., Graszkiewicz, Z., Taylor, M. and Nathan, R. (2014), Australian Rainfall and Runoff
Revision Project 6: Loss Models for Catchment Simulation, Phase 4, October 2014, Report
No. P6/S3/016B.
Hill, P.I., Graszkiewicz, Z., Loveridge, M., Nathan, R.J. and Scorah, M. (2015), Analysis of
loss values for Australian rural catchments to underpin ARR guidance. Hydrology and Water
Resources Symposium 2015, Hobart, 9-10 December 2015.
Hoang, T.M.T., Rahman, A., Weinmann, P.E., Laurenson, E. and Nathan, R. (1999) Joint
Probability description of design rainfall, Water 99 Joint congress, Brisbane.
Huff, A. (1967), Time distribution in heavy storms, Water Resources Research, 3:

1007-1019.
IEAust. (1977), Australian Rainfall and Runoff: flood analysis and design. The Institution of
Engineers Australia. Canberra.
Keifer, C.J. and Chu, H.H. (1957), Synthetic storm pattern for drainage design, ASCE
Journal of the Hydraulics Division, 83(4), 1-25.
100
Temporal Patterns
Kuczera, G., Lambert, M., Heneker, T., Jennings, S., Frost, A. and Coombes, P. (2003) Joint
Probability and design storms at the cross roads, Proceedings of the 28th International
Hydrology and Water Resources Symposium, Wollongong.
Loveridge, M., Babister, M., Retallick, M. and Testoni, I. (2015). Testing the suitability of
rainfall temporal pattern ensembles for design flood estimation, Proceedings of the 36th
Hydrology and Water Resources Symposium Hobart 2015.
Meighen, J. and Kennedy, M.R. (1995), Catalogue of Significant Rainfall Occurrences over
Southeast Australia , HRS Report No. 3, Hydrology Report Series, Bureau of Meteorology,
Melbourne, Australia, October.
Milston, A.K. (1979), The influence of temporal patterns of design rainfall in peak flood
discharge, Thesis (M.Eng.Sci.), University of New South Wales.
Nathan, R.J. and Weinmann, P.E. (1995), The estimation of extreme floods - the need and
scope for revision of our national guidelines. Aus J Water Resources, 1(1), 40-50.
Nathan, R.J., Weinmann, P.E. and Hill, P.I. (2002), Use of a Monte-Carlo framework to
characterise hydrologic risk. Proc., ANCOLD conference on dams, Adelaide.
Nathan, R., Weinmann, E., Hill, P. (2003) Use of Monte Carlo Simulation to Estimate the
Expected Probability of Large to Extreme Floods, Proc. 28th Int. Hydrology and Water Res.
Symp., Wollongong, pp: 1.105-1.112.
Pattison, A, (ed) 1977, Australian Rainfall and Runoff: Flood Analysis and Design, The
Institution of Engineers Australia
Phillips, P. and Yu, S. (2015), How robust are OSD and OSR Systems?, 2015 WSUD &
IECA, Sydney.
Phillips, B.C., Lees, S.J. and Lynch, S.J. (1994), Embedded Design Storms - an Improved
Procedure for Design Flood Level Estimation?. In: Water Down Under 94: Surface Hydrology
and Water Resources Papers; Preprints of Papers. Barton, ACT: Institution of Engineers,
Australia, 235-240. National conference publication (Institution of Engineers, Australia), No.
94/15.
Pilgrim, DH (ed) (1987) Australian Rainfall and Runoff - A Guide to Flood Estimation,
Institution of Engineers, Australia, Barton, ACT, 1987.
Pilgrim, D.H. and Cordery, I. (1975), Rainfall temporal patterns for design floods, Journal of
the Hydraulics Division, 101(1), 81-95.
Pilgrim, D.H., Cordery, I. and French, R. (1969), Temporal patterns of design rainfall for
Sydney, Institution of Engineers, Australia, Civil Eng. Trans., CE11: 9-14.
Podger, S., Babister, M., Brady, P. (2016), Deriving Temporal Patterns for Areal Rainfall
Bursts, Proceedings of the 37th Hydrology and Water Resources Symposium Queenstown
2016.
Rahman, A., Weinmann, P.E., Hoang, T.M.T. and Laurenson, E.M. (2002), Monte Carlo
simulation of flood frequency curves from rainfall. Journal of Hydrology.
Retallick, M., Babister, M., Varga, C., Ball, J. and Askew, A. (2009), Do Filtered patterns
resemble real patterns? Proceedings of the Hydrology and Water Resources Symposium,
Newcastle 2009, Engineers Australia.
101
Temporal Patterns
Rigby, E. and Bannigan, D. (1996), The Embedded design storm concept - A critical review,
23rd Hydrology and Water Resources Symposium Hobart.
Rigby, E., Boyd, M., Roso, S. and Van Drie, R. (2003), Storms, Storm bursts and flood
estimation- A need for review of the AR&R procedures, Proceedings of the 28th International
Hydrology and Water Resources Symposium, Wollongong.
Roso, S. and Rigby, E. (2006), The impact of embedded design storms on flows within a
catchment, Proceedings of the 30th Hydrology and Water Resources Symposium
Launceston 2006.
Sih, K., Hill, P. and Nathan, R. (2008), Evaluation of Simple Approaches to Incorporating
Variability in Design Temporal Patterns. In: Lambert, Martin (Editor); Daniell, TM (Editor);
Leonard, Michael (Editor). Proceedings of Water Down Under 2008. Modbury, SA: Engineers
Australia; Causal Productions, pp: 1049-1059.
Varga, C., Ball, J., Babister, M. (2009), An Alternative for Developing Temporal Patterns,
Proceedings of the Hydrology and Water Resources Symposium, Newcastle 2009,
Engineers Australia
WMAwater (2015a), Australian Rainfall and Runoff Revision Project 3: Temporal Patterns of
Rainfall, Part 1 - Development of an Events Database, Stage 3 Report, October 2015.
WMAwater (2015b), Australian Rainfall and Runoff Revision Project 3: Temporal Patterns of
Rainfall, Part 3 - Preliminary Testing of Temporal Pattern Ensembles, Stage 3 Report,
October 2015.
WMAwater (2015c), Australian Rainfall and Runoff Revision Project 3: Temporal Patterns of
Rainfall, Part 3 - Preliminary Testing of Temporal Pattern Ensembles, Stage 3 Report,
October 2015.
Webb McKeown and Associates, (2003), Assessment of the variability of rates of rise for
hydrographs on the Hawkesbury Nepean River, December 2003.
Weinmann, P., Rahman, A., Hoang, T., Laurenson, E., Nathan, R. (2000), Monte Carlo
Simulation of flood Frequency Curves, Proceedings of Hydro 2000, 3rd International
hydrology and water resources symposium, Perth, Inst. Of Engineers Australia, pp: 564-569.
Wood, J.F. and Alvarez, K. (1982), Review of Pindari Dam Design Inflow Flood Estimate,
Proceedings of the workshop on spillway design, Department of National Development and
Energy, AWRC Conference Series No.6.
102
Chapter 6. Spatial Patterns of Rainfall
Phillip Jordan, Alan Seed, Rory Nathan
6.1. Introduction
As discussed in Book 2, Chapter 2, the description of rainfall events used in most currently
applied design flood estimation methods is based on a reductionist approach, where the
temporal and spatial variations of rainfall within an event are represented separately by
typical temporal patterns and spatial patterns of event rainfall.
This chapter provides practitioners with recommendations on the derivation and application
of spatial patterns of rainfall for use in design flood estimation using representations of
varying complexity. This includes recommendations for reconstructing the space-time
patterns of rainfall for the observed events used in the calibration of hydrologic catchment
models.
There are a number of items where the authors recognise that the guidance adopted in this
chapter is uncertain and where benefit would be obtained from further research to better
quantify and potentially reduce the impact of those uncertainties on design flood estimation
practice. Section 6 lists and briefly discusses the residual uncertainties relating to various
aspects of the derivation and application of spatial and space-time patterns of rainfall, and
recommends potential areas of future investigation.
6.2. Methods for Deriving Spatial Patterns of Rainfall for

Events
6.2.1. Precipitation Observation Methods and Uncertainties
Associated with Reconstructing Space-Time Rainfall Patterns
There is no accepted method for determining the space-time pattern of rainfall that is not
influenced by the resolution and accuracy of networks of rainfall observations.
Rainfall gauges provide data on the rainfall depths observed at the point location of the
rainfall gauge over different periods of time. Daily reporting rainfall gauges provide rainfall
depths recorded at the gauge over the preceding day period, with the Bureau of
Meteorology’s typical practice being that these gauges report at 9:00 am local time on each
day. Pluviograph or tipping bucket rainfall gauges can provide rainfall depths observed at a
point location for sub-daily temporal resolution. They can provide rainfall depths with
temporal resolutions down to less than one minute.
Rainfall gauges are subject to some observational errors but they typically provide relatively
accurate measurements of the time series of rainfall recorded at a point location. They can
under-record rainfall during periods of high winds, particularly for snow or for when rainfall
intensities are low. There can be errors associated with estimating rainfall rates from tipping
bucket rainfall gauges over defined periods of time from the recorded times of the bucket
tips. During periods of very high rainfall intensity, tipping bucket rainfall gauges can be
subject to errors induced by the bucket failing to tip or tipping when it is partially full. Rainfall
103
Spatial Patterns of Rainfall
gauges can also be subject to errors in manual recording of the data or electronic
transmission of data from telemetered rainfall sites.
The chief uncertainty introduced by a network of rainfall gauges is in accurately observing

the space-time pattern of rainfall across an area because they cannot observe variations in
rainfall patterns between the gauges (Seed and Austin, 1990; Barnston, 1991; Bradley et al.,
1997).
Remote sensing approaches can provide estimates of rainfall intensity observed on a spatial
grid across a wide observation domain for a given period of time. The two most commonly
available remote sensing approaches for rainfall estimation are ground-based weather radar
and satellite observing systems.
Weather radars measure the reflectivity returned by rain drops, hail stones or snow, which
are converted into a rainfall intensity estimate. There are several different types of errors in
this process that degrade the accuracy of the radar rainfall measurement (Joss and
Waldvogel, 1990; Collier, 1996). The analysis of radar data to derive space-time patterns
requires specialist expertise that lies outside the scope of these guidelines. However, such
analysis could be considered for large or high risk studies which are able to secure the
specialist expertise required. Weather radar approaches typically provide estimates of
rainfall intensities that are more accurate on a relative basis within the space-time field of the
event than in absolute magnitude terms.
6.2.2. Data Availability

Rainfall gauge observations are available at many locations across Australia for very long
time periods, with data available at some sites since the middle of the 19th century (refer to
Book 2, Chapter 3). By contrast, reliable archives of data from remote sensing (weather
radar and satellite based instruments) are only available from about the mid 1990s. For
many events, the only data that will be available to the practitioner to reconstruct the space-
time pattern of rainfall will be from rainfall gauges.
The practitioner should make use of the available data to reconstruct the space-time pattern
of rainfall across the catchment or study area for the events that are to be utilised in model
calibration and design flood simulation. The practitioner should assess the suitability of the
rainfall data that is available for the event for reconstructing the space-time pattern, including
rainfall gauges and any data that is available from remote sensing.
The practitioner should consider the events that are to be used for constructing the space-
time patterns of rainfall. Factors that should be considered in selecting events are the:
• Number of sites and locations, relative to the catchment, of daily rainfall gauges;
• Number of sites and locations, relative to the catchment, of continuous rainfall gauges;
• Existence or otherwise of remotely sensed data;
• Likely accuracy of quantitative rainfall estimates derived from remotely sensed data;
• Purpose of estimating the space-time rainfall pattern, whether it is for hydrological model
calibration, deriving a space-time pattern or spatial pattern for inclusion in design flood
simulation or both;
• Existence and quality of recorded flood levels, flood extents and gauged flows for the
event, which make it a candidate for model calibration; and
104
• Estimated AEP of the flood event or the rainfall total for the event over the catchment of
interest, relative to the AEP of the design floods that are to be estimated.
The practitioner may need to make judgements between using the space-time pattern for an
older event, for which there is no remote sensing data and relatively poor coverage of rain
gauge data but which produced higher flood levels and a more recent event that has
remotely sensed data and/or better coverage of rain gauge data but which produced a
smaller flood.
6.2.3. Construction of Space-Time Patterns from Rainfall

Gauge Networks
When estimating the space-time pattern of rainfall from rainfall gauges there is an
uncertainty introduced to the estimates in the interpolation of the unobserved rainfall at
locations between the gauges (for example, Urbonas et al. (1992) and Ball and Luk (1998)).
The conventional approach applied in flood estimation for approximating the space-time
pattern of rainfall from gauge networks has been:
1. To estimate the spatial pattern of rainfall for the whole rainfall event; and
2. To disaggregate the rainfall accumulation for each part of the spatial domain, often a
model subarea or subcatchment, using the temporal pattern observed at a particular
rainfall gauge.
The conventional approach is a valid method in most situations but it may be that a more
sophisticated approach involving construction of different spatial patterns for different
increments of the event are required when spatial or temporal variability of the rainfall
pattern for the event is large (Umakhanthan and Ball, 2005). Considerations for application
of each of the two steps are discussed in Book 2, Chapter 6, Section 2 and Book 2, Chapter
6, Section 2. Potential alternative approaches are discussed in Book 2, Chapter 6, Section 2.
6.2.3.1. Construction of Spatial Patterns

The spatial pattern should be constructed using rainfall totals from daily rainfall gauges and
where available continuous rainfall gauges. Gauges should be obtained from both within the
catchment or study area and for a region around the catchment. As an indicative value, the
region used for constructing the spatial pattern should extend to include gauges that are
within at least 10 km of the catchment or study area boundary or further if internal catchment
gauges are further from the boundary.
There is no preferred technique for constructing a spatial pattern of rainfall for an event.
Hand drawing of rainfall contours informed by the rainfall totals at the gauges remains a valid
approach that will produce acceptable results for many rainfall events.
Spatial interpolation techniques using a computer usually involve interpolation between the
point observations onto a grid, defined in either a geographic or projected Cartesian
coordinate system. The grid resolution should be sufficiently fine to capture the spatial
variability in the rainfall field at a meaningful scale for the catchment. It is recommended that
the resolution of the grid should be 1 km (for a projected grid) or 0.01° (for a geographic grid)
or finer. There are many potential approaches that have been developed for spatial
interpolation (Verworn and Haberlandt (2011) and the references therein), including:
105
• Construction of Thiessen polygons, which is equivalent to adopting the rainfall depth from
the nearest neighbour rainfall gauge when applied using a grid;
• Weighting of rainfall using the inverse of the square of the distances to the gauges;
• Natural neighbours;
• Spline interpolation algorithms;
• Ordinary Kriging; and
• Variants on Kriging, such as indicator Kriging, regression Kriging and Kriging with external
drift.
Some Kriging and spline interpolation algorithms allow for the use of a covariate in the
interpolation algorithm, which may improve the accuracy of the interpolation. Either elevation
or design rainfall intensities for a relevant AEP and duration may provide appropriate
covariates that improve the accuracy of the interpolation, particularly in catchments or study
areas that are subject to appreciable and consistent orographic effects.
Gridded daily rainfall data sets are available from SILO and the Australian Water Availability
Project (Jones et al., 2009). These data sets may be useful for providing spatial patterns of
rainfall events but they should be used with caution as they were not derived with the
intention of being used for design flood estimation (refer to Book 1, Chapter 4, Section 9 and
Book 2, Chapter 7, Section 2).
Regardless of the approach that is used to produce the spatial pattern, the practitioner
should check the spatial pattern produced by mapping it against the point values observed
at rainfall gauges. If the mapping reveals anomalies in the interpolation approach, an
alternative method should be adopted. Where large gaps in rainfall gauge coverage exist
(particularly in mountainous areas) careful review of the simplifying assumptions made in the
interpolation procedure should be undertaken by the practitioner to avoid unrealistic spatial
patterns.
Rainfall totals for each model subarea should be estimated from the spatially interpolated
rainfall field for the event by averaging the rainfall totals at all grid cells that intersect with the
spatial extent of the model subarea. Mathematically, this is represented by:
∑�, � �� ∩ �, ��, �
�� = (2.6.1)
∑�, � �� ∩ �, �
where Ss is the rainfall depth for the total event applied to model subarea, s, �� ∩ �, �is the
area of overlap between grid cell at coordinate location (i,j) in the interpolated grid and the
model subarea s, and ��, �is the interpolated rainfall total for the grid cell.
For rainfall events that extend for a period longer than 24 hours, it may be useful to construct
spatial patterns for separate time periods of the event to investigate the temporal evolution of
the event.
106
6.2.3.2. Disaggregation Using Temporal Patterns from Observed

Pluviograph Data
A conventional approach adopted is to disaggregate the total rainfall for each subarea of a
rainfall runoff model using the temporal pattern recorded at a recording rain gauge, using the
formula:
��, �
��, � = (2.6.2)
∑� ��, �
Where Rs,i is the rainfall depth applied to model subarea, s, for model time increment, i; SS is
the total event rainfall for model subarea s; rg,i is the rainfall depth recorded at gauge, g , for
model time increment, i; and the summation is formed for all time increments over the event.
Where data is available from more than one recording rainfall gauge, this provides options to
the practitioner on which gauge to select to provide the temporal pattern for each subarea of
the model. An assumption is often made that the most appropriate pattern would be provided
by the recording rainfall gauge that is located closest, by horizontal distance, from the
centroid of the model subarea. Whilst the closest gauge by physical distance makes intuitive
sense, it is not necessarily the case that it must provide the most appropriate pattern for
allocating the temporal pattern of a particular subarea. The practitioner may consult other
information, such as catchment topography, data on wind velocities during the event or
remote sensing data to guide the selection of an alternative to the nearest gauge for
providing the temporal pattern. The practitioner may also use other information on the
meteorology of the event to justify use of an adjusted temporal pattern for disaggregation.
For example, if information was available that a particular rainfall event was moving in a
particular direction at an average velocity of 15 km/h and a model subarea was located 30
km downwind of a rainfall gauge, it may be justified to adjust the temporal pattern recorded
at this gauge by moving it backward in time by two hours, to represent the estimated travel
time of the storm from the rainfall gauge to the model subarea.
The time series of rainfall at each recording gauge should be checked before it is used for
disaggregation. The event total rainfall at each recording gauge should be checked, where
available, against the event total rainfall at other daily recording and continuously rainfall
gauges in the vicinity. A gauge should not be used if significant anomalies are identified in
the recorded data for the site.
6.2.3.3. Alternative Approaches to Construction of Space-Time

Patterns
A potential alternative approach to construct the space-time pattern of rainfall for an event
from rainfall gauge data only is to construct a three dimensional space-time pattern grid. In
this approach, the overall event spatial pattern would be interpolated onto a grid, using one
of the potential approaches discussed in Book 2, Chapter 6, Section 2. The total rainfall for
each grid cell would then be disaggregated using the temporal pattern from an assigned
rainfall gauge, using a similar method as discussed in Book 2, Chapter 6, Section 2. For
each time increment, a summation is formed for all of the grid cells that intersect spatially
with each model subarea. Each subarea would then have its own, potentially unique,
temporal pattern for the event.
In a catchment that is well instrumented with rainfall gauges, it is possible to perform a

spatial interpolation on to a grid for each time increment. The gridded rainfall for each time
107
increment would then be summed to produce a temporal pattern for each model subarea. If
this approach is used, the total rainfall across the event should be calculated for each
subarea and then compared to the rainfall computed from spatially interpolating the event
total rainfall only. Adjustments should be considered to the approach if the totals for any
subarea differ by more than 5%.
6.2.4. Space-Time Patterns for Calibration

