ASSESSMENT AND DEVELOPMENT OF

SYNTHETIC DESIGN STORMS FOR USE

IN URBAN ENVIRONMENTS:
GAUTENG PILOT STUDY
Report
to the Water Research Commission

by

Mouton J.v.S. 1, Loots I.1, Smithers J.C.2

1
University of Pretoria,
2
University of KwaZulu-Natal
Contact details of author for correspondence
[email protected]

Project No. C2020-2021-00657

April 2022
 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.
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

EXECUTIVE SUMMARY

This section provides a summary of the project background, aims, methodology and results.
Conclusions drawn from the results, as well as recommendations for the next phase of this
project, are also summarised.

BACKGROUND

Engineers have for centuries concerned themselves with redirecting and managing
stormwater runoff to address the persistent increase in demands associated with civilisation.
As a result, flood estimation methods were developed which have evolved into sophisticated
computer-aided stormwater simulation modelling. The hydrological and hydraulic behaviour
of an urban stormwater drainage network is frequently simulated using software such as the
Stormwater Management Model (SWMM), developed by the United States Environmental
Protection Agency (EPA). Simulations can either run on a continuous basis using observed
rainfall data or based on a single-event model simulation, using a hypothetical rainstorm
event (synthetic design storm) as input. Many urban stormwater and infrastructure designs
are based on single-event based modelling using synthetic design storms. However,
because of the abundance of methods to generate synthetic design storms, engineers
frequently base their method choice on familiarity with a method and preference rather than
sound evidence of the appropriateness of the selected method. The need for typical
synthetic design storms applicable to single-event based modelling of small urban
catchments in South Africa was identified.

AIMS

The aim of this study was to test the performance of the existing synthetic design storm
generation methods, and to identify the method most suited for single-event based modelling
of small urban catchments in Gauteng, using the 5-min interval rainfall records obtained
from the South African Weather Services (SAWS). The specific objectives to meet the aim
were to:
(a) Identify and assess the performance of currently available methods to generate
synthetic design storms used as input for single-event based modelling in the
selected pilot study area.
(b) Propose an improved procedure to generate a synthetic design storm that are
applicable to single-event based modelling of small catchment areas in the study
area.
(c) Disseminate the information to managers, designers and technicians involved in
urban stormwater planning and design.

_______________________________________________________________________
i
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
METHODOLOGY

In this project, existing methods for estimating temporal distributions of design rainfall were
analysed and tested against recently measured sub-hourly rainfall data, using the Gauteng
province in South Africa as a pilot study, to find the synthetic design storm most applicable
to single-event based modelling of small urbanised catchment areas, with the case study
area of Gauteng as the geographical focus.

This study commenced with a literature review and the identification of methods that are
frequently used in Gauteng to generate synthetic design storm events required for single
event-based modelling. The methods are broadly characterised into three distinct
categories, which is an adaptation of the four categories defined by Veneziano and Villani
(1999). The first category is based on Intensity Duration Frequency (IDF) curves, which
includes: (i) methods that use a single point on the IDF curve, and (ii) methods that use the
entire IDF curve. The second category utilises cumulative mass curves of observed rainfall
events. The third category entails the use of stochastic methods which are not covered by
this study.

Observed rainfall data recorded at 35 automatic rainfall stations situated in the Gauteng
province were collated and assessed in terms of completeness. The completeness of the
data sets was determined considering the periods of missing data and the period of available
data. The data sets of nine stations were of poor quality and were subsequently omitted
from the study. Five stations have good data sets with each having a data period of 26 years
and less than 5% of missing data during the rainy months. A further 17 stations were
identified with average quality data sets. Independent rainfall events were identified from the
observed rainfall data. Various Maximum Dry Period (MDP) criteria were used, namely 0,
15, 30, 60 and 120 minutes. After the separation of the data into idealised single events, a
minimum rainfall threshold and an intensity threshold was applied to each event to eliminate
insignificant events. The remaining events were classified as significant events which were
used in the assessments.

