0% found this document useful (0 votes)

8 views87 pages

WRC Deliverable 1.1 IL AK

Uploaded by

Jaco Mouton

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as DOCX, PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

8 views87 pages

WRC Deliverable 1.1 IL AK

Uploaded by

Jaco Mouton

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as DOCX, PDF, TXT or read online on Scribd

You are on page 1/ 87

DEVELOPMENT OF APPROPRIATE

SYNTHETIC DESIGN STORMS FOR

SMALL CATCHMENTS IN SOUTH AFRICA
Deliverable No. 1

Inception Report
to the Water Research Commission

Mouton J.v.S.1, Spooner J.D.1, Kruger A.2, Smithers J.C.3, Loots I.1

1
University of Pretoria,
2
South African Weather Service
3
University of KwaZulu-Natal
Contact details of author for correspondence
[email protected]

Project No. 2023/2024-01295

June 2023
DISCLAIMER

This report has been reviewed by the Water Research Commission (WRC) and
approved for publication. Approval does not signify that the contents necessarily reflect
the views and policies of the WRC, nor does mention of trade names or commercial
products constitute endorsement or recommendation for use.
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
EXECUTIVE SUMMARY

BACKGROUND

…………..

AIMS

……………

________________________________________________________________________
3
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

ACKNOWLEDGEMENTS

CONTENTS
EXECUTIVE SUMMARY................................................................................................................ii

ACKNOWLEDGEMENTS..............................................................................................................iii

CONTENTS.......................................................................................................................................iv

LIST OF FIGURES...........................................................................................................................vi

LIST OF TABLES...........................................................................................................................vii

ACRONYMS & ABBREVIATIONS............................................................................................viii

CHAPTER 1: BACKGROUND.............................................................................................1

1.1 INTRODUCTION......................................................................................................... 1
1.2 PROJECT AIMS.......................................................................................................... 2
1.3 REPORT STRUCTURE...............................................................................................2

CHAPTER 2: LITERATURE REVIEW...............................................................................4

2.1 INTRODUCTION......................................................................................................... 4
2.2 RAINFALL DATA QUALITY.........................................................................................5
2.2.1 Discordancy measure..........................................................................................6
2.2.2 Graphical methods.............................................................................................. 7
2.2.3 Rainfall data quality control.................................................................................8
2.2.3.1 Tipping rates.........................................................................................9
2.2.3.2 Identification of inaccurate accumulated totals...................................10
2.2.3.3 Identification of excessively long dry spells........................................10
2.2.3.4 Quality control flags............................................................................10
2.2.3.5 Global sub-daily rainfall quality control procedure..............................12
2.3 REGIONALISATION..................................................................................................16
2.3.1 Kӧppen climate classification system................................................................16
2.3.2 South African Weather Service system of climate regions................................18
2.3.3 Short duration design rainfall homogeneous regions........................................19
2.4 DATA INTERPRETATION.........................................................................................21
2.4.1 Frequency distribution analysis.........................................................................22
2.4.2 Minimum data period.........................................................................................26
2.4.3 Missing data...................................................................................................... 27
2.5 SPATIAL RAINFALL INTERPOLATION....................................................................28
2.5.1 Arithmetic mean.................................................................................................29
2.5.2 Normal-ratio.......................................................................................................29
2.5.3 Triangulation......................................................................................................30

________________________________________________________________________
5
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
2.5.4 Nearest neighbour.............................................................................................31
2.5.5 Natural neighbour..............................................................................................32
2.5.6 Inverse distance weighting................................................................................32
2.5.7 Kriging............................................................................................................... 33
2.5.7.1 Variogram...........................................................................................35
2.5.7.2 Neighbourhood...................................................................................36
2.5.8 Co-kriging.......................................................................................................... 37
2.6 STOCHASTIC RAINFALL GENERATION.................................................................37
2.6.1 Monte-Carlo simulations....................................................................................37
2.6.2 Bartlett-Lewis rectangular pulse models............................................................40
2.7 SYNTHETIC DESIGN STORM PARAMETERIZATION.............................................41
2.7.1 Background....................................................................................................... 41
2.7.2 Current Performance of Synthetic Design Storms.............................................43
2.7.2.1 Rectangular hyetograph.....................................................................43
2.7.2.2 Triangular hyetograph.........................................................................43
2.7.2.3 SCS and the SCS-SA method............................................................43
2.7.2.4 Huff curves..........................................................................................45
2.7.2.5 Chicago Design Storm (CDS).............................................................45
2.7.3 Sensitivity of Hydrological Models to Changes in Synthetic Design Storm
Parameters........................................................................................................46
2.7.3.1 Storm Duration....................................................................................47
2.7.3.2 Position of Peak Intensity...................................................................48
2.7.3.3 Time Step...........................................................................................50
2.7.4 Comparison between Single Event-Based and Continuous Simulation Modelling
.......................................................................................................................... 50
2.7.5 Assessment of an Ensemble Approach to Single Event-Based Modelling........52

CHAPTER 3: INCEPTION REPORT................................................................................53

3.1 OBJECTIVE............................................................................................................... 53
3.2 SCOPE AND LIMITATIONS...................................................................................... 53
3.3 PROPOSED METHODOLOGY.................................................................................53
3.3.1 Data quality assessment...................................................................................53
3.3.2 Design rainfall estimation.................................................................................. 55
3.3.3 Synthetic design storm evaluation.....................................................................55
3.3.4 Synthetic design storm evaluation tool..............................................................55
3.3.5 Knowledge dissemination workshop..................................................................55
3.4 TASKS AND DELIVERABLES...................................................................................56
3.5 CAPACITY BUILDING...............................................................................................57

CHAPTER 4: REFERENCES..............................................................................................58

EXECUTIVE SUMMARY................................................................................................................ii

________________________________________________________________________
6
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
ACKNOWLEDGEMENTS..............................................................................................................iii

CONTENTS.......................................................................................................................................iv

LIST OF FIGURES...........................................................................................................................vi

LIST OF TABLES...........................................................................................................................vii

ACRONYMS & ABBREVIATIONS............................................................................................viii

CHAPTER 1: BACKGROUND.............................................................................................1

CHAPTER 2: LITERATURE REVIEW...............................................................................8

2.1 INTRODUCTION......................................................................................................... 8
2.2 RAINFALL DATA QUALITY.........................................................................................9
2.2.1 Discordancy measure..........................................................................................9
2.2.2 Graphical methods............................................................................................ 11
2.2.3 Rainfall data quality control............................................................................... 13
2.2.3.1 Tipping rates.......................................................................................13
2.2.3.2 Identification of inaccurate accumulated totals...................................14
2.2.3.3 Identification of excessively long dry spells........................................14
2.2.3.4 Quality control flags............................................................................15
2.2.3.5 Global sub-daily rainfall quality control procedure..............................16
2.3 REGIONALISATION..................................................................................................20
2.3.1 Kӧppen climate classification system................................................................21
2.3.2 South African Weather Service system of climate regions................................22
2.3.3 Short duration design rainfall homogeneous regions........................................24
2.4 DATA INTERPRETATION.........................................................................................26
2.4.1 Frequency distribution analysis.........................................................................26
2.4.2 Minimum data period.........................................................................................30
2.4.3 Missing data...................................................................................................... 31
2.5 SPATIAL RAINFALL INTERPOLATION....................................................................32
2.5.1 Arithmetic mean.................................................................................................32
2.5.2 Normal-ratio.......................................................................................................33
2.5.3 Nearest neighbour.............................................................................................33
2.5.4 Natural neighbour..............................................................................................34
2.5.5 Triangulation......................................................................................................34
2.5.6 Inverse distance weighting................................................................................34
2.5.7 Kriging............................................................................................................... 35

________________________________________________________________________
7
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
2.5.7.1 Variogram...........................................................................................37
2.5.7.2 Neighbourhood...................................................................................38
2.5.8 Co-kriging.......................................................................................................... 39
2.6 STOCHASTIC RAINFALL GENERATION.................................................................39
2.6.1 Monte-Carlo simulations....................................................................................39
2.6.2 Bartlett-Lewis rectangular pulse models............................................................42
2.7 SYNTHETIC DESIGN STORM PARAMETERIZATION.............................................43
2.7.1 Background....................................................................................................... 43
2.7.2 Current Performance of Synthetic Design Storms.............................................45
2.7.2.1 Rectangular hyetograph.....................................................................45
2.7.2.2 Triangular hyetograph.........................................................................45
2.7.2.3 SCS and the SCS-SA method............................................................45
2.7.2.4 Huff curves..........................................................................................47
2.7.2.5 Chicago Design Storm (CDS).............................................................47
2.7.3 Sensitivity of Hydrological Models to Changes in Synthetic Design Storm
Parameters........................................................................................................48
2.7.3.1 Storm Duration....................................................................................49
2.7.3.2 Position of Peak Intensity...................................................................50
2.7.3.3 Time Step...........................................................................................52
2.7.4 Comparison between Single Event-Based and Continuous Simulation Modelling
.......................................................................................................................... 52
2.7.5 Assessment of an Ensemble Approach to Single Event-Based Modelling........54

CHAPTER 3: INCEPTION REPORT................................................................................56

3.1 OBJECTIVE............................................................................................................... 56
3.2 SCOPE AND LIMITATIONS...................................................................................... 56
3.3 PROPOSED METHODOLOGY.................................................................................56
3.3.1 Data quality assessment...................................................................................56
3.3.2 Design rainfall estimation.................................................................................. 58
3.3.3 Synthetic design storm evaluation.....................................................................58
3.3.4 Synthetic design storm evaluation tool..............................................................58
3.3.5 Knowledge dissemination workshop..................................................................58
3.4 TASKS AND DELIVERABLES...................................................................................59
3.5 CAPACITY BUILDING...............................................................................................60

CHAPTER 4: REFERENCES..............................................................................................62

________________________________________________________________________
8
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

LIST OF FIGURES

Figure 2.1: Categorization of synthetic design storm methods (Mouton et al., 2022)...........4
Figure 2.2: Double mass plot of daily rainfall for selected stations in the Cedara
catchments for the period October 1988 – September 1989 (Smithers and Schulze, 2000)8
Figure 2.3: Annual rainfall comparison for the Lanseria rainfall station with Irene and OR
Tambo stations (Mouton, 2023)........................................................................................... 8
Figure 2.4: The Kӧppen climatic classification of South Africa (Kruger and Mbatha, 2021)
........................................................................................................................................... 17
Figure 2.5: SAWS climatic regions of South Africa (Kruger and Mbatha, 2021)................19
Figure 2.6: Distribution of 15 clusters of relatively homogeneous extreme short duration (#
24 h) rainfall in South Africa (Smithers and Schulze, 2000)...............................................21
Figure 2.7: Typical Triangular Irregular Network (after Huisman and de By, 2009)...........31
Figure 2.8: Typical construction of Thiessen polygons (after Huisman and de By, 2009)..31
Figure 2.9: Typical Natural Neighbour................................................................................32
Figure 2.10: Tolerance angles and distance tolerances (Mert and Dag, 2017)..................37
Figure 2.11: Illustration of the inverse transformation method (Nathan and Weinmann,
2013).................................................................................................................................. 38
Figure 2.12: Flow chart for Monte-Carlo simulation (Kottegoda et al. 2014)......................40
Figure 2.13: Sensitivity of Peak Discharge to changes in Storm Duration for the Chicago
Design Storm (Baiti et al., 2017)........................................................................................ 47
Figure 2.14: Sensitivity of Peak Discharge to changes in Peak Position for the Chicago
Design Storm (Baiti et al., 2017)........................................................................................ 49
Figure 3.1: Data processing flow chart...............................................................................54
Figure 2.1: Categorization of synthetic design storm methods (Mouton et al., 2022)...........9
Figure 2.2: Double mass plot of daily rainfall for selected stations in the Cedara
catchments for the period October 1988 – September 1989 (Smithers and Schulze, 2000)
........................................................................................................................................... 12
Figure 2.3: Annual rainfall comparison for the Lanseria rainfall station with Irene and OR
Tambo stations (Mouton, 2023)......................................................................................... 12
Figure 2.4: The Kӧppen climatic classification of South Africa (Kruger and Mbatha, 2021)
........................................................................................................................................... 22
Figure 2.5: SAWS climatic regions of South Africa (Kruger and Mbatha, 2021)................23

________________________________________________________________________
9
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Figure 2.6: Distribution of 15 clusters of relatively homogeneous extreme short duration (#
24 h) rainfall in South Africa (Smithers and Schulze, 2000)...............................................25
Figure 2.6: Typical Triangular Irregular Network (after Huisman and de By, 2009)...........34
Figure 2.7: Tolerance angles and distance tolerances (Mert and Dag, 2017)....................39
Figure 2.8: Illustration of the inverse transformation method (Nathan and Weinmann, 2013)
........................................................................................................................................... 40
Figure 2.9: Flow chart for Monte-Carlo simulation (Kottegoda et al. 2014)........................42
Figure 2.10: Sensitivity of Peak Discharge to changes in Storm Duration for the Chicago
Design Storm (Baiti et al., 2017)........................................................................................ 49
Figure 2.11: Sensitivity of Peak Discharge to changes in Peak Position for the Chicago
Design Storm (Baiti et al., 2017)........................................................................................ 51
Figure 3.1: Data processing flow chart...............................................................................57

________________________________________________________________________
10
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

LIST OF TABLES

Table 1.1. Project aims........................................................................................................2

Table 2.1: Critical values for the discordancy statistic Di (Hosking and Wallis, 1997).........7
Table 2.2: Summary of automated quality control flags (Blenkinsop et al, 2016; Lewis et
al., 2018)............................................................................................................................ 10
Table 2.3 Summary of automated quality control flags of the global sub-daily rainfall quality
control procedure (Lewis et al. 2021).................................................................................12
Table 2.4: Summary of previous findings of most suitable probability distribution in design
rainfall estimation (after Mamoon and Rahman, 2017)......................................................24
Table 2.5: Abbreviations used for probability distributions.................................................25
Table 3.1. Work completed to date................................................................................... 56
Table 3.2. Deliverable due dates and deliverables submitted to date...............................56
Table 3.3. Registered students.......................................................................................... 57
Table 1.1. Project aims........................................................................................................2
Table 2.1: Critical values for the discordancy statistic Di (Hosking and Wallis, 1997).......11
Table 2.2: Summary of automated quality control flags (Blenkinsop et al, 2016; Lewis et
al., 2018)............................................................................................................................ 15
Table 2.3 Summary of automated quality control flags of the global sub-daily rainfall quality
control procedure (Lewis et al. 2021).................................................................................17
Table 2.4: Summary of previous findings of most suitable probability distribution in design
rainfall estimation (after Mamoon and Rahman, 2017)......................................................28
Table 3.1. Work completed to date................................................................................... 59
Table 3.2. Deliverable due dates and deliverables submitted to date...............................59
Table 3.3. Registered students.......................................................................................... 60

________________________________________________________________________
11
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

ACRONYMS & ABBREVIATIONS

Acronym Description
ABS Absolute relative error
Average Intermediate Curve per Recurrence
AIC_T
Interval
AIC_t Average Intermediate Curve per time step
AMC Antecedent Moisture Content
AMS Annual Maximum Series
ARD Average Relative Difference (%)
ARS Automatic Rainfall Station
AWS Automatic Weather Station
CDS Chicago Design Storm
CTMM City of Tshwane Metropolitan Municipality
D Total rainfall volume (mm)

DDF Depth Duration Frequency

DRESA Design Rainfall Estimation in South Africa

EPA Environmental Protection Agency

GEV General Extreme Value
GOF Goodness-of-Fit
GRG Generalised Reduced Gradient
i Storm intensity (mm/hour)
ia Intensity after the peak intensity (mm/hour)
iav Average intensity (mm/hour)
ib Intensity before the peak intensity (mm/hour)
IC Intermediate Curve
ip Peak intensity (mm/hour)
IDF Intensity Duration Frequency
IDW Inverse Distance Weighting
MAP Mean annual precipitation (mm)

________________________________________________________________________
12
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
MARE Mean Absolute Relative Error
MDP Maximum Dry Period (min)
MRD Minimum Rainfall Depth (mm)
MRI Minimum Rainfall Intensity (mm/hour)
MS Microsoft
NOAA National Oceanic and Atmospheric Administration
NRCS National Resources Conservation Service
P Total rainfall depth (mm)
Cumulative rainfall depth after the peak intensity
Pa
(mm)
Cumulative rainfall depth before the peak
Pb
intensity (mm)
PD Probability Distribution
Storm advancement coefficient. The ratio of the
r storm duration of the peak intensity relative to the
total storm duration
RE Relative Error (%)
RD Relative Difference (%)
RI Recurrence Interval (1:year)
RIM Recurrence Interval Maximum
RIR Recurrence Interval Ratio
RMSE Root Mean Square Error
R2 Coefficient of determination
SAWS South African Weather Service
SCS Soil Conservation Services
SCS curves adapted for Southern African
SCS-SA
conditions
SWMM Stormwater Management Model
ta Storm duration after the peak intensity (min)
tb Storm duration before the peak intensity (min)
Tc Critical storm duration (hour)
Td Total storm duration (min)
Tp Time when peak intensity occurs (min)
USA United States of America

________________________________________________________________________
13
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
USDA United States Department of Agriculture
WO Weather Office

CHAPTER 1: BACKGROUND

1.1 INTRODUCTION

Civilization is unequivocally reliant on the effective management of stormwater runoff. This

efficiency is initiated during the design of the stormwater infrastructure, which requires
relevant experience, adequate resources, and reliable information. The Reliable input data
is crucial during design,required to accurately design infrastructure need to be reliable in
order to limit the risk of failure, the risk to loss of life, or over-expenditure on hydraulic
structures (Johnson et al., 2021). The social and economic benefit provided by the
completed infrastructure to society, is balanced against the intrinsic, but acceptable, risk of
failure. Because the balance between risk and benefit could be compromised by unreliable
information, the consequences of failure might not be anticipated and the impact on local
economies, the environment, and society could be disastrous.

The uncertainty in hydrological modelling originates at instrumental and human errors (Al
Mamoon and Rahman, 2014). The sSampling uncertainties, caused by a short data period
and missing data, further reduces the reliability and quality of the observed data. The
spatial and temporal variability in rainfall, the degree of homogeneity of an assumed
homogeneous region, the choice of probability distribution model, and the assumingly
stationary data which are challenged by climate change, adds further layers of uncertainty
(Al Mamoon and Rahman, 2014). Significant deviations were found in the annual rainfall of
some stations in Gauteng compared to reliable stations, but the cause of these deviations
was uncertain, and the quality of their data were believed to be unreliable for design
rainfall estimation (Mouton, 2023). The impact of inaccuracies was also investigated at a
daily timescale by Singh (2021), but the impact of these errors on short duration rainfall
data is uncertain.