Simulation of historical flood events for calibration purposes includes reconstruction of the
space-time pattern of rainfall over the catchment by the practitioner. The values of the
apparent optimum set of model parameters for a given event can be influenced by the
approach taken to estimate the space-time pattern of rainfall for the event. For example, if a
catchment is simulated using an initial loss-continuing loss runoff generation model, then if
all other parameters and inputs are the same a lower continuing loss rate is likely to be
required to generate the same volume of runoff if a uniform spatial pattern is adopted
compared with a non-uniform spatial pattern. The practitioner should consider and articulate
the influence of assumptions made in deriving the space-time pattern of rainfall for the event
on the values of the runoff-routing model parameters calibrated for the event.
The spatial coverage of pluviographs around a catchment may be relatively sparse. This
introduces uncertainty into the estimation of the actual temporal pattern of rainfall for any
given subarea or grid cell of a model. Whilst a reasonable assumption may be that the
temporal pattern for a given model subarea would be defined by the pluviograph that is
nearest in horizontal distance, this may not necessarily produce the most accurate temporal
pattern for the model subarea. As discussed in Book 2, Chapter 6, Section 2, a more
accurate representation of the space-time rainfall pattern over a model subarea may be
produced by judicious adjustment of the temporal pattern observed at a gauge or selection
of a temporal pattern from a gauge that is not physically closest to the subarea centroid.
Adjusting the assignment of temporal patterns to subareas in this manner may assist the
practitioner in achieving a more robust calibration of the model parameters to the event.
Adjustment of temporal patterns and temporal pattern assignment is therefore allowable,
particularly if meteorological evidence is provided to support the decision.
The spatial or space-time pattern should be interpolated on to a regular grid that is

constructed over the catchment. The resolution of the grid should be sufficiently fine to allow
for spatial variations to be adequately represented. In most situations, the grid resolution
should be selected so that there are at least 4 grid cells overlapping with the smallest
subcatchment to be adopted in the model.
Remote sensing data, where available, may be used to estimate the space-time rainfall field
of an event for catchment modelling system calibration. If used, the space-time rainfall field
should be corrected using data from rainfall gauges, using a recommended approach for
adjusting for the mean field bias.
6.3. Spatial and Space-Time Patterns for Design Flood

Estimation
The aim of a design flood estimation should be to provide a probability neutral transformation
between the design rainfall inputs and design flood characteristics
The space-time pattern or set of space-time patterns adopted for design flood estimation
should be chosen in a manner that, when coupled with other aspects of the catchment
108
modelling system, preserves the AEP of the design flood when derived from its causative
rainfall.
6.3.1. Guidance for Catchments up to and Including 20 km2:

Single Uniform Spatial Pattern
Catchments with areas up to and including 20 km2 are sufficiently small that there is little
available data to derive a spatial pattern. For these catchments, it is usually acceptable to
adopt a uniform spatial pattern.
If there is sufficient density of continuously rainfall gauges that have recorded a number of
rainfall events, using this data to derive alternative (non-uniform) design spatial patterns may
be considered.
6.3.2. Guidance for Catchments Greater than 20 km2: Single

Non-Uniform Spatial Pattern
As a minimum, it is recommended that a single non-uniform spatial pattern is applied to
catchments with an area greater than 20 km2. The non-uniform spatial pattern should be
derived with the aim of replicating the systematic variation in spatial variability that would be
expected across the catchment during rainfall events of similar AEP to the design floods that
are being estimated.
For estimation of design flood events more frequent than and including the 1% AEP event,
the spatial pattern should be estimated using the spatial pattern derived from the design
rainfall grids (as discussed in Book 2, Chapter 3) across the catchment for the relevant IFD
surface for the AEP and duration. In many cases there will be little relative variation in spatial
distribution between probabilities or adjacent duration. Different spatial patterns could be
applied for different durations. Alternatively, one spatial pattern may be estimated for the
critical duration and this single spatial pattern may then be applied for all durations.
For estimation of design flood events rarer than 1% AEP with durations of 6 hours and less
on catchment areas less than 1000 km2, the spatial pattern should be derived in accordance
with Woolhiser (1992) for the relevant duration. Use of different spatial patterns for different
AEP ranges may introduce inconsistencies at the adjacent limits of each method, and if this
is the case then any such inconsistencies should be smoothed in an appropriate fashion.
For estimation of design flood events rarer than 1% AEP with durations of 9 hours and
greater or on catchment areas greater than 1000 km2, the spatial pattern should be derived
from the Topographic AdjustmentFactor (TAF) database derived from the generalised PMP
method that is relevant for zone that the catchment is located in. Use of different spatial
patterns for different AEP ranges may introduce inconsistencies at the adjacent limits of
each method, and if this is the case then any such inconsistencies should be smoothed in an
appropriate fashion.
For large studies and particularly for large catchments the practitioner should investigate and
analyse the variability in spatial patterns between events. Where topography is dominant or
large events are generally produced by a single rainfall mechanism there is likely to be only
moderate variability between events but for some catchments there can be significant
variations in space-time patterns between events. The practitioner should prepare and
examine maps of the spatial pattern of rainfall for each event as a whole and for time slices,
for example each 24 hour period, using an approach described in Book 2, Chapter 6,
Section 2. These spatial patterns should be compared to rainfall accumulations from
109
Intensity Frequency Duration analysis for a relevant duration and AEP (refer to Jordan et al.,
2015 for an example of this approach). Consistency in spatial patterns between events may
reveal that it is acceptable to apply a single spatial pattern for all design flood estimates,
particularly if it is consistent with the design rainfall analysis.
As discussed in Book 2, Chapter 4, Section 3, partial area storms should always be explicitly
considered for catchments with an area exceeding 30 000 km2 and it should be considered
for catchments larger than 5000 km2.
6.3.3. Alternative Approach: Monte Carlo Sampling from

Separate Populations of Spatial and Temporal Patterns
A more advanced approach that may be justified would be Monte Carlo simulation by
sampling from a set of space-time rainfall patterns across the catchment of interest. There
are two potential options that may be considered for implementing this approach: (1)
sampling from separate populations of spatial and temporal patterns for the catchment; or
(2) sampling from a single set of “linked” space-time patterns for the catchment.
The first approach requires the assembly of:
• A set (or population) of spatial patterns across the catchment of interest from a number of
observed rainfall events; and
• A set of temporal patterns from a number of observed rainfall events.
The spatial pattern of rainfall should be assembled for each event in the population, in
accordance with the methods discussed in Book 2, Chapter 6, Section 2. The catchment
average rainfall accumulation should be computed for each event and continuously rainfall
gauges located in or near the catchment should be used to estimate the duration over which
most of the total rainfall accumulation was likely to have fallen in the catchment. The
estimated catchment average depth for the event and the estimated rainfall event duration
should be used with the table of design rainfall estimates for the catchment, after application
of the applicable ARF, to estimate the AEP of the rainfall for each event.
Similarly, the temporal pattern of rainfall should be assembled for each event in the
population. The temporal pattern may be assembled at a single continuously rainfall gauge
or from a combination of a number of continuously rainfall gauges located in the vicinity of
the catchment or study area. The temporal pattern should be analysed to extract the
maximum burst for a number of different durations. The rainfall accumulations over these
bursts should be compared to the design rainfall estimates at the location of the rainfall
gauge, without the application of the ARF, to estimate the AEP of the rainfall for each event.
A sample of patterns for use in the Monte Carlo simulation should be selected from the set of
historical events that are available. A sufficient number of events should be selected to allow
for a meaningfully large sample in the Monte Carlo simulation. It is expected that a minimum
of five patterns would be required each of the sets of spatial and temporal patterns.
However, events should only be selected for inclusion in the sample if they are relatively
similar in terms of the AEP of the rainfall to the range of AEP that design flood estimates are
being produced. Ideally, the spatial patterns and temporal patterns of events selected for the
Monte Carlo sample should have an estimated AEP that is between 1/10 and ten times the
AEP of the design flood event to be simulated. Adding more historic spatial patterns to an
ensemble does not necessarily improve the simulation accuracy of a Monte Carlo model, as
the additional patterns that are most likely to be added would be at the more common end of
the AEP range. In many catchments, design floods of interest are caused by rainfall events
110
with a specific hydrometeorological mechanism, which is then associated with a range of

space-time rainfall patterns that are different to those observed in rainfall events caused by
more commonly occurring hydrometeorological conditions.
Unless there is hydrometeorological evidence to the contrary, all potential spatial and
temporal patterns in the sets available for sampling should be given equal probability of
selection in the Monte Carlo simulation.
After the spatial and temporal patterns for the design rainfall burst have been selected
stochastically, the patterns should be scaled so that the catchment average rainfall depth for
the design rainfall burst matches the depth generated stochastically by the sampling
scheme.
There has been limited assessment on methods for selection of space-time patterns for use
in Monte Carlo simulation for design flood estimation. Further research should be conducted
in this area to provide more robust guidance on the minimum number of temporal and spatial
patterns in the sampling populations, the range of AEP represented by the populations of
spatial and temporal patterns to be sampled compared to the AEP of the depth of the rainfall
burst and the relative probabilities to be applied in the selection of spatial and temporal
patterns.
6.3.4. Alternative Approach: Monte Carlo Sampling from Single

Population of Space-Time Patterns
The foregoing approach ignores the potential dependency that exists between the temporal
and spatial characteristics of storms. According, an alternative approach would involve
sampling the space-time pattern for the event from a single population of space-time rainfall
patterns over the catchment. The sample of space-time patterns may be assembled from
space-time patterns of rainfall observed during historical rainfall events in the catchment.
The space-time patterns should be assembled in accordance with the methods discussed in
Book 2, Chapter 6, Section 2. The estimated catchment average depth for different burst
durations within the event and the estimated rainfall event duration should be used with the
table of design rainfall estimates for the catchment, after application of the applicable ARF, to
estimate the AEP of the rainfall for each event.
It may be an option to transpose space-time rainfall patterns from an area with a good
observational network for rainfall to a catchment with a poorer observational network. If this
is done, the practitioner should only transpose (non-dimensional) space-time rainfall patterns
from an area that is subject to rainfall events that are driven by similar hydrometeorological
processes. The transposition region should be subject to similar orographic influences. In
some cases, the space-time patterns may need to be rotated to maintain consistency
between the spatial gradients in the space-time patterns and orographically influenced
gradients in the design rainfall gridded data.
6.3.5. Spatial Patterns for Pre-Burst and Post-Burst Rainfall

If pre-burst or post-burst rainfall is to be applied, it is recommended that, unless there is
evidence to the contrary, the spatial pattern applied to the pre-burst or post-burst rainfall
should be the same as the spatial pattern applied for the design rainfall burst.
6.3.6. Spatial Patterns for Continuous Rainfall Series

Book 2, Chapter 7 discusses the production of continuous rainfall time series for production
of design flood estimation using continuous simulation approaches. Book 2, Chapter 7
111
includes an approach to post-process the sequence of generated rainfall data for the
catchment of interest so that the characteristics of the large and extreme rainfall events in
the sequence reflect the Intensity Frequency Duration (IFD) statistics for the catchment of
interest. If such an adjustment is conducted then it is recommended that the IFD statistics
used as the basis of the adjustment are calculated for the catchment of interest after
multiplying by the ARF that is applicable for the catchment area.
Book 2, Chapter 7 recommends that the IFD adjustment is applied for a set of target
durations of either 6 minutes, 1 hour and 3 hours or for a set of durations of 6 minutes, 30
minutes, 1 hour, 3 hours, 6 hours and 12 hours. It is recommended therefore that the
following procedure is adopted:
1. The IFD statistics for each of the target durations are calculated as the average of the
point IFD statistics across the catchment, for each of the standard AEP (1EY to 1% AEP);
2. The ARF are computed for each of the target durations, at each of the standard AEP, for
the total area of the catchment to be modelled;
3. The catchment IFD statistics are computed for each of the target durations, at each of the
standard AEP, as the product of the point IFD statistics (from step 1) and the ARF (from
step 2);
4. The catchment IFD statistics (from step 3) are applied in the modification procedure as
the depths at the selected target durations.
The practitioner may be adopting a catchment model that allows for spatial distribution of the
simulated rainfall sequence across the catchment. If this is the case, it is recommended that
the generated sequence of rainfall is scaled for each portion of the model (subcatchment or
grid cell as applicable to the particular model) to reflect the spatial distribution of rainfall that
would be typically observed across the catchment. Parts of the catchment that are typically
wetter would have rainfall depths applied in the model that are larger than the generated
mean rainfall depth across the catchment but with the same timing and sequencing.
Conversely, parts of the catchment that are typically drier would have rainfall depths applied
in the model that are smaller than the generated mean rainfall depth across the catchment
but with the same timing and sequencing.
Selection of an appropriate means of deriving the spatial pattern for a continuous simulation
model that includes spatial distribution depends upon the AEP of the design events that are
of most interest and the flood response characteristics of the catchment:
• When the focus is on estimation of floods with relatively frequent AEP (around 10% or
more common) and for catchments with large moisture stores having significant relation
between antecedent rainfall and the annual maximum flood, it is recommended that the
spatial pattern applied in the model should be estimated from contours of mean annual
rainfall;
• However when the focus is on estimation on floods with rarer AEP (5% or rarer) and for
catchments where the influence of large moisture stores are less significant, it is
recommended that the spatial pattern should be selected in a manner that is consistent
with the recommendation for design event simulation (refer to Book 2, Chapter 6, Section
3 and Book 2, Chapter 6, Section 3 above).
112
6.4. Potential Influences of Climate Change on Areal

Reduction Factors, Spatial and Space-Time Patterns
There is very little credible guidance on how climate change is projected to influence ARF,
spatial patterns or space-time patterns of rainfall events used in design flood estimation.
Abbs and Rafter (2009) used dynamic downscaling using a regional climate model to identify
that increases in rainfall intensity are likely to be greater in those areas of south-east
Queensland that are subject to orographic enhancement than those areas that are not.
There is insufficient evidence to confirm whether this projection is an artefact of the
downscaling approach and whether it would still apply if the dynamic downscaling model
were forced with Global Circulation Model results from the more recent Intergovernmental
Panel on Climate Change (2013). Even if it were proven for Southeast Queensland, it is not
clear that the guidance would be more generally applicable to other parts of Australia.
It is recommended that until credible further studies are completed showing otherwise, for
simulations applying projected climate change ARF, spatial patterns and space-time patterns
of design rainfall should be the same as derived under existing climatic conditions.
6.5. Worked Examples

All of the worked examples in this chapter use data for the Stanley River catchment, which is
in the upper part of the Brisbane River basin in Southeast Queensland. Worked example 1
demonstrates three different mathematical algorithms for estimating the spatial pattern of
design rainfall for a particular rainfall event that occurred in January 2013. Worked examples
2 through 4 demonstrate the process for design flood estimation, from calculation of ARF
and catchment average design rainfall estimates (Worked example 2), through calculation of
a representative spatial pattern (Worked example 3), to production of design flood estimates
using a runoff-routing model (Worked example 4).
6.5.1. Catchment Used for Worked Examples

The Stanley River catchment drains into Somerset Dam, which is in the upper part of the
Brisbane River basin in South-east Queensland. Figure 2.6.1 shows a map of the 1324 km2
shows the catchment area, with Somerset Dam located in the southwestern corner. For this
worked example, the Stanley River catchment is modelled using a runoff-routing model with
76 subcatchments, with subareas as shown in Figure 2.6.1.
Design rainfall estimates are developed for the Stanley River at Woodford, which has a
catchment area of 245 km2. The catchment to Woodford is the north-eastern portion of the
catchment to Somerset Dam and for this worked example it includes fifteen subcatchments
in the runoff-routing model.
A significant feature of the Stanley River catchment is the appreciable gradient in rainfall that
is typically observed during large rainfall events. Tropical cyclones, ex-Tropical Cyclones,
East Coast Lows and other rainfall producing systems typically feed moisture into the
catchment from the Pacific Ocean. Since the north-eastern part of the catchment is only 20
km from the coast but the western side of the catchment is almost 70 km from the coast, the
typical direction of storm movement and typical direction of flow of warm moist air from the
ocean results in a gradient of rainfall totals that reduce from east to west across the
catchment in most rainfall events. The strength of the rainfall gradient is enhanced by
orographic effects with the highest totals typically also occurring in the north-eastern part of
the catchment.
113
Figure 2.6.1. Stanley River Catchment, Showing Runoff-routing Model Subcatchments and
the Locations of Daily rainfall and pluviograph gauges
6.5.2. Worked Example 1: Interpolation of Spatial Patterns for

an Event Using Various Methods
Tropical Cyclone Oswald generated heavy rainfall in the Stanley River catchment between
23 and 29 January 2013, generating flooding in the catchment. Rainfall totals were observed
at 20 continuous rainfall gauges around the Stanley River catchment, as shown in
Figure 2.6.2 (rainfall data supplied by SeqWater). The blue circles are scaled in proportion to
the rainfall depth recorded at the gauge.
114
Figure 2.6.2. Rainfall totals (mm) Recorded at Rainfall Gauges for the January 2013 Event in
the Vicinity of the Stanley River Catchment
The January 2013 rainfall event was used in this worked example to demonstrate various
approaches to interpolation of spatial patterns of historical rainfall events, for the purpose of
calibration of runoff-routing models. For all of the algorithms, the rainfall totals were first
interpolated onto a 0.5 km resolution grid over the catchment. Rainfall totals for each of the
76 runoff-routing model subcatchments were then computed from the average of the rainfall
totals at the grid cells that overlapped each subcatchment.
Rainfall totals were spatially interpolated using:
1. Thiessen polygons, as shown in Figure 2.6.3. Observed totals at gauges are shown as
blue circles in both panels. The top panel shows interpolation to a 0.5 km grid (red
shading), whilst the bottom panel shows calculated subcatchment average depths in mm
(red circles);
2. Inverse distance weighting, as shown in Figure 2.6.4. Observed totals at gauges are
shown as blue circles in both panels. The top panel shows interpolation to a 0.5 km grid
(red shading), whilst the bottom panel shows calculated subcatchment average depths in
mm (red circles); and
3. Ordinary Kriging, as shown in Figure 2.6.5. Ordinary Kriging was applied using a linear
semi-variogram that was fitted to observed rainfall totals at the 20 gauges from the
January 2013 event, as shown in Figure 2.6.6.Observed totals at gauges are shown as
blue circles in both panels. The top panel shows interpolation to a 0.5 km grid (red
shading), whilst the bottom panel shows calculated subcatchment average depths in mm
(red circles).
115
Figure 2.6.3. Application of Thiessen Polygons- Rainfall Totals for the January 2013 Event -
Stanley River Catchment
116
Figure 2.6.4. Application of Inverse Distance Weighting - Rainfall Totals for the January 2013
Event - Stanley River Catchment
117
Figure 2.6.5. Application of Ordinary Kriging - Rainfall Totals for the January 2013 Event -
Stanley River Catchment
118
Figure 2.6.6. Observed Semi-variogram and Fitted Linear Semi-variogram for the January
2013 Rainfall Event for Stanley River catchment, Applied in the Ordinary Kriging Algorithm
6.5.3. Worked Example 2: Calculation of Catchment Average

Design Rainfall Depths and Areal Reduction Factors
Design rainfall intensities were extracted at the centroids of each of the fifteen runoff-routing
model subcatchments in the catchment of the Stanley River to Woodford from for the 1%
AEP and 24 hour duration. The weighted average of the point rainfall depths was computed,
as shown in Table 2.6.1.
The catchment area for the Stanley River to Woodford is 245.07 km2. The ARF for 24 hour
duration was computed by applying Equation (2.4.2), with the relevant coefficients for the
East Coast North region (from the 1st row of Table 2.4.2). For the 1% AEP event the relevant
ARF is given by:
��

= �� 1, [1 − �(�� − �log10��)��−�
+��(0.3 + log10��) (2.6.3)
��
��
1440
+ℎ10 (0.3 + log10��)]
119
��

= �� 1, [1 − 0.327(245.070.241 − 0.448log10(1440))1440−0.36
+0.00096 x 245.070.481440−0.21(0.3 + log10(0.01)) (2.6.4)
1440
−0.001 x 245.07 x
1440
+0.012 x 10 (0.3 + log100.01)]
�� (24ℎ��) = 0.929 (2.6.5)
Table 2.6.1. Calculation of Weighted Average of Point Rainfall Depths for the 1% AEP 24
hour Design Rainfall Event for the Stanley River at Woodford
Centroid Centroid Area (km2) 1% AEP, 24 hour Design Depth x