The correlation between the total storm duration and total rainfall depth related to different
MDPs was consistent with Ramlall’s (2020) finding, who found an increase in correlation by
decreasing the MDP from 6-hour to 1-hour. However, this study is based on a MDP
associated with the reaction time of a typical small urban catchment. It was argued that any
rainfall after a 15-min MDP would not impact the peak discharge from the previous event,
but it is rather seen as the start of the next event. Appropriate storm parameters were
determined. These storm parameters were subsequently applied to the Chicago Design
Storm (CDS) and Triangular (TRI) methods, respectively, to generate synthetic design
storms for the assessment. The significant events were further used to determine
standardised mass curves (Huff curves) which were also used as synthetic design storms,
but it was shown that the design rainfall of shorter time steps are underestimated. This will

_______________________________________________________________________
ii
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
result in an underestimation of the peak discharge and therefore, this method was deemed
inappropriate for single-event based modelling of small urban catchments with rapid
response times.

The design rainfall ratios of the Soil Conservation Service method adapted for South Africa
(SCS-SA) curves were compared with the ratios determined from the Design Rainfall
Estimation for South Africa (DRESA) software’s design rainfall. The sub-daily rainfall ratios
in relation to the 24-hour rainfall were determined from the design values estimated using
the observed data and compared with the ratios of the SCS-SA curves. The maximum within
this range was determined for each station and assigned to that station. The inverse
distance weighting (IDW) interpolation technique was adopted with a power coefficient of
2.5 for interpolation. The concept of an intermediate curve type was developed which could
be used in combination with the interpolated map of Gauteng. The design rainfall ratios of
the SCS curves were compared with the SCS-SA curves.

Distribution Curves (DC) were developed for the five best stations in Gauteng from the 1:5
year (DC5), 1:10 year (DC10) and 1:20 year (DC20) at-site design rainfall. The adopted
procedure to develop a 24-hour DC was applied, as well as the procedure to extract events
of < 24-hour durations. Both procedures were adopted from the National Engineering
Handbook (NRCS, 2019), and the adaptations were documented.

The synthetic design storm evaluation was conducted by comparing the synthetic design
storms with the observed rainfall events. Two aspects were evaluated, namely the shape of
the mass curves, as well as the average intensities embedded in each synthetic design
storm. The Goodness-Of-Fit (GOF) was determined using the Mean Absolute Relative Error
(MARE) technique. An evaluation of the simulated runoff was undertaken following the
synthetic design storm evaluation. This was achieved by simulating a hypothetical
catchment with an area of 11.6 ha.

RESULTS AND DISCUSSION

The performance of various synthetic design storm methods, including: (a) the CDS method;
(b) the SCS-SA Type 2 and 3 curves; (c) the DC5, DC10 and DC20 curves developed from
the at-site design rainfall; (d) the REC method; and (e) the TRI method was assessed. This
was achieved by comparing the synthetic design storms with the observed rainfall events
which has proved that synthetic design storms do not exist in nature. A comparison of
simulated and observed runoff was undertaken following the synthetic design storm
evaluation. Various observations were made from these results. For example, the suggested
values for the initial deficit associated with the Green-Ampt infiltration parameters did not
result in good simulations. The CDS method was also shown to provide consistent results
compared to continuous simulation. However, a sensitivity analysis must be conducted to