The cascading uncertainties associated with design rainfall estimation are carried forward
and are adopted into the synthetic design storm generation process on which single event
event-based simulation modelling depends on. The parameterization of the synthetic
design storm is also contributingcontributes to the already compounded list of
uncertainties. The timing of the peak intensity and the storm duration parameters of a
synthetic design storm could impact the peak discharge on which the infrastructure design
is based. A synthetic design storm with a duration of two hours would generally be
sufficient to exceed the longest time of concentration of a typical urban catchment
(Watson, 1981), but the effect of different durations on the peak discharge is uncertain.
The rapid decline of infiltration rates of certain soil types during the early part of a
rainstorm event could be contributingcontribute to this uncertainty. A representative storm
advancement coefficient was determined for Gauteng using the methodology proposed by
________________________________________________________________________
1
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Keifer and Chu (1957),, which considerings only portions with the highest rainfall in an
event (Mouton et al., 2022). However, if the entire event is considered, a much smaller
coefficient was found to be better representative of observed rainstorm events.

Hence, a more decisive guideline for the selection of a storm duration and an
advancement coefficient for a synthetic design storm is required. Rainfall distribution
curves which are intermediary to the standard SCS-SA curves, were developed in order to
generate more accurate synthetic design storms (Mouton et al., 2022). The regionalisation
was based on geographical interpolation using the Inverse Distance Weighting method,
but the applicability of the method as well as an appropriate power coefficient required by
the method was uncertain.

As a result of these accumulating uncertainties in hydrological modelling, this study seeks

to investigate the reliability of the available short duration rainfall data in South Africa, the
parameterization of synthetic design storms applicable to single event event-based
simulation modelling, as well as the regionalisation of intermediate SCS-SA rainfall
distribution curves for South Africa.

1.2 PROJECT AIMS

The following are the aims of the project:specific aims of the project are described in Table
1.1.

Table 1.11.2. Project aims.

No. Aim
1 To assess the quality of the currently available short duration and daily rainfall
data, improve the data sets where possible and develop a classification criterion
with relevant thresholds for selection of stations to use in further analyses.
2 To assess the performance of selected methods to estimate synthetic design
storms applicable to all South African regions with available data (Sensitivity
analysis).
3 To develop, assess and investigate the regionalisation of improved synthetic
design storms applicable to small catchments in South Africa.
4 To develop a user-friendly tool that provides the user with an opportunity to
determine a synthetic design storm for a specific location or region.
5 To disseminate the information to managers, designers and technicians involved in
planning and design of urban stormwater systems and in small rural catchments.

________________________________________________________________________
2
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1.3 REPORT STRUCTURE

This document consists of the following chapters thus far: Chapter 1 provides an
introduction and the aims of this study; Chapter 2 contains a detailed literature review
pertaining to the quality assessment of rainfall data, geographical interpolations, and
synthetic design storm parameterization; Chapter 3 contains the inception report that
describes the proposed project methodology; and Chapter 3 4 contains the references
used to understand the topic and plan the methodological approach.

[1.4] SCOPE AND LIMITATIONS

The observed rainfall data collection was limited to automatic recording rainfall stations
with 5-min interval measurements operated by the SAWS within the boundaries of South
Africa. Even though many synthetic design storm methods are available, this study
focusses on the two methods that were found appropriate for single even based simulation
modelling during a pilot study in Gauteng, namely the Chicago Design Storm method and
the SCS-SA rainfall distribution curves (Mouton et al. 2022).

[1.5] TASKS AND DELIVERABLES

Table 1.3 summarises the work done to date. Table 1.4 shows the list of deliverables and
indicates the status of each deliverable.

Table 1.3. Work completed to date.

No. Task Summary of work to date
1 Literature review This deliverable describes the literature review on rainfall
data quality, geographical interpolation methods, and
synthetic design storm parameterization.

Table 1.4. Deliverable due dates and deliverables submitted to date.

No. Deliverable Status Due date
1 Literature review and inception report Completed 31 Jul 2023
2 Annual progress report 2023/2024 Outstanding 28 Feb 2024
3 Data collection and quality assessment Outstanding 30 Jun 2024
4 Annual progress report 2024/2025 Outstanding 30 Nov 2024
5 Synthetic design storm evaluation Outstanding 31 Jul 2025
6 Knowledge dissemination workshop Outstanding 31 Jan 2026
7 Final report Outstanding 31 Jan 2026

The complete list of tasks required to meet these deliverables include:

[a)] Literature review

[i)] Data quality control
[ii)] Data interpretation

________________________________________________________________________
4
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
[iii)] Geographical interpolation (SCS-SA, possibly IDF regression coefficients –
CDS, Inverse Distance Weighting, Kriging, linear)
[iv)] Synthetic design storm parameterization
[b)] Data quality assessment
[i)] Data capturing - human and instrumental errors.
[ii)] Reprocessing database – flag inconsistent data, climatological review.
[c)] Data interpretation
[i)] Data period.
[ii)] Missing data.
[iii)] AMS, n-largest approach.
[iv)] Monte Carlo – confidence levels, extrapolation
[d)] Design rainfall
[i)] At-site (GEV LM)
[ii)] DRESA (Relative differences, motivation for DRESA overhaul)
[iii)] Intermediate SCS-SA curves (IC values)
[iv)] Possible regionalised IDF regression coefficients (CDS)
[e)] Synthetic design storm parameterization
[i)] Storm duration.
[ii)] timing of peak intensity.
[iii)] infiltration models.
[iv)] catchment characteristics – size, slope, materials.
[v)] Peak discharge, runoff volume.
[vi)] Reaction times.
[vii)] Guideline for parameterization.
[f)] Synthetic design storm comparisons
[i)] Shape
[ii)] Intensity
[iii)] RI variations

Quality Control
(5-min Rainfall)

Climatological
Internal review (Replace
consistency with missing data)

Tipping rates Data patching

Accumulated Stochastic
totals simulations

Dry spells Probability

distribution

Regional
consistency

Discordancy Intermediate
measure SCS-SA curves

CDS
Dry period
coefficients

Accumulated Synthetic
daily rainfall design storms

Figure 1.1: Data processing flow chart

Capacity building on this project includes two students registered currently for a topic
directly associated with the project title and one student registered with a topic in a related
field. A prospective student is expected to be register in 2024. The details of the three
registered students and the one prospective student are provided in Table 1.5.

Table 1.5. Registered students.

Surname. Initials Gender Race Degree Institution

Registered students:
Loots I Female White PhD UKZN
Spooner JD Male White MEng UP
??? ??? ??? ??? ??? ???

Prospective student:
Mouton JvS Male White PhD UP

Knowledge dissemination is planned to take place in the form of a workshop. The results
will also be presented at ……….

[CHAPTER 2:] LITERATURE REVIEW

[2.1] INTRODUCTION

Civil Engineers are concerned with urban stormwater and flood management that is
associated with rapid urbanisation that is occurring globally (Zhang, 2016). Computer-
aided stormwater rainfall-runoff models, such as the US Environmental Protection
Agency’s Stormwater Management Model (SWMM), are popular tools used in urban flood
forecasting and stormwater design around the world (Rossman, 2010; Rabori and
Ghazavi, 2018; Jiang, Chen and Wang, 2015) and in South Africa (City of Cape Town
Development Service, 2002; Barnard et al., 2019).

Hydrological models, like SWMM, require the generation of a rainfall hyetograph that is
routed through the model (Prodanovic and Simonovic, 2004). This hyetograph can either
be obtained from historical rainfall data, or by generating a synthetic design storm for a
given storm duration and peak rainfall intensity that captures the rainfall features of a
particular locality (Te Chow, Maidment and Mays, 1988). In South Africa, continuous and
reliable rainfall data is increasingly hard to come by and so the use of synthetic design
storms are essential (Mouton, Loots and Smithers, 2022).

The design storm concept has been around for nearly 100 years, and by the early 1930s
the association of frequencies with rainfall intensities and the derivation of intensity-
duration-frequency (IDF) curves from historical rainfall records had already become well
known. The combination of the Rational Formula with the IDF curves led to the
development of the rectangular hyetograph with a constant intensity. This became the first
“design storm” and is still used today (Adams and Howard, 1986). Today there are many
types of synthetic design storms that are available to use, with synthetic design storms
being broadly divided into three categories, namely Intensity Duration Frequency (IDF)
based methods, standardised mass curves, and stochastic based methods, as
summarised by Mouton et al. (2022) (Figure 2.2). The process of generating an IDF IDF-
based synthetic design storm is but the last step in creating an artificial hyetograph for a
dynamic rainfall-runoff simulation model. This process is preceded by several other
processes associated with the development of IDFs. The first process is the measuring of
the actual rainfall using, for example, a Tipping Bucket Rain Gauge (TBRG). Although a
TBRG must be subjected to regular calibration and maintenance to ensure that the
recorded information is an accurate representation of the true rainfall depth, this study will
not go into any further details about the accuracy of a TBRG.

________________________________________________________________________
9
Stochastically based
Considerable number of models not covered in this study
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Figure 2.2: Categorization of synthetic design storm methods (Mouton et al., 2022)

The next process entails the reviewing of the recorded information by a professional
climatologist. and iIf the rainfall is consistent with other climatological information, the
information is stored in the observed rainfall data base. Even though this process is meant
to filter out all discrepancies in the data, doubt about the reliability of the observed rainfall
data because still exists due to the possibility of the human error still exists. This could
result in inaccurately high observed rainfall values or rainfall recorded as zero values
instead of missing data. Even though this study will also not focus on this process in any
further detail, it does recognise that uncertainties that originates at these processes, could
also manifest itself themselves in a synthetic design storm. Therefore, this literature review
will consider, among other topics, methods to identify inconsistencies in the observed
rainfall data that could have originated during the rainfall measuring and the data reviewing
processes. Once the inconsistent periods within stations are identified, the observed
rainfall data could be changed to missing data which could then be dealt with accordingly.

Uncertainties about sampling are also affecting the reliability of an IDF curve because the
estimates of higher order moments become unstable, especially if high outliers are present
(Al Mamoon and Rahman, 2014). This includes the uncertainties about the consequences
of the data period as well as the amount of missing data within the observed rainfall data
base. The effects of missing data on the results of a Probability Distribution (PD) analysis
in terms of the amount that is allowable are uncertain. The literature will consider the
minimum data period required to conduct a Probability Distribution (PD) analysis, as well
as methods to improve the reliability of results obtained from a short data period. The
literature review will also consider the topic of missing data and methods of replacing
missing data with estimated values (imputation). Further uncertainties include the
underlying distribution of the sample and the degree of heterogeneity of an assumed
homogeneous region.

The final uncertainty this literature review will consider is the parameterization of a
synthetic design storm and in particular the effect it has on the results of a dynamic
rainfall-runoff simulation model. The two parameters which are of interest include (a) the
duration of a storm event and (b) the timing of the peak intensity.

Rainfall data may contain substantial errors, which could be due to the incorrect recording
of the data or due to circumstantial changes in the collection of the data (Smithers and
Schulze, 2000). Therefore, the screening of the datadata screening is necessary to identify
stations with gross data errors in their data. To establish a data base of high quality rainfall
data on which further research will be conducted, methods to identify rainfall stations with
unreliable data sets are discussed in this section.

1.4.1[2.2.1] Discordancy measure

Hosking and Wallis (1997) proposed the use of the discordancy measure to flag stations
with inconsistent data in a group of stations. The discordancy measure is calculated using
the L-moments from each station. Once a station with inconsistent data was is identified,
then the data can be further scrutinised. Smithers and Schulze (2000) used this method to
screen the data of two research catchments, consisting of 10 and 12 rainfall stations,
respectively. The discordancy measure was determined for the 16 standard time steps and
the results indicated that the 10-min time step of one of the rainfall stations was discordant
from the rest of the stations. The results were verified graphically by comparing the 10-min
AMS of all the stations, and also by conducting accumulated daily rainfall plots. Based on
this finding, it was concluded that the discordancy measure was an effective way of
detecting discrepancies in the data. The discordancy measure was also applied
successfully in many other international studies … Therefore, it was considered to be
appropriate for this study. The discordancy measure is calculated using Equations 2.1 to
2.4 as follows:

ui = [ ❑ ❑4 ]
(i ) (i ) ( i) T
❑3 [ 2.1 ]
where:
ui = column vector matrix of L-moment ratios of station i ,
T = transposition of a vector matrix,
τ = L-CV (Coefficient of L-variation),
τ3 = L-skewness, and
τ4 = L-kurtosis.

N
u=N −1
∑ ui [ 2.2 ]
i=1

where:
u = unweighted group average, and
N = number of stations.

where:
A = Matrix (3 x 3) of sums of squares and cross-products.

1 T −1
D i= N ( ui −u ) A ( ui −u ) [ 2.4 ]
3
where:
Di = Discordancy measure for station i .

Hosking and Wallis (1997) have defined a station to be discordant if Di exceeds the critical
values summarised in Table 2.6.

Table 2.62.7: Critical values for the discordancy statistic Di (Hosking and Wallis, 1997)
Number of sites in region Critical value
5 1.333
6 1.648
7 1.917
8 2.140
9 2.329
10 2.491
11 2.632
12 2.757
13 2.869
14 2.971
≥15 3.000

1.4.2[2.2.2] Graphical methods

One graphical method is the double mass plot of daily rainfall. This was successfully used
by Smithers and Schulze (2000) to verify the erroneous errors identified using the
discordancy measure discussed in subsection 2.2.1. Figure 2.3 depicts one example of
such verification where Smithers and Schulze (2000) identified station C163 to be
discordant with the remainder of the stations in the Cedara catchments. In the study
conducted by Mouton (2023), the compared annual rainfalrainfall of each station was
compared with l for different stations withthe annual rainfall at nearby stations with
confirmed good quality data to identify d. Discrepancies in annual totals were identified.

Figure 2.3: Double mass plot of daily rainfall for selected stations in the Cedara catchments
for the period October 1988 – September 1989 (Smithers and Schulze, 2000)

Mouton (2023) compared annual rainfall for different stations with annual rainfall at nearby
stations with confirmed good quality data. Discrepancies in annual totals were identified as
shown in Figure 2.4.

Irene 5-min data (Annual rainfall) O.R Tambo 5-min data (Annual rainfall)
Station 5-min data (Annual rainfall) Missing data (%)

________________________________________________________________________
13
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Figure 2.4: Annual rainfall comparison for the Lanseria rainfall station with Irene and OR
Tambo stations (Mouton, 2023)

1.4.3[2.2.3] Rainfall data quality control

In the absence of a standard approach for the Quality Control (QC) of sub-daily rainfall
data, Blenkinsop et al. (2016) developed a semi-automatic QC procedure to identify hourly
rainfall data from a single site that could contain errors. Their procedure contributed to
addressing the need of developing QC procedures with the capability to of identifying false
extreme rainfall values, as highlighted by Alexander et al. (2016), but to simultaneously
preserve the true extreme rainfall values (Lewis et al. 2018). The procedure identifies
errors associated with erroneously high tipping rates of the Tipping Bucket Rain Gauge
(TBRG), inaccurate accumulated totals, and inaccurate dry spells. Lewis et al. (2018) has
built on the work done by Blenkinsop et al. (2016) by adding checks using the nearest
neighbouring stations.

1.4.3.1[2.2.3.1] Tipping rates

According to Upton and Rahimi (2003) as cited by Blenkinsop et al. (2016), an excessively
high tipping rate could be due to equipment disturbance, the incorrect bucket positioning,
or rapid snow melt. Based on the distribution of the rainfall intensity statistic ( λ k ), calculated
using Equation 2.5, Upton and Rahimi (2003) suggested a maximum of 5 before the data
is rejected.

| ( )|
λ k = ln
τk
τ k−1
[ 2.5 ]

where:
λk = rainfall intensity statistic of time interval k ,
τk = Inter-tip time of time interval k (sec), and
τ k−1 = Inter-tip time of time interval k −1 (sec).

The slow tipping rate of a TBRG was not considered by Blenkinsop et al. (2016), but
because blockages caused by bird droppings or leaves are always a possibility, the slow
tipping rate was considered to be appropriate for this study. Upton and Rahimi (2003)
suggested that a slow tipping rate is a typical result of a partially blocked gauge where
rainfall accumulates in the funnel rather than be channelled into the bucket. The rainfall
trickles through at a rate that will be a function of the pressure head immediately above the
blockage. Although Upton and Rahimi (2003) did not provide the standard orifice equation,
it is presented as Equation 2.6 for clarity with the units and descriptions amended to
explicitly relate to the trickling rate.

________________________________________________________________________
14
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
q=C d ab √ 2 gh [ 2.6 ]
where:
q = trickling rate into bucket (unit volume per time interval),
Cd = discharge coefficient, and
ab = effective opening in the blockage (unit area),
g = gravitational acceleration (9.81 m/s²), and
h = static head (unit distance),

Upton and Rahimi (2003) reported a nearly monotonically increasing sequence in the inter-
tip time and suggested that a TBRG could be blocked if a sequence of twelve or more time
intervals are positive as per Equation 2.7.

{
1 if τ k > τ k−1
δ k = 0 if τ k =τ k−1 [ 2.7 ]
−1 if τ k < τ k−1
where:
δk = inter-tip change index

1.4.3.2[2.2.3.2] Identification of inaccurate accumulated totals

Blenkinsop et al. (2016) used the accumulated rainfall total in the United Kingdom (UK) to
automatically flag data periods as inconsistent if any one of the following procedures
returned a positive result:

i) If the accumulated 1-hour and 24-hour rainfalls exceeds the maximum recorded
values of the United Kingdom (UK), by 20%,
ii) If the accumulated 24-hour rainfall exceeds two times the mean daily rainfall of that
month,
iii) If the accumulative rainfall of two consecutive hours is equal, and equal to two times
the mean daily rainfall of that month,

Blenkinsop et al. (2016) also considered the accumulative rainfall during winter months,
which are associated with synoptic scale events in the UK by using the following
procedures to flag inconsistent data if multiple procedures returned positive results:

i) If the maximum 1-hour intensity in October to April exceeds 80% of the maximum
recorded values of the UK,
ii) If only a single 1-hour value exceeds two times the mean daily rainfall for that
month, and
iii) If the maximum intensity occurs frequently at specific dates.

________________________________________________________________________
15
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1.4.3.3[2.2.3.3] Identification of excessively long dry spells

TBRGs could also be malfunctioning that could lead to incorrect recording of zero values
for long periods of time (Blenkinsop et al. 2016). Therefore, excessively long dry spells
were identified by Blenkinsop et al. (2016) to be > more than 6 months for some stations.

1.4.3.4[2.2.3.4] Quality control flags

The eleven quality control flags for a single station, developed by Blenkinsop et al. (2016),
as well as the additional four regional quality control flags that were later added by Lewis
et al. (2018), are summarised in Table 2.8.