Latitude (°) Longitude (°) Design Point Area (ML)
Rainfall Depth
at Centroid
(mm)
-26.8467 152.8510 5.66 511.4 2896.0
-26.8060 152.8320 17.31 518.5 8973.2
-26.7895 152.8776 16.84 570.4 9605.7
-26.8201 152.8719 16.55 530.9 8787.9
-26.8631 152.8756 16.25 517.7 8414.3
-26.8110 152.8000 15.10 527.0 7956.5
-26.8135 152.9077 15.01 551.3 8276.7
-26.8998 152.7978 23.45 467.1 10954.8
-26.8828 152.8459 22.74 493.3 11218.6
-26.9171 152.7620 17.78 454.2 8076.1
-26.8557 152.7656 16.34 490.0 8007.1
-26.9190 152.8552 16.24 502.6 8164.3
-26.8568 152.8185 15.62 484.6 7571.4
-26.9354 152.8096 15.12 465.2 7031.9
-26.8445 152.7889 15.04 508.9 7654.0
Totals 245.07 123588.6
Weighted Average = 123588.6 / 245.07 504.3
The catchment average design rainfall depth for 1% AEP, 24 hour duration for the Stanley
River at Woodford was therefore computed by multiplying the ARF by the weighted average
of the design point rainfall depths:
Catchment ave. design rainfall depth (24 ℎ��, 1%��)

(2.6.6)
= ARF (24 ℎ�) × Weighted ave. of design point rainfall depths (24 ℎ�, 1%��)
Catchment average design rainfall depth(24ℎ��, 1%��)=0.929 × 504.3 mm (2.6.7)
Catchment average design rainfall depth(24ℎ��, 1%��)=468.6 mm (2.6.8)
120
The calculation was repeated for the catchment of the Stanley River to Woodford for each
combination of standard durations between 3 and 72 hours and the 1 Exceedance per Year
to the 1% AEP. These computations are shown for the Stanley River catchment to Woodford
in Table 2.6.2.
The catchment area for the Stanley River to Somerset Dam is 1324 km2. The calculation of
catchment average design rainfall intensities, after application of areal reduction factors, is
shown in Table 2.6.3. Comparing the top panels of Table 2.6.2 and Table 2.6.3, for the
corresponding AEP and durations the weighted averages of the point rainfall depths for the
catchment to Somerset Dam are less than those for Woodford, due to the gradient in the IFD
grids. Comparing the middle panels of Table 2.6.2 and Table 2.6.3, for the corresponding
AEP and durations the ARF catchment to Somerset Dam are less than those for Woodford
because the catchment area to Somerset Dam is larger. Hence comparing the bottom
panels of Table 2.6.2 and Table 2.6.3, for the corresponding AEP and durations the
catchment average design rainfall depths to Somerset Dam are less than those for
Woodford.
Table 2.6.2. Stanley River Catchment to Woodford: Calculation of Catchment Average

Design Rainfall Depths (bottom panel) from Weighted Average of Point Rainfall Depths (top
panel) and Areal Reduction Factors (middle panel)
Weighted Average of Point Rainfall Depths (mm)

Duration 1 Annual Exceedance Probability
(hours) Exceeda
nce per
Year
50% 20% 10% 5% 2% 1%
3 54.1 61.6 85.7 102.8 120.0 143.5 162.1
6 70.5 81.5 117.3 142.8 168.7 204.6 233.3
12 94.4 110.8 164.9 203.8 243.8 299.5 344.5
24 128.1 152.0 231.4 289.5 349.8 434.8 504.3
48 170.2 202.2 310.5 391.2 476.1 598.0 699.3
72 195.5 231.7 355.3 448.5 547.6 691.4 812.1
Areal Reduction Factor

(hours) Exceeda
nce per
Year
50% 20% 10% 5% 2% 1%
3 0.841 0.835 0.812 0.794 0.776 0.753 0.735
6 0.879 0.876 0.864 0.854 0.845 0.832 0.823
12 0.909 0.907 0.901 0.896 0.891 0.884 0.879
24 0.945 0.944 0.940 0.938 0.935 0.932 0.929
48 0.959 0.959 0.957 0.955 0.954 0.951 0.950
72 0.966 0.966 0.964 0.963 0.962 0.961 0.959
Catchment Average Design Rainfall Depth (mm)
121

(hours) Exceeda
nce per
Year
50% 20% 10% 5% 2% 1%
3 45.5 51.4 69.6 81.6 93.1 108.0 119.2
6 62.0 71.4 101.3 122.0 142.5 170.2 191.9
12 85.8 100.5 148.5 182.6 217.1 264.8 302.9
24 121.1 143.4 217.6 271.5 327.1 405.2 468.6
48 163.3 193.9 297.1 373.6 454.0 569.0 664.2
72 188.8 223.7 342.6 432.0 526.8 664.2 779.2
Table 2.6.3. Stanley River Catchment to Somerset Dam: Calculation of Catchment Average
Design Rainfall Depths (bottom panel) from Weighted Average of Point Rainfall Depths (top
panel) and Areal Reduction Factors (middle panel)
Weighted Average of Point Rainfall Depths (mm)
(hours) Exceeda
nce per
Year
50% 20% 10% 5% 2% 1%
3 48.3 54.8 75.7 90.2 104.7 124.2 139.5
6 61.0 70.0 99.2 119.7 140.3 168.5 190.9
12 79.0 91.9 133.9 163.9 194.4 236.6 270.4
24 103.8 121.9 182.2 226.1 271.4 335.1 387.0
48 134.5 158.7 240.6 301.7 366.0 458.6 535.5
72 153.0 180.4 274.2 345.2 420.9 531.1 624.0
Areal Reduction Factor

(hours) Exceeda
nce per
Year
50% 20% 10% 5% 2% 1%
3 0.735 0.727 0.694 0.669 0.644 0.611 0.586
6 0.796 0.792 0.774 0.761 0.748 0.731 0.718
12 0.843 0.841 0.832 0.826 0.826 0.811 0.804
24 0.900 0.899 0.896 0.894 0.892 0.889 0.887
48 0.924 0.924 0.921 0.920 0.918 0.916 0.914
72 0.936 0.935 0.933 0.932 0.930 0.928 0.926
Catchment Average Design Rainfall Depth (mm)
(hours) Exceeda
122
nce per
Year
50% 20% 10% 5% 2% 1%
3 35.5 39.9 52.5 60.4 67.4 75.9 81.8
6 48.6 55.4 76.8 91.1 105.0 123.2 137.0
12 66.6 77.3 111.4 135.3 159.3 191.8 217.4
24 93.3 109.6 163.3 202.2 242.1 298.1 343.4
48 124.3 146.6 221.7 277.5 336.0 420.0 489.5
72 143.2 168.7 255.9 321.6 391.5 492.9 578.1
6.5.4. Worked Example 3: Calculation of Spatial Pattern for

Design Flood Estimation
Book 2, Chapter 6, Section 3 recommends that estimation of design flood events of 1% AEP
and more frequent, the spatial pattern for design event should be estimated using the spatial
pattern derived from the design rainfall grids across the catchment for the 1% AEP and for a
duration that is anticipated to correspond to the duration of the rainfall burst that is likely to
be critical at the specified location.
For the Stanley River catchment, this approach was demonstrated using the 24 hour
duration IFD data. For the catchment to Woodford, point rainfall depths at each of the
subcatchment centroids were divided by the weighted average of the point rainfall depths to
derive the non-dimensional spatial pattern, as computed in Table 2.6.4 and mapped in the
top panel of Figure 2.6.7. To model the 1% AEP 24 hour design flood event for the
catchment, the non-dimensional spatial pattern was multiplied by the catchment average
design rainfall depth to Woodford for this duration (468.6 mm, after application of the ARF),
as computed in Table 2.6.4 and mapped in the top panel of Figure 2.6.8.
The process was repeated for the Stanley River to Somerset Dam, with the map of the non-
dimensional spatial pattern shown in the bottom panel of Figure 2.6.7 and the map of the
design depths for the 1% AEP, 24 hour duration event in the bottom panel of Figure 2.6.8.
Table 2.6.4. Calculation of Design Spatial Pattern for Stanley River at Woodford
Centroid Centroid 1% AEP, 24 hour Point Rainfall Depth to be

Latitude (°) Longitude (°) Design Point Depth Divided Applied to
Rainfall Depth by Weighted Model 1% AEP,
at Centroid Average of 24 hour Design
(mm) Point Rainfall Event (mm)
Depths (%)
-26.8467 152.8510 511.4 101.4 475.2
-26.8060 152.8320 518.5 102.8 481.8
-26.7895 152.8776 570.4 113.1 530.0
-26.8201 152.8719 530.9 105.3 493.3
-26.8631 152.8756 517.7 102.7 481.0
-26.8110 152.8000 527.0 104.5 489.7
-26.8135 152.9077 551.3 109.3 512.3
123
Centroid Centroid 1% AEP, 24 hour Point Rainfall Depth to be

Latitude (°) Longitude (°) Design Point Depth Divided Applied to
Rainfall Depth by Weighted Model 1% AEP,
at Centroid Average of 24 hour Design
(mm) Point Rainfall Event (mm)
Depths (%)
-26.8998 152.7978 467.1 92.6 434.0
-26.8828 152.8459 493.3 97.8 458.4
-26.9171 152.7620 454.2 90.1 422.0
-26.8557 152.7656 490.0 97.2 455.3
-26.9190 152.8552 502.6 99.7 467.0
-26.8568 152.8185 484.6 96.1 450.3
-26.9354 152.8096 465.2 92.2 432.3
-26.8445 152.7889 508.9 100.9 472.9
Weighted Average 504.3 100.0 468.6
124
Figure 2.6.7. Non-dimensional Spatial Pattern (percentage of catchment average design

rainfall depths) for Events with AEP of 1% and more Frequent for Stanley River to Woodford
(top panel) and Stanley River to Somerset Dam (bottom panel)
125
Figure 2.6.8. Design Spatial Pattern of Design Rainfall Depths 1% AEP 24 hour Event for
Stanley River to Woodford (top panel) and Stanley River to Somerset Dam (bottom panel)
126
6.5.5. Worked Example 4: Application to Design Flood

Estimation
Design flood peak estimates were produced for the Stanley River at its outlet (inflow to
Somerset Dam) using a RORB runoff-routing model of the catchment. A more complete
description of this case study is contained in Jordan et al. (2015).
Design peak flow estimates at Somerset Dam inflow were produced from a number of Monte
Carlo simulations that were implemented within RORB. There were a number of common
elements to all of these simulations:
• all adopted the same catchment average design IFD information multiplied by the areal
reduction factor for the applicable duration from Jordan et al. (2013);
• all were run using the stratified Monte Carlo sampling scheme that is implemented within
RORB (Laurenson et al., 2010);
• all were run for rainfall burst durations of 6, 9, 12, 18, 24, 36 and 48 hours, with the peak
flow defined by the highest flow from among these durations at each AEP;
• all simulations sampled from the same non-dimensional probability distribution of initial
loss values defined by Ilahee (2005), scaled by a median initial loss of 40 mm;
• all adopted a constant continuing loss rate of 1.7 mm/hour across all subcatchments;
• all adopted a RORB non-linearity parameter, m, value of 0.8; and
• all simulations adopted RORB delay parameter, kc , values of 20 for the catchment
upstream of Peachester, 20 for the catchment between Peachester and Woodford, 16 for
the catchment upstream of Mount Kilcoy and 45 for the residual catchment to Somerset
Dam inflow.
The Monte Carlo simulations differed from one another in their approach to sampling of
spatial, temporal and space-time patterns across the catchment, as shown in Table 2.6.5.
Table 2.6.5. RORB Model Scenarios Run for Worked Example on Stanley River Catchment
to Somerset Dam
Case Spatial Pattern(s) Temporal Pattern(s)
1 Single spatial pattern derived Random sampling from a set
from IFD analysis, 1% AEP of 13 temporal patterns for
24 hour spatial pattern each duration, derived from
the bursts of the
corresponding duration within
the 13 selected events listed
in Table 1 of Jordan et al.
(2015)
2 Random sampling from a set of 13 space-time patterns for
each duration, derived from the bursts of the corresponding
duration within the 13 selected events listed in Table 1 of
Jordan et al. (2015)
Sinclair Knight Merz (2013) fitted a Generalised Extreme Value (GEV) distribution to the
estimated annual maxima inflows to Somerset Dam over the period between 1955 and 2013.
127
The estimated inflow flood peak for the 1893 flood of 6200 m³/s was included as a censored
flow in the analysis. The distribution fitted to the estimated observed inflows was used to test
the performance of the RORB model simulations.
Figure 2.6.9 shows that both cases of RORB model simulations all provide an excellent
match to the fitted flood frequency quantiles across the range between 5% and 0.2% AEP.
Design peak inflow floods to Somerset Dam were insensitive to whether space-time patterns
are randomly sampled or only temporal patterns are randomly sampled in the Monte Carlo
simulation (case 1 versus case 2).
Figure 2.6.9. Flood Frequency Curves for Stanley River at Somerset Dam Inflow Derived
from Analysis of Estimated Annual Maxima and from RORB Model Simulations
6.6. Recommended Further Research

6.6.1. Deriving Spatial and Space-Time Patterns of Rainfall for
Events
The capacity to collect and archive remotely sensed rainfall estimates and to provide that
information to practitioners is growing. It is recommended that the Bureau of Meteorology
continues to invest in routinely archiving remotely sensed rainfall data, particularly from its
network of ground based weather radars. It is recommended that the Bureau of Meteorology
continues to expand the provision of quality controlled and bias corrected space-time rainfall
estimates to practitioners, for use across the industry. It is recommended that tools should
be further developed and disseminated to practitioners to facilitate the use of remotely
sensed rainfall data.
128
It is recommended that further research is conducted into quality control of remotely sensed
estimates of the space-time pattern of rainfall.
It is recommended that further research is conducted to improve methods for mean field bias
correction of remotely sensed rainfall data. The recommendations on the approaches that
should be adopted for mean field bias correction should be updated in these guidelines in
accordance with the findings from this research.
At the time of writing, there was not an agreed optimum method for deriving space-time
rainfall patterns from rainfall gauge data for Australian catchments, although Verworn and
Haberlandt (2011) provide reasonable guidance. It is recommended that further research is
conducted to identify a superior method (or set of potential methods) that are demonstrated
to reliably produce more accurate estimation of the space-time rainfall field from gauge
observations. It may be that the optimum method depends upon meteorological
characteristics of the storm, density of rainfall gauges, orographic characteristics of the
region or other factors. It is recommended that further research is conducted to explore
these influences on the selection of optimum spatial and space-time interpolation methods
for flood model calibration and design flood estimation.
6.6.2. Space-Time Patterns for Calibration of Rainfall Runoff

Models to Historical Floods
As discussed in Book 2, Chapter 6, Section 2, the calibrated parameter values for a rainfall
runoff model for a particular flood event may be sensitive to the method used to derive the
space-time rainfall field for the event, particularly where the field is interpolated from a
network of rain gauges only. It is recommended that further research is conducted into the
sensitivity of rainfall-runoff routing model parameter estimation to assumptions made in the
process of estimating the space-time rainfall field gauged rainfall data.
6.6.3. Spatial and Space-Time Patterns for Design Flood

Estimation
It is recommended that further research is conducted into hydrometeorological drivers for
space-time rainfall patterns that lead to flood events across different regions of Australia.
The research should be used to inform practitioners on how they may choose between the
space-time patterns of rainfall from different historical rainfall events to form the populations
of space-time, spatial and temporal patterns in design flood simulation schemes. Research
may investigate seasonal influences on space-time patterns of rainfall for use in design flood
estimation.
There has been limited assessment on methods for selection of space-time patterns for use
in Monte Carlo simulation schemes for design flood estimation. Further research should be
conducted in this area, to provide more robust guidance on:
• The minimum number of space-time or temporal and spatial patterns in the sampling
population(s);
• The range of AEP represented by the populations of space-time or spatial and temporal
patterns to be sampled compared to the AEP of the depth of the rainfall burst; and
• The relative probabilities to be applied in the selection of patterns from the relevant
populations.
129
It is recommended that further research is conducted into the validity of transposing space-
time patterns from one location to another. The research should assist in defining valid
regions over which transposition of space-time patterns is acceptable and conversely
boundaries between regions over which transposition should not occur. The research should
also consider other aspects of transposition, such as the validity or otherwise of rotating
space-time patterns and the maximum recommended angles for rotation.
Further research should be conducted into methods for stochastic generation of space-time
rainfall patterns. The research should investigate how orographic influences should be
incorporated into the stochastic generation algorithms in a way that replicates the space-time
variability of rainfall observed in historic rainfall events. Research should also develop more
definitive guidance on appropriate statistical tests to demonstrate that the stochastically
generated space-time rainfall patterns replicate the space-time statistical characteristics of
historical rainfall events that are sufficiently large to have caused flood events.
6.6.4. Potential Influences of Climate Variability and Climate

Change
Climatic variability at inter-decadal scales is likely to influence the relative occurrence and
severity of different types of heavy rainfall events. Hydrometeorological understanding of the
connection between storm types and ARFs may enable predictions of the future trend in
ARFs that will occur as the climate changes over coming decades. Hydrometeorological
understanding of the connection between storm types and space-time rainfall patterns may
also allow for more accurate guidance to practitioners on the potential changes in space-
time patterns that is predicted as a result of climate change.
6.7. References
downscaling, CSIRO.
Ball, J.E. and Luk, K.C. (1998), Modelling spatial variability of rainfall over a catchment, J.
Hydrologic Engineering, 3(2), 122-130.
Barnston, A.G., (1991), An empirical method of estimating raingauge and radar rainfall
measurement bias and resolution, J. Applied Meterology, 30: 282-296.
Bradley, S.G., Gray, W.R., Pigott, L.D., Seed, A.W., Stow, C.D. and Austin, G.L. (1997),
Rainfall redistribution over low hills due to flow perturbation, J. Hydrology, 202, 33-47.
Collier, C.G. (1996), Applications of Weather Radar Systems: A Guide to Uses of Radar in
Meteorology and Hydrology. 2d ed. John Wiley, p: 383
Ilahee, M. (2005), Modelling Losses in Flood Estimation, A thesis submitted to the School of
Urban Development Queensland University of Technology in partial fulfilment of the
requirements for the Degree of Doctor of Philosophy, March 2005.
Intergovernmental Panel on Climate Change, (2013), Climate Change 2013: The Physical
Science Basis, Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change [Stocker,T.F., D. Qin, G.-K. Plattner, M. Tignor,
S.K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley (eds.)]. Cambridge
University Press, Cambridge, United Kingdom and New York, NY, USA, p: 1535.
130
Jones, D., Wang, W. and Fawcett, R. (2009), High-quality spatial climate data-sets for
Australia, Australian Meteorological and Oceanographic Journal, 58: 233-248.
Jordan, P., Nathan, R. and Seed, A. (2015), Application of spatial and space-time patterns of
design rainfall to design flood estimation. Engineers Australia, Hydrology and Water
Resources Symposium 2015.
Jordan, P., Weinmann, P.E., Hill, P. and Wiesenfeld, C. (2013), Collation and Review of Areal
Reduction Factors from Applications of the CRC-FORGE Method in Australia, Final Report,
Australian Rainfall and Runoff Revision Project 2: Spatial Patterns of Rainfall, Engineers
Australia, Barton, ACT.
Joss, J. and Waldvogel, A. (1990), Precipitation measurement and hydrology, in Radar in

Meteorology, D. Atlas, Ed., American Meteorological Society, pp: 577-606.
Laurenson, E.M., Mein, R.G. and Nathan, R.J. (2010), RORB Version 6, Runoff Routing
Program, User Manual, Version 6.14, Monash University and Sinclair Knight Merz,
Melbourne, Victoria.
Seed, A.W. and Austin, G.L. (1990), Sampling errors for raingauge derived mean-areal daily
and monthly rainfall, J. Hydrology, 118: 163-173.
Sinclair Knight Merz, (2013), Brisbane River Catchment Dams and Operational Alternatives
Study, Generation of Inflow Hydrographs and Preliminary Flood Frequency Analysis,
Revision 1, Brisbane, Queensland.
Umakhanthan, K. and Ball, J.E. (2005), Rainfall models for catchment simulation. Australian
Journal of Water Resources, 9(1), 55-67.
Urbonas, B.R., Guo, J.C.Y. and Janesekok, M.P. (1992), Hyetograph density effects on
urban runoff modelling, Proc. International Conference on Computer Applications in Water
Resources, Tamkang University, Tamsui, Taiwan, pp: 32-37.
Verworn, A. and Haberlandt, U. (2011), Spatial interpolation of hourly rainfall - effect of

additional information, variogram inference and storm properties, Hydrology and Earth
System Sciences, 15: 569-584.
Woolhiser, D.A. (1992), Modeling daily precipitation - progress and problems, in Statistics in
the environmental and earth sciences, edited by A.T. Walden and P. Guttorp, p: 306, Edward
Arnold, London, U.K.
131
Chapter 7. Continuous Rainfall
Simulation
Ashish Sharma, Ratnasingham Srikanthan, Raj Mehrotra, Seth Westra,
Martin Lambert
7.1. Use of Continuous Simulation for Design Flood