_______________________________________________________________________
iii
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
determine the effect of the advancement coefficient and total storm duration on the peak
discharge.
The design rainfall ratios of the SCS-SA curves were compared with the ratios determined
from the Design Rainfall Estimation for South Africa (DRESA) software’s design rainfall. It
was observed that the Gauteng stations are generally closer to the Type 2 curve rather than
Type 3. Further comparisons were undertaken by estimating design rainfall using a
Probability Distribution (PD) analysis on the observed rainfall data of Gauteng’s stations.
The General Extreme Value distribution was adopted for this analysis. The sub-daily rainfall
ratios in relation to the 24-hour rainfall were determined from the design values estimated
using the observed data and compared with the ratios of the SCS-SA curves. On average
the ratios of Gauteng’s stations compared well with the Type 2 curve, although a wide
variation in ratios was observed. Due to the rapid response time of small urban catchments,
emphasis was placed on the 5 to 30-min ratios. The maximum within this range was
determined for each station and assigned to that station. The inverse distance weighting
(IDW) interpolation technique was adopted with a power coefficient of 2.5 for interpolation.
The concept of an intermediate curve type was developed which could be used in
combination with the interpolated map of Gauteng. Further verifications and sensitivity
analyses are, however, recommended. The design rainfall ratios of the SCS curves were
compared with the SCS-SA curves and concluded that the SCS curves should be applied
with caution in practice. The SCS Type II curve should be applied in areas with a maximum
intermediate curve Type of 2.31 for a 30-min reaction time. Other maximum values apply to
different durations, for example, the intermediate curve type for the 5-min duration is 1.63.
The minimum and maximum for the SCS Type III for the 5-min to 30-min durations was 0.79
and 1.63 respectively. Therefore, the Type III is not recommended for single-event based
modelling of small urban catchments in Gauteng.

In the synthetic design storm evaluation, the Rectangular (REC) method was found to be
the worst representation of observed events. The performance of the methods concerned
with the IDF curve were initially also poor but improved after the location of the peak intensity
was manipulated. The variation of the Recurrent Interval (RI) relative to the average
intensities of the standard time steps, also contributes to the poor performance. This
analysis has therefore provided sound evidence that synthetic design storms do not exist in
nature.

Various observations were made from the SWMM runoff comparison of which the most
important was that the suggested values for the initial deficit associated with the Green-
Ampt infiltration parameters did not result in good simulations, the result of the DC5, DC10
and DC20 curves provided an independent verification of the reaction time of the catchment,
and the CDS method was also shown to provide consistent results compared to continuous
simulation. However, a sensitivity analysis must be conducted to determine the effect of the
advancement coefficient and total storm duration on the peak discharge.

_______________________________________________________________________
iv
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
CONCLUSIONS

It is, in general, concluded that synthetic design storms, applied to a single event-based
model, provides the engineer with the ability to assess the complex hydrological and
hydraulic characteristics of an urban stormwater network. Despite its unrealistic
assumptions, shortcomings, and the criticism the synthetic design storm concept has
received, applying it to a single event-based model has resulted in good peak discharge and
runoff volume estimates. Three methods, to generate synthetic design storms, were
identified that could be applied to a single event-based model. They are the REC, SCS-SA
and CDS methods.

The REC method also provides a means of evaluating the response time of an urban
catchment, but the initial deficit defined as the difference between the porosity and field
capacity, associated with the Green-Ampt infiltration method, did not result in good
simulations. The design rainfall ratio comparisons from both the at-site and DRESA design
rainfall, provided the bases to conclude the inappropriateness of the SCS-SA Type 3 curve
for Gauteng. However, it was also concluded that an interpolation between the standard
type curves is needed for better single event-based simulation results. The methodology
that was used to determine the CDS regression coefficients from the DRESA design rainfall
was sufficient and resulted in good results when applied to a single event-based simulation.

It can be concluded that all three project aims were achieved, namely:

(a) The performance of currently available methods to estimate synthetic design storms
was assessed and used as input for single-event modelling in the selected pilot study
area.
(b) An improved procedure to generate a synthetic design storm for the study area was
proposed. An improved procedure to generate a synthetic design storm applicable to
small catchment areas in the study area is proposed through the development of
intermediate SCS-SA curve types. The intermediate curves could be used in
combination with the interpolated map of Gauteng with lower values to the north of
the province and higher values to the south.
(c) The results of this project were disseminated to managers, designers and technicians
involved in urban stormwater planning and design through a WRC workshop
presented in conjunction with the annual National Flood Studies Programme’s
workshop, as well as at the 2022 UP Flood Hydrology course.