Table 2.82.9: Summary of automated quality control flags (Blenkinsop et al, 2016; Lewis et
al., 2018)
Acronym Description Flag applied

QC1 (Threshold) 1-hour record (92 mm) with separate flags showing exceedance by:
< 20%, 1
≥ 20%, 2
33%, 3
or 50% 4

QC2 (Threshold) Intense ‘winter’ rainfall with 1-hour total > 80% of 1-hour record (92 1
mm) in October-April period.
QC3 (Threshold) 24-hour record (279 mm) with separate flags showing exceedance
by:
< 20%, 1
≥ 20%, 2
33%, 3
or 50% 4

QC4 (Non-threshold) Suspect daily accumulations at 0900 or 1200 flagged where a

recorded rainfall amount at these times is preceded by 23-hour with
no rain. A threshold of 2 x the mean wet day amount for the
corresponding month is applied to increase the chance of
identifying accumulated values at the expense of genuine,
moderate events:
Accumulation at 0900 1
Accumulation at 1200 2

QC5 (Non-threshold) Suspect consecutive daily accumulations at 0900 or 1200 flagged

recorded rainfall amounts at these times are preceded by 23-hour
with no rain on consecutive days with not threshold to the wet hour
amount applied.
Accumulation at 0900 1
Accumulation at 1200 2

________________________________________________________________________
16
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
QC6 (Non-threshold) Suspect monthly accumulations. Identified where only one hourly 1
value is reported over a period of a month and that value exceeds
the mean wet hour amount for the corresponding month (a lower
threshold than in QC4 is used here as a dry month is much more
unlikely than a dry day in GB).
QC7 (Non-threshold) Total consecutive large values. Large values are defined as those 1
exceeding 2 x the mean wet hour amount for the corresponding
month for the corresponding month. This threshold was applied as
genuine duplicate small to moderate rainfall amounts occur
legitimately.
QC8 (Non-threshold) Frequent tipping (only applied to EA TBR data using Upton and 1
Rahimi (2003) algorithm).
QC9 (Non-threshold) Additional suspect data identified by manual inspection and 1
comparison with documentary records e.g., British Rainfall or the
CEH-GEAR daily data.
QC10 Terminal dry spell at start / end of gauge record ≥ 31-day duration 1
QC11 Dry period:
≥ 31-day duration, 1
≥ 45-day duration 2
QC12 Neighbourhood check – dry periods > 14 days. Winter (NDJFMA):
90th, 1
95th, 2
99th bounds 3
QC13 Neighbourhood check – dry periods > 14 days. Summer (MJJASO):
90th, 1
95th, 2
99th bounds 3
QC14 (Threshold) Neighbourhood check – daily precipitation total. Winter (NDJFMA):
90th, 1
95th, 2
99th bounds 3
QC15 (Threshold) Neighbourhood check – daily precipitation total. Summer
(MJJASO):
90th, 1
95th, 2
99th bounds 3

1.4.3.5[2.2.3.5] Global sub-daily rainfall quality control procedure

The automated quality control flags of Lewis et al. (2018) were generalised by Lewis et al.
(2021) to identify inconsistent gauges across various geographical, climatic and
instrumental conditions, and called the Global Sub-Daily Rainfall Quality Control (GSDR-
QC) procedure. The automated quality control checks conducted by the GSDR-QC

________________________________________________________________________
17
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
procedure consist of 25 checks as summarised in Table 2.10, which is broadly divided into
five groups:
i) The first group of checks (QC1 to QC7) identifies inconsistencies in large portions
of the data of a single station.
[ii)] The second group (QC8 to QC11) uses thresholds to identify inconsistently high hourly,
and annual total rainfall of a single station.
ii)[iii)] The third group (QC12 to QC15) is a combination of long dry periods, and
accumulations of daily and monthly total rainfall checks on a single station.
iii)[iv)] The fourth group (QC16 to QC17) considers the total daily rainfall of neighbouring
stations to identify inconsistencies in high rainfall.
iv)[v)] The fifth group (QC18 to QC25) is a combination of long dry periods and total rainfall
on a daily and monthly scale from the neighbouring stations.

Table 2.102.11 Summary of automated quality control flags of the global sub-daily rainfall
quality control procedure (Lewis et al. 2021)
ID Short Name Description Flag

QC1 Percentiles Identifies years where the 95th and 99th percentiles of List of years.
rainfall are zero.
QC2 K-largest Identifies years where the 1, 5 and 10 largest hourly List of years.
rainfall amounts are zero.
QC3 Days of week Two-sided t-test on the distribution of mean rainfall over 1 if outside the
days of the week. 99% confidence
interval.

QC4 Hours of day Two-sided t-test on the distribution of mean rainfall over 1 if outside the
hours of the day. 99% confidence
interval.

QC5 Intermittency Return years where more than 5 no data periods List of years.
bounded by zeros.
QC6 Breakpoints Pettitt test to identify possible break points in median
annual wet hour rainfall, with indication of pre-/post-
breakpoint difference in magnitudes:
< 50% 0
≥ 50% 1
≥ 100% 2
≥ 200% 3
≥ 500% 4
or ≥ 1000% 5

QC7 Minimum value Count of instances where minimum non-zero rainfall Count and list of
change value changes from one year to the next (i.e., indication years
of data resolution change indicating possible equipment
change and therefore inhomogeneity).
QC8 R99p maximum ETCCDI R99p value (based A flag for each
on location) with separate flags showing exceedance by: year
(Neighbouring
________________________________________________________________________
18
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
R99p value used)
< 20% 1 (5)
≥ 20% 2 (6)
≥ 30% 3 (7)
or ≥ 50% 4 (8)
QC9 PRCPTOT Annual exceedance of the maximum ETCCDI PRCPTOT A flag for each
value (based on location) with separate flags showing year
exceedance by: (Neighbouring
PRCPTOT value
used)
< 20% 1 (5)
≥ 20% 2 (6)
≥ 33% 3 (7)
or ≥ 50% 4 (8)
QC10 World Record rainfall world record (world record = 401 mm in 1 h) with
separate flags showing exceedance by:
< 20% 1
≥ 20% 2
≥ 33% 3
or ≥ 50% 4
QC11 Rx1 day Exceedance of the maximum ETCCDI Rx1day value (Neighbouring
(based on location) by hourly values with separate flags Rx1day value
showing exceedance by: used)
< 20% 1 (5)
≥ 20% 2 (6)
≥ 33% 3 (7)
or ≥ 50% 4 (8)
QC12 CDD Identify suspicious long dry periods through exceedance (Neighbouring
of the maximum ETCCDI CDD value (based on location) CDD value used)
with separate flags showing exceedance by:
< 20% 1 (5)
≥ 20% 2 (6)
≥ 33% 3 (7)
or ≥ 50% 4 (8)
QC13 Daily Possible daily accumulations (where an hour with a
Accumulations recorded rainfall amount is preceded by 23 h with no
rain) are flagged in the following cases:
(a) Potential daily accumulations exceed the mean wet 1 (2 if
day amount (maximum ETCCDI variable SDII, based on neighbouring
location) to increase the chance of identifying true SDII value used)
accumulated values
(b) Periods of ≥2 days with any non-zero potential daily 3 (4 if
accumulations neighbouring
SDII value used)
(c) Periods of zeros in between potential daily 6

________________________________________________________________________
19
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
accumulations flagged by (a) and (b)
QC14 Monthly Potential monthly accumulations flagged where an hour
Accumulations with a recorded rainfall amount is preceded by one
month with no rain. A threshold of 2 x the mean wet day
amount (maximum ETCCDI variable SDII based on
location) is applied. Separate flags indicate cases where
the day following the potential accumulation is:
(a) dry (i.e. isolated high value, which may be more likely 1 (2 if
to be in error) neighbouring
SDII value used)
(b) wet (i.e. potentially more legitimate, as multiple wet 3 (4 if
hours following dry period) neighbouring
SDII value used)
QC15 Streaks Suspect repeated consecutive values are flagged in the
following cases:
(a) Streaks of 2 or more repeated large hourly rainfall 1 (2 if
values exceeding 2 x mean wet day rainfall (from the neighbouring
maximum ETCCDI variable SDII based on location) SDII value used)
(b) Streaks of 12 or more repeated hourly values greater 3
than data resolution (i.e. relatively long periods of uniform
intensity rainfall)
(c) Streaks of 24 or more repeated hourly values greater 4
than zero (i.e. very long streaks at data resolution)
(d) Periods of zeros bounded by streaks of ≥24 repeated 5
values (where length of zeros is a multiple of 24, i.e. to
find only parts of record where uniform disaggregation or
repetition of daily totals is likely to have been applied)

QC16 Daily Identify suspicious large values by comparison to

neighbours neighbouring daily gauges with separate flags indicating
(wet) difference between gauge and all neighbours exceeds
the following percentiles (of the distribution of gauge-
neighbour differences):
≥ p95 1
≥ p99 2
≥ p99.9 3
QC17 Hourly As per QC16, but identifying suspicious large values by
neighbours comparison to neighbouring hourly gauges (after
(wet) aggregation to daily interval) with separate flags as per
QC16 indicating difference between gauge and all
neighbours exceeding:
≥ p95 1
≥ p99 2
≥ p99.9 3
QC18 Daily Identify suspicious dry periods by comparison to
neighbours (dry) neighbouring daily gauges with separate flags indicating
the average number of wet days for all of the neighbours
during a 15-day dry spell at the gauge:
1 wet day 1

________________________________________________________________________
20
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
2 wet days 2
≥ 3 wet days 3
QC19 Hourly As per QC18, but identifying suspicious dry periods by
neighbours (dry) comparison to neighbouring hourly gauges with separate
flags indicating the average number of wet days for all of
the neighbours during a 15 day dry spell at the gauge:
1 wet day 1
2 wet days 2
≥ 3 wet days 3
QC20 Monthly Identify suspicious monthly totals by comparison to
neighbours neighbouring monthly gauges with separate flags
indicating percentage difference between gauge and all
neighbours by:
−100% (i.e. gauge dry but neighbours not) -3
≤ −50% -2
≤ −25% -1
≥ 25% 1
≥ 50% 2
≥ 100% 3
Flags equal to 3 may be upgraded using additional
checks on whether month total is:
≥1.25 x record maximum for all neighbours 4
≥2 x record maximum for all neighbours 5
QC21 Timing offset For whole time series, identify suspicious timing of data
by testing AI and r2 between gauge and nearest daily
gauge and returning optimum matching at a lag of
-1 day -1
0 days (i.e., no offset) 0
+1 day 1
QC22 Pre-QC affinity AI calculated between gauge and nearest daily AI (between 0
index (AI) neighbour to identify a generally suspicious gauge. AI is and 1)
a wet/dry day matching statistic.
QC23 Pre-QC Pearson r² calculated between gauge and nearest daily neighbour r² (between −1
correlation to identify a generally suspicious gauge. and 1)
coefficient
QC24 Daily factor Mean factor difference between gauge and nearest daily scalar
neighbour to identify unit errors.
QC25 Monthly factor Time series of factor difference between gauge and
nearest monthly neighbour to identify unit errors of:
~10 x greater than neighbour monthly total 1
~25.4 x greater than neighbour monthly total 2
~2.54 greater than neighbour monthly total 3
~10 x smaller than neighbour monthly total 4
~25.4 x smaller than neighbour monthly total 5
~2.54 x smaller than neighbour monthly total 6

________________________________________________________________________
21
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

[2.3] REGIONALISATION

As part of the rainfall data quality control, neighbouring stations need to be considered, but
because South Africa consists of different climatic regions, appropriate neighbouring
stations need to be selected. Therefore, the regionalization of South African rainfall was
reviewed.

1.4.4[2.3.1] Kӧppen climate classification system

The Kӧppen (also known as Kӧppen-Geiger) climate classification system is based on the
monthly temperature and precipitation threshold values and seasonality (Beck et al. 2018).
The Kӧppen climate classification system is the most well-known system According to
(Kruger and Mbatha, (2021), and is also the most widely used system (Kottek et al. 2006;
Rubel and Kottek, 2011) the Kӧppen climate classification system is the most best-well-
known system used globally. The system was originally developed by the German-
Russian climatologist, Wladimir Peter Kӧppen (1846-1940), who published his first version
in 1884. After working throughout his life on a number of versions, he published his last
version in 1936 (Rubel and Kottek, 2011). Further updates of the Kӧppen classification
system were later published by Rudolf Geiger (1894-1981) which has ever since became
also known as the Kӧppen-Geiger climate classification system. According to this system,
South Africa consists of eight regions as depicted in Figure 2.5. The regions are described
as follows:
1) Arid (BW),
2) Semi-arid (BS),
3) Winter rain with hot summers (Csa),
4) Winter rain with cool summers (Csb),
5) Summer rain with hot summers (Cwa),
6) Summer rain with cool summers (Cwb),
7) All-year rain with hot summers (Cfa), and
8) All-year rain with cool summers (Cfb).
.

________________________________________________________________________
22
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 2.5: The Kӧppen climatic classification of South Africa (Kruger and Mbatha, 2021)

1.4.5[2.3.2] South African Weather Service system of climate regions

A more detailed vegetation-based climatic classification system was developed for South
Africa by the South African Weather Service (SAWS) which is based on the relationship
between the Kӧppen classification system and the vegetation biomes in South Africa
(Kruger and Mbatha, 2021). The system consists of 24 regions as depicted on Figure 2.6.
The system which consists of the following:

1) Northern Arid Bushveld

2) Central Bushveld
3) Lowveld Bushveld
4) South-Eastern Thornveld
5) Lowveld Mountain Bushveld
6) Eastern Coastal Bushveld
7) KwaZulu-Natal Central Bushveld
8) Kalahari Bushveld
9) Kalahari Hardveld Bushveld
________________________________________________________________________
23
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
10)Dry Highveld Grassland
11)Moist Highveld Grassland
12)Eastern Grassland
13)South-Eastern Coast Grassland
14)Eastern Mountain Grassland
15)Alpine Heathland
16)Great and Upper Karoo
17)Eastern Karoo
18)Little Karoo
19)Western Karoo
20)West Coast
21)North-Western Desert
22)Southern Cape Forest
23)South-Western Cape
24)Southern Cape.

Figure 2.6: SAWS climatic regions of South Africa (Kruger and Mbatha, 2021)

1.4.6[2.3.3] Short duration design rainfall homogeneous regions

Smithers and Schulze (2000) regionalised data from 172 rainfall stations with data record
periods of at least 10 years to form 15 homogeneous regions in South Africa. The

________________________________________________________________________
24
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
regionalisation was based on the cluster analysis proposed by Hosking and Wallis (1997).
The clusters were then used for the Regional Linear- Moment Algorithm and Scale
Invariance (RLMA&SI) procedure, developed by Smithers and Schulze (2000), to estimate
short duration design rainfall for any location in South Africa. The site characteristics of the
stations were used for the regionalisation. which This enables the independent testing of
the homogeneity of the clusters by using their statistics (Hosking and Wallis, 1997). The
site characteristics that were used by Smithers and Schulze (2000) to conduct the cluster
analysis, included the latitude, longitude, altitude, concentration of precipitation, mean
annual precipitation, seasonality, and distance from sea. The homogeneity of the clusters
was determined using the heterogeneity measure proposed by Hosking and Wallis (1997).
This entails the simulation of a region having the same data period as the observed data
(Hosking and Wallis, 1997). Each simulation had the 4-parameter kappa distribution as its
underlying distribution using the regional average Linear moment (L-moment) ratios (
R R R
l ,t ,t 3 , t 4 ). The regional average L-CV was weighted proportionally to the record length
of each site using Equation 2.8. This was followed by the calculation of the weighted
standard deviation of the as-site sample L-CV using Equation 2.9, and then the
heterogeneity measure using Equation 2.10.

N N
t R =∑ ni t (i) / ∑ ni [ 2.8 ]
i=1 i=1

where:
R
t = Regional average L-CV,
(i)
t = sample L-moment ratio L-CV for station i,
ni = record length of for station i , and
N = number of stations in the cluster.

{∑ }
N N 1 /2
R 2
V= n i ( t −t ) / ∑ n i
(i)
[ 2.9 ]
i=1 i=1

where:
V = weighted standard deviation of the as-site sample L-CVs,

(V −μ v )
H= [ 2.10 ]
σv
where:
H = Heterogeneity measure,
V = weighted standard deviation of the as-site sample L-CVs,
μv = mean of V from a large number of simulation experiments, and
σv = standard deviation of V from the large number of simulation experiments.

The 24-hour Annual Maximum Series (AMS) was used to determine the heterogeneity
measure for which the majority of the regions (clusters 1, 3, 4, 6, 8, 9, 10, 11, 13, and 14)
________________________________________________________________________
25
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
were “acceptably homogeneous” with H < 1. The remainder of the regions (clusters 2, 5, 7,
12, and 15) were “possible heterogeneous” with 1 ≤ H < 2. The 15 homogeneous regions
identified by Smithers and Schulze (2000) are depicted in Figure 2.7. Because the focus of
this study is on small urban catchments with relatively rapid response times, the
heterogeneous measure will be determined using the 5-min as well as the 24-hour AMSs
of all stations eligible to be included for further analyses.

Figure 2.7: Distribution of 15 clusters of relatively homogeneous extreme short duration (# 24

h) rainfall in South Africa (Smithers and Schulze, 2000)

[2.4] DATA INTERPRETATION

A literature review ofThis section describes the current analytical methods that are used to
analyse observed rainfall data methods used for observed rainfall analysis and to estimate
short duration design rainfall estimation, is provided in this section. South Africa has seen
a declining trend in the number of active rainfall stations since 1980 (Singh, 2021). Some
________________________________________________________________________
26
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
stations have short data periods (< 10 years), and many stations are affected by missing
data. The research question that is asked is how can the available observed short duration
rainfall data in South Africa be used optimally to estimate design rainfall, considering a
short data period and missing data? To address this uncertainty, the following was
investigated in this literature review:
i) The use of frequency distribution analysis for estimating short duration design
rainfall,
ii) The impact of a short data period on the estimates,
iii) The impact of missing data on the estimates, and
iv) The use of stochastic based methods, like the Monte Carlo simulation technique
since it is often used in hydrology to quantify model output uncertainties (Al
Mamoon and Rahman, 2014), and the Bartlett-Lewis rectangular pulse models.

1.4.7[2.4.1] Frequency distribution analysis

Archaeological findings suggests that board games existed, and by implication the random
element of chance, at least 3500 BC, but it is from the correspondence between two
French mathematicians in the sixteenth century that the fascinating development of the
modern probability theory originated from (David, 1962). In his famous letter, Baise Pascal
posed some mathematical questions on games of chance to Pierre de Fermat (David,
1962; Debnath and Basu, 2015). This led to the discovery of the normal distribution by
Abraham de Moivre but he could only provide an approximate expression of the
distribution (Borovcnik and Kapadia, 2014; Debnath and Basu, 2015). The breakthrough
came in the nineteenth century when Carl Friedrich Gauss developed a mathematical
equation that accurately represents the normal distribution (Stahl, 2006). Henceforth, the
normal distribution also became known as the Gaussian distribution. With the normal
distribution as the foundation stone for modern statistical reasoning, extreme value
statistics (also known as extreme value theory) were developed in the early twentieth
century (Kinnison, 1983). Since then, various developments in the theory of probability
were made, from which several probability distributions became available to describe the
possible value of a random variable. Examples of common distributions used in
hydrological studies are the logLog-normalNormal (LN), General Extreme Value (GEV),
generalised Pareto (GP), Pearson type Type III (P3), logLog-Pearson type Type III (LP3),
to name but a few common distributions. The modern era has seen developments in
parameter estimations, from the method of moments by Karl Pearson in 1902 (Chow et al.
1988), Maximum Likelihood Estimation (MLE) by Fisher (1922) as cited by Aldrich (1997),
Probability-Weighted Moments (PWM) by Greenwood et al. (1979), and Linear Moments
(LM) by Hosking (1990). The advantage of LMs being they are less sensitive to outliers,
capable of identifying the underlying probability distribution from a small sample, and yield
more efficient parameter estimates (Hosking, 1990).