Estimation
Design floods can be estimated based on either historical flood data using at-site Flood
Frequency Analysis or Regional Flood Frequency Estimation, or derived using rainfall data
and a suitable hydrologic model to simulate flows. When using a hydrologic model, two
options exist. The first is the event-based approach which converts the design rainfall storm
to a corresponding design flood using a hydrologic model. The second is the continuous
simulation approach, which converts a continuous rainfall time series to a flow time series
using a hydrologic model, followed by the application of a frequency analysis on the flows to
estimate the design flood. The generation of the rainfall time series for this latter approach is
the focus of this chapter.
While a clear case is often present when deciding between a Flood Frequency Analysis and
an event-based approach for estimating the design flood, it is less clear when a continuous
simulation approach should be used in place of an event-based approach. In general, the
primary benefits of continuous simulation approaches arise when the relationship between
the catchment’s antecedent moisture stores and the flood-producing rainfall event are not
independent of each other, or change over time (Blazkova and Beven, 2002; Boughton and
Droop, 2003; Cameron et al., 2000; Lamb and Kay, 2004). Continuous simulation allows an
explicit representation of the joint probability of antecedent moisture conditions and flood-
producing rainfall data, which can be challenging for event-based approaches. Therefore
key areas where continuous simulation approaches are likely to be useful include the
following:
• Catchments with large moisture stores which have a significant relationship between
antecedent rainfall and the annual maximum flood (Pathiraja et al., 2012);
• Examining the joint probability of flooding arising from the confluences of streams which
are subject varying spatial rainfall distributions;
• Situations where the relationship between historical antecedent conditions and flood-
producing rainfall are not representative of the design period. This is may occur as a result
of climate change, but may also be relevant when calibrating over a period that is over-
represented in terms of El Niño or La Niña events (e.g. Pui et al. (2011));
• Situations requiring a quantification of the uncertainty of flood quantities, where

continuous simulation approaches can provide a natural method for representing the
dependence between flood-producing rainfall and the antecedent conditions;
• Situations where the initial level of flood and reservoir storages are unknown and these
influence the resulting downstream flood flows; and
132
Continuous Rainfall Simulation
• When using a virtual laboratory to test proposed simpler event-based approaches.
Consider the example in Figure 2.7.1 which uses data for a South Australian catchment to
illustrate the workings of the three approaches used for design flood estimation. While
Figure 2.7.1 uses the 90 day antecedent rainfall to illustrate its relation with the extreme
rainfall, similar joint relationships could exist between antecedent rainfall for longer periods,
or other more subtle rainfall characteristics that are difficult to summarise using a simple
metric.
The first two panels illustrate the working of a flood frequency or event-based modelling
approach for design flood estimation. The last panel illustrates a continuous simulation
model that attempts to capture the strong relationship in extreme rainfall with the 90 day
Antecedent rain.
Figure 2.7.1. Flood Events for a Typical Australian Catchment - Scott Creek, South Australia
As highlighted in Figure 2.7.1, continuous simulation approaches for design flood estimation
require continuous rainfall sequences as the primary data input. Although continuous rainfall
data exist in some locations for periods of several decades or longer, for most locations in
Australia the continuous data is either unavailable, too short or of insufficient quality to
support continuous rainfall-runoff modelling. This chapter therefore presents the basis and
techniques for stochastically generating continuous rainfall records in a catchment. Also
discussed are:
i. Generic issues regarding the accuracy of rainfall observations and methods for identifying
errors in rainfall time series;
133
ii. Approaches for infilling rainfall data at a point location;
iii. When to generate multi-site data as compared to lumped or single site rainfall;
iv. Approaches for generating data at locations where rainfall records are not available; and
v. Implications of non-stationarity in the rainfall record as a result of urbanisation and climate

change.
Book 2, Chapter 7, Section 2 discusses the approaches used to prepare rainfall data for use
in stochastic generation or other modelling studies. Book 2, Chapter 7, Section 3 discusses
a conceptual framework that underlies stochastic rainfall generation at point or multiple
locations. Alternatives for generation of daily rainfall are discussed Book 2, Chapter 7,
Section 4, while Book 2, Chapter 7, Section 4 discusses alternatives for disaggregation of
daily rainfall to sub-daily time scales. Alternatives that generate continuous rainfall
sequences without reference to a daily total at point and multiple locations conclude the
presentation. Worked examples illustrating the applications of some of the models presented
are included to assist with practical implementations (Refer to Book 2, Chapter 7, Section 4).
7.2. Rainfall Data Preparation

7.2.1. Errors in Rainfall Measurements
Stochastic rainfall generation aims to generate continuous rainfall sequences that are
representative of the underlying climate. Hence, it is important that the observed rainfall is a
true representation of the underlying climate, and is not influenced by potential measurement
or sampling inaccuracies that may lead to biased rainfall sequences. The first step of
stochastic rainfall generation is to identify and correct for noticeable errors in the observed
rainfall record.
Rainfall measurements can be susceptible to a range of errors:
• Effect of wind, wetting, evaporation and splashing on daily rainfall measurements – The
World Meteorological Organisation (World Meteorological Organisation, 1994) states that
these factors can result in the measured daily rainfall being less than the true rainfall by
anywhere between three and 30%.
• Errors in tipping bucket measurements – Tipping bucket rainfal gauges are the preferred
means of continuous rainfall measurement over the world. While reasonably accurate at
low rainfall intensities, tipping bucket rainfall gauges can underestimate the rainfall when
intensities are high due to the water lost as a result of the tipping motion of the rainfall
gauge. Typical errors for intensities greater than 200 mm/hr can range from 10-15% of the
true rainfall (La Barbera et al., 2002). A simple model for characterising gauge
measurement errors was proposed by Ciach (2003), marking them inversely proportional
to the measured rainfall intensity.
• Homogeneity of rainfall measurements - The double mass curve is a commonly used

technique to identify and correct for changes in the exposure or location of the gauge,
changes in the manner in which data is collected, or any other changes that result in a
systematic bias in the measurements compared to the general trend in nearby locations.
An example of such change is illustrated in Figure 2.7.2 for a hypothetical rainfall record,
where the change appears to have occurred around 1955 with the slope of the mass curve
changing from 0.95 to 0.75 from that point onwards. Changes such as those illustrated in
Figure 2.7.2 should be investigated in greater detail and corrective measures (e.g.
134
multiplicative scaling) may need to be used. Note that similar comparative checks can also
be used in the context of identifying ‘odd’ rainfall gauge locations from the regional
average (slope of the double mass curve will be significantly different to 1).
• Homogeneity of gridded rainfall – An important source of error in gridded rainfall data is

related to the use of varying number of rainfall gauges over time in the process of
reconstructing records. (Hasan et al., 2014) investigated at the extent of this variability in
the context of radar rainfall estimation expressed this as a function of the grid size and the
density of gauges within the grid. Using daily gauge data for Sydney, the coefficient of
variation of the rainfall for a 1 km x 1 km grid cell having a single gauge was estimated as
1.35, with reductions in this value as more gauges were included, and increases when
extended to larger grid sizes. This error was found to be considerably larger than the
measurement error discussed before (Ciach, 2003). While there is no clear way of
addressing this error, its variation over time can be factored in the specification of any
model that is developed using this as inputs (Chowdhury and Sharma, 2007).
• Effect of untagged multi-day accumulations in daily rainfall data – As nearly a third of the
long-term daily rainfall records are recorded at Post Offices and other public buildings, the
occurrence of multi-day readings (representing Saturday to Monday) recorded on the first
working day after the weekend is frequent. An example of one such station is illustrated in
Figure 2.7.3. Viney and Bates (2004) outline a hypothesis test for identifying the periods in
a rainfall record that reflect significant multi-day accumulations. While there is no simple
corrective procedure that can be employed, common-sense alternatives such as
comparing with data at nearby locations (after ascertaining that they do not suffer from the
same problem), and using the persistence structure of the non-accumulated data to
disaggregate the accumulated values, should be adopted. It should be noted that while
such accumulations may not affect calculations in yield or water balance studies, their
implications in flood estimation studies can be significant.
135
Figure 2.7.2. Double Mass Curve Analysis for Rainfall at Station A (from World
Meteorological Organisation (1994))
136
Figure 2.7.3. Total Rainfall Amounts for Rainfall Station 009557 over the Period 1956-1962
(from Viney and Bates (2004))
7.2.2. Options for Catchments with no Rainfall Records

One of the advantages of event-based approaches for design flood estimation, is the
availability of design rainfall data in different parts of Australia. These data are derived
through spatial interpolation of Intensity Frequency Duration (IFD) parameters, with
assumptions on the changes one would expect from gauged to ungauged locations. In
contrast, continuous simulation either requires observed rainfall time series at each location
of interest, or a procedure to generate such series based on data from nearby locations. For
situations where observed rainfall data are not available, the following alternatives can be
considered:
• Use of gridded rainfall products -Given the need to use catchment averages of rainfall and
potential evapotranspiration in a range of hydrologic studies, datasets of gridded rainfall
and temperature have been produced for Australia and elsewhere. Two gridded datasets
used routinely in Australia are the SILO and the Australian Water Availability Project
(AWAP) daily rainfall 5 km x 5 km gridded datasets. The SILO project (Jeffrey et al., 2001)
by the Queensland Centre for Climate Applications, Department of Natural Resources,
aimed to develop a comprehensive archive of key meteorological variables (Maximum and
Minimum Temperature, Rainfall, Class-A pan Evaporation, Solar Radiation and Vapour
Pressure) through interpolation on a 0.05° grid extending from latitude 10°S to 44°S and
longitude 112°E to 154°E. The project has also resulted in a patched daily rainfall series at
4600 locations extending back to 1890. In addition, the AWAP dataset was produced by
the CSIRO and the Bureau of Meteorology (Jones et al., 2009) at the same resolution
using a different averaging procedure. These datasets have been compared (Beesley et
137
al., 2009) and found to be similar in many respects, while still resulting in a dampening of
high extremes due to averaging, as well as a over-simulation of the number of wet days in
a year. Similar biases occur in the representation of persistence attributes, possibly
distorting the specification of antecedent conditions prior to large rainfall events. Care
must be taken when using such datasets, especially if the intention is to simulate flow
extremes for the catchment.
• Use of radar or satellite derived rainfall measurements - While the above mentioned
gridded data are based on spatial interpolation of gauged rainfall alone, another option
that has been pursued with success is to combine gauge and remotely sensed rainfall,
which is known to improve accuracy especially in remote locations with limited gauge
coverage. Examples of approaches that have produced and assessed such combined
datasets include (Chappell et al., 2013). While they suffer from the same problems as
other gridded datasets, the advantages they offer in remote locations should be taken into
consideration.
• Use of statistical interpolation techniques based on nearby daily and sub-daily gauge
records - Refer to the alternatives for continuous simulation at ungauged locations
presented later in the chapter. These alternatives use separate approaches for daily and
sub-daily continuous generation at ungauged locations. The daily alternative amounts to
identifying nearby gauges that “mirror” key characteristics that would be expected of daily
rainfall at the location of interest. These nearby gauge records are then transformed to the
current location by adjusting for any difference in their annual mean. Each nearby gauge is
assigned a probability depending on how “similar” it may be to the location of interest,
which allows characterisation of the uncertainty associated with this procedure. In the sub-
daily case, a second step is adopted. Once the daily record has been generated, it is
disaggregated using data on sub-daily fragments based on a different set of
characteristics that define the sub-daily climate of the location. More details on these
procedures are presented later.
7.2.3. Missing Rainfall Observations

Rainfall records often contain missing observations that need to be filled using appropriate
techniques. This problem is often compounded when records from multiple sites are to be
used for analysis. The World Meteorological Organisation (World Meteorological
Organisation, 1994) expresses caution against filling more than 10% of the rainfall records
as the aggregate rainfall information may be influenced by interpretation. Some of the
methods recommended for filling short gaps in the rainfall record are as follows:
• Normal Ratio Method – This method estimates the missing rainfall �� at gauge � as a
weighted average of the measured rainfall at nearby rainfall gauges:
�─�
∑�� = 1 �
�─� � (2.7.1)
�� =
�
─ ─
where G represent the total number of rainfall gauges, �� and �� the average annual rainfall
at gauges g and i respectively, and �� the rainfall at gauge i for the time period being filled.
Care must be taken to ensure that the “host” rainfall gauges have similar climatic conditions
as the gauge where the missing observations are being infilled.
• Quadrant Method – This method is related to the Normal Ratio method, but aims to
account for the proximity of the rainfall gauges to the target location. The missing rainfall
138
�� at gauge g is estimated as a weighted average of observations at four rainfall gauges,

one in each quadrant using north-south and east-west lines that intersect the location of
gauge g. The rainfall is estimated as:
1
∑4� = 1 ��
�2�
�� = 1 (2.7.2)
∑4� = 1 2
��
where di is the Euclidean distance between gauges i and g.
• Isohyetal Method – This method involves drawing isohyets (lines of equal rainfall) for the
storm duration over the network of rainfall gauges available, and inferring the rainfall at the
missing rainfall gauge by interpolation. The accuracy of the Isohyetal method depends
significantly on the number of rainfall gauges used and the interpolation algorithm being
used to construct the isohyets.
• Copula based interpolation – Bárdossy and Pegram (2014) presented an alternative for
interpolating existing data to infill missing values at a station of interest. They used a
copula-based specification of the conditional probability distribution of the missing rainfall
based on values at nearby gauges. They compared their approach with both regression
and other spatial interpolation based alternatives and found it to perform better using daily
rainfall data from South Africa. Another advantage of their approach is that it can include
conditioning on exogenous variables which could include atmospheric fields that are
common to all stations in the area of interest, thereby allowing additional information on
the nature of precipitation.
The above methods are fairly intuitive and modifications of the basic logic outlined are
common. For instance, in situations where data from nearby rainfall gauges are hard to find,
the interpolation is often from previous years of record at the same rainfall gauge, the period
being chosen to represent the same season and similar antecedent rainfall conditions.
The methods suggested above should be used with care, with consideration for the
distributional changes that occur as a result of the interpolation. For instance, if the stations
used for spatial averaging are at significant distances to the station where the interpolation is
required, then the interpolated rainfall is likely to be ‘smoother’ than the rainfall that would
have occurred at that location, potentially leading to an overestimation of wet days and an
underestimation of peak rainfall. Similarly, if the interpolation is performed at each time step
independently, the dependence of rainfall from one time to the next may not be accurately
represented. These considerations attain importance particularly when short time steps
(daily and sub-daily) are considered, and when the missing periods are a significant portion
of the overall record.
Missing data within historical rainfall records can be a serious problem, the amount of which
can affect the type of model structure considered. Few researchers explain adequately how
this is dealt with. Cowpertwait (1991) described a replacement strategy to handle missing
data but it is not apparent that this approach will be adequate with significant missing or
rejected data. Katz and Parlange (1995) and Gyasi-Agyei (1999) ignore and discard months
with any missing data. As a result, valuable information could be lost, particularly if there is
limited data in the first place. For some months of the year Gyasi-Agyei (1999) discarded up
to half of the available data. With an event-based approach, discarding storm events or inter-
event times containing missing intervals should introduce no significant bias into the
139
calibration, provided the occurrence of this corrupted data is random. Therefore, if part of a
month of data is missing it does not invalidate the remaining good quality data in that month.
7.3. Stochastic Rainfall Generation Philosophy

Stochastic generation of daily or sub-daily rainfall sequences requires the specification of a
probabilistic model of rainfall over time. Such a probability model should account for the
following features of daily or sub-daily rainfall:
• The significant probability mass for zero values (no rain);
• The seasonality of a range of rainfall statistics, including wet/dry days, averages and
extremes;
• The low-frequency variability, which causes below- or above-average rainfall to persist for
multiple consecutive years;
• The short-range (day-to-day and within-day) persistence of wet and dry periods; and
• The highly skewed distribution of rainfall, with the rainfall features often of most interest in
a design flood estimation context being located at the tail of the distribution.
Simulation of these aspects of rainfall requires careful formulation of the rainfall generation
model, often by using conditional variables that enforce this variability at multiple timescales.
Finally, although the current chapter does not discuss the case of stochastic generation at
multiple locations, this added consideration would require the specification of multivariate
conditional probability distributions characterising both the temporal evolution of the process,
as well its links in space.
In general, single site rainfall generation approaches fall into the following categories:
1. Daily rainfall generation;
2. Sub-daily rainfall generation; and
3. Sub-daily rainfall generation through disaggregation of daily rainfall.
Many alternative models exist for each of these categories, as do their extensions to
ungauged or partly gauged locations. Readers are referred to Sharma and Mehrotra (2010)
for a review on these alternatives. A subset of these alternatives is discussed in Book 2,
Chapter 7, Section 4. It should be noted that some of the sub-daily models simulate daily
rainfall very well when aggregated to daily (refer to Frost et al. (2004)).
7.4. Rainfall Generation Models

7.4.1. Daily Rainfall Generation
7.4.1.1. Overview of Daily Rainfall Generation Techniques
Generation of daily rainfall sequences requires the formulation of procedures for generating
rainfall occurrences (wet or dry) and amounts (for the wet days). As rainfall occurs in bursts,
it is important to represent the day-to-day persistence in the rainfall. This can be
accomplished by assuming rainfall is a Markovian process, with the nature of persistence
140
defined by the order of Markovian dependence. A first-order Markovian process assumes

rainfall depends only on the rainfall (amount or occurrence) on the previous day.
Assuming first or low-order dependence can result in the number of wet days in a year being
similar from one year to the next. This is contrary to the nature of rainfall in Australia and
elsewhere, with considerable variations from one year to the next often modulated by low-
frequency climatic anomalies such as the El Nino Southern Oscillation (ENSO)
phenomenon. This inability of rainfall generation models to simulate observed variability at
aggregated (annual or longer) scales is referred to as “over-dispersion” (Katz and Parlange,
1998).
In addition to the representation of persistence, a rainfall generation model needs to also

allow for seasonal variation. This is often accomplished by allowing model parameters to be
estimated on a seasonal or monthly basis. While these distributions are characterised by
sample parameter estimates at locations with sufficient observational records, these can
also be regionalised for use in ungauged locations.
Table 2.7.1 (adapted from Sharma and Mehrotra (2010)) summarises the approaches used
for generation of daily rainfall. The higher-order Markov approaches listed are especially
relevant for Australia, given the significant low-frequency variability that characterises
Australian rainfall. Misrepresentation of this variability can have serious implications in the
representation of pre-burst antecedent conditions, as well as the relationship between the
rainfall extremes and the longer-range antecedent rainfall, given both are known to be
modulated by climatic anomalies responsible for such variability in rainfall time series.
Table 2.7.1. Alternative Methods for Stochastic Generation of Daily Rainfall

Model Description/Advantages/ References
Drawbacks
Daily Rainfall Occurrence Generation
Low-order Markov Chain Based on wet day (Buishand, 1977; Caskey,
Models probabilities. For some 1963; Feyerherm and Bark,
regions generates rainfall 1965; Feyerherm and Bark,
series with too few long dry 1967; Gabriel and Neumann,
spells. 1962; Hopkins and Robillard,
1964; Racsko et al., 1991;
Selvalingam and Miura,
1978; Stern and Coe, 1984;
Wilks, 1998; Chapman,
1997)
Higher-order Markov Chain Based on wet day (Coe and Stern, 1982;
Models probabilities of few Dennett et al, 1983; Gates
consecutive days. The and Tong, 1976; Jones and
approach increases the Thornton, 1997; Mehrotra
length of the Markov model’s and Sharma, 2007a;
‘memory’ of antecedent wet Mehrotra and Sharma,
and dry days. The number of 2007b; Mehrotra et al., 2012;
parameters (i.e., transition Pegram, 1980; Singh and
probabilities) required Kripalani, 1986)
increases exponentially as
the order increases, being 2k
for a kth-order chain. These
141