The objective of the study was therefore successfully achieved by the identification of the
three methods that are suitable for single event-based modelling. These methods are the
REC hyetograph, the CDS with location specific regression coefficients and the SCS-SA
curves with intermediate types that was developed for better modelling of different regions
in Gauteng.
_______________________________________________________________________
v
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
RECOMMENDATIONS

Based on the results of this pilot study, various recommendations are made for future
phases of this project. They can be summarised as follows:

(a) The impact the advancement coefficient and total storm duration have on the peak
discharge must be investigated by conducting a sensitivity analysis. The parameters
that could be considered for the analysis includes the size of catchment, slope,
roughness, soil types, and the initial deficit.
(b) The current DRESA software, be further developed to provide the user with an
opportunity to determine a synthetic design storm for a specific location.
(c) The impact that missing data has on the design rainfall estimation and by implication
the design rainfall ratios, in the context of this study, should be investigated.
(d) The discrepancy between the dates of the daily and 5-min data, and the annual
maximum daily rainfalls from the daily and the 5-min data should be investigated.
(e) The inconsistency between the design rainfall estimated using data from the Unisa
and Proefplaas, compared to their neighbouring stations, should be investigated.
(f) The appropriateness of the GEV PD for short duration rainfall should be re-confirmed.
(g) An appropriate power coefficient for the IDW interpolation technique should be
investigated.
(h) The suggested values for the initial deficit associated with the Green-Ampt infiltration
parameters should be investigated.
(i) The relevance of the CDS, SCS-SA and REC methods for generating synthetic
design storms applicable to single-event based modelling of small urban catchments
must be expanded on a national scale, with the possibility of following a regional
approach and ensemble modelling investigated.

_______________________________________________________________________
vi
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
ACKNOWLEDGEMENTS

The project team wishes to thank the following people for their contributions to the project.

Name Role Affiliation

Mr W Nomquphu Reference Group member Water Research Commission
(Chair)
Mr S Dunsmore Reference Group member Fourth Element Consulting
Mr C Brooker Reference Group member CBA Specialist Engineers
Mr R Males Reference Group member DRA
Ms N Smal Reference Group member Ekurhuleni Metropolitan
Municipality
Prof JG Ndiritu Reference Group member University of Witwatersrand
Prof J Gericke Reference Group member Central University of Technology
Prof JA du Plessis Reference Group member Stellenbosch University
Mr P Shepard Reference Group member SRK
Mr Musa Mkhwanazi Provision of rainfall data South African Weather Services
Mr Gerhard Munro Assistance with code Oracle Certified Java SE
development Programmer

_______________________________________________________________________
vii
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
CONTENTS
Page
EXECUTIVE SUMMARY ...................................................................................................... i
ACKNOWLEDGEMENTS ................................................................................................. vii
CONTENTS ..................................................................................................................... viii
LIST OF FIGURES ............................................................................................................. xi
LIST OF TABLES .............................................................................................................. xv
ACRONYMS AND ABBREVIATIONS.............................................................................. xvi
BACKGROUND ......................................................................................... 1
INTRODUCTION........................................................................................................ 1
PROJECT AIMS......................................................................................................... 1
SCOPE AND LIMITATIONS ...................................................................................... 3
LITERATURE REVIEW .............................................................................. 4
BACKGROUND ......................................................................................................... 4
RECTANGULAR HYETOGRAPH .............................................................................. 5
THE CHICAGO DESIGN STORM (CDS) ................................................................... 6
HUFF CURVES........................................................................................................ 11
SCS CURVES .......................................................................................................... 13
2.5.1 Standard SCS temporal distribution curves .................................................. 13
2.5.2 SCS-SA curves ............................................................................................. 15
2.5.3 NOAA Atlas 14 curves .................................................................................. 17
TRIANGULAR HYETOGRAPH ................................................................................ 18
DAILY RAINFALL DISAGGREGATION MODEL FOR SOUTH AFRICA ................. 23
SUMMARY AND CRITICAL EVALUATION OF METHODS .................................... 24
DATA COLLECTION AND STORM EVENT IDENTIFICATION .............. 27
SAWS RAINFALL DATA SOURCE.......................................................................... 27
DATA COLLATION .................................................................................................. 29
DATA PROCESSING ............................................................................................... 30
MISSING DATA ANALYSIS ..................................................................................... 32
STORM EVENT IDENTIFICATION .......................................................................... 36
3.5.1 Maximum Dry Period (MDP) ......................................................................... 36
3.5.2 Minimum Rainfall Depth (MRD) .................................................................... 37
3.5.3 Minimum Rainfall Intensity (MRI) .................................................................. 38