________________________________________________________________________
27
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The design of stormwater infrastructure is generally dependent on design rainfall
estimates. which isThese are determined from observed rainfall data using a suitable
underlying distribution. Finding the most suitable underlying distribution of observed rainfall
data and other hydrological data, considering different parameter estimation methods, is a
topic frequently found in the literature. Different categories of methods were used by Calitz
(2020) to determine the most suitable distribution for flood frequency analyses, namely
graphical, goodness-of-fit, model selection criterion, and predictive performance methods.
Graphical methods include, plotting positions, product moment diagrams, and linear
moment ratio diagrams, whereas. gGoodness-of-fit methods considered by Calitz (2020)
includes, Chi-squared, Kolmogorov-Smirnov, Cramer von Mises, and Modified Anderson-
Darling. Calitz (2020) used an iterative approach to identify the most suitable distribution
for flood frequency analysis in South Africa. Firstly, linear moment ratio diagrams were
used to identify the Log-Pearson Type IIILP3 distribution, generalized ParetoGP
distribution, and the three parameter Kappa distribution to be the most favourable
distributions. Thereafter, the goodness-of-fit, model selection criterion, and predictive
performance methods were applied to the three favourable distributions. Finally, each
distribution was ranked considering the results of all four methods, respectively. The best
performing method was ranked first and the poorest third. The overall best performing
distribution was the generalised ParetoGP distribution, which had the lowest total score.

In an earlier study, Smithers and Schulze (2000) used the z-test statistics and LM ratio
diagrams to find the most suitable distribution for short duration design rainfall estimation
in South Africa. Ten probability distributions were considered. and bBased on the number
of homogeneous regions in which the candidate distributions gave an acceptable fit to the
24-hour annual maximum seriesAMS, the GEV distribution performed the best. Mamoon
and Rahman (2017) considered fourteen different distributions and the Kolmogorov–
Smirnov, Anderson–Darling and Chi-squared goodness-of-ﬁt tests. They found the GEV
distribution to be the most suitable distribution for Qatar. Various studies were conducted
internationally to find the most suitable distributions for design rainfall estimations. A
summary of previous findings, adopted from Smithers and Schulze (2000) and Mamoon
and Rahman (2017), and as well as expanded with more findings from this review, are
provided in Table 2.12. A summary of the abbreviations used for the distributions are
provided in Table 2.13.

________________________________________________________________________
28
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Table 2.12: Summary of previous findings of most suitable probability distribution in design
rainfall estimation (after Mamoon and Rahman, 2017)
Suitable
Publication Data used dDistributio Country
n
Publications cited by Smithers and Schulze (2000)
Canterford et al. (1987a) LP3 Australia
Canterford et al. (1987b) 6-min to 72-hour LP3,/LN2 Australia
James et al. (1987) EV1 India
Sendil and Salih (1987) 10-min to 12-hour EV1 Saudi Arabia
Ferreri and Ferro (1990) EV1 Sicily
Schaefer (1990) 2, 6, and 24-hour GEV USA
Shuy (1990) EV1 Singapore
Buishand (1991) GEV
Griffiths and Pearson
EV1,/KAP New Zealand
(1993)
Naghavi et al. (1993) LP3 USA
Guttman (1992) LP3 USA
Cannarozzo et al. (1995) TCEV Sicily
Publications cited by Mamoon and Rahman (2017)
Thailand, India,
Phien and Ajirajah (1984) 1-hour to 31-day LP3
Laos and USA
Subyani and Al-Amri Al-Madinah City,
24-hour EV1, PE3
(2015) Saudi Arabia
15, 30, and 60-min and 1, 2, 3, 6,
Tortorelli et al. (1999) GLO Oklahoma, USA
12, and 24-hour, and 1, 3, and 7-day
Green et al. (2012) 1 to 12-hour, 1 to 7-day GEV Australia
Sen and Eljadid (1999) Monthly rainfall GAM Lybia
Ogunlela (2001) Daily and monthly rainfall LP3 Nigeria
Zalina et al. (2002) 1-hour GEV Malaysia
Wan Zin et al. (2008) 24-hour GLO Malaysia
Southern Quebec,
Tao et al. (2002) 5-min and 1-hour GEV
Canada
Lee (2005) 24-hour LP3 Taiwan
Kwaku and Duke (2007) 1, 2, 3, 4 and 5-day LN Ghana
Sharma and Singh
24-hour LN Pantnagar, India
(2010)
Johnson et al. (2012) 5-min to 72-hour GEV Australia
Fadhilah et al. (2007) 1-hour MEX Malaysia
Mamoon et al. (2014) 24-hour PE3 Qatar
Publications from this study
Smither and
24-hour GEV South Africa
SchulzeAlam et al. 2018
GEV and
Sharma and Singh
1-week to 17-week various India
(2010)
others
Alam et al. 2018 Monthly GEV Bangladesh
Sharma and Singh, 2010
________________________________________________________________________
29
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Ahmed El-Sayed (2011) 5-min to 24-hour LN , LP3 Egypt
Ewea et al. (2016) 10-min to 24-hour EV1 Saudi Arabia

Table 2.13: Abbreviations used for probability distributions

Abbreviation Probability Distribution Abbreviation Probability Distribution
EV1 Extreme Value Type 1 LN3 3 parameter Log-Normal
GAM Gamma LP3 Log-Pearson Type III
GEV General Extreme Value PE3 Pearson Type III
GPA Generalised Paretor MEX Mixed exponential
GLO Generalised Logistic TCEV Two Component Extreme Value
KAP Kappa NOR Normal
LN2 2 parameter Log-Normal

From Hosking and Wallis (1997), the GEV distribution is defined by Equations 2.11 to 2.13
as follows:

{
, y= −k log { 1−k x −ξ /α } ,∧k ≠ 0
−1
−1 −1 ( 1−k ) y−e
f ( x )=α e
−y
( )
[ 2.11 ]
( x−ξ ) /α ,∧k=0
−y

F ( x )=e−e [ 2.12 ]

{
x ( F )= ξ +α {1−(−log F ) }/k ,∧k ≠ 0
k
[ 2.13 ]
ξ−α log (−log F ) ,∧k =0

where:
f (x) = Probability density function,
F(x) = Cumulative distribution function,
x (F ) = Quantile function,
α = scale parameter,
k = shape parameter, and
ξ = location parameter

The L-moments are given by Equations 2.14 to 2.17;

λ 1=ξ+ α { 1−Γ ( 1+ k ) } /k [ 2.14 ]

λ 2=α ( 1−2−k ) Γ ( 1+k ) /k [ 2.15 ]

________________________________________________________________________
30
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

τ 3 =2 ( 1−3−k ) / ( 1−2−k )−3 [ 2.16 ]

τ 4= {5 ( 1−4 ) −10 ( 1−3 ) +6 ( 1−2 ) } / ( 1−2 )

−k −k −k −k
[ 2.17 ]

Here Γ ( . ) denotes the gamma function as per Equation 2.18;

∞
Γ ( x )=∫ t
x−1 −t
e dt [ 2.18 ]
0

The parameters as given by Equations 2.19 to 2.21.

2 log 2
2
k ≈ 7.8590 c +2.9554 c , c= − [ 2.19 ]
3+ τ 3 log 3

λ2k
α= [ 2.20 ]
( 1−2−k ) Γ (1+ k )

ξ=λ 1−α { 1−Γ ( 1+k ) } /k [ 2.21 ]

The L-moment ratios are given by Equations 2.22 to 2.24;

λ2
τ= [ 2.22 ]
λ1

λ3
τ3= [ 2.23 ]
λ2

λ4
τ 4= [ 2.24 ]
λ2

Detection of non-stationarity in precipitation extremes using a max-stable process model

(Westra and Sisson, 2011).

1.4.8[2.4.2] Minimum data period

In quantitative statistics it is generally accepted that larger samples approximate the

population more closely, which improves statistical inferences, but. However, increasing to
increase the data period of observed hydrological data could become challenging, since
________________________________________________________________________
31
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
the unmeasured value will be forever unknown. According to Smithers and Schulze
(2000), more rainfall stations with short data periods (<10 years) in South Africa exists
than stations with long data periods (≥10 years). If no other information or technique is
available, then the design rainfall estimates have to be estimated from the short data
period (Smithers and Schulze, 2000). In this section, the minimum data periods used in
previous hydrological studies, concerning extreme value statistics, were investigated, as
well as some methods to determine the minimum required sample size.

From a literature review conducted by Smithers and Schulze (2000) it was evident that a
minimum data period of 10 years has generally been adhered to. Hence, they used rainfall
stations in South Africa which had a data period of 10 or more years for design rainfall
estimation. For the stations with short data periods, they showed by means of two case
studies, that design rainfall is better estimated from synthetic rainfall data series generated
by the Bartlett-Lewis Rectangular Pulse Gamma Model (BLRPGM), than on estimating the
design rainfall directly from the short data period. In another study, Calitz (2020) used a
minimum data period of 20 years to conduct flood frequency analyses in South Africa.
Although no justification was provided for the selection of this period, it was recognised
that long data periods are required to provide reliable results.

Two methods to determine the minimum sample size required for statistical inference were
reviewed. Cai and Hames (2010) developed a general method for determining the
minimum sample size for a GEV distribution based on the asymptotic distribution of MLEs.
They argued that because the MLE of the sample statistics is normally distributed by
generating multiple bootstrap samples of a certain size, the results of the recurrence
interval should also be normally distributed (Cai and Hames, 2010). Therefore, the
Shapiro-Wilk test for normality was used to reject the null hypothesis that the recurrence
interval is normally distributed to conclude that the sample was too small. Their method
can be used to determine the minimum sample size of any quantile, and by applying the
method to two case studies, Cai and Hames (2010) got a minimum sample size of
approximately 40 for a recurrence interval of 1:100 years.

Mamoon and Rahman (2017) used the annual maximum seriesAMS for the 24-hour
duration rainfall of 29 rainfall gauges in Qatar in their study. The data period ranged from
24 to 49 years and according to Mamoon and Rahman (2017), this was too short for a
meaningful statistical inference. Bootstrapping with replacement was applied to each of the
selected stations to generate 10,000 random samples. Data periods of 10 years up to 500
years were generated with the bootstrapping method and the minimum sample size that
resulted in acceptable sampling errors was determined that resulted in acceptable
sampling errors. The results indicated that the standard deviation was highly sensitive to
data period, but the skewness appeared to be most sensitive to the data period.

________________________________________________________________________
32
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1.4.9[2.4.3] Missing data

Apart from a short data period that could affect the confidence in the result of a probability
distribution analysis, missing data could also skew the result. Therefore, the handling of
missing data and the effect on a probability distribution analysis was investigated and the
findings are presented in this section.

Missing data is a topic frequently found in many quantitative statistical research fields and
although this research document is applicable to rainfall data, some techniques of handling
missing data are universal. The mechanisms of missing data formulated by Rubin (1976)
is an important consideration when handling missing data (Jadhav et al. 2019). For
example, Missing Completely at Random (MCAR) is the mechanism where the
missingness of a variable is not related to itself or any other variable (Jadhav et al. 2019;
Soley-Bori, 2013). Other mechanisms that are used to describe the randomness of
missing data are Missing at Random (MAR) which has a slightly weaker assumption of
randomness than MCAR, and Not Missing at Random (NMAR), which is dependent on
unobserved values (Soley-Bori, 2013). Although not mentioned in the literature, the MCAR
missing data mechanism would generally be applicable to missing rainfall data as a result
of the following:

[(a)] Missing data caused by equipment malfunctions like power failures, theft, or
vandalism.
[(b)] Missing data caused by the malfunctioning of equipment during a storm event or
periods with high likelihood of rainfall, for which the inconsistencies in the data
would be identified by considering other meteorological data, and then removed if it
is deemed unreliable.
[(c)] Prolonged periods of missing data resulting from a delayed response to rectify
malfunctions.

The complete exclusion of a data set is the simplest conventional method of handling
MCAR data (Soley-Bori, 2013) and can be used without the risk of being biased (Batista
and Monard, 2002). Sometimes this method would be applied when the missingness has
reached a pre-determined level of significance. This was the methodology followed by
Smithers and Schulze (2000) since rainfall data can sometimes be partly missing. They
used one station as a case study and extracted Partial Duration Series (PDS) for all
standard time steps followed by Annual Maximum Series (AMS), from which the design
rainfall estimates were determined. The largest value was randomly excluded from 10% to
50% of the years and the t-test statistic used to reject the null hypothesis that the result will
be within 5% when all the largest values were present. It was concluded that the exclusion
of the largest values should be limited to 20% to prevent a significant difference in design
rainfall estimates.

________________________________________________________________________
33
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The second conventional method of handling missing data in order to provide a complete
dataset, for modelling purposed in particular (Moeletsi et al. 2016), is the substitution of the
missing data with educated estimates, called imputation. The missing data are imputed
using spatial interpolation which are based on known relationships identified from the
available data (Batista and Monard, 2002). Various methods with varying degree of
difficulty is are available in the literature. which These are discussed separately in the
following section.

1.5[2.5] SPATIAL RAINFALL INTERPOLATION

The spatial interpolation of observed rainfall data is often used to estimate missing rainfall
data (Fung et al. 2022). and inIt is also applied in some computer hydrological models,
which is are an essential tools for hydrologists and engineers (Ly et al. 2013). Many other
environmental studies also require a continuous surface of spatially interpolated rainfall
data for water resource planning (Brunsdon et al. 2001). The spatial interpolation methods
are broadly categorized into two groups. The first is deterministic methods, for example the
Arithmetic mean (ARM), Normal-Ratio (NRA), Nearest Neighbour (NEN), Natural
Neighbour (NAN), Triangulation (TRI), and Inverse Distance Weighting (IDW). The second
category is geostatistical methods, for example Ordinary Kriging (ORK), and Co-Kriging
(COK). The methods from both these categories are presented individually in the following
sections.

1.5.1[2.5.1] Arithmetic mean

Replacing missing data with the arithmetic mean value of the observed data is the simplest
(e.g. Chow et al. 1988) and the most common imputation technique (Jadhav et al. 2019).
The missing value according to this technique is estimated using Equation 2.25.

n
1
pt = ∑ x i [ 2.25 ]
n i=1
where:
pt = estimated value of the missing rainfall (mm),
xi = observed rainfall at neighbouring station (mm), and
n = number of neighbouring stations (Radi et al. 2015; Pinthong et al. 2022).

Using the arithmetic mean can also be applied to a single data series. The technique is
simple to apply, but it can change the shape of the data’s underlying distribution. This
which can lead to biased results. As more instances of missing data are replaced with the
mean value, the standard deviation of the imputed data set decreases, since all missing
data will have the same value (Suthar et al. 2012), but. t The results can be improved by
dividing the data into subgroups (Jadhav et al. 2019). An alternative approach is to use the

________________________________________________________________________
34
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
mean value above and below the missing value to replace the missing value. Noor et al
(2014) achieved better results using this approach for filling the missing observations in air
pollution datasets compared to replacing the missing values with the overall mean value.

1.5.2[2.5.2] Normal-ratio

Paulhus and Kohler (1952) developed the normal ratio technique to impute missing
monthly and annual rainfall. It is similar to the arithmetic mean technique, but they
introduced the annual rainfall as a weighting factor. Three index stations are selected that
are evenly distributed about the station with missing data and the missing value calculated
using Equation 2.26.

Pt =
1
3 [( ) ( ) ( )
Nt
N1
N N
P1 + t P2 + t P3
N2 N3 ] [ 2.26 ]

where:
Pt = estimated value of the missing rainfall at station of interest (mm),
P1 ,2 , 3 = observed rainfall at neighbouring stations (mm),
Nt = annual rainfall amount at the target station (mm), and
N 1 ,2 , 3 = annual rainfall amount at neighbouring stations (mm).

Because of the subjectiveness of selecting the three neighbouring stations, Young (1992)
modified the normal-ratio technique using the square of the t-statistic. The three selected
stations were weighted and the three stations with the highest correlation coefficients were
selected (Young, 1992). Using the weighted factor showed a slight improvement in the
resulting estimates. The weighting factor is calculated using Equation 2.27.

2
r i ( r i−2 )
w t= 2 [ 2.27 ]
1−r i
where:
wt = weighting coefficient by the square of the t-statistic,
ri = correlation coefficient between the station of interest and the i-th
neighbouring station, and
ni = is the number of points on which the correlation coefficient is based on.

1.5.3[2.5.3] Triangulation

The Triangular Irregular Network (TIN) model, first introduced by Franklin (1973) as cited
by Franklin and Gousie (1999), is commonly used in Geographical Information Systems
(GIS) (Huisman and de By, 2009). The TIN model is used to present a continuous field
from irregular spaced known points with values of some other variable, like elevation. The

________________________________________________________________________
35
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
known points are connected to form a tessellation of triangles, which is a pattern with no
gaps or overlaps (Huisman and de By, 2009). However, for a set of points there are many
possible triangulations. Figure 2.8 illustrates two possibilities with two different estimated
values for the unknown point P. Huisman and de By (2009) have provided two properties
of an optimal triangulation. The first being the sides of each triangle must be as equilateral
(equal-sided) as possible, and the second being the circumcircle of each triangle must not
contain any other known point, also known as Delaunay triangulation (Huisman and de By,
2009). This is illustrated in Figure 2.8 (b).

980 980

1550 P
P
1250 1250
Circumcircle

1340

(a) Triangulation with many (b) Triangulation with nearly

‘stretched’ triangles; P = 1078 equilateral triangles (Delaunay
triangulation); P = 1193

Figure 2.8: Typical Triangular Irregular Network (after Huisman and de By, 2009)
……………..

1.5.4[2.5.4] Nearest neighbour

T…………….he Nearest Neighbour (NEN) method entails the construction of Thiessen

polygons around the stations with known values after which all locations inside the polygon
adopt the same known value (Li and Heap, 2008). The Thiessen polygons were introduced
by Alfred H. Thiessen (1872-1956) which is a way of representing rainfall data from
irregularly spaced stations (Brassel and Reif, 1979). The process consists of constructing
Delaunay triangles, similar to the TIN model, and then perpendicular bisecting lines are
drawn between the stations which produces polygons with one station at the centre of
each polygon. This process is illustrated in Figure 2.9

________________________________________________________________________
36
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

980 980

1550 P 1550 P
1250 1250

(a) Delaunay triangulation (b) Thiessen polygons; P = 980

Figure 2.9: Typical construction of Thiessen polygons (after Huisman and de By, 2009)

1.5.5[2.5.5] Natural neighbour

The Natural Neighbour (NAN) method, introduced by Sibson (1981), is a combination of

the TIN model and the NEN method (Webster and Oliver, 2007). The process of
constructing Thiessen polygons for stations with known values is repeated with the
location with the ungauged site included. The portion of the new polygon that is located in
the original polygon is determined and the weight of the particular station calculated
according to Equation 2.29 (Webster and Oliver, 2007). The NAN process is illustrated in
Figure 2.10.……………

❑❑
❑❑ ❑
[ 2.28 ]
∑
❑
❑❑
where:
❑❑ = weight of neighbouring station i,
❑❑ = area located in neighbouring polygon of station i,
❑❑ = area located in neighbouring polygon of station k , and
= number of neighbouring stations.