Drawbacks
models improves the
representation of observed
interannual variance in the
simulations but still fell short
of observed climatic
variability on average.
‘Hybrid-order’ Markov Models The Markov ‘memory’ (Stern and Coe, 1984; Wilks,
extends further back in time 1999a)
for the dry spells only.
Alternating Renewal Process These spell-length models (Buishand, 1977; Racsko et
Based Models operate by fitting probability al., 1991; Roldan and
distributions to observed Woolhiser, 1982; Woolhiser,
relative frequencies of wet 1992; Wilks, 1999a)
and dry spell lengths. The
approach may not be suited
in arid regions or in cases
with less than 25 years of
observations.
Non-parametric wet-dry spell (Lall et al., 1996; Sharma and
length models. Lall, 1999)
Daily Rainfall Generation including Amount
Parametric Precipitation Based on some distribution (Wilks, 1999b; Coe and
Amounts Models like a two parameter gamma Stern, 1982; Richardson,
distribution, exponential and 1981; Woolhiser and
mixed exponential Pegram, 1979; Woolhiser
distribution. These models and Roldán, 1982; Woolhiser
assume that precipitation and Roldán, 1986)
amounts on wet days are
independent, and follow the
same distribution.
Wet Spell Based These models allow different (Chapman, 1997; Buishand,
Precipitation Amount Models probability distributions for 1977; Chin and Miller, 1980;
precipitation amounts Cole and Sheriff, 1972;
depending on that day’s Wilks, 1999b)
position in a wet spell
(separate models for start,
mid and end of a wet spell).
Non-parametric Precipitation A non-parametric kernel (Harrold et al., 2003a;
Amount Models density estimation based Mehrotra and Sharma, 2006;
procedure is used to simulate Oriani et al., 2014)
the rainfall conditional on
previous time step value of
rainfall and/or other
variables.
Multi-state Markov Models These Markov models (Boughton, 1999; Gregory et
simulate both precipitation al., 1993; McMahon and
occurrence and amounts, by Srikanthan, 1983; Srikanthan
142

Drawbacks
defining different ranges of and McMahon, 1985; Haan
precipitation amounts as et al., 1976)
constituting distinct states.
The outcome of this
approach depends on the
choice of the number of
states, their ranges and on
the distributions used for wet
day amounts in any given
state. These models involve
comparatively large numbers
of parameters, and thus
require quite long data
records in order to be
estimated well.
Cluster Based Point Rainfall process is described (Evin and Favre, 2012;
Processes Models using cluster of rectangular Kavvas and Delleur, 1981;
pulses. In the approach, Kim et al., 2014; Leblois and
storms arrive according to a Creutin, 2013; Leonard et al.,
Poisson process and are 2008; Onof et al., 2000;
represented by clusters of Ramirez and Bras, 1985;
rainfall cells temporally Rodriguez-Iturbe et al., 1984;
displaced from the storm Rodriguez-Iturbe et al., 1987;
centre. Rodriguez-Iturbe et al., 1988;
Waymire and Gupta, 1981a;
Waymire and Gupta, 1981b;
Wheater et al., 2000)
Copula Theory Based Multi-variate copulas are (Bárdossy and Pegram,
Models used to describe the spatial 2009; Serinaldi, 2009)
structure of rainfall amounts
and occurrences.
Multi-fractal Simulation These models characterise (Seed et al., 1999; Menabde
Techniques rainfall by scale invariant et al, 1997; Jha et al., 2015)
(scaling) and fractal
properties.
Conditioning on Co-variates Monthly statistics of rainfall, (Jones and Thornton, 1997;
long-range forecasts of the Katz et al., 2003; Wilks,
monthly statistics, random 1999a)
numbers or a ‘hidden’
mixture approach to capture
some inter-annual variability.
Conditioning on Previous Rainfall occurrences and (Harrold et al., 2003b;
Time History of Simulated amounts are simulated Harrold et al., 2003a;
Rainfall conditional on the recent past Mehrotra and Sharma,
rainfall behaviour. 2007a; Mehrotra and
Sharma, 2007b; Sharma and
O'Neill , 2002)
143

Drawbacks
Conditioning on Some Using the Lamb Weather (Katz and Parlange, 1998;
Aspect of Large-scale Type weather classification, Hay et al., 1991; Charles et
Atmospheric Circulation/ monthly Southern Oscillation al., 1999; Hughes and
Weather Patterns Index (SOI), North Atlantic Guttorp, 1994; Woolhiser,
Oscillation Index (NAOI), 1992; Wilby et al., 1998; Kim
North Atlantic sea surface et al., 2014; Kim et al., 2012;
temperature (SST) anomalies Wallis and Griffiths, 1997;
and other atmospheric Bárdossy and Plate, 1992;
predictors. Serinaldi and Kilsby, 2014;
Kleiber et al., 2012; Carey-
Smith et al., 2014; Heaps et
al., 2015)
Model Nesting at Multiple Rainfall amounts are (Boughton, 1999; Srikanthan
Time Scales adjusted at monthly/seasonal and Pegram, 2009; Wang
and annual time scales to and Nathan, 2007; Lambert
maintain the desired et al., 2003; Thyer and
variability at higher time Kuczera, 2003a; Thyer and
scales. Kuczera, 2003b)
The daily generation models in Table 2.7.1 are often formulated using high quality observed
rainfall records, and then regionalised for use anywhere. Regionalisation of a rainfall
generation model is accomplished either by interpolating model parameters for use at
ungauged locations, or by sampling data from other locations as representative for the
location of interest. In the discussion that follows, two methods - the regionalised Nested
Transition Probability Model (N-TPM) and the Regionalised Modified Markov Model
(RMMM), are summarised due to their widespread use in Australia and the availability of
software to facilitate implementation within the country.
7.4.1.2. Nested Transition Probability Matrix Approach

The Transition Probability Model (TPM) offers a simple and effective characterisation of
Markov order-one persistence in the daily rainfall generation process (Srikanthan et al.,
2003). In the TPM, the daily rainfalls are divided into a maximum of seven states. State 1 is
dry (no rainfall) and the other states are wet. The rainfall amounts in the largest state are
generated using a Gamma distribution. The model operates by estimating the transition
probability of sampling a state given the state of the preceding time step. Hence, if seven
states are used, a 7 x 7 transition probability matrix needs to be estimated from the data. As
only Markov order-one dependence is assumed, a correction is needed to ensure that
simulated rainfall exhibits sufficient variability at an annual time scale. This correction occurs
by rescaling of the daily rainfall amounts, thereby inflating the variability of rain on each day,
while keeping the fraction of wet days in a year constant.
The TPM has been applied in a number of studies, and exists in a regionalised form for use
anywhere in Australia. The computer program for the TPM can be obtained from the
Stochastic Climate Library as part of the e-Water Toolkit (http://toolkit.net.au/Tools/SCL ).
Parameters for major city centres and recommendations for ungauged locations are
provided within the software. Table 2.7.2 and Table 2.7.3 present the number of states and
the rainfall amount associated with highest state used for major city centres in Australia. If
the number of states is less than seven the upper limit of the last state is infinite. Figure 2.7.4
144
provides regional extents that are used in applying the method to other locations not
included in the tables.
Table 2.7.2. Number of States used for Different Rainfall Stations in the Transition Probability
Model (Srikanthan et al., 2003)
Station Latitud Longitu J F M A M J J A S O N D
e °S de °E
Melbour 37 49 144 58 6 6 6 6 6 6 6 6 6 6 6 6
ne
Lerderd 37 30 144 22 6 6 6 6 6 6 6 6 6 6 6 6
erg
Monto 24 51 151 01 6 6 6 6 6 6 6 6 6 6 6 6
Cowra 33 49 148 42 6 6 6 6 6 6 6 6 6 6 6 6
Adelaid 34 56 138 35 6 6 6 6 6 6 6 6 6 6 6 6
e
Perth 31 57 115 51 6 6 6 6 6 6 6 6 6 6 6 6
Sydney 33 52 151 12 7 7 7 7 7 7 7 7 7 7 7 7
Brisban 27 28 121 06 7 7 7 7 7 7 7 7 7 7 7 7
e
Mackay 21 06 149 06 7 7 7 7 7 7 7 7 7 7 7 7
Kalgoorl 30 47 21 27 5 5 5 5 5 5 5 5 5 5 5 5
ie
Alice 23 49 133 53 4 4 4 4 4 4 4 4 4 4 4 4
Springs
Onslow 21 40 115 07 4 4 4 3 4 3 4 3 3 3 3 3
Bambo 22 03 119 38 6 6 6 5 5 5 2 2 2 2 2 2
o
Springs
Broome 17 57 122 15 7 7 7 3 3 3 3 3 3 3 3 4
Darwin 12 27 130 48 7 7 7 7 3 2 2 2 3 7 7 7
145
Figure 2.7.4. Rainfall Stations used in Table 2.7.1 for the Transition Probability Model
(Srikanthan et al., 2003)
Table 2.7.3. State Boundaries for Rainfall Amounts in the Transition Probability Model
State Number Upper State Boundary Limit (mm)
1 0.0
2 0.9
3 2.9
4 6.9
5 14.9
6 30.9
7 ∝
As the Transition Proability Method requires a correction for the misrepresentation of low-
frequency variability, several alternatives have been developed to address this limitation.
The Nested Transition Probability Method (Srikanthan and Pegram, 2009) operates by
aggregating the sequences of rainfall from the TPM to first a monthly and then to an annual
time scale. Once aggregated, rainfall is modelled as a Markov order-one process at the
aggregated time scale, accounting for the lag-one auto-correlation and variability that is
manifested in the aggregated process. This offers an effective means of correcting variability
in rainfall across a range of time scales, making the generated series more useable for
hydrological applications. As with the TPM, the computer program for the Nested TPM can
be obtained from the Stochastic Climate Library as part of the e-Water Toolkit (http://
toolkit.net.au/Tools/SCL ).
146
7.4.1.3. Regionalised Modified Markov Model

The Regionalised Modified Markov Model (RMMM) offers a non-parametric basis for daily
rainfall generation at any location in Australia in a manner that ensures generated
sequences mimic observed rainfall in its representation of distributional features as well as
low-frequency variability. The RMMM is a regionalised version of the Modified Markov Model
(MMM) (Mehrotra and Sharma, 2007a) which simulates rainfall by characterising the rainfall
occurrence by a variable-order Markovian process that is designed to simulate low-
frequency variability. This variable-order Markov process is defined by assuming that daily
rainfall occurrence depends on the rainfall state on the previous day as well as the
aggregated rainfall for the past 30 and 365 days. The use of the aggregated rainfall
conditioning variables allows the generated sequences to reflect the dependence there
exists in observed rainfall across different temporal scales. Furthermore, use of aggregated
variables allows invoking of the Central Limit Theorem and approximating their probability
distribution as a Gaussian distribution, thereby simplifying parameter estimation and
implementation. As a result of using the aggregated variables, the number of wet days in a
year exhibit variability that is consistent with the observed record, in contrast to the Nested
TPM approach that offers similar variability with rainfall amounts alone. Once the rainfall
occurrences have been generated, rainfall amounts are generated using a non-parametric
kernel density estimation approach.
The algorithm for generating daily rainfall using the Modified Markov Model is presented in
Algorithm for step-wise daily rainfall generation using Modified Markov Model (Mehrotra and
Sharma, 2007a) .
147
Algorithm for step-wise daily rainfall generation using Modified Markov Model
(Mehrotra and Sharma, 2007a)
1. For all calendar days of the year calculate the transition probabilities of the standard
first-order Markov model using the observations falling within the moving window of
31 days centered on each day. Denote these transition probabilities as p11 for
previous day being wet and p01 for previous day being dry.
2. Also estimate the means, variances and co-variances of the higher time scale
predictor variables separately for occasions when current day is wet/day and
previous day is wet/dry. Mehrotra and Sharma (2007a), identified 2 variables
namely, previous 30 and 365 days wetness state)
3. Consider a day. Ascertain appropriate critical transition probability to the day t based
on previous day’s rainfall state of the generated series. If previous day is wet, assign
critical probability p as p11 otherwise assign p01.
4. Calculate the values of the 30 and 365 days wetness state for the day t and the
available generated sequence (Jo). To have values of wetness state in the beginning
of the simulation randomly pickup a year from the historical record and calculate
values of 30 and 365 days wetness states.
5. Modify the critical transition probability p of step 3 using the following equation and,
conditional means, variances, co-variances and tth day value of higher time scale
predictors for the generated day t. Denote the modified transition probability as � .
1 −1 � − �
− � − �1, � �1,
2 � � � 1, � ’
�
det �1, �
� = �1i 1 1
− �� − �1, � �−1 −1 � − �
1, � �� − �1, � ’ − �� − �1, � �1,
2 2 � � 1, � ’
� �
�1i + 1 − �1i
det �1, � det �1, �
where �� is the predictor set at time t, the �1, � parameters represent the mean
� �� = 1, �� − 1 = � and �1, � is the corresponding variance-co-variance matrix.
Similarly, �0, � and �0, � represent, respectively, the mean vector and the variance-co-
variance matrix of � when �� − 1 = � and �� = 0 . The �1i parameters represent the
baseline transition probabilities of the first order Markov model defined by
� �� = 1 �� − 1 = � and det represents the determinant operation.
6. Compare � with the uniform random variate �� for station k. If �� ≤ � , assign

rainfall occurrence, Jo� for the day t as 1 otherwise zero.
7. Move to the next date in the generated sequence and repeat steps 2-5 until the
desired length of generated sequence is obtained.
Readers are referred to Mehrotra and Sharma (2007a) for details of the Modified Markov
Model rainfall generation algorithm. A R-package to generate daily rainfall at multiple
locations given observed rainfall time series has been developed Mehrotra et al. (2015) and
is available for download from Hydrology@UNSW Software website (http://
148
www.hydrology.unsw.edu.au/download/software/multisite-rainfall-simulator). The package

exists as a Multi-site Rainfall Simulator (abbreviated MRS), offering the capability to
generate rainfall at multiple locations of interest while maintaining their spatial dependence
attributes in sequences, but simplifies to the Modified Markov Model when used to generate
rainfall for a single location.
7.4.1.3.1. Regionalisation
The Modified Markov Modelrequires a representative sample of daily rainfall for generation
to proceed. This restricts its application only to locations having long-length observed
records. An attempt to regionalise the Modified Markov Model was presented in Mehrotra et
al. (2012), using a similar approach to the regionalised sub-daily generation model Westra et
al. (2012) in Book 2, Chapter 7, Section 4. Unlike the regionalised version of the Nested
TPM (Book 2, Chapter 7, Section 4), here the regionalisation involved identifying rainfall
records for locations deemed ‘similar’ to the target, followed by rescaling to adjust for
changed climatology, and then pooling to take account of relative similarities each nearby
location bears to the target location. This pooled record was then used as the basis for
generating the daily rainfall sequences.
As the regionalised approach relies on using data from nearby rainfall stations, it is
necessary to:
1. identify metrics to determine whether two stations are ‘similar’; and
2. predict the probability that stations within a ‘neighbourhood’ of the target site are similar
by regressing against physiographic indicators such as the difference in latitude,
longitude, elevation and relative distance to coast between station pairs.
The relative distance to coast is obtained by dividing the difference in distance to coast
between two stations by the distance to coast of the target site. This is done to account for
the fact that the relative influence of distance to coast is likely to be greater for two stations
having greater proximity to the coastline.
‘Similarity’ between any two sites was assessed based on the similarity in the bivariate
probability distributions of a daily-scale attribute of interest, and the annual rainfall total.
Table 2.7.3 outlines the attributes used in formulation of the RMMM. Each of the attributes
listed were used to define similarity between stations based on a two sample, 2 dimensional
Kolgomorov-Smirnov test (Fasano and Fanceschini, 1987). The resulting classification of
similarity (‘1’ for similar and ‘0’ for dissimilar) for each attribute was pooled in a logistic
regression framework, using the difference in latitude for the two stations, difference in
longitude, and difference in the relative distance to coast as covariates.
Table x presents Daily scale attributes used to define similarity between locations. Each of
these variables were estimated for each location and each year of record, and then paired to
assess the best basis for defining ‘similarity’ between stations. Using 2708 separate rain
gauge stations with at least 25 years of data, this resulted in a total of 3,665,278 station
pairs.
Table 2.7.4. Daily Scale Attributes used to Define Similarity between Locations
Daily Maxima Daily Maximum rainfall for DJF, MAM, JJA,

SON
7 Day Maxima Maximum 7-day total rainfall for each season
149
Wet/Dry Spell Lengths Maximum Wet/Dry spell length for each

season
Rainfall Amount per Spell Total Rainfall in maximum wet spell for each
season
Daily Averages • Average rainfall amount per wet day for
year
• Average rainfall amount per day (wet or

dry) for each season
Number of Wet Days • Total number of wet days each year
• Total number of wet days each season
This logistic regression framework can then be used to determine the similarity between any
two stations for the attribute of interest. Therefore, for a given target location where rainfall is
to be generated, one can now rank stations with data from most similar to least similar for
each attribute. The approach adopted in RMMM is to form an average rank using all
attributes for all nearby stations, and use the lowest S ranks to identify the stations to use as
the basis of rainfall generation. To account for the relative similarity across these S stations,
each station is selected with a probability equal to:
1
��
�� = 1 (2.7.3)
∑�� = 1 ��
where wi represents the weight associated with the ith station and ri the rank associated with
that station, used as the basis for probabilistically selecting nearby stations in the Modified
Markov Model. Lower ranked stations, which, by definition have rainfall attributes which are
most statistically similar to the target site, attain higher weight and therefore a higher
probability of being used in MMM. This rationale is summarised in Figure 2.7.5.
150
Figure 2.7.5. Identification of “Similar” Locations for Daily Rainfall Generation using RMMM
Once the ‘similar’ S stations have been identified, the generation of rainfall sequences at the
target location proceeds as per the generation algorithm for MMM in Algorithm for step-wise
daily rainfall generation using Modified Markov Model (Mehrotra and Sharma, 2007a), with
the inclusion of two additional steps. The first of these steps involves a rescaling of the
“similar” locations identified as described in Figure 2.7.5. The second of these steps is a
probabilistic selection of the “similar” locations, based on the weights associated with each
location. These steps are summarised in Figure 2.7.6.
151
Figure 2.7.6. Generation of Daily Rainfall Sequences using the Regionalised Modified
Markov Model Approach
It should be noted that the low frequency variable states (30 and 365 day wetness states in
MMM) are ascertained based on the generated sequence, and hence represent the
probabilistic average from the collation of the locations that have been selected as “similar”
for the generation procedure. The software first identified “similar" locations to the target
location of interest, and then estimates the parameters of the MMM for these locations. As
the criterion for selecting “similar” locations is defined as a function of differences in latitude,
longitude, elevation and rescaled distances from the coast, a new location with daily
observations can be included for the procedure to work. The parameters of the logistic
regression model have been ascertained using high quality daily rainfall observations, and
will be updated with significant updates in the daily rainfall datasets available in Australia.
It should also be noted that use of actual rainfall data from similar locations is followed by a
rescaling approach to account for changed climatology results in maximal use of observed
rainfall. The use of MMM has been shown to produce generated rainfall with low frequency
variability and extremes that are consistent with observations. Given not one but multiple
similar locations are used, the likelihood of over sampling rainfall attributes from a
misclassified similar location is reduced. An assessment by Mehrotra et al. (2012) indicates
that the method is able to capture the important attributes that define daily rainfall in both
gauged and ungauged locations in Australia.
152
7.4.2. Sub-daily Rainfall Generation

While considerable research has been done on the generation of daily rainfall sequences for
design flood estimation, information is often required at a sub-daily time scale. Sub-daily
rainfall sequences are generated using two approaches. In the first case, rainfall is
generated assuming a model formulated based on sub-daily rainfall observations. In the
second case, sub-daily rainfall is generated conditional to daily rainfall through a
disaggregation algorithm, the aim being to utilise the value of the much longer daily rainfall
data and adopt a sensible approach to convert it to finer time steps. Table 2.7.5 summarises
many of the sub-daily rainfall generation approaches available in the literature.
Table 2.7.5. Commonly used Sub-daily Rainfall Generation Models