_______________________________________________________________________
viii
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
3.5.4 Identification of single storm events .............................................................. 39
CHAPTER SUMMARY ............................................................................................. 42
METHODOLOGY AND DATA ANALYSIS............................................... 43
STORM PARAMETERS .......................................................................................... 43
4.1.1 Selection of an appropriate MDP .................................................................. 43
4.1.2 Storm advancement coefficient – CDS method ............................................ 45
4.1.3 Dimensionless time to peak - TRI method .................................................... 47
DEPTH-DURATION-FREQUENCY (DDF) CURVES ............................................... 49
INTENSITY-DURATION-FREQUENCY (IDF) REGRESSION COEFFICIENTS ...... 52
RAINFALL DISTRIBUTION CURVE (DC) ................................................................ 55
4.4.1 Development of a 24-hour DC ...................................................................... 55
4.4.2 Development of a DC for durations < 24-hours ............................................ 58
SCS-SA DESIGN RAINFAL RATIO COMPARISONS ............................................. 60
4.5.1 Comparison with DRESA design rainfall ....................................................... 60
4.5.2 Comparison with at-site design rainfall ......................................................... 65
4.5.3 Recommended intermediate curve types for Gauteng.................................. 69
SCS CURVE COMPARISON ................................................................................... 71
STANDARDISED MASS CURVES .......................................................................... 73
CHAPTER SUMMARY ............................................................................................. 75
EVALUATION OF SYNTHETIC DESIGN STORMS ................................ 77
MASS CURVE COMPARISON ................................................................................ 77
ADJUSTMENT OF THE PEAK INTENSITY’S POSITION ....................................... 80
AVERAGE INTENSITY COMPARISON ................................................................... 82
VARIATION IN RI OF AVERAGE INTENSITIES ..................................................... 85
EVENT-BASED AND CONTINUOUS SIMULATION ............................................... 86
5.5.1 Peak discharge comparison ......................................................................... 87
5.5.2 Runoff volume .............................................................................................. 91
5.5.3 Critical and minimum durations..................................................................... 94
5.5.4 SCS-SA ratios for 2-hour events................................................................... 96
5.5.5 Consistency of DC ........................................................................................ 97
CHAPTER SUMMARY ............................................................................................. 97
DISCUSSION, CONCLUSIONS AND RECOMMENDATIONS ................ 99
OBJECTIVE ............................................................................................................. 99
DISCUSSION ........................................................................................................... 99
CONCLUSIONS ..................................................................................................... 102
_______________________________________________________________________
ix
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
RECOMMENDATIONS .......................................................................................... 103
REFERENCES ....................................................................................... 104