Figure 2.10: Typical Natural Neighbour

________________________________________________________________________
37
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

[2.5.6] Inverse distance weighting

The Inverse Distance Weighting (IDW) function was developed by Shepard (1968) and is a
two-dimensional interpolation technique that is used to create a continuous surface from
irregular spaced data points (Shepard, 1968). The IDW function is presented in Equation
2.29.

( )
P
1
Pi
n
di
Pt =∑ n [ 2.29 ]
∑ ( d1 )
P
i=1

i=1 i

where:
Pu = Predicted value at ungauged location u,
Pi = Known rainfall value at station i ,
di = Euclidean distance between ungauged location u and station i ,
n = Number of stations,
P = IDW exponent,

The main assumption of the IDW function is that the influence from a station decreases the
further away the ungauged point is from the station (Moeletsi et al. 2016). The magnitude
of influence is severely affected by the exponent which significantly changes the
interpolated surface. As the exponent decreases, sudden changes in the interpolated
values appear near stations in order to attain the true value of each station, whereas. fFor
an increased exponent, sudden changes appear between the stations (Shepard, 1968).
From empirical tests conducted by Shepard (1968), an exponent of 2.0 produces
reasonable results. Moeletsi et al. (2016) found both the value of 2.0 and 2.5 to produced
optimal results for estimating daily and decadal rainfall in the Free State Province, South
Africa.

The function is also commonly used for estimating missing rainfall data (Pinthong et al.
2022). Moeletsi et al. (2016) found the IDW function to estimate low daily rainfall (≤ 5mm)
accurately with an average Mean Absolute Error (MAE) value of 1.02 mm, but its accuracy
decreased for high daily rainfall (>20mm) with an MAE value of 14.73 mm. This was
attributed to high rainfall events that are associated with distinctly localised convective
clouds within most regions in South Africa (Moeletsi et al. 2016). Despite the high MAE
value for high daily rainfall, it was concluded that the IDW function performs well and thus,
it is recommended to estimate missing data in the Free State Province.

________________________________________________________________________
38
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Shepard (1968) recognised that for the IDW function to be intuitively reasonable, the
elevation and direction also needs to be considered….

1.5.6[2.5.7] Kriging

A prevalent, and probably the most favoured, interpolation method in geostatistics is

kriging (Oliver and Webster, 2014). The method was originally invented by a South African
mining engineer, Danie Krige, to estimate the boundary between rich and poor ores. The
method was later formalised by a French mathematician and engineer, Georges Matheron
(Chilès and Desassis, 2018; Oliver and Webster, 2014). The method is also used in many
other scientific domains, and since its development, it continues to be actively researched.
The method is researched to the extent that the period between 2010 to 2016 has seen a
total of 1 076 articles on kriging published by the journals of the Scopus database (Chilès
and Desassis, 2018). The kriging method was therefore an obvious choice to consider for
this research. It is much more complicated compared to, for example, to the IDW method
and involves a number of calculations which needs to be repeated for every ungauged
site. The model is often left to specialised computer software programs to solve without the
necessary knowledge to make appropriate decisions (Oliver and Webster, 2014). Thus,
the calculation process of the ordinary kriging model was reviewed and the necessary
equations to appropriately apply the method is provided.

According to the kriging model, the estimated rainfall value at an ungauged site is
calculated using the weighted average of all the stations in a neighbourhood as per
Equation 2.30 which is known as the kriging estimator.

n
^Z ( x 0 ) =∑ λi ∙ Z ( xi ) [ 2.30 ]
i=1

where:
^Z ( x 0 ) = estimated value at the ungauged site,
n = the number of stations nearest to the ungauged site,
λi = weight assigned to station i with the total weights equal to 1, and
Z ( xi ) = known rainfall value at station i .

The weights assigned to each station are chosen to minimize the prediction error variance
as per Equation 2.31 and together with the kriging estimator, it is known as the ordinary
kriging model.

n n

∑ λi ∙ γ ( x i−x j ) +ψ (x 0 )=γ ( x j−x 0 ) ; ∑ λ i=1 [ 2.31 ]

i=1 i=1

where:
λi = weight assigned to station i with the total weights equal to 1,

________________________________________________________________________
39
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
γ ( x i−x j ) = semi-variance between data points i and j ,
γ ( x j−x 0 ) = semi-variance between data points j and the ungauged site x 0, and
ψ (x 0 ) = Lagrange multiplier introduced to minimise the error variance.

From Equation 2.31, a set of N+1 linear equations with an equally number of unknowns
are produced which can be solved by matrix algebra as per Equation 2.32.

A × λ=b →

[ ][ ][ ]
γ ( x 1−x 1 ) γ ( x 1−x 2 ) ⋯ γ ( x1 −x n) 1 λ1 γ ( x 1−x 0 )
γ ( x 2−x 1 ) γ ( x 2−x 2 ) ⋯ γ ( x2 −x n) 1 λ2 γ ( x 2−x 0 )
[ 2.32 ]
⋮ ⋮ ⋱ ⋮ ⋮ × ⋮ = ⋮
γ ( x n− x1 ) γ ( x n−x 2 ) ⋯ γ ( xn −x n ) 1 λn γ ( x n−x 0 )
1 1 ⋯ 1 0 ψ ( x 0) 1
where:
A = (n+1) x (n+1) matrix for the semi-variance between data points i and j ,
λ = (n+1) vector matrix for the weights to be applied to each station, and
b = (n+1) vector matrix for the semi-variances between the ungauged site and
data points i and j .

To solve for the weights in Equation 2.32, the semi-variances between the stations, and
the ungauged site are calculated using Matheron’s method of moments (Oliver and
Webster, 2015) as per Equation 2.33.

m (h )
1
∑
2
γ ( x i−x j )=
2 m(h) i=1
{ γ ( x i )−γ ( x i−h ) } [ 2.33 ]

where:
γ ( x i−x j ) = semi-variance between data points i and j ,
m(h) = number of paired comparisons at lag h , and

The semi-variances are then used as the experimental variogram, and a function is
selected that fits the data best in order to populate Equation 2.32. The variogram and other
aspects to consider when an ordinary kriging model is used to estimate the rainfall at an
ungauged site is discussed separately in more detail in the following sections.

1.5.6.1[2.5.7.1] Variogram

Using the kriging model to estimate the rainfall at an ungauged site is the second last step
in a series of calculations with the graphical displaying of the results on a map being the
last step (Oliver and Webster, 2014). During the initial stages, the rainfall data is
processed to calculate a variogram, comprising of an experimental variogram and a model
variogram. The experimental variogram is determined from the rainfall data by calculating
________________________________________________________________________
40
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
the semi-variances at discrete lag distances between pairs of stations using Equation 2.33
(Oliver and Webster, 2014). Because of the discreteness of the lag distances and the
irregular spacing of stations, a lag interval, bin width, and sometimes the directions, must
be chosen to group stations together. The function that best fits the experimental
variogram becomes the model variogram which is continuous for all lag distances. A
number of functions are available to represent the data. The most popular functions are
the spherical function as per Equation 2.34, the exponential function as per Equation 2.35,
the Gaussian function as per Equation 2.36, and the power function as per Equation 2.37
(Oliver and Webster, 2014).

Spherical :

{ ( ) }for 0<h ≤ r
3
3h 1 h
γ ( h )=c0 + c +
2r 2 r [ 2.34 ]
¿ c 0 +c for h>r
¿ 0 for h=0
where:
c = spatially correlated variance,
c0 = uncorrelated nugget component, and
r = range.

Exponential :

{
γ ( h )=c0 + c 1−exp (−ha )}for h> 0 [ 2.35 ]

¿ 0 for h=0
where:
a = distance parameter.

Gaussian :

{
γ ( h )=c0 + c 1−exp (−ha )}for h> 0 [ 2.36 ]

¿ 0 for h=0
where:
a = distance parameter.

Power :
η
γ ( h )=c0 + b h for h> 0 [ 2.37 ]
¿ 0 for h=0
where:
a = distance parameter.

________________________________________________________________________
41
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1.5.6.2[2.5.7.2] Neighbourhood

…………….

Northing (Y)

ce
0)

w um
an

th
nd im
4
st

id
g
di

ba ax
La
um

M
im
ax
M

g 3
La
2g
La
1g
La

Tolerance, (dα)
Direction, (α)
Easting (X)
0)

Figure 2.11: Tolerance angles and distance tolerances (Mert and Dag, 2017)

1.5.7[2.5.8] Co-kriging

…………….

1.6[2.6] STOCHASTIC RAINFALL GENERATION

1.6.1[2.6.1] Monte-Carlo simulations

In order to determine the uncertainty pertaining to the data period and missing data, the
Monte-Carlo Simulation technique was considered for this project because it is often used
in hydrology to quantify model output uncertainties (Al Mamoon and Rahman, 2014). It has
the advantage of investigating the impacts of many possible combinations of the input
variables and model parameters in design rainfall estimation (Al Mamoon and Rahman,
2014). It also offers an alternative approach to the design event concept used by
deterministic peak discharge calculations (Nathan and Weinmann, 2013). This approach
attempts to simulate the natural variation in rainfall by recognising that the peak discharge
could be because of a variety of random variables (Nathan and Weinmann, 2013). Nathan

________________________________________________________________________
42
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
and Weinmann (2013) compiled a report that describes the use of Monte-Carlo simulations
in a practical way so that it can be implemented in a spreadsheet. Hence, the review on
the use of Monte Carlo simulations provided in this section originates largely from Nathan
and Weinmann (2013).

The Monte-Carlo simulation approach for stochastic sampling is based on the inverse
transformation method where a random number is generated uniformly between 0 and 1.
The stochastic generated rainfall values are then calculated using the inverse of the
cumulative density function of the underlying distribution. The approach is depicted in
Figure 2.12 which illustrates three randomly generated numbers that are equally spaced
along the horizontal axis of the cumulative distribution function. As the numbers are
transformed the spacing of the rainfall values along the vertical axis are unequal in
accordance with the adopted distribution (Nathan and Weinmann, 2013).

x x

U: 1 2 3

F(x) f(x)

Figure 2.12: Illustration of the inverse transformation method (Nathan and Weinmann, 2013)

The inverse functions of the Normal , log-Normal, Beta, and Gamma distributions are pre-
programmed into standard spreadsheets (Nathan and Weinmann, 2013), like Microsoft
Excel, but the inverse function of the GEV distribution is not. Smithers and Schulze (2000)
found the GEV distribution to be appropriate for short duration design rainfall in South
Africa. Therefore, the inverse function of the GEV distribution is provided in Equation 2.38.

{
F−1 ( x )= ξ +α {( ln x ) −1 }/k ,∧k ≠ 0
−k
[ 2.38 ]
ξ−α ln (−ln x ) ,∧k =0

Kottegoda et al. (2014) developed a simple model to simulate rainfall hyetographs in two
regions of northern Italy using Monte-Carlo simulations. Hyetographs were stochastically
generated in 5-min intervals and validated using the Depth-Duration-Frequency (DDF)
curves generated from observed rainfall data. A minimum data period of 10 years was
used which generated accurate DDF curves. The potential of using their model at an
ungauged site was also illustrated. The parameters of the model were estimated from
observed rainfall data. A maximum dry period criterion of 10-hour was used to separate
________________________________________________________________________
43
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
the data into individual rainfall events. The criteria to identify significant events were a
minimum storm duration of 20-min and a cumulative rainfall of 25 mm.

Some statistical properties of each significant event were determined. This included the
total duration of the event, the durations of wet and dry spells, the timing of the peak
intensity, and the number of times the intensity increased or decreased from the previous
interval in contrary to the trend. According to Kottegoda et al. (2014) the number of events
in a year were normally distributed. The number of spells within each event, and the
lengths of wet and dry spells, had geometric distributions. The cumulative rainfall was
directly related to the duration of the wet spells. Likewise, the timing of the peak intensity
with a lognormally distribution. The cumulative rainfall before the peak intensity was
related to its timing. The cumulative rainfall before and after the peak intensity had beta
distributions with its parameters being gamma distributed, and the relationship between
the alpha and beta parameters being linear.

Once the statistical properties of the observed hyetographs were determined, then
hyetographs were stochastically generated using the appropriate distributions (normal, log-
normal, and geometrical).

________________________________________________________________________
44
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
‘

Calibration

Simulate number Simulate number

Input number
of events per year of separate Simulate single
of year in
using normal hyetographs per hyetograph
simulation
distribution events

Simulate duration of
Simulate length of dry run
rainfall
End of
event
Simulate position of
maximum

Simulate
Simulate peak
cumulative rainfall End of
intensity of rainfall
of hyetograph year

For rising limb Simulate point

Simulate
simulate rainfalls for
reversals
cumulative rainfall limb

End of
simulation
For falling limb evaluate
of cumulative rainfall as Simulate point
Simulate
a difference between rainfalls for
reversals
cumulative rainfall and falling limb
peak intensity

Draw
hyetograph
Draw depth-duration curves for different return periods

Figure 2.13: Flow chart for Monte-Carlo simulation (Kottegoda et al. 2014)

1.6.2[2.6.2] Bartlett-Lewis rectangular pulse models

……………………

________________________________________________________________________
45
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1.7[2.7] SYNTHETIC DESIGN STORM PARAMETERIZATION

1.7.1[2.7.1] Background

Due to inherent complexities in rainfall and runoff generating mechanisms that are both
nonlinear and multivariate, the use of a single design storm as a unique climatic input into
a hydrological model is not realistic (Balbastre-Soldevila, García-Bartual and Andrés-
Doménech, 2019). It is often incorrectly assumed that the return period of the design storm
hyetograph is the same as that of the runoff hydrograph produced by it, where this is
almost never the case (Adams and Howard, 1986). The use of historical records and
synthetic events derived from stochastic rainfall models are shown, in some cases, to be
more accurate and representative, covering a wide range of possible scenarios (Larsen,
1981; Willems, 1999; Maßmann, Krämer, Fuchs, Herrmann, Kuchenbecker, Birkholz,
Sympher, Haberlandt, Morales and Eisele, 2019).

Despite the limitations, the use of design storms remains very popular as an efficient
engineering tool in urban hydrology studies and for the planning and design of stormwater
management facilities (Balbastre-Soldevila et al., 2019; Guo and Zhuge, 2008). Design
storms are still needed and will remain a fundamental component of infrastructure
development and implementation in the years to come (Markolf, Chester, Helmrich and
Shannon, 2021).

Some of a synthetic design storm’s inherent advantages are highlighted below:

i. It guarantees a uniform level of quality and operation standards across different
urban catchments (Balbastre-Soldevila et al., 2019).
ii. It simplifies and reduces the hydrologic-hydraulic calculations compared to
employing historical or stochastic rainfall methods. These methods imply a much
higher computational cost that is not justified in many practical instances (Marsalek
and Watt, 1984). As pointed out by Strecker and Huber (2008), synthetic design
storms are standardized and easy to construct.
iii. Scarcity of data is a frequent problem experienced in real applications of urban
hydrology engineering, and in this case the design storm approach is the most
efficient way to tackle the design problem (Balbastre-Soldevila et al., 2019).
iv. Another advantage is that if a design storm is parameterised in a simple form, then
regional approaches that include the mapping of design storm parameters can be
very useful for urban drainage design purposes (Balbastre-Soldevila et al., 2019).

Results of rainfall-runoff modelling shows that hyetograph shape has a strong effect on the
shape of a catchment’s outflow hydrograph and total flood volume (De Lima and Singh,
2002; Marani, Grossi, Napolitano, Wallace and Entekhabi, 1997; Alfieri, Laio and Claps,
2008). Baiti, Bouziane, Ouazar and Hasnaoui (2017) found that design peak discharges
are not necessarily affected by the maximum intensity value of a design storm, but rather
________________________________________________________________________
46
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
by their structure or hyetograph shape, and Wenzel Jr and Voorhees (1984) confirmed that
the hyetograph used for a design storm can significantly affect the return period of the
resulting design. Careful consideration of the available synthetic design storms is required
to ensure that the stormwater infrastructure designed for an urban catchment is sufficient
to deal with the resulting runoff (Prodanovic and Simonovic, 2004).

The wide variety of design storms available has created confusion as to which should be
adopted in design practises to obtain correct flood peak estimates (Alfieri et al., 2008).
Mouton et al. (2022) state that due to the abundance of methods available, engineers
usually base their choice of method on familiarity and preference, rather than on evidence
of the appropriateness of the selected method. The need for typical synthetic design
storms applicable to single event-based modelling of urban catchments in South Africa
was identified by Mouton et al. (2022).

1.7.2[2.7.2] Current Performance of Synthetic Design Storms

The synthetic design storms applicable to South Africa, as covered in the research done
by Mouton et al. (2022) include:
i. Rectangular hyetograph (Mulvaney, 1851),
ii. Triangular hyetograph (Yen and Chow, 1980),
iii. USDA’s SCS method (SCS, 1973),
iv. SCS-SA method (Schulze and Arnold, 1984),
v. Huff curves (Huff, 1967) and,
vi. Chicago Design Storm (CDS) (Keifer and Chu, 1957).

1.7.2.1[2.7.2.1] Rectangular hyetograph

The rectangular hyetograph is the simplest design storm and is closely related to the
Rational Method (Balbastre-Soldevila et al., 2019). This method assumes that the rainfall
intensity is constant for the duration of the storm, which has been shown to give an
incorrect picture of a hyetograph (Niemczynowicz, 1982). Pilgrim and Cordery (1993)
suggest that the rectangular hyetograph is most widespread due to its ease of use.

However, this method has been shown to underestimate the total precipitation volume,
peak flow and total flood volume across various different scenarios and locations (Arnell,
1982; Alfieri et al., 2008). This makes it likely that stormwater infrastructure designed using
this method could be under designed and not adequate to effectively deal with future storm
events. Mouton et al. (2022) found the rectangular hyetograph method to be the worst
representation of observed events when evaluating synthetic design storms in Gauteng,
South Africa.

________________________________________________________________________
47
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1.7.2.2[2.7.2.2] Triangular hyetograph

The triangular hyetograph was developed by Yen and Chow (1980). Although this method
is also simple and intuitive, it does not have a strong conceptual basis and may produce
biased flow estimates (Veneziano and Villani, 1999).

1.7.2.3[2.7.2.3] SCS and the SCS-SA method

The SCS temporal distribution curves were first developed by the Soil Conservation
Service (SCS), which later became the Natural Resources Conservation Service (NRCS)
in the United States (SCS, 1973). These curves were developed from generalised rainfall
depth-duration relationships across the United States. This method has primarily been
used in the design of small dams, although it has been applied in many urban and rural
catchments over the years (Prodanovic and Simonovic, 2004). The SCS hyetograph (and
the derived SCS-SA hyetographs) are characterised by a “peaky” shape, with a high
intensity burst near the centre of the event. Measured rainfall during tropical cyclones or
similar extreme events is the basis for the high intensity burst (Strecker and Huber, 2008).
Prodanovic and Simonovic (2004) recommended that this method performed best for a
catchment that experiences high intensity precipitation over a short duration.