Drawbacks
Poisson Cluster Process Represents rainfall events as (Cowpertwait, 2010;
Based Models clusters of rain cells where Cowpertwait et al., 2002;
each cell is considered a Leonard et al., 2008;
pulse with a random duration Koutsoyiannis et al., 2003;
and random intensity. A Onof and Townend, 2004;
rainfall generation model, Wheater et al., 2000; Gyasi-
however, can also be used Agyei, 2013; Rodriguez-
for rainfall disaggregation. Iturbe et al., 1987;
Rodriguez-Iturbe et al., 1988;
Eagleson, 1978; Heneker et
al., 2001; Koutsoyiannis and
Pachakis, 1996; Menabde
and Sivapalan, 2000)
Scale Invariance Theory Utilises the moment scaling (Waymire and Gupta, 1981a;
Based Models function and an appropriate Menabde et al, 1997; Seed et
probability distribution for the al., 1999; de Lima and
weights. Grasman, 1999; Deidda et
al., 1999; Gupta and
Waymire, 1993; Schertzer
and Lovejoy, 1987;
Sivakumar et al., 2001;
Lovejoy and Schertzer, 1990;
Molnar and Burlando, 2005;
Olsson and Berndtsson,
1998; Over and Gupta, 1996)
Parametric and non- Based on disaggregation of (Arnold and Williams, 1989;
parametric stochastic daily rainfall based on Connolly et al., 1998;
disaggregation models distribution of sub-daily Cowpertwait et al., 1996;
rainfall statistics/ rainfall Econopouly et al, 1990;
values. Hershenhorn and Woolhiser,
1987; Sharma and
Srikanthan, 2006; Westra et
al., 2012; Koutsoyiannis,
2001; Koutsoyiannis and
Onof, 2000)
153
Book 2, Chapter 7, Section 4 and Book 2, Chapter 7, Section 4. discuss two approaches
recommended for use in Australia. Both approaches exist in a regionalised form and can be
adopted at any location within the country. The Disaggregated Rainfall Intensity Pulse
approach represents a sub-daily rainfall generator that is calibrated using sub-daily data and
parameters regionalised for use anywhere, while the Regionalised Method of Fragments is a
daily to sub-daily disaggregation approach that relies on either the observed daily rainfall or
a generated daily rainfall sequence to convert to a sub-daily scale.
7.4.2.1. Disaggregated Rainfall Intensity Pulse

Also known as ‘alternating renewal’ or ‘profile-based’ models, event-based models break the
rainfall process into a series of events characterised by inter-arrival time, storm duration and
mean storm intensity. Early work on such models by Eagleson (1978) involved simulating
rainfall arrivals using a Poisson distribution, the time between events and the event duration
using an exponential distribution, and the storm event depth using a gamma distribution.
Since this time these models have undergone significant development, including the
elucidation of the self-similarity concept, in which storms are found to exhibit similar internal
structure despite differing durations and storm depths, thus providing a basis for the
disaggregation of storm events into within-storm temporal patterns (Koutsoyiannis, 2001;
Garcia-Guzman and Aranda-Oliver, 1993), and the development of a generalised
exponential distribution for representing inter-storm and storm durations (Lambert and
Kuczera, 1998).
The Disaggregated Rectangular Intensity Pulse (DRIP) model was developed by (Heneker
et al., 2001) with the view to addressing several perceived deficiencies in existing event-
based models, particularly with regard to the simulation of extreme rainfall and aggregation
statistics. The DRIP modelling process is divided into two stages. The generation stage
(Figure 2.7.7) is represented by three random variables: dry spell or inter-event time ta, the
wet spell or storm duration td, and the average intensity i, with ta and tdboth described by a
generalised exponential distribution and the intensity (i) described by a Generalised Pareto
distribution. In the second stage, the individual events are disaggregated through a
constrained random walk (Figure 2.7.7b) to represent the rainfall temporal pattern for each
event.
154
Figure 2.7.7. Disaggregated Rectangular Intensity Pulse Model (extracted from Heneker et
al. (2001))
The random walk through a non-dimensional time-depth space is illustrated in Figure 2.7.2.
This is then used to disaggregate the rectangular pulse to time steps of the order of six or
�
fewer minutes. Time during the storm is non-dimensionalised by � = �� where t is the time
� �
since the start of the storm and depth is non-dimensionalised by � = �� where � � is the
cumulative rainfall up to time t. The random walk progresses in discrete time intervals ��
from coordinate (0,0) to (1,1) in Figure 2.7.8, always with a non-negative slope. There are
two possibilities for a jump from � to � + ��:
1. An internal dry spell (represented by a horizontal segment in Figure Figure 2.7.8) whose
probability of occurrence is defined by a probability distribution; or
2. A rainfall burst (represented by a sloping segment in Figure Figure 2.7.8) whose non-
dimemsional depth �� is sampled from a probability distribution.
155
Figure 2.7.8. Schematic of Non-dimensional Random Walk used in DRIP disaggregate

pulses
To fit a probability distribution to the observed inter-event time ta and storm duration
tdpopulations, a procedure was employed to extract independent events from the continuous
historical record. After analysis of correlation results, Heneker et al. (2001) adopted a
minimum inter-event time of 2 hours to distinguish independent storms and inter-storm
periods. This value provides a balance between ensuring consecutive events are sufficiently
independent and the need to have as much calibration storm data as possible within a fixed
length historical record. While different minimum inter-event times have been reported (e.g.
(Grace and Eagleson, 1966; Sariahmed and Kisiel, 1968; Koutsoyiannis and Xanthopoulos,
1990; Heneker et al., 2001)), Heneker et al. (2001) showed that 2 hours was shown to
assure independence of storm events across numerous Australian sites.
The generalised exponential distribution developed by (Lambert and Kuczera, 1998) was
used to model the distributions of inter-event time and storm duration. The generalised
exponential distribution takes the form:
156
−�� , ��
� � �� = � � ≤ � �� = 1 − � ,� > 0 (2.7.4)
where X is an independently distributed random variable and �� is a parameter vector which

may be dependent ont defined as the time at the start of the storm or inter-event time, and
�� is a kernel function. The kernel chosen by Heneker et al. (2001) to best fit the data
was a combination of functions based on the Generalised Pareto Distribution (Rosjberg et
al., 1992) and the power law:
1 � �
log� 1 − � � �� = −� �, �� = log� 1 − �1 − �3� 4, �1 < 0, �2, �3, �4 > 0 (2.7.5)
�1 �2
The parameter vector �� is estimated using maximum likelihood techniques. The DRIP
parameters are usually calibrated for each month of the year to capture seasonal variability
in the rainfall process. Figure 2.7.9 and Figure 2.7.10 illustrate observed and fitted
probability distributions for inter-event and storm durations for Melbourne for select months
and demonstrate the good fit typically achieved by the generalized exponential distribution.
Noting that exponentially distributed data would plot as a straight line in Figure 2.7.9 and
Figure 2.7.10, the use of an exponential distribution for inter-event and storm durations
would be clearly inappropriate. A detailed comparison of the DRIP model with other point
process models is given in Frost et al. (2004).
Figure 2.7.9. Heneker et al. (2001) Model Fitted to Monthly Inter-event Time Data for
Melbourne in January
157
Figure 2.7.10. Heneker et al. (2001) Model Fitted to Monthly Storm Duration Data for
Melbourne in May
Recently, DRIP has been extended to any location where sufficient daily data is available,
thus greatly augmenting the domain of the approach. The basis of this regionalisation is a
‘master-target’ scaling relationship in which model calibration is undertaken at a ‘master’ site
with a long pluviograph record which is then updated and scaled to the ‘target’ site of interest
using the information from either a short pluviograph or daily rainfall record (Jennings, 2007),
with testing providing encouraging results for separations of up to 190 km between the
master and the target.
The software for DRIP is available via the Stochastic Climate Library as part of the e-Water
Toolkit (http://toolkit.net.au/Tools/DRIP).
7.4.2.2. Regionalised Method of Fragments

The regionalised method of fragments offers a mechanism to disaggregate observed or
generated daily rainfall sequences to a sub-daily time scale. The disaggregation rationale for
the method is patterned after the Method of Fragments (Boughton, 1999) that resamples the
observed near-continuous fractions (or fragments) of daily accumulated rainfall for use with
any daily total that is closest in magnitude. This approach assumes that the sub-daily rainfall
structure depends solely on the daily rainfall, an assumption that can lead to discontinuities
in the generated sub-daily sequences between two adjacent days. Taking this on board,
Westra et al. (2012) modified the basic Method of Fragments approach in two ways. The first
modification was to the traditional fragments approach to work at ungauged locations. The
second modification was the use of a “state-based” conditioning approach (Sharma and
Srikanthan, 2006) that makes use of information about the state of rainfall on the preceding
and the next day, in an attempt to reduce the disconnect in sub-daily rainfall attributes across
daily boundaries.
158
Figure 2.7.11 illustrates the rationale behind the regionalised version of the state-based
Method of Fragments procedure used in Westra et al. (2012). A sub-daily time-step of 6-
minutes is used in Figure 2.7.11, although no change in the procedure is needed if an
alternate sub-daily time-step is to be adopted. Here, I(Rt) represents the state (wet or dry) of
the rainfall on day t. Conditioning the selection of a “similar” day in the historical record
involves selecting from a subset of days that (a) fall within a calendar window representative
of the season (chosen equal to +/-15 days in Westra et al. (2012)), and (b) represent the
same state � �� − 1 , � �� = 1 , � �� + 1 . Once these sub-sets of days are identified, they are
ranked based on their similarity with the rainfall amount that is sought to be disaggregated.
This forms the sample of days the fragments can be resampled from. Resampling proceeds
probabilistically using the k-nn resampling approach of Lall and Sharma (1996). Once the
fragments have been resampled, they are scaled back to rainfall amounts by multiplication
with the daily rainfall total for the day being disaggregated.
Figure 2.7.11. State-based Method of Fragments Algorithm used in the Regionalised Method
of Fragments Sub-daily Rainfall Generation Procedure
The logic used to regionalise the method is similar to that adopted in case of the
Regionalised Modified Markov Model (RMMM) (Book 2, Chapter 7, Section 4). Here, the
importance of regionalisation is all the more given the paucity of sub-daily rainfall records in
most parts of Australia (and the world). However, here the aim of the regionalisation is not to
159
identify locations having similar rainfall attributes as the target, but a similar daily to sub-daily
disaggregation relationships. As with the daily rainfall generation, a range of criteria were
used to characterise this relationship. These are listed in Table 2.7.6. Each of these
variables were estimated for each location, and then paired to assess the best basis for
defining ‘similarity’ between stations.
Table 2.7.6. Sub-daily Attributes used to Define Similarity between Locations
Maximum Sub-daily Intensity Maximum Intensity Fraction for each day for
6, 12, 30, 60, 120, 180 and 360 minute
durations.
Fraction of Zeroes Fraction of zero rainfall time-steps within
each day at a 6 minute time scale.
Timing of Maximum Intensity The timing associated with the maximum
intensity fraction for the day for 6, 12, 30, 60,
120, 180 and 360 minute time steps.
Using 232 separate rain gauge stations with at least 30 years of data, a total of 26 796
station pairs were formulated for each attribute. The similarity in each attribute across each
pair was then assessed using a two sample two dimensional Kolmogorov Smirnov test.
Using a significance level of 5%, this allowed the identification of pairs where the attributes
were similar. This then allowed the identification of covariates that could be used to
distinguish “similar” locations to allow the regionalisation to proceed. Use of attributes
pertaining to the maximum sub-daily fractions at multiple durations, as well as the timing of
the maximum, allowed similarity to be defined taking both diurnal pattern characterisation
and rainfall magnitudes into account. The use of fraction of zeroes allowed distinction
between locations having dominantly convective extremes from those that were spread over
the day.
The results of the significance testing described above were used as the basis for
formulating a logistic regression relationship for each attribute, with regression coefficients
being allowed to vary with season. The predictor variables found to be significant in defining
the relationship were the differences in latitude, longitude, elevation and the relative distance
to the coast. Based on this relationship, given any location in Australia, the user can identify
a subset of sub-daily locations having attributes that are most similar to the target location
sequences are needed at. This information is expressed as a probability, which is then used
to identify a defined number of sub-daily locations for use in the RMOF procedure.
The logistic regression of the binomial (0 for insignificant and 1 for a significant test outcome)
response for each sub-daily attribute can be expressed as:
��
Pr � = 1 = logit � = (2.7.6)
�� + 1
The logit function transforms the continuous predictor variables in Table 2.7.7 to the range
[0,1] as required when modelling a binomial response. In this equation, z is defined as:
� = �0 + �1�1 + ... + �5�5 (2.7.7)
with β representing the regression coefficients in Table 2.7.7 for the five predictor variables
used.
160
Table 2.7.7. Logistic Regression Coefﬁcients for the Regionalised Method of Fragments Sub-
daily Generation Modela
Logistic Regression Coefficients
Season Sub-daily Intercept Latitude Longitude Latitudex Distance Elevation
Rainfall Longitude Coast
Attribute
DJF 6 min 0.426 -0.345 -0.0377 0.0064 -0.186 -0.00089
intensity
DJF 1 hr 0.823 -0.333 -0.0425 0.0093 -0.231 -0.00075
intensity
DJF Fraction -0.375 -0.253 -0.0318 0.0075 -0.242 -0.00065
of zeros
DJF 6 min 0.0979 -0.137 -0.0099 0.0022 -0.453 -0.00141
time
MAM 6 min -0.067 -0.192 -0.0065 NS -0.218 -0.00130
intensity
MAM 1 hr 0.308 -0.178 -0.0074 NS -0.107 -0.00098
intensity
MAM Fraction -0.806 -0.157 -0.0105 0.0025 -0.165 -0.00060
of zeros
MAM 6 min 1.256 -0.140 -0.0226 -0.0034 -0.227 -0.00092
time
JJA 6 min -0.197 -0.097 -0.0110 0.0034 -0.096 -0.00198
intensity
JJA 1 hr 0.471 -0.0102 -0.0204 0.0033 NS -0.00335
intensity
JJA Fraction -0.365 -0.073 -0.0171 0.0031 -0.101 -0.00116
of zeros
JJA 6 min 2.078 -0.098 -0.0321 0.0037 -0.156 -0.00069
time
SON 6 min 0.474 -0.387 -0.0722 0.0129 NS -0.00146
intensity
SON 1 hr 0.824 -0.325 -0.0835 0.0135 NS -0.00132
intensity
SON Fraction -0.382 -0.239 -0.0623 0.0104 -0.087 -0.00095
of zeros
SON 6 min 1.028 -0.162 -0.0287 0.0042 -0.317 NS
time
aAll predictors were found to be statistically significant (usually with a p-value <0.001 level), with the exception of
several predictors labelled as NS (Not Significant). Seasons include December-January-February (DJF), March-
April-May (MAM), June-July-August (JJA) and September-October- November (SON).
This allows the identification of the most to least similar sub-daily locations for each attribute
of interest, which forms the basis for identification of a subset of locations used to sample
the fragments. As multiple sub-daily attributes are considered in this choice, this subset is
selected based on a common rank averaged across all the attributes for each season. The
161
number of locations the fragments are pooled from depend on their respective data lengths
as a total of 500 years of data (including zeroes) is needed for the approach to work.
7.4.3. Identifying ‘Nearby’ Stations - Application to Sydney

Airport
Book 2, Chapter 7, Section 4 provides a demonstration of a single application of the
approach at one location: Sydney Airport (Gauge number 066037). This location represents
a relatively long-record pluviograph station, and therefore provides a useful record for
verification of the method.
The approach to identifying ‘nearby’ stations is as follows:
• For all the 1396 pluviograph stations in Australia (excluding the Sydney Airport gauge),
calculate each of the regression predictors; namely, difference in latitude, longitude,
latitude*longitude, elevation and normalised distance to coast, relative to the Sydney
Airport station;
• Having developed the 1396 x 5 predictor matrix, apply the regression model presented in
Equation (2.7.6) and Equation (2.7.7) using the regression coefficients shown in
Table 2.7.2 for each season and attribute to calculate the probability Pr(u =1);
• Separately for each season and attribute, rank the probabilities from highest to lowest;
• For each season calculate the average rank for each station across all attributes;
• Select the S lowest ranked stations for inclusion in the disaggregation model.
This algorithm yields different choices of stations for each season, as physiographic
influences may vary depending on the dominant synoptic systems occurring and different
times of the year. It is noted that the selection of the size of S represents a somewhat
subjective decision, as larger values of S increase the probability of selecting stations which
are statistically different to the target station, whereas smaller values of S will result in small
sample sizes. For this case we a total of 500 years of data (including zero rainfalls)
distributed over the 13 stations (S=13).
These lowest ranked 13 stations for the summer season are shown in Figure 2.7.12. As
expected, the lowest ranked stations (i.e. those with the greatest chance of being ‘similar’ to
Sydney Airport, brown dots) are those which are most proximate to this station, generally
within a small distance to coast, and all are at low coastal elevations. In this case, therefore,
the stations appear to be selected over a wide range of latitudes, which is probably due to
the strong increases in elevation and relative distance to coast with changing longitude.
162
Figure 2.7.12. Sydney Airport and nearby pluviograph stations
It should be noted that the RMOF approach can be expanded to use more sub-daily data
than the 1306 stations used in the example for Sydney Airport presented above. New data
can be included without the need to update the coefficients of the logistic regression model
unless these inclusions are substantial enough to change the distributional characteristics of
the data being used. This allows improvements in the representativeness of the continuous
simulations as more data over time.
It should also be noted that the RMOF can be used at completely locations having no sub-
daily or daily rainfall observations, or to disaggregate daily rainfall records at locations where
sub-daily data is not available. In the first case, the use of a daily generation approach is
recommended such as the RMMM to generate daily sequences that should then be
disaggregated using RMOF. In the latter case, the observed daily sequence can be used
directly as the basis for disaggregation.
The software for the RMOF approach is available on request from the authors at this stage,
and will be uploaded to the Hydrology@UNSW Software website after a formal review
process (http://www.hydrology.unsw.edu.au/download/software/). This document will be
updated to reflect the full location once the download of the software is completed.
7.4.4. Modification of Generated Design Rainfall Attributes

The stochastic nature of the algorithms described in previous sections mean that the
stochastic sequences will also produce stochastic estimates of design rainfall. In many
cases this is a desirable outcome of the approach, as it enables the representation of
uncertainty associated with design rainfall. However, in some situations, it may be desirable
to post process the design rainfall characteristics obtained through stochastic generation in
order to reflect published Intensity Frequency Duration curves. This is likely to be particularly
useful when conducting comparisons between the outputs of continuous versus event-based
163
models, or when seeking to understand the role of a catchment’s antecedent moisture

content conditional to pre-specified design rainfall features.
For cases where it is necessary to have consistency between the Bureau of Meteorology
IFDs and the IFDs derived from continuous simulation, a modification in the generation
algorithms for RMMM and RMOF was proposed. The main steps involved as illustrated in
Figure 2.7.13. First, annual extreme rainfall is corrected at multiple durations so that the IFD
based on the generated rainfall matches up with the observed IFD (henceforth referred as
‘target IFD’). Second, the non-extreme rainfall (i.e. rainfall that is not part of the annual
extreme series) is corrected in such a way that the cumulative rainfall before and after
correction is maintained. The dry periods are kept the same before and after bias correction,
hence no correction is required for dry periods. As the majority of the data is in the non-
extreme category, the corrections are markedly smaller for the non-extreme case.
Due to the inter-dependence of the extreme rainfall across various durations, it is necessary
to apply the above corrections in a recursive manner, with each recursion repeating the
above steps using a new set of durations exhibiting the maximum difference between the
generated and target intensities. This recursion is applied until the following objective
function reaches a minimum:
IFD�AEP − IFD�AEP
RMAEAEP = (2.7.8)
IFD�AEP
where the objective to be minimised is a dimensionless standardised error measure referred

as Relative Mean Absolute Error (RMAE) for consistent comparison across various durations
and exceedance probabilities. The RMAE at each of the Annual Exceedance Probabilities
(AEP) is estimated through the mean of the absolute difference between the target IFD
� �
(IFDAEP ) and generated IFD (IFDAEP ) scaled by the target IFD.
Minimisation of the RMAE in Equation (2.7.8) requires the specification of the set of target
durations to be used in its adjustment. The choice of durations is governed by the
dependence that the extremes for one duration have with the extremes for another. For
instance, it is more likely for 6 minute extremes to be a subset of 30 minute extremes than 6
hour extremes (say). In such a case, the durations should be selected keeping an interval
that maximises the independence between the extremes being evaluated. In practice, the
procedure uses two recursions with separate durations. For both recursions, three target
durations, i.e. D = 6 min, 1 hr and 3 hrs are considered, which keeps the distance between
the durations far enough to reduce the dependence between them. Options exist to use a
broader set of durations in the second iteration (6 min, 30 min, 1 hr, 3 hr, 6 hr, 12 hr)
although assessment with data for selected city centres in Australia indicated the benefits
from this were not significant.
164
Figure 2.7.13. Main Steps Involved in the Adjustment of Raw Continuous Rainfall
Sequences to Preserve the Intensity Frequency Duration relationships
The software for the post-processing approach described above is available on request from
the authors at this stage, and will be uploaded to the Hydrology@UNSW Software website
after a formal review process (http://www.hydrology.unsw.edu.au/download/software/). This
document will be updated to reflect the full location once the download of the software is
completed.
165
7.4.5. Example of Daily and Sub-Daily Rainfall Generation