These standard SCS curves were adopted for use in South Africa by Schulze and Arnold
(1979), resulting in the development of the SCS-SA temporal distribution curves. The
SCS-SA method was refined, and by observing rainfall stations across South Africa,
Weddepohl (1988) was able to group South Africa into 4 regions with their own SCS-SA
distributions (Type I, II, III, IV).

The main appeal of this category of methods that uses standardised profiles obtained
directly from rainfall records is that the resulting hyetograph is based on the actual data of
intense regional precipitation. Furthermore, when evaluating hydrological extremes and
risks in a catchment, the advantage of these methods is that precipitation of return periods
exceeding 100 years can easily be used because these methods do not rely on IDF data
(Prodanovic and Simonovic, 2004). However, due to a large number of uncertainties
surrounding the physical variability of rainfall and the definition of a storm, temporal
smoothing needs to be performed for these methods, which may miss some of the
important features of the rainfall at the locality of interest (Prodanovic and Simonovic,
2004).

During the comparison between various synthetic design storm methods, Baiti et al. (2017)
found that the peak discharge generated using the SCS Type 1 design storm generated
similar results to the peak discharge generated from historical rainfall data for an
experimental catchment in Tangier, Morocco. Kang, Goo, Song, Chun, Her, Hwang and
Park (2013) also found the SCS model to provide an acceptable method, especially for
urban areas with significant impervious land cover.
________________________________________________________________________
48
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

A weakness of these methods is that large sample sets of data are required for the
construction of regional profiles (Prodanovic and Simonovic, 2004). It is therefore not
feasible to construct regional profiles for each specific catchment locality, and the four
regions across South Africa with their own SCS-SA distributions are used to provide
indicative regional profiles for any locality across the country. However, research done in
the Gauteng Province by Males, Braune, Van Bladeren and Mahlangu (2004), showed that
the SCS-SA Type 3 curve applicable to this area yielded peak discharges that were too
high for certain sites in Gauteng, when applied to a SWMM model.

Mouton et al. (2022) compared the design rainfall ratios of the SCS-SA curves to the ratios
determined from the Design Rainfall Estimation for South Africa (DRESA) software’s
design rainfall (Smithers and Schulze, 2003). It was observed that the Gauteng weather
stations were generally closer to the SCS-SA Type 2 curve, rather than the Type 3 curve
stipulated by the SCS-SA method developed by Schulze and Arnold (1979). An improved
procedure was proposed to generate a synthetic design storm applicable to small
catchments in Gauteng through the development of intermediate SCS-SA curve types. It
was recommended that this improved procedure be expanded for use across the whole of
South Africa in future studies.

1.7.2.4[2.7.2.4] Huff curves

The Huff curves presented by Huff (1967) were created by separating independent storms
using certain criteria, and then by non-dimensionalising each mass curve in terms of the
total rainfall volume and storm duration. This methodology was, however, never
documented, which led to the revised and formulated methodology presented by Bonta
(1997).

Adamson (1981) suggested that Huff curves may be representative of regions in Southern
Africa with rainfall climate and topography similar to that of the Mid-West of the United
States. Mouton et al. (2022) applied the Huff curve method in the Gauteng Province but
showed that this method underestimated the design rainfall for short time steps. This
would have resulted in the underestimation of peak discharge and so the method was
deemed inappropriate for single event-based modelling of urban catchments with rapid
response times, which are typical across South Africa.

1.7.2.5[2.7.2.5] Chicago Design Storm (CDS)

The CDS method was first developed by Keifer and Chu (1957). This method considers
antecedent rainfall before the maximum intensity duration, the average intensity during the
maximum period, and the location of the peak intensity. This method’s intended application
was the sizing of stormwater networks with a storm duration of 3 hours, although there is
nothing in the method that limits it to only these applications (Prodanovic and Simonovic,
________________________________________________________________________
49
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
2004). The CDS method uses point rainfall that applies to a particular site and so the
unique characteristics and variability of the rainfall patterns are automatically embedded in
the CDS method (Mouton et al., 2022). Mouton et al. (2022) also noted that methods that
use the whole IDF curve (such as the CDS), to produce a synthetic design storm, have
great potential for use in South Africa due to the regionalisation of design rainfall produced
by Smithers and Schulze (2000).

Balbastre-Soldevila et al. (2019) found that the Alternating Blocks method (Te Chow et al.,
1988), a variant of the CDS, performed very well when compared to other design storms
for a single-event based simulation. It was found that the Alternating Blocks method
performed the best of all the design storms obtained from IDF curves and presents the
advantage of being the most widespread and internationally used.

It has, however, been found across a variety of studies that the CDS tends to overestimate
peak discharge as it produced an unrealistic single event (Veneziano and Villani, 1999;
Malik and James, 2007; Alfieri et al., 2008).

When comparing the simulated peak flow using the CDS and the real historical peak flow,
Alfieri et al. (2008) found that the percentage error calculated (EQ) was very consistent
across a number of different catchment and climatic scenarios, which showed that the
CDS is characterised by a very stable bias. This was observed for a CDS with a midpoint
peak and an advancement coefficient of 0.5 (r = 0.5). This allowed for the application of a
correction factor to remove the bias. This multiplicative correction factor for the CDS was
found to be 0.94.

During the comparison of various synthetic design storm methods, Baiti et al. (2017) also
showed that the CDS allowed for the determination of peak discharges with minor
deviations compared to the peak discharges generated based on observed rainfall data.
Watson (1981) compared the CDS method to two real storms in Johannesburg and
concluded that the CDS was an adequate technique for predicting peak discharge.

Mouton et al. (2022) found that the CDS method provided consistent results when
compared to observed rainfall for both single events and a continuous simulation. It was
recommended that a sensitivity analysis be conducted to determine the effect of the
advancement coefficient and total storm duration on the peak discharge of a model.
Mouton et al. (2022) also highlighted a number of advantages of the application of the
CDS method in South Africa, namely:
i. The IDF coefficients can be accurately determined from the design rainfall obtained
using the Design Rainfall Estimation in South Africa (DRESA) software developed
by Smithers and Schulze (2003).
ii. The location of the hyetograph’s peak intensity in relation to the total event duration
can be adjusted to be anywhere between the start and end of the storm event.

________________________________________________________________________
50
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1.7.3[2.7.3] Sensitivity of Hydrological Models to Changes in Synthetic Design Storm
Parameters

As shown by Mouton et al. (2022), both the SCS-SA and CDS methods show promising
results for use across South Africa, however it was recommended that a sensitivity
analysis be conducted to determine the effect of various design storm input parameters on
the peak discharge of a model.

1.7.3.1[2.7.3.1] Storm Duration

Alfieri et al. (2008) considered an “ideal” river basin with a linear and time-invariant
hydrologic response. They estimated the basin outflow hydrograph using an instantaneous
unit hydrograph (IUH) with a particular distribution. For simulations using this simplified
hydrological model it was observed that the peak discharge generated using the Chicago
Design Storm appeared to be independent of the storm duration.

Baiti et al. (2017) found that the CDS maximum intensity remained constant despite
changes to the storm duration. However, although the maximum intensity of the CDS
hyetograph remaining constant, changes in the storm duration did affect the peak
discharge generated by the CDS. As shown in Figure 2.14, reducing the storm duration by
50% led to a 15% reduction in the peak discharge, while increasing the storm duration by
200% led to a 18% increase in the peak discharge simulated. Baiti et al. (2017) also
observed that the storm duration is the only parameter that significantly influences the
hydrograph volumes produced, for all design storm methods that were investigated. The
hydrograph volume increased as the storm duration was increased, and decreased as the
storm duration was decreased. This is expected as an increase in the storm duration does
not change the maximum rainfall intensity, but will increase the total rainfall volume which,
in turn, should increase the total runoff/hydrograph volume.

________________________________________________________________________
51
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
20%

15%

% Change in Peak Discharge 10%

0%
-50% 0% 50% 100% 150% 200%

-5%

-10%

-15%

-20%
% Change in Storm Duration

Figure 2.14: Sensitivity of Peak Discharge to changes in Storm Duration for the Chicago
Design Storm (Baiti et al., 2017).

Knighton and Walter (2016) highlighted that engineers either assume that the storm
duration is linked to the catchment’s time of concentration, or they consider the critical
duration of a storm to be the storm duration that maximizes the runoff at a given
frequency. Kang et al. (2013) suggested that estimating the time of concentration should
only be a first approximation for the storm duration, and that the critical storm duration was
a crucial factor, along with the design rainfall amount, in estimating and designing
hydraulic structures. However, Knighton and Walter (2016) proposed that both the time of
concentration and critical storm duration approaches oversimplify design flood estimation,
and that researchers should consider the critical characteristics of other event statistics
and their joint probability of occurrence when calculating event frequency and storm
duration.

Guo and Zhuge (2008) proposed that a time of dispersion caused by a channel reach, be
added to the upstream catchment’s time of concentration to account for the peak
attenuation effect of the channel reach, when selecting a suitable duration for a design
storm. This time of dispersion for a detention pond was recommended to be two times the
detention time that the pond provides.

1.7.3.2[2.7.3.2] Position of Peak Intensity

The position of the peak intensity for the CDS is set using the advancement coefficient
parameter (r), which is the ratio of the storm duration preceding the peak intensity relative

________________________________________________________________________
52
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
to the total storm duration (Keifer and Chu, 1957). This parameter considers the
antecedent rainfall volume that occurs before the peak intensity.

As explained by Eagleson (1972), cumulative hyetographs of rainfall in convective or

frontal type storms have their peak rainfall rates early in a storm, while cyclonic storms
tend to have their peak towards the middle of the storm event. As is well known and
documented, South Africa’s rainfall consists of frontal storms sweeping across the coast
from East to West, with convectional thunderstorms across the inland plateau of the
country (Wellington, 1929). Therefore, with this in mind, the development of a synthetic
design storm specific to South Africa may require an advanced peak position.

The sensitivity analysis conducted by Alfieri et al. (2008) found that the peak position of
the CDS did affect the peak discharge estimation, with errors between the simulated
results and actual measured data ranging from -10% to 8%. The CDS with a midpoint
peak was always found to overestimate the peak discharge results but produced the most
stable results. Alfieri et al. (2008) deemed this to be the most suitable of the peak position
variations as the overestimations could be accounted for by a multiplicative correction
factor of 0.94, as highlighted in Chapter 2.7.2.5.

Baiti et al. (2017) used a midpoint peak position, with the advancement coefficient (r) for
the CDS set to 0.5 as the reference case. When conducting a sensitivity analysis for the
peak position of the CDS it was found that for an advanced storm with r = 0.25, the peak
discharge was observed to decrease by 20%, while for the CDS with r = 0.75, the peak
discharge was observed to increase by 20%. This is shown in Figure 2.15. Baiti et al.
(2017) also showed that the total runoff volume was not affected by changes to the peak
position for both the CDS and triangular hyetograph.

________________________________________________________________________
53
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
25%

20%

15%
% Change in Peak Discharge
10%

0%
-50% -40% -30% -20% -10% 0% 10% 20% 30% 40% 50%
-5%

-10%

-15%

-20%

-25%

% Change in Peak Position (r)

Figure 2.15: Sensitivity of Peak Discharge to changes in Peak Position for the Chicago Design
Storm (Baiti et al., 2017).

When investigating a sensitivity analysis for parameters used in the generation of a

stochastic rainfall model and design storm, Knighton and Walter (2016) found that the
interevent time of precipitation was a sensitive parameter for the peak discharge
simulated. This showed that the catchment runoff was a function of the precipitation as
well as the antecedent conditions of the catchment. The antecedent conditions of a
catchment can have as much effect on the simulated peak discharge as the hyetograph
shape, according to Wenzel Jr and Voorhees (1984), and they proposed that the average
antecedent conditions are used for design rather than extreme conditions. This is
consistent with the CDS methodology originally proposed by Keifer and Chu (1957), who
calculated the advancement coefficient using the average antecedent rainfall volume for
the city of Chicago.

The study by Knighton and Walter (2016) also highlighted the importance of the event
temporal distribution when defining design hyetographs. It was observed that the temporal
loading of the flood inducing precipitation events were skewed to the front of the storms,
i.e., they were advanced storms with r < 0.5. This observation was confirmed by the fact
that for 60% of the flood producing event, 75% of the event rainfall volume occurred before
the halfway point of the storm.

Keifer and Chu (1957) originally determined the advancement coefficient for the city of
Chicago to be 0.386 when considering the antecedent rainfall volume, while when ignoring
the antecedent rainfall before the maximum duration, the advancement coefficient was
determined to be 0.375. Watson (1981) determined the advancement coefficient for
Johannesburg to be 0.28 and 0.22 for a two and three hour storm duration respectively,
________________________________________________________________________
54
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
while Weesakul, Chaowiwat, Rehan and Weesakul (2017) determined the advancement
coefficient for Bangkok, Thailand to be between 0.20 and 0.49, based on the recorded
rainfall data in the area. Mouton et al. (2022) found that for the Gauteng Province of South
Africa, the average advancement coefficient across all of the weather stations observed
was 0.38, which is similar to the coefficient proposed by Keifer and Chu (1957).

1.7.3.3[2.7.3.3] Time Step

The study conducted by Baiti et al. (2017) showed that the maximum intensity of CDS
decreased with an increase in time step. However, the intensities of the other methods
investigated, which included the rectangular and triangular hyetographs, remained
constant when the time step was changed. It was therefore shown that the CDS maximum
intensity is dependent on the time step used, but independent on the storm duration.
Interestingly, only slight changes in the peak discharge and flood volume were observed (-
5% to +5%) when changes to the time step was made for the CDS by Baiti et al. (2017).

1.7.4[2.7.4] Comparison between Single Event-Based and Continuous Simulation Modelling

Strecker and Huber (2008) identify the major categories of precipitation inputs and related
modelling analyses to be:
i. Synthetic Design Storms,
ii. Continuous Simulations, and
iii. Derived runoff-based statistics of input rainfall.

Although there are many positive attributes of synthetic design storms, as discussed in
Chapter 2.7.1, the main limitation is that it is wrongly assumes that the return period of the
hyetograph is the same as the return period of the resulting hydrograph. This is patently
wrong as it neglects the effects of the antecedent conditions which can cause the same
runoff peak to result from many combinations of different rainfall events (Strecker and
Huber, 2008; Adams and Howard, 1986).

Design methods based on design storms developed from IDF curves cannot reveal
information on the long-term, average performance of a drainage system because they
depict only specific, selected events of intermediate frequency (Adams and Howard,
1986). Strecker and Huber (2008) also noted that synthetic design storms are thought to
be conservative for the purpose of flood control design, but that it was not possible to
perform a “thoughtful over-design” with these techniques because there is no way to get a
more accurate estimate of the return period for the simulated runoff. Nnadi, Kline, Wray Jr
and Wanielista (1999) also noted the importance of correctly sizing a stormwater
management structure to deal with potential flood risks, while not imposing unnecessary
expenses on the developer due to an overly conservative design.

________________________________________________________________________
55
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Adams and Howard (1986) stated that given a record of sufficient length and quality, and
an adequate simulation model, the statistics of continuous simulation model output and the
drainage system performance would be reliable. Adams and Howard (1986) also
recommended that continuous analysis should, at the very least, be used as a screening
process for selecting more detailed analyses to test system designs, and should, in the
future, penetrate deeper into the design process.

Strecker and Huber (2008) outlined the process of performing a continuous simulation in
the following steps:
i. Obtain long term historical precipitation data,
ii. Obtain other required meteorological data, such as evaporation data,
iii. Use a calibrated model capable of accepting such data and execute the simulation,
iv. Evaluate the output of long-term hydrographs, which is often enormous.

Strecker and Huber (2008) also highlighted a number of advantages of continuous

simulation, including:
i. Antecedent conditions are accounted for,
ii. The design event is chosen on this basis of the return period of the output of
interest, such as peak discharge, runoff volume or pollutant load,
iii. The choice of an arbitrary storm duration is not required,
iv. Citizens can know that a storm from memory (e.g., The KZN floods of April 2022)
will be controlled by the design, whereas the design for a hypothetical event is not
as clear.

Some disadvantages and theoretical issues of continuous simulations have also been
highlighted during various studies. Firstly, the time series of historical rainfall must be
available, and for hyetograph intervals short enough to be appropriate for urban areas with
short response times (Strecker and Huber, 2008; Nnadi et al., 1999). This makes its
application challenging when there is limited data available (Hossain, Hewa and Wella-
Hewage, 2019). Another disadvantage of continuous simulations is that it becomes a
challenge to determine large return period hydrographs from a limited historical record,
and it was pointed out that scaling up a historical rainfall event is little better than using a
synthetic design storm (Strecker and Huber, 2008).

Chu and Steinman (2009) compared event-based modelling to a continuous approach in

West Michigan, USA, and found that the event-based model performed better than the
continuous simulation. Azmat, Qamar, Ahmed, Hussain and Umair (2017) found the event-
based model to be more accurate than the continuous simulation for a high altitude
catchment in the Western Himalayas. Sampath, Weerakoon, Mishra and Herath (2014)
used both the event-based and continuous simulation approach to predict streamflow
hydrographs in Tittawella, Sri Lanka, and the event-based model was found to perform
better. De Silva, Weerakoon and Herath (2014) also found that the event-based model

________________________________________________________________________
56
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
performed better when simulating extreme events in the Kelani River basin, Sri Lanka. In
the review of Australian Rainfall-Runoff computed by Ball, Babister, Nathan, Weinmann,
Weeks, Retallick and Testoni (2016), the continuous simulation model was shown to have
limited ability to represent rural catchment behaviour.

Hossain et al. (2019) compared the performance of event-based modelling to continuous

simulation modelling for two catchments in South Australia using the SWMM rainfall-runoff
model. For the two case study catchments used, it was found that event-based modelling
performed better in reproducing both the direct runoff and total runoff hydrographs.

In contrast, Cunderlik and Simonovic (2005) found that continuous simulations produced
more reliable hydrographs than event-based models. The continuous simulation approach
was also successfully used to estimate design floods and carry out real time flood
forecasting by Brocca, Melone and Moramarco (2011), Boughton and Droop (2003) and
Paquet, Garavaglia, Garçon and Gailhard (2013).

Strecker and Huber (2008) compared the SCS synthetic design storm to a continuous
simulation and found that the continuous simulation was more accurate in producing the
measured peak discharge in a catchment, and that the synthetic design storm generated
very conservative results. It was therefore their conclusion that synthetic design storms are
better suited to flood control, while continuous simulation, where possible, was better
suited to assigning realistic frequencies to runoff peaks, volumes and other parameters of
interest for more frequent storm events.

Hossain et al. (2019) highlighted that there is generally a clear distinction between
hydrographs generated by urban and rural catchments, and therefore the performance of
event-based and continuous simulation models can vary from one catchment to another.
Strecker and Huber (2008) pointed out that practitioners resist continuous simulation
because of resource and money restraints, and it was noted by Mouton et al. (2022) that in
South Africa, continuous and reliable rainfall data is increasingly hard to come by and so
the use of event-based synthetic design storms are essential.

1.7.5[2.7.5] Assessment of an Ensemble Approach to Single Event-Based Modelling

Dlamini, N. S. 2019. Development and assessment of an ensemble joint probability event

based approach for design flood estimation in South Africa.

________________________________________________________________________
57
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

CHAPTER 2: INCEPTION REPORT

2.1 OBJEECTIVE

The objective of this study is to test the performance of the existing design storm shapes,
and to develop, assess and investigate the regionalisation of improved synthetic design
storms applicable to small catchments in South Africa. The specific aims that will be
applied to meet this objective are summarised in Table 1.1.

2.2 SCOPE AND LIMITATIONS

2.3 PROPOSED METHODOLOGY

The research aims will be achieved by assessing the performance of existing methods and
testing them against recently measured sub-daily rainfall data across South Africa to
develop, assess and investigate the regionalisation of improved synthetic design storms
applicable to small catchments in South Africa. The proposed methodology is described in
the following paragraphs.

2.3.1 Data quality assessment

The quality of the rainfall data will be assessed and the effect of applied quality control
procedures on the design rainfall estimation will be investigated and interpreted. The
assessment of the quality of climarte data is imperitave in any climatologically based
research, particularly in instances where extreme data values are the major input.
Because a large fraction of the data sets consist of relatively short time series, it follows
that erroneous extreme values could have relatively large effects on the reliability of the
subsequent statistics developed from the data. This need was identified because of the
scarcity of short duration rainfall data and the short data period of the available data limits
the possibilities and reliability of the research that is conducted using the data. Figure xx
presents a spatial representation of an initial audit of sub-daily data available for the
research.

________________________________________________________________________
58
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure xx. Initial audit of SAWS Automatic Weather- and Rainfall Stations (AWSs and
ARSs) available for the research.

The comprehensive assessment of usable data entails an audit of the completeness of

individual station data sets and subsequently the application of a set of quality control
procedures to investigate the general quality of the data as well as the accurate
measurement of individual storms used in the estimation of the eventual statistics. The
quality of the available data is further downgraded by missing data and possible
inaccuracies. Internal and regional consistency will be evaluated, as shown in e 3.17.
Furthermore, the possible severity of inaccuracies will be determined. Currently sub-daily
data are quality controlled by SAWS through the identification and verification of outliers in
the 5-minute data, as well as monthly checks on the spatial consistency of rainfall totals.
These checks do not automatically remove dubious data values from the data sets, and
therefore the likelihood of erroneous values in the data is significant. Figure 3.1 presents
an initial proposed approach to do the necessary quality checks on the data. The actual
checks on the data will be done broadly following the checks as developed by Lewis et al.,
(2021) of which the details are presented in Table 2.3. These checks cover most of the
potential errors that can be present in sub-daily data sets. After identification of possible
erroneous values, re-checking will be done visually, as well as verification of reports of
extreme storms in the region of measurement. If feasible and necessary, a classification
criterion with relevant data quality thresholds will be developed to enable easy
identification and assessment of a station’s quality of data and the possibility of data
patching, investigated.

________________________________________________________________________
59
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Quality Control
(5-min Rainfall)

Climatological
Internal review (Replace
consistency with missing data)

Tipping rates Data patching

Accumulated Stochastic
totals simulations

Dry spells Probability

distribution

Regional
consistency

Discordancy Intermediate
measure SCS-SA curves

CDS
Dry period
coefficients

Accumulated Synthetic
daily rainfall design storms

Figure 3.16: Data processing flow chart

________________________________________________________________________
60
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
2.3.2 Design rainfall estimation

The at-site design rainfall will be used to determine the sub-daily rainfall ratios in relation to
the 24-hour rainfall and compared with the ratios of the SCS-SA curves to develop
intermediate design rainfall curves. The maximum values within selected duration ranges
will be determined for each station, and the inverse distance weighting interpolation
method, or another method that was identified in this project to be suitable for this
application, will be used for the geographical interpolation between stations. The concept
of intermediate curves will be expanded to all relevant regions in South Africa. The
Chicago Design Storm method and its coefficients will be further refined, and possible
regionalization of the regression coefficients will be investigated.

2.3.3 Synthetic design storm evaluation

The synthetic design storms will be evaluated by comparing them with the observed
rainfall events, considering the shape of the mass curves, as well as the average
intensities embedded in the synthetic design storm. Each significant storm event, that will
be separated with an appropriate maximum dry period (e.g. 15-min), will be evaluated
using the Mean Absolute Relative Error (MARE) technique to determine the goodness-of-
fit of the synthetic storm event compared to the observed storm event at 5-min intervals.
The location of the peak intensity and the variation in recurrence interval of average
intensities will also be assessed. This investigation will provide a measure to determine the
extent of idealisation of a synthetic design storm which will, (1) contextualise the result of
the next part of this study, and (2) provide some groundwork for the development of a
potential new stochastic method for daily rainfall disaggregation (e 3.17).

2.3.4 Synthetic design storm evaluation tool

Practitioners often revert to familiar synthetic design storm methods rather than an
appropriate method because of the challenge and effort of mastering an unfamiliar
method. Therefore, the need for an open-source tool to generate regional or site-specific
synthetic design storms in South Africa was identified. This project will investigate the
development of software (e.g. an Excel spreadsheet) which will provide the basis for
further development in a future phase, where the tool and methodology will be extended
by incorporating additional functionality in the DRESA software to generate a site-specific
synthetic design storm, and possibly the overhauling of the software.

2.3.5 Knowledge dissemination workshop

A training session is planned to transfer the necessary skills to managers, designers,

technicians and operators involved in planning and design of urban and rural drainage
infrastructure to understand and successfully apply design storm shapes.
________________________________________________________________________
61
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
2.4 E 3.17:TASKS AND DELIVERABLES

Table 3.14Table 1.3 summarises the work done to date. Table 3.16Table 1.4 shows the
list of deliverables and indicates the status and due date of each deliverable.

Table 3.143.15. Work completed to date.

No. Task Summary of work to date
1 Literature review and This deliverable describes the literature review on rainfall
inception report data quality, geographical interpolation methods, and
synthetic design storm parameterization. The inception
report describes the proposed methodology, tasks and
deliverables and capacity building of this project.

Table 3.163.17. Deliverable due dates and deliverables submitted to date.

The complete list of tasks required to meet these deliverables include:

a) Literature review (included in Chapter 2 of this document)

i) Data quality control
ii) Data interpretation
iii) Regionalisation and geographical interpolation
iv) Synthetic design storm parameterization
b) Data quality assessment
i) Data capturing - human and instrumental errors.
ii) Reprocessing database – flag inconsistent data, climatological review.
c) Data interpretation
i) Data period.
ii) Missing data.
iii) AMS, n-largest approach.
iv) Monte Carlo – confidence levels, extrapolation
d) Design rainfall
________________________________________________________________________
62
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
i) At-site (GEV LM)
ii) DRESA (Relative differences, motivation for DRESA overhaul)
iii) Intermediate SCS-SA curves (IC values)
iv) Possible regionalised IDF regression coefficients (CDS)
e) Synthetic design storm parameterization
i) Storm duration
ii) timing of peak intensity
iii) infiltration models
iv) catchment characteristics – size, slope, materials
v) Peak discharge, runoff volume
vi) Reaction times
vii) Guideline for parameterization
f) Synthetic design storm comparisons
i) Shape
ii) Intensity
iii) RI variations

2.5 CAPACITY BUILDING

Table 3.183.19. Registered students.

Surname. Initials Gender Race Degree Institution

Registered students:
Loots I Female White PhD UKZN
Spooner JD Male White MEng UP
Mbatha S Male African MSc UP

Prospective student:
Mouton JvS Male White PhD UP

Knowledge dissemination is planned to take place in the form of a workshop. The results
will also be presented at the UP Flood Modelling course.

________________________________________________________________________
63
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

[CHAPTER 3:] REFERENCES

Adams, B. J. and Howard, C. D. 1986. Design storm pathology. Canadian Water

Resources Journal, Vol 11, No 3, pp 49-55.

Adamson, P. 1981. TR102 Southern African Storm Rainfall. Department of Environment

Affairs, Directorate of Water Affairs.

Al Mamoon, A. and Rahman, A., 2014. Uncertainty in design rainfall estimation: A

review. J. Hydrol. Environ. Res, 2(1), pp.65-75.

Aldrich, J., 1997. RA Fisher and the making of maximum likelihood 1912-1922. Statistical
science, 12(3), pp.162-176.

Alexander, L., Zhang, X., Hegerl, G., Seneviratne, S., Behrangi, A., Fischer, E., Martius,
O., Otto, F., Sillmann, J. and Vautard, R., 2016. Implementation Plan for WCRP
Grand Challenge on Understanding and Predicting Weather and Climate Extremes–
the “Extremes Grand Challenge”. Version, June.

Alfieri, L. Laio, F. and Claps, P. 2008. A simulation experiment for optimal design
hyetograph selection. Hydrological Processes: An International Journal, Vol 22, No 6,
pp 813-820.

Arnell, V. 1982. Rainfall data for the design of sewer pipe systems. Chalmers Tekniska
Hogskola (Sweden).

Azmat, M. Qamar, M. Ahmed, S. Hussain, E. and Umair, M. 2017. Application of HEC-

HMS for the event and continuous simulation in high altitude scarcely-gauged
catchment under changing climate. European water, Vol 57, pp 77-84.

Baiti, H. B. Bouziane, A. Ouazar, D. and Hasnaoui, M. D. 2017. Storm water management

model sensitivity to different design storm types and parameters: the case of tangier
experimental Basin, Morocco. Journal of Sustainable Watershed Science and
Management, pp 130-142.

Balbastre-Soldevila, R. García-Bartual, R. and Andrés-Doménech, I. 2019. A comparison

of design storms for urban drainage system applications. Water, Vol 11, No 4, pp
757.

Ball, J. Babister, M. Nathan, R. Weinmann, P. Weeks, W. Retallick, M. and Testoni, I.

2016. Australian Rainfall and Runoff-A guide to flood estimation.) Commonwealth of
Australia.

________________________________________________________________________
64
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Batista, G.E. and Monard, M.C., 2002. A study of K-nearest neighbour as an imputation
method. His, 87(251-260), p.48.

Blenkinsop, S., Lewis, E., Chan, S.C. and Fowler, H.J., 2017. Quality‐control of an hourly
rainfall dataset and climatology of extremes for the UK. International Journal of
Climatology, 37(2), pp.722-740.

Bonta, J. 1997. Proposed use of Huff curves for hyetograph characterization. In Climate
and Weather Research Workshop Proceedings.).

Borovcnik, M. and Kapadia, R., 2014. A historical and philosophical perspective on

probability. Probabilistic thinking: presenting plural perspectives, pp.7-34.

Boughton, W. and Droop, O. 2003. Continuous simulation for design flood estimation—a
review. Environmental Modelling & Software, Vol 18, No 4, pp 309-318.

Brocca, L. Melone, F. and Moramarco, T. 2011. Distributed rainfall‐runoff modelling for

flood frequency estimation and flood forecasting. Hydrological processes, Vol 25, No
18, pp 2801-2813.

Brunsdon, C., McClatchey, J. and Unwin, D.J., 2001. Spatial variations in the average
rainfall–altitude relationship in Great Britain: an approach using geographically
weighted regression. International Journal of Climatology: A Journal of the Royal
Meteorological Society, 21(4), pp.455-466.

Cai, Y. and Hames, D., 2010. Minimum sample size determination for generalized extreme
value distribution. Communications in Statistics—Simulation and
Computation®, 40(1), pp.87-98.

Calitz, J.P., 2020. Development and assessment of regionalised approaches to design

flood estimation in South Africa.

Chilès, J.P. and Desassis, N., 2018. Fifty years of kriging. Handbook of mathematical
geosciences: Fifty years of IAMG, pp.589-612.

Chu, X. and Steinman, A. 2009. Event and continuous hydrologic modeling with HEC-
HMS. Journal of Irrigation and Drainage Engineering, Vol 135, No 1, pp 119-124.

Cunderlik, J. M. and Simonovic, S. P. 2005. Hydrological extremes in a southwestern

Ontario river basin under future climate conditions/Extrêmes hydrologiques dans un
basin versant du sud-ouest de l'Ontario sous conditions climatiques futures.
Hydrological Sciences Journal, Vol 50, No 4.

David, F.N., 1962. Games, gods and gambling: The origins and history of probability and
statistical ideas from the earliest times to the Newtonian era.

________________________________________________________________________
65
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
De Lima, J. and Singh, V. 2002. The influence of the pattern of moving rainstorms on
overland flow. Advances in Water Resources, Vol 25, No 7, pp 817-828.

De Silva, M. Weerakoon, S. and Herath, S. 2014. Modeling of event and continuous flow
hydrographs with HEC–HMS: case study in the Kelani River Basin, Sri Lanka.
Journal of Hydrologic Engineering, Vol 19, No 4, pp 800-806.

Debnath, L. and Basu, K., 2015. A short history of probability theory and its
applications. International Journal of Mathematical Education in Science and
Technology, 46(1), pp.13-39.

Eagleson, P. S. 1972. Dynamics of flood frequency. Water Resources Research, Vol 8, No

4, pp 878-898.

Fung, K.F., Chew, K.S., Huang, Y.F., Ahmed, A.N., Teo, F.Y., Ng, J.L. and Elshafie, A.,
2022. Evaluation of spatial interpolation methods and spatiotemporal modeling of
rainfall distribution in Peninsular Malaysia. Ain Shams Engineering Journal, 13(2),
p.101571.

Greenwood, J.A., Landwehr, J.M., Matalas, N.C. and Wallis, J.R., 1979. Probability
weighted moments: definition and relation to parameters of several distributions
expressable in inverse form. Water resources research, 15(5), pp.1049-1054.

Guo, Y. and Zhuge, Z. 2008. Analytical probabilistic flood routing for urban stormwater
management purposes. Canadian Journal of Civil Engineering, Vol 35, No 5, pp 487-
499.

Hosking, J.R., 1990. L‐moments: Analysis and estimation of distributions using linear
combinations of order statistics. Journal of the royal statistical society: series B
(methodological), 52(1), pp.105-124.

Hosking, J.R.M. and Wallis, J.R., 1997. Regional frequency analysis (p. 240).

Hossain, S. Hewa, G. A. and Wella-Hewage, S. 2019. A comparison of continuous and

event-based rainfall–runoff (RR) modelling using EPA-SWMM. Water, Vol 11, No 3,
pp 611.

Huff, F. A. 1967. Time distribution of rainfall in heavy storms. Water Resources Research,
Vol 3, No 4, pp 1007-1019.

Jadhav, A., Pramod, D. and Ramanathan, K., 2019. Comparison of performance of data
imputation methods for numeric dataset. Applied Artificial Intelligence, 33(10),
pp.913-933.

________________________________________________________________________
66
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Jiang, L. Chen, Y. and Wang, H. 2015. Urban flood simulation based on the SWMM
model. Proceedings of the International Association of Hydrological Sciences, Vol
368, pp 186-191.

Johnson, K.A., Smithers, J.C. and Schulze, R.E., 2021. A review of methods to account for
impacts of non-stationary climate data on extreme rainfalls for design rainfall
estimation in South Africa. Journal of the South African Institution of Civil
Engineering, 63(3), pp.55-61.

Kang, M. Goo, J. Song, I. Chun, J. Her, Y. Hwang, S. and Park, S. 2013. Estimating
design floods based on the critical storm duration for small watersheds. Journal of
Hydro-environment Research, Vol 7, No 3, pp 209-218.

Keifer, C. J. and Chu, H. H. 1957. Synthetic storm pattern for drainage design. Journal of
the hydraulics division, Vol 83, No 4, pp 1332-1-1332-25.

Kinnison, R.R., 1983. Applied extreme-value statistics.

Knighton, J. O. and Walter, M. T. 2016. Critical rainfall statistics for predicting watershed
flood responses: rethinking the design storm concept. Hydrological processes, Vol
30, No 21, pp 3788-3803.

Kottegoda, N.T., Natale, L. and Raiteri, E., 2014. Monte Carlo Simulation of rainfall
hyetographs for analysis and design. Journal of hydrology, 519, pp.1-11.

Larsen, T. 1981. Real Rainfall Time Series for Storm Sewer Design.

Lewis, E., Pritchard, D., Villalobos-Herrera, R., Blenkinsop, S., McClean, F., Guerreiro, S.,
Schneider, U., Becker, A., Finger, P., Meyer-Christoffer, A. and Rustemeier, E., 2021.
Quality control of a global hourly rainfall dataset. Environmental Modelling &
Software, 144, p.105169.

Lewis, E., Quinn, N., Blenkinsop, S., Fowler, H.J., Freer, J., Tanguy, M., Hitt, O., Coxon,
G., Bates, P. and Woods, R., 2018. A rule based quality control method for hourly
rainfall data and a 1 km resolution gridded hourly rainfall dataset for Great Britain:
CEH-GEAR1hr. Journal of Hydrology, 564, pp.930-943.

Loots, I. and Smithers, J. 2020. Design Flood Estimation in Urban Areas in South Africa:
Preliminary Results from Tshwane Case Studies.

Ly, S., Charles, C. and Degré, A., 2013. Different methods for spatial interpolation of
rainfall data for operational hydrology and hydrological modeling at watershed scale:
a review. Biotechnologie, agronomie, société et environnement, 17(2).

________________________________________________________________________
67
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Males, R. Braune, M. Van Bladeren, D. and Mahlangu, D. 2004. Regional Hydrological
Modeling of City of Tshwane Municipality using VisuaiSWMM. Journal of Water
Management Modeling.

Malik, U. and James, W. 2007. Reliability of design storms used to size urban stormwater
system elements. Journal of Water Management Modeling.

Mamoon, A.A. and Rahman, A., 2017. Selection of the best fit probability distribution in
rainfall frequency analysis for Qatar. Natural hazards, 86, pp.281-296.

Marani, M. Grossi, G. Napolitano, F. Wallace, M. and Entekhabi, D. 1997. Forcing,

intermittency, and land surface hydrologic partitioning. Water Resources Research,
Vol 33, No 1, pp 167-175.

Markolf, S. A. Chester, M. V. Helmrich, A. M. and Shannon, K. 2021. Re-imagining design

storm criteria for the challenges of the 21st century. Cities, Vol 109, pp 102981.

Marsalek, J. and Watt, W. 1984. Design storms for urban drainage design. Canadian
Journal of Civil Engineering, Vol 11, No 3, pp 574-584.

Maßmann, S. Krämer, S. Fuchs, L. Herrmann, O.-C. Kuchenbecker, A. Birkholz, P.

Sympher, K.-J. Haberlandt, U. Morales, B. and Eisele, M. 2019. Applicability of
stochastically generated synthetic rainfall time series for urban drainage modelling. In
Geophysical Research Abstracts.), Vol 21.