This example presents the generation of daily, sub-daily and corrected sub-daily rainfall for
Alice Springs, including the case where it is assumed that data for the location is not
available. Hence, the results here represent a typical example practitioners may face when
generating rainfall sequences for any ungauged location in Australia.
Alice Springs is an arid region with average annual rainfall of 280 mm. The observed record
at Alice Springs Airport exists for 67 years (1942-2008) and the sub-daily record for 57 years
(1951-2007, with missing periods). Each of the statistics presented are based on 100
realisations of length 67 years.
7.4.5.1. Daily Rainfall Generation

For daily rainfall generation, two options are considered;
a. observed rainfall record at the location is available (at-site generation); and
b. no daily rainfall record is available i.e. location is ungauged (regionalised generation).
Table 2.7.8. Statistical Assessment of Daily Rainfall from RMMM for Alice Springs using 100
Replicates 67 years Long
Attribute Observed Simulated
At-site Regionalised
Average Annual Wet 41 40 31
Days (Nos)
Average Annual 279 297 306
Rainfall (mm)
Average Standard 13 12 15
Deviation of Annual
Wet Days (Nos)
Average Standard 152 160 189
Deviation of Annual
Rainfall (mm)
Figure 2.7.14 presents annual rainfall simulations for Alice Springs using 100 replicates. The
probability distribution of annual rainfall is well represented even in the case of the
regionalised simulation where at-site data was not used. This indicates a reasonable
representation of the inter-annual variability that characterises Australian rainfall.
166
Figure 2.7.14. Annual Rainfall Simulations for Alice Springs using 100 Replicates
As with the annual rainfall in Figure 2.7.13, the extremes are reasonably well simulated even
for the regionalised case, except for the most extreme event on record, which the model
under-simulates in the regionalised setting (Figure 2.7.15).
Figure 2.7.15. Intensity Frequency relationship for 24 hour Duration.
It should be noted that the results of the RMMM approach use 2708 daily rainfall stations
with long records, instead of the complete daily rainfall observation dataset for Australia.
One can expect better representation of underlying rainfall attributes as better and longer
datasets are used.
7.4.5.2. Sub-Daily Rainfall Generation

Sub-daily rainfall generation is based on the Regionalised Method of Fragments approach
(RMOF) in Book 2, Chapter 7, Section 4. Keeping in mind data availability scenarios for sub-
daily rainfall generation, the following generation options are possible:
A. Daily and sub-daily rainfall record at the location of interest is available- Daily time series
are disaggregated using available at-site sub-daily time series. To obtain multiple
simulations, the same daily rainfall time series is used (at-site daily and at-site sub-daily).
167
B. Only a daily rainfall record at the location of interest is available - Daily time series is
disaggregated using sub-daily time series from nearby locations (regionalised sub-daily).
To obtain multiple simulations, the same daily rainfall time series is used (at-site daily and
regionalised sub-daily).
C. No daily or sub-daily rainfall record at the location of interest is available - First multiple
realisations of daily time series are obtained using regionalised daily model. In the second
step, each daily time series is disaggregated using sub-daily time series from nearby
locations (regionalised sub-daily) (regionalised daily and regionalised sub-daily).
Selected results from this assessment are presented in Table 2.7.9 and Figure 2.7.15. The
deterioration in the representation of extremes in the shorter duration case, is observed,
when regionalised options are considered, especially for the smallest duration (6 minute).
Table 2.7.9. Performance of extremes and representation of zeroes (for 6 minute time-steps)
from the sub-daily rainfall generation using RMOF for at-site generation using observed sub-
daily data (option 1), at-site disaggregation using observed daily data (option 2), and the
purely regionalised case (option 3).
Average Annual Maximum Rainfall (mm) in Spell of
Duration Observed Option A Option B Option C
6 min 5.5 6.75 6.77 8.02
30 min 16.71 18.07 18.23 20.97
1 hr 22.14 24.19 24.17 26.56
3 hr 32.58 34.77 33.56 34.94
6 hr 39.61 41.73 39.79 40.74
12 hr 48.18 47.65 46.5 46.78
Percentage of 98.54 98.62 98.78 98.68
zeros
Top panel presents results for option A, at-site daily rainfall and fragments, middle panel
presents results for option B, regionalised daily rainfall at-site fragments, while the bottom
panel presents results for option C, regionalised results using ‘nearby’ daily as well as sub-
daily records. Dark Blue dots represent observed data, the solid line represents the median
of 100 simulations, and dashed lines represent the 5 and 95 percentile simulated values
168
Figure 2.7.16. 6 minute (left column) and 6 hour (right column) Annual Maximum Rainfall
against Exceedance Probability for Alice Springs.
7.4.5.3. Post-Processing of Continuous Rainfall to Correct for

Intensity Frequency Duration Biases
As illustrated in Figure 2.7.16, in cases where observed rainfall datasets used for continuous
simulation are of poor quality or are pooled from locations that are dis-similar to the target,
the RMMM and RMOF approaches will simulate sequences with different IFD attributes
compared to those published by the Bureau of Meteorology Intenisty Frequency Duration
data. This is addressed using a post-processing step that involves scaling of the continuous
sequences to alter extremes while attempting to maintain the average annual rainfall for the
location of interest.
Results from this post-processing step for the continuous rainfall sequences from RMOF for
Alice Springs are presented in Figure 2.7.17. The broken lines (blue and green) indicate the
5 and 95 percentiles for raw and bias corrected data, respectively. The continuous series
that was generated has not used rainfall data from Alice Springs for the purpose of
generation. In addition to representing low frequency variability characteristics through the
proper simulation of daily rainfalls, these continuous sequences are able to mimic actual
IFDs and annual rainfall totals, thus making them suitable for continuous flow simulation.
169
Figure 2.7.17. Intensity Duration Frequency Relationships for Target and Simulated Rainfall
before and after Bias Correction at Alice Springs
7.5. Implications of Climate Change

The implications of climate change on design rainfall have been discussed in Book 1,
Chapter 6 and are not repeated here. The focus of this section is to discuss how the
procedures for continuous simulation described here may be altered to account for climate
change. This may be particularly important if there are changes in extreme rainfall, in
antecedent rainfall, or in the dependence between the two, all of which will have significant
impacts on the resulting design flood.
Both the daily and the sub-daily continuous simulation alternatives discussed here will be
affected by climate change. Practitioners may need to use daily rainfall sequences that are
representative of future warmer climates, and are referred to the statistical downscaling
extensions of the RMMM daily generation approach discussed in Mehrotra and Sharma
(2010) for an alternative for generating daily sequences for any location of interest. This
generation requires selection of appropriate Global Climate Models (GCMs) and
atmospheric predictors, followed by sensible correction of GCM simulations to remove
known biases. Practitioners are referred to (Sharma et al., 2013) for a review of the
approaches used commonly for these purposes.
Generation of sub-daily sequences will require modification of the RMOF to alternatives that
take into account changes to extremes at sub-daily timescales (Westra et al., 2014) as well
as changes to associated temporal patterns (Wasko and Sharma, 2015). An alternative that
can be used to accommodate these changes is presented in (Westra et al., 2013). In
general, approaches for stochastically generating continuous (sub-daily) rainfall sequences
under a future climate are a rapidly evolving area of research, and detailed advice on theory
and approaches for continuous simulation under a future climate are outside of the scope of
this document.
7.6. References
Arnold, J.G. and Williams, J.R. (1989), Stochastic Generation of Internal Storm Structure at
a Point, Transactions of the ASAE, 32(1), 161-0167.
Beesley, C.A., Frost, A.J. and Zajaczkowski, J. (2009), A comparison of the BAWAP and
SILO spatially interpolated daily rainfall datasets, paper presented at 18th World IMACS /
MODSIM Congress, MSSANZ, Cairns, Australia.
170
Blazkova, S. and Beven, K. (2002), Flood frequency estimation by continuous simulation for
a catchment treated as ungauged (with uncertainty), Water Resources Research, Volume
38(8).
Boughton, W.C. (1999), A daily rainfall generating model for water yield and flood
studiesRep. Report 99/9, 21pp pp, CRC for Catchment Hydrology, Monash University,
Melbourne.
Boughton, W. and Droop, O. (2003), Continuous simulation for design flood estimation - a
review, Environmental Modelling & Software, 18(4), 309-318.
Buishand, T.A. (1977), Stochastic modeling of daily rainfall sequences, Meded.

Landbouwhogesch. Wageningen, 77(3), 211.
Bárdossy, A. and Pegram, G. (2014), Infilling missing precipitation records - A comparison of

a new copula-based method with other techniques, Journal of Hydrology, 519, Part A, pp:
1162-1170.
Bárdossy, A. and Pegram, G.G.S. (2009), Copula based multisite model for daily
precipitation simulation, Hydrol. Earth Syst. Sci., 13(12), 2299-2314.
Bárdossy, A. and Plate, E.J. (1992), Space-time model for daily rainfall using atmospheric
circulation patterns, Water Resources Research, 28(5), 1247-1259.
Cameron, D., Beven, K., Tawn, J. and Naden, P. (2000), Flood frequency estimation by
continuous simulation (with likelihood based uncertainty estimation), Hydrology and Earth
System Sciences, 4(1), 23-34.
Carey-Smith, T., Sansom, J. and Thomson, P. (2014), A hidden seasonal switching model for
multisite daily rainfall, Water Resources Research, 50(1), 257-272.
Caskey, J.E. (1963), A MARKOV CHAIN MODEL FOR THE PROBABILITY OF

PRECIPITATION OCCURRENCE IN INTERVALS OF VARIOUS LENGTH, Mon. Wea. Rev.,
91(6), 298-301.
Chapman, T.G. (1997), Stochastic models for daily rainfall in the Western Pacific,
Mathematics and Computers in Simulation, 43: 351-358.
Chappell, A., Renzullo, L.J., Raupach, T.H. and Haylock, M. (2013), Evaluating geostatistical
methods of blending satellite and gauge data to estimate near real-time daily rainfall for
Australia, Journal of Hydrology, 493: 105-114.
Charles, S.P., Bates, B.C. and Hughes, J.P. (1999), A spatiotemporal model for downscaling
precipitation occurrence and amounts, Journal of Geophysical Research-Atmospheres,
104(D24), 31657-31669.
Chin, E.H. and Miller, J.F. (1980), On the Conditional Distribution of Daily Precipitation
Amounts, Mon. Wea. Rev., 108(9), 1462-1464.
Chowdhury, S. and Sharma, A. (2007), Mitigating Parameter Bias in Hydrological Modelling

due to Uncertainty in Covariates, Journal of Hydrology, 340(3-4), 197-204.
Ciach, G.J. (2003), Local random errors in tipping-bucket rain gauge measurements, Journal
of Atmospheric and Oceanic Technology, 20(5), 752-759.
171
Coe, R. and Stern, R.D. (1982), Fitting Models to Daily Rainfall Data, Journal of Applied
Meteorology, 21(7), 1024-1031.
Cole, J.A. and Sherriff, J.D.F. (1972), Some single- and multi-site models of rainfall within
discrete time increments, Journal of Hydrology, 17(1-2), 97-113.
Connolly, R.D., Schirmer, J. and Dunn, P.K. (1998), A daily rainfall disaggregation model,
Agricultural and Forest Meteorology, 92(2), 105-117.
Cowpertwait, P.S.P. (1991), Further developments of the Neyman‐Scott clustered point

process for modeling rainfall. Water Resources Research, 27: 1431-1438.
Cowpertwait, P.S.P. (2010), A spatial-temporal point process model with a continuous

distribution of storm types, Water Resources Research, 46(12).
Cowpertwait, P.S.P., Kilsby, C.G. and O'Connell, P.E. (2002), A space-time Neyman-Scott
model of rainfall: Empirical analysis of extremes - art. no. 1131, Water Resources Research,
38(8), 1131-1131.
Cowpertwait, P.S.P., O'Connell, P.E., Metcalfe, A.V. and Mawdsley, J.A. (1996), Stochastic
point process modelling of rainfall. II. Regionalisation and disaggregation, Journal of
Hydrology, 175(1-4), 47-65.
Das, D., Kodra, E., Obradovic, Z. and Ganguly, A.R. (2012), Mining extremes: Severe rainfall
and climate change, in Frontiers in Artificial Intelligence and Applications, edited, pp:
899-900.
Deidda, R., Benzi, R. and Siccardi, F. (1999), Multifractal modeling of anomalous scaling
laws in rainfall, Water Resources Research, 35(6), 1853-1867.
Dennett, M.D., Rodgers, J.A. and Stern, R.D. (1983), Independence of Rainfalls through the
Rainy Season and the Implications for the Estimation of Rainfall Probabilities, Journal of
Climatology, Volume 3(4), pp.375-384.
Eagleson, P.S. (1978), Climate, Soil, and Vegetation 2. The Distribution of Annual
Precipitation Derived From Observed Storm Sequences, Water Resources Research, 14(5),
713-721.
Econopouly, T.W., Davis, D.R. and Woolhiser, D.A. (1990), Parameter transferability for a
daily rainfall disaggregation model, Journal of Hydrology, 118(1-4), 209-228.
Evin, G., and Favre, A.C. (2012), Further developments of a transient Poisson-cluster model
for rainfall, Stochastic Environmental Research and Risk Assessment, 27(4), 831-847.
Fasano, G. and Franceschini, A. (1987), A Multidimensional Version of the Kolmogorov-

Smirnov Test, Mon Not R Astron Soc, 225(1), 155-170.
Feyerherm, A.M. and Bark, L.D. (1965), Statistical Methods for Persistent Precipitation
Patterns, Journal of Applied Meteorology, 4(3), 320-328.
Feyerherm, A.M. and Bark, L.D. (1967), Goodness of Fit of a Markov Chain Model for
Sequences of Wet and Dry Days, Journal of Applied Meteorology, 6(5), 770-773.
Frost, A., R. Srikanthan, and P. Cowpertwait (2004), Stochastic generation of point rainfall
data at subdaily timescales: a comparison of DRIP and NSRP, Technical report 04/09,
Cooperative Research Centre - Catchment Hydrology, Melbourne, Australia.
172
Gabriel, K.R. and Neumann, J. (1962), A Markov chain model for daily rainfall occurrence at
Tel Aviv, Q.J Royal Met. Soc., 88(375), 90-95.
Garcia-Guzman, A., and Aranda-Oliver, E. (1993), A stochastic model of dimensionless

hyetograph, Water Resources Research, 29(7), 2363-2370.
Gates, P. and Tong, H. (1976), On Markov Chain Modeling to Some Weather Data, Journal
of Applied Meteorology, 15(11), 1145-1151.
Grace, R.A. and Eagleson, P.S. (1966). The Synthesis of Short-Time-Increment Rainfall
Sequences, Hydrodynamics Lab. Report 91, Dept. of Civil Engineering, Massachusetts
Institute of Technology, Cambridge, MA.
Gregory, J.M., Wigley, T.M.L. and Jones, P.D. (1993), Application of Markov models to area-
average daily precipitation series and interannual variability in seasonal totals, Clim. Dyn., 8:
299-310.
Gupta, V.K. and Waymire, E.C. (1993), A Statistical Analysis of Mesoscale Rainfall as a
Random Cascade, Journal of Applied Meteorology, 32(2), 251-267.
Gyasi‐Agyei, Y. (1999), Identification of regional parameters of a stochastic model for rainfall

disaggregation. Journal of Hydrology. 223: 148-163.
Gyasi-Agyei, Y. (2013), Evaluation of the effects of temperature changes on fine timescale

rainfall, Water Resources Research, 49(7), 4379-4398.
Haan, C.T., Allen, D.M. and Street, J.D. (1976), A Markov chain model of daily rainfall, Water
Resour. Res., 12: 443-449.
Harrold, T.I., Sharma, A. and Sheather, S.J. (2003a), A nonparametric model for stochastic
generation of daily rainfall amounts - art. no. 1343, Water Resources Research, 39(12),
1343-1343.
Harrold, T.I., Sharma, A. and Sheather, S.J. (2003b), A nonparametric model for stochastic
generation of daily rainfall occurrence - art. no. 1300, Water Resources Research, 39(10),
1300-1300.
Hasan, M.M., Sharma, A., Johnson, F., Mariethoz, G., Seed, A. (2014), Correcting bias in
radar Z-R relationships due to uncertainty in point raingauge networks. Journal of Hydrology
519(B), 1668-1676.
Hay L.E., McCabe G.J., Wolock D.M. and Ayers M.A. (1991), Simulation of precipitation by
weather type anaylsis. Water Resources Research, 27: 493-501.
Heaps, S.E., Boys, R.J. and Farrow, M. (2015), Bayesian modelling of rainfall data by using
non-homogeneous hidden Markov models and latent Gaussian variables, Journal of the
Royal Statistical Society: Series C (Applied Statistics), 64(3), 543-568.
Heneker, T.M., Lambert, M.F. and Kuczera, G. (2001), A point rainfall model for risk-based
design, Journal of Hydrology, 247(1-2), 54-71.
Hershenhorn, J. and Woolhiser, D.A. (1987), Disaggregation of daily rainfall, Journal of

Hydrology, 95(3-4), 299-322.
Hopkins, J.W. and Robillard, P. (1964), Some Statistics of Daily Rainfall Occurrence for the
Canadian Prairie Provinces, Journal of Applied Meteorology, 3(5), 600-602.
173
Hughes, J.P. and Guttorp, P. (1994), Incorporating Spatial Dependence and Atmospheric
Data in a Model of Precipitation, Journal of Applied Meteorology, 33(12), 1503-1515.
Jeffrey, S.J., Carter, J.O., Moodie, K.B. and Beswick, A.R. (2001), Using spatial interpolation
to construct a comprehensive archive of Australian climate data, Environmental Modelling &
Software, 16(6), 309-330.
Jennings S. (2007). A High Resolution Point Rainfall Model Calibrated to Short Pluviograph
or Daily Rainfall Data, PhD Thesis, The University of Adelaide, Australia.
Jha, S.K., Mariethoz, G., Evans, J., McCabe, M.F. and Sharma, A. (2015), A space and time
scale dependent non-linear geostatistical approach for downscaling daily precipitation and
temperature, Water Resources Research, available at: http://onlinelibrary.wiley.com/doi/
10.1002/2014WR016729/full accesses [20 July 2015].
Jones, P.G. and Thornton, P.K. (1997), Spatial and temporal variability of rainfall related to a
third-order Markov model, Agricultural and Forest Meteorology, 86(1-2), 127-138.
Jones, D., Wang, W.and Fawcett, R. (2009), High-quality spatial climate data-sets for
Australia, Australian Meteorological and Oceanographic Journal, 58: 233-248.
Katz, R. and Parlange, M.B. (1995), Generalizations of chain-dependent processes:

Application to hourly precipitation. Water Resour. Res., 31: 1331-1341.
Katz, R.W. and Parlange, M.B. (1998), Overdispersion phenomenon in stochastic modeling
of precipitation, Journal of Climate, 11(4), 591-601.
Katz, R.W., Parlange, M.B. and Tebaldi, C. (2003), Stochastic modeling of the effects of
large-scale circulation on daily weather in the southeastern US, Climatic Change, 60(1-2),
189-216.
Kavvas, M.L. and Delleur, J.W. (1981), A stochastic cluster model of daily rainfall sequences,
Water Resources Research, 17(4), 1151-1160.
Kim, D., Kim, J. and Cho, Y.-S. (2014), A Poisson Cluster Stochastic Rainfall Generator That
Accounts for the Interannual Variability of Rainfall Statistics: Validation at Various
Geographic Locations across the United States, Journal of Applied Mathematics, 2014, pp:
1-14.
Kim, Y., Katz, R.W. Rajagopalan, B., Podestá, G.P. and Furrer, E.M. (2012), Reducing
overdispersion in stochastic weather generators using a generalized linear modeling
approach, Climate Research, 53(1), 13-24.
Kleiber, W., Katz, R.W. and Rajagopalan, B. (2012), Daily spatiotemporal precipitation
simulation using latent and transformed Gaussian processes, Water Resources Research,
48(1).
Koutsoyiannis, D. (2001), Coupling stochastic models of different timescales, Water

Resources Research, 37(2), 379-391.
Koutsoyiannis, D. and Onof, C. (2000), A computer program for temporal rainfall

disaggregation using adjusting procedures, paper presented at 25th General Assembly of
the European Geophysical Society, EGS, Nice.
174
Koutsoyiannis, D. and Pachakis, D. (1996), Deterministic chaos versus stochasticity in

analysis and modeling of point rainfall series, Journal of Geophysical Research, 101(D21),
26441-26451.
Koutsoyiannis, D. and Xanthopoulos, T. (1990). A dynamic model for short-scale rainfall

disaggregation, Hydrological Sciences Journal, 35(3), 303-322.
Koutsoyiannis, D., Onof, C. and Wheater, H.S. (2003), Multivariate rainfall disaggregation at
a fine timescale - art. no. 1173, Water Resources Research, 39(7), 1173-1173.
La Barbera, P., Lanza, L.G. and Stagi, L. (2002), Tipping bucket mechanical errors and their
influence on rainfall statistics and extremes, Water Science and Technology, 45(2), 1-10.
Lall, U. and Sharma, A. (1996), A nearest neighbor bootstrap for time series resampling,
Water Resources Research, 32(3), 679-693.
Lall, U., Rajagopalan, B. and Tarboton. D.G. (1996), A nonparametric wet/dry spell model for
resampling daily precipitation, Water Resources Research, 32(9), 2803-2823.
Lamb, R., and Kay, A.L. (2004), Confidence intervals for a spatially generalized, continuous
simulation flood frequency model for Great Britain, Water Resources Research, 40(7).
Lambert, M., and Kuczera, g. (1998), Seasonal generalized exponential probability models
with application to interstorm and storm durations, Water Resources Research, 34(1),
143-148.
Lambert M.F., Whiting J.P. and Metcalfe, A.V. (2003). A non-parametric hidden Markov
model for climate state identification. Hydrology and Earth Systems Sciences, 7(5), pp.
652-667.
Leblois, E., and Creutin J.-E. (2013), Space-time simulation of intermittent rainfall with
prescribed advection field: Adaptation of the turning band method, Water Resources
Research, 49(6), 3375-3387.
Leonard, M., Lambert, M.F. Metcalfe, A.V. and Cowpertwait, P.S.P. (2008), A space-time
Neyman-Scott rainfall model with defined storm extent, Water Resources Research, 44(9).
Lovejoy, S. and Schertzer, D. (1990), Multifractals, universality classes and satellite and
radar measurements of cloud and rain fields, J. Geophys. Res., 95(D3), 2021-2034.
McMahon, T.A. and Srikanthan, R. (1983), Statistical characteristics of Australian Rainfall

and EvaporationRep.
Mehrotra, R., and Sharma, A. (2006), A nonparametric stochastic downscaling framework for
daily rainfall at multiple locations, Journal of Geophysical Research-Atmospheres,
111(D15101), doi:10.1029/2005JD006637.
Mehrotra, R., and Sharma, A. (2010), Development and Application of a Multisite Rainfall
Stochastic Downscaling Framework for Climate Change Impact Assessment, Water
Resources Research, 46.
Mehrotra, R., and Sharma, A. (2007a), Preserving low-frequency variability in generated

daily rainfall sequences, Journal of Hydrology, 345(1-2), 102-120.
175
Mehrotra, R. and Sharma, A. (2007b), A semi-parametric model for stochastic generation of

multi-site daily rainfall exhibiting low-frequency variability, Journal of Hydrology, 335(1-2),
180-193.
Mehrotra, R., Westra, S., Sharma, A. and Srikanthan, R. (2012), Continuous rainfall
simulation: 2 - A regionalised daily rainfall generation approach, Water Resources Research,
48(1), W01536
Mehrotra, R., Li, J., Westra, S. and Sharma, A. (2015), A programming tool to generate
multi-site daily rainfall using a two-stage semi parametric model Environmental Modelling &
Software, 63: 230-239.
Menabde, M. and Sivapalan, M. (2000), Modeling of rainfall time series and extremes using
bounded random cascades and Levy-stable distribution, Water Resources Research, 36(11),
3293-3300.
Menabde, M., Seed, D., Harris and Austin, G. (1997), Self-similar random fields and rainfall
simulation, J. Geophys. Res., 102(D12), 13509-13515.
Micevski, T. and Kuczera, G. (2009), Combining Site and Regional Flood Information Using
a Bayesian Monte Carlo Approach, Water Resources Research(doi:
10.1029/2008WR007173), available at: online on http://www.agu.org/journals/wr/
papersinpress.shtml.
Molnar, P. and Burlando, P. (2005), Preservation of rainfall properties in stochastic

disaggregation by a simple random cascade model, Atmospheric Research, 77(1-4),
137-151.
Olsson, J. and Berndtsson, R. (1998), Temporal rainfall disaggregration based on scaling

properties, Water Science and Technology, 37(11), 73-79.
Onof, C. and Townend, J. (2004), Modelling 5-minute rainfall extremes, paper presented at
British Hydrological Society International Conference, London.
Onof, C., Chandler, R.E. Kakou, A., Northrop, P., Wheater, H.S. and Isham, V. (2000),
Rainfall modelling using Poisson-cluster processes: a review of developments, Stochastic
Environmental Research and Risk Assessment, 14(6), 384-411.
Oriani, F., Straubhaar, J., Renard, P. and Mariethoz, G. (2014), Simulation of rainfall time
series from different climatic regions using the direct sampling technique, Hydrology and
Earth System Sciences, 18(8), 3015-3031.
Over, T. M. and Gupta, V.K. (1996), A space-time theory of mesoscale rainfall using random
cascades, J. Geophys. Res., 101(D21), 26319.
Pathiraja, S., Westra, S. and Sharma, A. (2012), Why Continuous Simulation? The Role of
Antecedent Moisture in Design Flood Estimation, Water Resources Research, 48(W06534).
Pegram, G.G.S. (1980), An Autoregressive Model for Multilag Markov Chains, Journal of
Applied Probability, 17(2), 350.
Pui, A., Lal, A. and Sharma, A. (2011), How does the Interdecadal Pacific Oscillation affect
design floods in Australia?, Water Resources Research, 47(5).
176
Pui, A., Sharma, A., Santoso, A. and Westra, S. (2012), Impact of the El Niño Southern
Oscillation, Indian Ocean Dipole, and Southern Annular Mode on daily to sub-daily rainfall
characteristics in East Australia, Monthly Weather Review, 140: 1665-1681.
Racsko, P., Szeidl, L. and Semenov, M. (1991), A serial approach to local stochastic weather
models, Ecological Modelling, 57(1-2), 27-41.
Ramirez, J. A. and Bras, R.L. (1985), Conditional Distributions of Neyman-Scott Models for
Storm Arrivals and Their Use in Irrigation Scheduling, Water Resources Research, 21(3),
317-330.
Richardson, C.W. (1981), Stochastic Simulation of Daily Precipitation, Temperature, and

Solar-Radiation, Water Resources Research, 17(1), 182-190.
Rodriguez-Iturbe, I., Gupta, V.K. and Waymire, E. (1984), Scale considerations in the
modeling of temporal rainfall, Water Resources Research, 20(11), 1611-1619.
Rodriguez-Iturbe, I., Cox, D.R. and Isham, V. (1987), Some Models for Rainfall Based on
Stochastic Point Processes, Proceedings of the Royal Society A: Mathematical, Physical
and Engineering Sciences, 410(1839), 269-288.
Rodriguez-Iturbe, I., Cox, D.R. and Isham, V. (1988), A Point Process Model for Rainfall:
Further Developments, Proceedings of the Royal Society A: Mathematical, Physical and
Engineering Sciences, 417(1853), 283-298.
Roldan, J., and Woolhiser, D.A. (1982), Stochastic daily precipitation models 1. A
comparison of occurrence models, Water Resources Research, 18(5), 1451-1459.
Rosjberg, D., Madsen, H. and Rasmussen, P. F. (1992). Prediction in Partial Duration Series
with Generalized Pareto-Distirbuted Exceedences, Water Resources Research, 28(11),
3001-3010.
Sariahmed, A. and Kisiel, C.C. (1968). Synthesis of sequences of summer thunderstorm

volumes for the Atterbury watershed in the Tucson area. Proceedings International
Association Hydrologic Sciences Symposium on Use of Analog and Digital Computers in
Hydrology, 2: 439-447.
Schertzer, D. and Lovejoy, S. (1987), Physical modeling and analysis of rain and clouds by
anisotropic scaling multiplicative processes, J. Geophys. Res., 92(D8), 9693.
Seed, A.W., Srikanthan, R. and Menabde, M. (1999), A space and time model for design
storm rainfall, Journal of Geophysical Research-Atmospheres, 104(D24), 31623-31630.
Selvalingam, S. and Miura, M. (1978), STOCHASTIC MODELING OF MONTHLY AND

DAILY RAINFALL SEQUENCES, J Am Water Resour As, 14(5), 1105-1120.
Serinaldi, F. (2009), A multisite daily rainfall generator driven by bivariate copula-based

mixed distributions, J. Geophys. Res., 114(D10).
Serinaldi, F. and Kilsby, C.G. (2014), Simulating daily rainfall fields over large areas for
collective risk estimation, Journal of Hydrology, 512: 285-302.
Sharma, A. and Lall, U. (1999), A nonparametric approach for daily rainfall simulation,
Mathematics and Computers in Simulation, 48(4-6), 361-371.
177
Sharma, A. and Mehrotra, R. (2010), Rainfall Generation, in Rainfall: State of the Science,
edited by F.Y. Testik and M. Gebremichael, p: 287, American Geophysical Union, San
Francisco.
Sharma, A. and O'Neill, R. (2002), A nonparametric approach for representing interannual

dependence in monthly streamflow sequences - art. No.1100, Water Resources Research,
38(7), 1100-1100.
Sharma, A. and Srikanthan, R. (2006), Continuous Rainfall Simulation: A Nonparametric

Alternative, paper presented at 30th Hydrology and Water Resources Symposium, Institution
of Engineers (Australia), Launceston.
Sharma, A., Mehrotra, R. and Johnson, F. (2013), A New Framework for Modeling Future
Hydrologic Extremes: Nested Bias Correction as a Precursor to Stochastic Rainfall
Downscaling (invited), in Climate Change Modeling, Mitigation, and Adaptation, edited by
R.Y. Surampalli, T.C. Zhang, C.S.P. Ojha, B.R. Gurjar, R. D. Tyagi and C.M. Kao, p: 698,
American Society of Civil Engineers, Reston, Virginia.
Singh, S.V. and Kripalani, R.H. (1986), Analysis of persistence in daily monsoon rainfall over
India, Journal of Climatology, 6(6): 625-639.
Sivakumar, B., Berndtsson, R., Olsson, J. and Jinno, K. (2001), Evidence of chaos in the
rainfall-runoff process, Hydrological Sciences Journal, 46(1), 131-145.
Srikanthan, R. and McMahon, T.A. (1985), Stochastic generation of rainfall and evaporation
data, Tech. Paper No. 84, Aust. Water Resour. Council, Canberra, 1985, Australian Water
Resources Council, Technical Paper No.84.
Srikanthan, R. and Pegram, G. (2009), A Nested Multisite Daily Rainfall Stochastic

Generation Model, Journal of Hydrology, Available at: http://www.science-direct.com/ until
publication.
Srikanthan, R., McMahon, T., Harrold, T. and Sharma, A. (2003), Stochastic Generation of
Daily Rainfall DataRep., CRC for Catchment Hydrology Technical Report 03, Melbourne,
Australia.
Steinschneider, S., and Lall, U. (2015), A hierarchical Bayesian regional model for
nonstationary precipitation extremes in Northern California conditioned on tropical moisture
exports, Water Resources Research, 51(3), 1472-1492.
Stern, R.D. and Coe, R. (1984), A Model Fitting Analysis of Daily Rainfall Data, Journal of
the Royal Statistical Society. Series A (General), 147(1), 1-34.
Thyer, M. and Kuczera, G. (2003a), A hidden Markov model for modelling long-term
persistence in multi-site rainfall time series 1. Model calibration using a Bayesian approach
Journal of Hydrology, 275(1-2), 12-26.
Thyer, M. and Kuczera, G. (2003b), A hidden Markov model for modelling long-term
persistence in multi-site rainfall time series. 2. Real data analysis Journal of Hydrology,
275(1-2), 27-48.
Viney, N.R. and Bates, B.C. (2004), It never rains on Sunday: The prevalence and
implications of untagged multi-day rainfall accumulations in the Australian high quality data
set, International Journal of Climatology, 24: 1171-1192.
178
Wallis, T.W.R. and Griffiths, J.F. (1997), Simulated meteorological input for agricultural
models, Agricultural and Forest Meteorology, 88(1-4), 241-258.
Wang, Q. J. and Nathan, R.J. (2007), A method for coupling daily and monthly time scales in
stochastic generation of rainfall series, Journal of Hydrology, 346(3-4), 122-130.
Waymire, E. and Gupta, V.K. (1981a), The mathematical structure of rainfall representations:
1. A review of the stochastic rainfall models, Water Resources Research, 17(5), 1261-1272.
Waymire, E., and Gupta, V.K. (1981b), The mathematical structure of rainfall
representations: 2. A review of the theory of point processes, Water Resources Research,
17(5), 1273-1285.
Westra, S., Mehrotra, R., Sharma, A. and Srikanthan, R. (2012), Continuous rainfall
simulation: 1 - A regionalised sub-daily disaggregation approach, Water Resources
Research, 48(1), W01535.
Westra, S., Fowler, H.J., Evans, J.P., Alexander, L.V., Berg, P., Johnson, F., Kendon, E.J.,
Lenderink, g. and Roberts, N.M. (2014), Future changes to the intensity and frequency of
short-duration extreme rainfall, Rev Geophys, 52.
Westra, S., Evans, J.P., Mehrotra, R. and Sharma, A. (2013), A conditional disaggregation
Hydrology, 479: 86-99.
Wheater, H. S., Isham, V.S., Cox, D.R., Chandler, R.E., Kakou, A., Northrop, P.J., Oh, L.,
Onof, C. and Rodriguez-Iturbe, I. (2000), Spatial-temporal rainfall fields: modelling and
statistical aspects, Hydrology and Earth System Sciences, 4(4), 581-601.
Wilby, R. L., Wigley, T.M.L., Conway, D., Jones, P.D., Hewitson, B.C., Main, J. and Wilks,
D.S. (1998), Statistical Downscaling of General Circulation Model Output - a Comparison of
Methods, Water Resources Research, 34(11), 2995-3008.
Wilks, D.S. (1998), Multisite generalization of a daily stochastic precipitation generation

model, Journal of Hydrology, 210: 178-191.
Wilks, D.S. (1999a), Multisite downscaling of daily precipitation with a stochastic weather
generator, Climate Research, 11(2), 125-136.
Wilks, D.S. (1999b), Simultaneous stochastic simulation of daily precipitation, temperature

and solar radiation at multiple sites in complex terrain, Agricultural and Forest Meteorology,
96(1-3), 85-101.
Woolhiser, D.A. (1992), Modeling daily precipitation - progress and problems, in Statistics in
the environmental and earth sciences, edited by A.T. Walden and P. Guttorp, p: 306, Edward
Arnold, London, U.K.
Woolhiser, D.A. and Pegram, G.G.S (1979), Maximum Likelihood Estimation of Fourier
Coefficients to Describe Seasonal Variations of Parameters in Stochastic Daily Precipitation
Models, Journal of Applied Meteorology, 18(1), 34-42.
Woolhiser, D.A. and Roldán J. (1982), Stochastic daily precipitation models: 2. A comparison
of distributions of amounts, Water Resources Research, 18(5), 1461-1468.
179
Woolhiser, D.A. and Roldán, J. (1986), Seasonal and Regional Variability of Parameters for
Stochastic Daily Precipitation Models: South Dakota, U.S.A, Water Resources Research,
22(6), 965-978.
World Meteorological Organisation (WMO) (1994), WMO Guide to Hydrological Practices,

World Meteorological Organisation, Geneva, p: 770.
de Lima, M.I.P. and Grasman, J. (1999), Multifractal analysis of 15-min and daily rainfall from
a semi-arid region in Portugal, Journal of Hydrology, 220(1-2), 1-11.
180
11 national Circuit
BARTON ACT 2600
e | arr_admin@arr.org.au
www.arr.org.au

ARR 190514 Book2

Uploaded by

Copyright:

Available Formats

ARR 190514 Book2

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

ARR 190514 Book2

Uploaded by

Copyright:

Available Formats

A GUIDE TO

BOOK 2 - RAINFALL ESTIMATION

Please give attribution to: © Commonwealth of Australia (Geoscience Australia) 2019.

How to reference this book:

How to reference Book 9: Runoff in Urban Areas:

One of the major responsibilities of the National Committee on Water Engineering of

ARR Technical Committee:

Chair: Mark Babister

ARR Editorial Team:

Associate Editors for Book 9 - Runoff in Urban Areas

Editorial assistance: Mikayla Ward

What is new in ARR 2019?

Geoscience Australia, on behalf of the Australian Government, asked the National

Update ARR 2016 ARR 2019

Book 9 Available as “rough” draft Peer reviewed and completed

Guideline Epub version Following practitioner feedback, a pdf version of ARR

User Limited functionality in web-based version Additional pdf format available

As of May 2019, this version is considered to be final.

3.8. Uncertainty in Design Rainfalls ............................................................................. 49

5.9.8. Region Considerations .............................................................................. 96

7.4.4. Modification of Generated Design Rainfall Attributes .............................. 163

2.5.7. Temporal Pattern Regions ......................................................................................... 81

1.1. Scope and Intent

1.2. Application of these Guidelines

1.3. Climate Change

Chapter Status Final

2.2. Space-Time Representation of Rainfall Events

Figure 2.2.1. Conceptual Diagram of Space-Time Pattern of Rainfall

However, there are often hydrometeorological drivers, as discussed in Book 2, Chapter 4

2.3. Orographic Enhancement and Rain Shadow Effects on

• Upward deflection of large scale horizontal flow by the topography; or

• Anabatic or upward vertical propagation of moist air up an orographic slope caused by

2.4. Conceptualisation of Design Rainfall Events

• Duration of storm or burst event;

• Temporal distribution (or pattern) of rainfall during the event.

2.4.1. Event Definitions

2.4.2. Rainfall Event Duration

2.4.3. Event Rainfall Depth (or Average Intensity)

2.4.4. Temporal Patterns of Rainfall

2.4.5. Spatial Patterns of Rainfall

In the following, a number of different approaches to model the space-time characteristics of

2.5. Spatial and Temporal Resolution of Design Rainfall

Variability in hydrographs introduced by space-time variability in rainfall will be accentuated

• significant variations in soil type;

• significant variations in vegetation type, such as forested and cleared areas;

• significant variations in stream channel and floodplain roughness;

• significant variations in slope of stream channels and floodplains;

• variations in elevation, snowpack depth, density and temperature in those catchments

2.6. Applications Where Flood Estimates are Required at

2.7. Climate Change Impacts

Woods and Sivapalan (1999), A synthesis of space-time variability in storm response:

3.2. Design Rainfall Concepts

Table 2.3.1. Classes of Design Rainfalls

Figure 2.3.1. Classes of Design Rainfalls

3.3. Climate Change Impacts

3.4. Frequent and Infrequent Design Rainfalls

3.4.2. Rainfall Database

3.4.2.1. Types of Rainfall Data

3.4.2.1.1. Daily Read Rainfall Gauges