Mert, B.A. and Dag, A., 2017. A computer program for practical semivariogram modeling
and ordinary kriging: a case study of porosity distribution in an oil field. Open
Geosciences, 9(1), pp.663-674.

Moeletsi, M.E.A.R.C., Shabalala, Z.P.A.R.C., De Nysschen, G.A.R.C., Moeletsi, M.E. and

Walker, S. 2016. Evaluation of an inverse distance weighting method for patching
daily and dekadal rainfall over the Free State Province, South Africa. Water SA, Vol
42, No 3, pp 466-474.

Mouton, J. Loots, I. and Smithers, J. 2022. ASSESSMENT AND DEVELOPMENT OF

SYNTHETIC DESIGN STORMS FOR USE IN URBAN ENVIRONMENTS:
GAUTENG PILOT STUDY.

Mouton, J. v. S. 2023. Development of appropriate synthetic design storms for Gauteng,

South Africa. University of Pretoria, Department of Civil Engineering, Pretoria, South
Africa.

Mouton, J.v.S, Loots, I. and Smithers, J.C., 2022. Assessment and development of
synthetic design storms for use in urban environments: Gauteng pilot study.

________________________________________________________________________
68
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Mulvaney, T. J. 1851. On the use of self-registering rain and flood gauges in making
observations of the relations of rainfall and flood discharges in a given catchment.
Proceedings of the institution of Civil Engineers of Ireland, Vol 4, pp 19-31.

Nathan, R. and Weinmann, E., 2013. Australian rainfall and runoff discussion paper:
monte carlo simulation techniques. Australia: Engineering House.

Niemczynowicz, J. 1982. Areal intensity-duration-frequency curves for short term rainfall

events in Lund. Hydrology Research, Vol 13, No 4, pp 193-204.

Nnadi, F. Kline, F. Wray Jr, H. and Wanielista, M. 1999. COMPARISON OF DESIGN

STORM CONCEPTS USING CONTINUOUS SIMULATION WITH SHORT
DURATION STORMS 1. JAWRA Journal of the American Water Resources
Association, Vol 35, No 1, pp 61-72.

Noor, M.N., Yahaya, A.S., Ramli, N.A. and Al Bakri, A.M., 2014. Mean imputation
techniques for filling the missing observations in air pollution dataset. Key
Engineering Materials, 594, pp.902-908.

Oliver, M.A. and Webster, R., 2014. A tutorial guide to geostatistics: Computing and
modelling variograms and kriging. Catena, 113, pp.56-69.

Oliver, M.A. and Webster, R., 2015. Basic steps in geostatistics: the variogram and kriging.

Paquet, E. Garavaglia, F. Garçon, R. and Gailhard, J. 2013. The SCHADEX method: A

semi-continuous rainfall–runoff simulation for extreme flood estimation. Journal of
Hydrology, Vol 495, pp 23-37.

Paulhus, J.L. and Kohler, M.A., 1952. Interpolation of missing precipitation

records. Monthly Weather Review, 80(8), pp.129-133.

Pilgrim, D. and Cordery, I. 1993. Flood Runoff. Chapter 9. Handbook of Hydrology.

McGraw-Hill, New York, NY.

Pinthong, S., Ditthakit, P., Salaeh, N., Hasan, M.A., Son, C.T., Linh, N.T.T., Islam, S. and
Yadav, K.K., 2022. Imputation of missing monthly rainfall data using machine
learning and spatial interpolation approaches in Thale Sap Songkhla River Basin,
Thailand. Environmental Science and Pollution Research, pp.1-17.

Prodanovic, P. and Simonovic, S. P. 2004. Generation of synthetic design storms for the
Upper Thames River basin. Department of Civil and Environmental Engineering, The
University of Western ….

Rabori, A. M. and Ghazavi, R. 2018. Urban flood estimation and evaluation of the
performance of an urban drainage system in a semi‐arid urban area using SWMM.
Water Environment Research, Vol 90, No 12, pp 2075-2082.
________________________________________________________________________
69
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Radi, N.F.A., Zakaria, R. and Azman, M.A.Z., 2015, February. Estimation of missing
rainfall data using spatial interpolation and imputation methods. In AIP conference
proceedings (Vol. 1643, No. 1, pp. 42-48). American Institute of Physics.

Rossman, L. A. 2010. Storm water management model user's manual, version 5.0.
National Risk Management Research Laboratory, Office of Research and ….

Rubin, D.B., 1976. Inference and missing data. Biometrika, 63(3), pp.581-592.

Sampath, D. Weerakoon, S. Mishra, B. and Herath, S. 2014. HEC-HMS model for runoff
simulation from Tittawella tank catchment.

Schulze, R. and Arnold, H. 1979. Estimation of volume and rate of runoff in small
catchments in South Africa, based on the SCS technique. . University of Natal,
Pietermaritzburg, Dept Agric. Eng., ACRU Report, No 8.

Schulze, R. and Arnold, H. 1984. Hydrological models for application to small rural
catchments in Southern Africa. Dept. of Agricultural Engineering, University of Natal.

SCS. 1973. A method for estimating volume and rate of runoff in small watersheds.) SCS
Washington, DC.

Shepard, D. 1968. A two-dimensional interpolation function for irregularly-spaced data. In

Proceedings of the 1968 23rd ACM national conference, pp. 517-524.
https://doi.org/10.1145/800186.810616

Singh, K.R., 2021. Detecting and assessing the impacts of outlier events and data
availability on design rainfall and flood estimation in South Africa.

Smithers, J. and Schulze, R. 2000. Long duration design rainfall estimates for South
Africa. Water Research Commission Pretoria.

Smithers, J. C. and Schulze, R. 2003. Design rainfall and flood estimation in South Africa.
Water Research Commission Pretoria.

Smithers, J. C. and Schulze, R. E. 2000. Development and evaluation of techniques for

estimating short duration design rainfall in South Africa. Water Research
Commission. WRC Report No. 681/1/00. Pretoria.

Soley-Bori, M., 2013. Dealing with missing data: Key assumptions and methods for applied
analysis. Boston University, 4(1), pp.1-19.

Stahl, S., 2006. The evolution of the normal distribution. Mathematics magazine, 79(2),
pp.96-113.

________________________________________________________________________
70
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Strecker, E. and Huber, W. 2008. Comparison of Design Event Selection Options for
Urban Watersheds.

Suthar, B., Patel, H. and Goswami, A., 2012. A survey: classification of imputation
methods in data mining. International Journal of Emerging Technology and
Advanced Engineering, 2(1), pp.309-12.

Te Chow, V., Maidment, D.R. and Mays, L.W., 1988. Applied hydrology.

Upton, G.J.G. and Rahimi, A.R., 2003. On-line detection of errors in tipping-bucket
raingauges. Journal of Hydrology, 278(1-4), pp.197-212.

Veneziano, D. and Villani, P. 1999. Best linear unbiased design hyetograph. Water
Resources Research, Vol 35, No 9, pp 2725-2738.

Watson, M. D. 1981. Application of ILLUDAS to stormwater drainage design in South

Africa. University of the Witwatersrand, Johannesburg.

Weddepohl, J. P. 1988. Design rainfall distributions for Southern Africa.) University of

Natal (Pietermaritzburg).

Weesakul, U. Chaowiwat, W. Rehan, M. M. and Weesakul, S. 2017. Modification of a

design storm pattern for urban drainage systems considering the impact of climate
change. Engineering and Applied Science Research, Vol 44, No 3, pp 161-169.

Wellington, J. H. 1929. Some physical influences in the human geography of South Africa.
South African Journal of Science, Vol 26, No 12, pp 81-94.

Wenzel Jr, H. and Voorhees, M. 1984. An evaluation of urban design storm sensitivity.
Water Science and Technology, Vol 16, No 8-9, pp 219-236.

Westra, S. and Sisson, S.A., 2011. Detection of non-stationarity in precipitation extremes

using a max-stable process model. Journal of Hydrology, 406(1-2), pp.119-128.

Willems, P. 1999. Stochastic generation of spatial rainfall for urban drainage areas. Water
Science and Technology, Vol 39, No 9, pp 23-30.

Yen, B. C. and Chow, V. T. 1980. Design hyetographs for small drainage structures.
Journal of the hydraulics division, Vol 106, No 6, pp 1055-1076.

Young, K.C., 1992. A three-way model for interpolating for monthly precipitation
values. Monthly Weather Review, 120(11), pp.2561-2569.

Zhang, X. Q. 2016. The trends, promises and challenges of urbanisation in the world.
Habitat international, Vol 54, pp 241-252.

________________________________________________________________________
71
Development of appropriate synthetic design storms of small catchments in South Africa
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

________________________________________________________________________
72

Research Proposal For Mba Thesis
84% (25)
Research Proposal For Mba Thesis
41 pages
ED2 Sample
No ratings yet
ED2 Sample
53 pages
Inventory Management Proposal
67% (12)
Inventory Management Proposal
9 pages
Scoping Report-Kilombero Bridge
100% (3)
Scoping Report-Kilombero Bridge
35 pages
Research Proposal PHD 32
56% (9)
Research Proposal PHD 32
21 pages
An Evaluation of Fishing Activities at Kasenyi Fish Landing Site Along The Northern Shores of L.victoria (2000-2004)
100% (4)
An Evaluation of Fishing Activities at Kasenyi Fish Landing Site Along The Northern Shores of L.victoria (2000-2004)
46 pages
Rationalization of Hydrometric Station Network in Ambon (English) R4-Check Frank-1Dec2020
100% (1)
Rationalization of Hydrometric Station Network in Ambon (English) R4-Check Frank-1Dec2020
137 pages
Inception Report-Teru F
100% (1)
Inception Report-Teru F
211 pages
69EG4218 Dissertation Template
No ratings yet
69EG4218 Dissertation Template
15 pages
GIS and Remote Sensing Techniques (Preview)
75% (4)
GIS and Remote Sensing Techniques (Preview)
29 pages
OLD DX
No ratings yet
OLD DX
40 pages
DIPLOMA Project
No ratings yet
DIPLOMA Project
25 pages
Soy Milk An Attributional Life Cycle Assessment Examining The Potential Environmental Impact of Soy Milk PDF
No ratings yet
Soy Milk An Attributional Life Cycle Assessment Examining The Potential Environmental Impact of Soy Milk PDF
23 pages
24C1ECO50101201 Nhóm 14
No ratings yet
24C1ECO50101201 Nhóm 14
77 pages
final project proposal-kelvin mbuchi
No ratings yet
final project proposal-kelvin mbuchi
20 pages
final project proposal-kelvin mbuchi (1)
No ratings yet
final project proposal-kelvin mbuchi (1)
22 pages
LUCIA - IRDP MWANZA
No ratings yet
LUCIA - IRDP MWANZA
21 pages
Nkumba Last 2
No ratings yet
Nkumba Last 2
25 pages
Shaft Works
No ratings yet
Shaft Works
22 pages
Asumina R. Masue-1
No ratings yet
Asumina R. Masue-1
36 pages
Serka Prop
No ratings yet
Serka Prop
51 pages
Distribution Network Design Under Uncertain Conditions: A Research Proposal On
No ratings yet
Distribution Network Design Under Uncertain Conditions: A Research Proposal On
15 pages
Yasmin Research
No ratings yet
Yasmin Research
36 pages
Application of 6-Sigma For Service Improvement-AT International Management Institute Canteen
100% (2)
Application of 6-Sigma For Service Improvement-AT International Management Institute Canteen
31 pages
An Individual's "Know How"
No ratings yet
An Individual's "Know How"
86 pages
G2 ResearchReport
No ratings yet
G2 ResearchReport
39 pages
Jazz Dias IT SBA Full Final
No ratings yet
Jazz Dias IT SBA Full Final
28 pages
Addis Sinior Essay A
No ratings yet
Addis Sinior Essay A
83 pages
Dropout Report in USE Schools 2012
No ratings yet
Dropout Report in USE Schools 2012
18 pages
Fayol Mislang Bernadeth Chapter 1 5
No ratings yet
Fayol Mislang Bernadeth Chapter 1 5
26 pages
UKZN 2005 - Environmental History - Zero Waste Project
No ratings yet
UKZN 2005 - Environmental History - Zero Waste Project
150 pages
Key Issues Influencing The Level of Community Participation in Projects
100% (6)
Key Issues Influencing The Level of Community Participation in Projects
35 pages
Application of 6-Sigma For Error Reduction-AT Advance Informatics PVT LTD
No ratings yet
Application of 6-Sigma For Error Reduction-AT Advance Informatics PVT LTD
15 pages
Innocentus Wambua Project (1)
No ratings yet
Innocentus Wambua Project (1)
62 pages
Thesis Template
No ratings yet
Thesis Template
24 pages
AN ANALYSIS ON-WPS Office
No ratings yet
AN ANALYSIS ON-WPS Office
7 pages
Shear Wave Velocity As Function of SPT Penetration Resistance and Vertical Effective Stress at California Bridge Sites
No ratings yet
Shear Wave Velocity As Function of SPT Penetration Resistance and Vertical Effective Stress at California Bridge Sites
80 pages
Root Canal Treatment
No ratings yet
Root Canal Treatment
24 pages
REPORT Ground Water Quality Assesment in Sulur Region Coimbatore 2023
No ratings yet
REPORT Ground Water Quality Assesment in Sulur Region Coimbatore 2023
55 pages
Proposal
50% (2)
Proposal
30 pages
GVI Pez Maya Quarterly Report Oct-Dec 2010 Final
No ratings yet
GVI Pez Maya Quarterly Report Oct-Dec 2010 Final
42 pages
It Sba
No ratings yet
It Sba
18 pages
Fuzzy Logic Controller For Automated Guided Vehicle
67% (3)
Fuzzy Logic Controller For Automated Guided Vehicle
69 pages
Taxation System
No ratings yet
Taxation System
55 pages
s225474115 - Denga A.K - LELV111 Report
No ratings yet
s225474115 - Denga A.K - LELV111 Report
19 pages
hayat_final_proposal_1[1]
No ratings yet
hayat_final_proposal_1[1]
27 pages
Dimensional Modeling
No ratings yet
Dimensional Modeling
22 pages
Mwswat Setup
100% (1)
Mwswat Setup
77 pages
43706109073, Mose Joseph Moranga
No ratings yet
43706109073, Mose Joseph Moranga
53 pages
Proposal Work Final
71% (7)
Proposal Work Final
23 pages
TB HIV Research Proposal Harish
100% (4)
TB HIV Research Proposal Harish
23 pages
DLV WP2 - 12 - Feasibility Study - Published
No ratings yet
DLV WP2 - 12 - Feasibility Study - Published
39 pages
Estimation of Rainfall-Runoff For Umiam Watershed Using SCS-CN Method and Gis Technique
No ratings yet
Estimation of Rainfall-Runoff For Umiam Watershed Using SCS-CN Method and Gis Technique
18 pages
Project Report Water Logging Thane
78% (9)
Project Report Water Logging Thane
79 pages
Ishaya Nathaniel EKutso Research Project 2021
No ratings yet
Ishaya Nathaniel EKutso Research Project 2021
70 pages
Managerial Economics...
No ratings yet
Managerial Economics...
26 pages
Thesis Journal Format As of 01-11-2019
No ratings yet
Thesis Journal Format As of 01-11-2019
28 pages
02-Instrument Types and Performance Characteristics PDF
100% (2)
02-Instrument Types and Performance Characteristics PDF
28 pages
MAF551 Budget&BudgetaryControl TimeSeries&SeasonalVariationsAnalysis
No ratings yet
MAF551 Budget&BudgetaryControl TimeSeries&SeasonalVariationsAnalysis
10 pages
Is 436 2
100% (1)
Is 436 2
17 pages
Interpretation of DSC Curves in Polymer Analysis 2000 - Toledo
No ratings yet
Interpretation of DSC Curves in Polymer Analysis 2000 - Toledo
0 pages
Understanding Process Vacuum Condensers: Issue 3
No ratings yet
Understanding Process Vacuum Condensers: Issue 3
4 pages
Statistics Study Guide: Measures of Central Tendancy
No ratings yet
Statistics Study Guide: Measures of Central Tendancy
2 pages
Predicting Crash Likelihood and Severity
No ratings yet
Predicting Crash Likelihood and Severity
10 pages
Lecture 19: Applications of Feature Measurements: Measurement
No ratings yet
Lecture 19: Applications of Feature Measurements: Measurement
8 pages
Anticipated Sea-Level Rise Coastal Louisiana Technical Report Version 1.4
No ratings yet
Anticipated Sea-Level Rise Coastal Louisiana Technical Report Version 1.4
121 pages
ASTM D792 Densidad y Peso Específico
100% (1)
ASTM D792 Densidad y Peso Específico
6 pages
LBOLYTC Quiz No. 1 - T3 2023
No ratings yet
LBOLYTC Quiz No. 1 - T3 2023
13 pages
Manual de Diseño de Cimentaciones de Puentes PDF
100% (1)
Manual de Diseño de Cimentaciones de Puentes PDF
592 pages
Weakening of Indian Summer Monsoon Rainfall Due To Changes in Land Use Land Cover
No ratings yet
Weakening of Indian Summer Monsoon Rainfall Due To Changes in Land Use Land Cover
10 pages
Cheeto
No ratings yet
Cheeto
7 pages
Chapter 6 DPD
No ratings yet
Chapter 6 DPD
47 pages
tmp8086 TMP
No ratings yet
tmp8086 TMP
24 pages
Physics 73 Review
No ratings yet
Physics 73 Review
17 pages
Studies On Moisture Content in Cotton Fibre Assembly (Lint) in Relation To Variations in Temperature and Relative Humidity
No ratings yet
Studies On Moisture Content in Cotton Fibre Assembly (Lint) in Relation To Variations in Temperature and Relative Humidity
1 page
Pearson Product-Moment Correlation: Correlation - Html#Ixzz293U1Fdye
No ratings yet
Pearson Product-Moment Correlation: Correlation - Html#Ixzz293U1Fdye
3 pages
A Tutorial On Hidden Markov Models and Selected Applications in Speech Recognition
No ratings yet
A Tutorial On Hidden Markov Models and Selected Applications in Speech Recognition
30 pages
Probablity Questions
100% (1)
Probablity Questions
14 pages
Business Analytics Data Analysis and Decision Making Albright 5th Edition Solutions Manual
100% (46)
Business Analytics Data Analysis and Decision Making Albright 5th Edition Solutions Manual
2 pages
Part 2 Project-Basic Inferential Stat
No ratings yet
Part 2 Project-Basic Inferential Stat
6 pages
Calculating Calibration Uncertainties in An Automated Temperature Calibration System
No ratings yet
Calculating Calibration Uncertainties in An Automated Temperature Calibration System
3 pages
Topic 6. Two-Way Designs: Randomized Complete Block Design
No ratings yet
Topic 6. Two-Way Designs: Randomized Complete Block Design
18 pages
Learning Tableau - Sample Chapter
100% (1)
Learning Tableau - Sample Chapter
28 pages
Tigist Mideksa
No ratings yet
Tigist Mideksa
70 pages
AH Lec - 6
No ratings yet
AH Lec - 6
16 pages