_______________________________________________________________________
x
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
LIST OF FIGURES
Page
Figure 2.1: A rainfall hyetograph showing the three most important characteristics
affecting the peak rate of runoff (Keifer and Chu, 1957) ................................ 6
Figure 2.2: Development of a synthetic storm pattern from the Intensity-Duration-
Frequency curve (Keifer and Chu, 1957) ....................................................... 8
Figure 2.3: Chicago Design Storm (Watson, 1981) ............................................................. 9
Figure 2.4: Dimensionless storm mass curve intersections with isopleths connecting
equal probabilities of dimensionless storm depths (i.e., Huff curves) for
a sample size of 322 May and June storms at Invercargill (base/smooth
curves) (Bonta and Shahalam, 2003) .......................................................... 12
Figure 2.5: Huff curves (Huff, 1990) ................................................................................... 12
Figure 2.6: Approximate geographic boundaries for SCS rainfall distributions (NRCS,
1986)............................................................................................................ 13
Figure 2.7: SCS Types I, IA, II and III curves (NRCS, 1986) ............................................. 14
Figure 2.8: Map of States with updated synthetic rainfall distributions as of January
2016 (NRCS, 2019). .................................................................................... 15
Figure 2.9: Regionalisation of synthetic rainfall distributions in southern Africa
(Weddepohl, 1988) ...................................................................................... 16
Figure 2.10: Time distributions of accumulated rainfall depth divided by total rainfall
depths (Schmidt and Schulze, 1987) ........................................................... 17
Figure 2.11: Typical 6-Hour curves for the Interior Highlands region (Perica et al., 2018)
..................................................................................................................... 18
Figure 2.12: Example Hyetograph (Yen and Chow, 1980) ................................................ 19
Figure 2.13: Triangular representation of the hyetograph (Yen and Chow, 1980) ............. 20
Figure 2.14: Mean values of ao for nondimensional triangular hyetographs for Boston,
Massachusetts (Yen and Chow, 1980) ........................................................ 22
Figure 2.15: Dimensionless cumulative rainfall hyetographs for runoff producing storms
having 0 to 24-hour and 24 to 72-hour durations computed by the
triangular hyetograph model for Texas (Asquith et al, 2007) ....................... 22
Figure 2.16: Regionalised map of the mean maximum hourly fraction (Knoesen, 2005)
..................................................................................................................... 23
Figure 2.17: Categorization of synthetic design storm methods covered in the literature
review .......................................................................................................... 25
Figure 3.1: SAWS stations with short duration rainfall data in Gauteng ............................. 27
_______________________________________________________________________
xi
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Figure 3.2: Data processing flow chart .............................................................................. 31
Figure 3.3: Average monthly rainfall for O.R. Tambo, Irene and Jhb Bot Gardens ............ 32
Figure 3.4: Total number of stations according to quality classification criteria.................. 34
Figure 3.5: Rainfall difference between daily and 5-min data on a daily scale at O.R.
Tambo station .............................................................................................. 34
Figure 3.6: Annual maximum daily rainfall from the 5-min and daily rainfall data for the
O.R. Tambo station ...................................................................................... 35
Figure 3.7: Annual missing data for the O.R. Tambo station ............................................. 35
Figure 3.8: Idealised independent events (after Restrepo and Eagleson, 1982) ............... 37
Figure 3.9: Typical design rainfall estimation results for O.R. Tambo International
Airport obtained from the DRESA software.................................................. 38
Figure 3.10: Typical depth-frequency relationship for O.R. Tambo International Airport
..................................................................................................................... 39
Figure 3.11: Typical maximum recurrence intervals per standard time step for the storm
event that occurred on 13 March 2011 at the O.R. Tambo International
Airport .......................................................................................................... 39
Figure 3.12: Total number of storm events and insignificant events identified based on
different MDP’s ............................................................................................ 40
Figure 3.13: Frequency of events based on different MDP’s ............................................. 41
Figure 4.1 Correlation between total rainfall and duration for different MDP criterion ........ 44
Figure 4.2: Typical location of the peak intensity within a 30-min duration ........................ 46
Figure 4.3 Storm advancement coefficients, based on Keifer and Chu’s (1957) second
approach ...................................................................................................... 47
Figure 4.4: Correction of an obtuse triangle such as for the storm event at O.R. Tambo
on 4 January 1997 at 00:50 ......................................................................... 48
Figure 4.5 Time to peak intensity using method proposed by Yen and Chow (1980) ........ 48
Figure 4.6: Rainfall stations in Gauteng that were used to develop IDF curves ................. 49
Figure 4.7: Average RE for each standard time step (ARE_t) ........................................... 51
Figure 4.8: Average RE for each rainfall station (ARE_S) ................................................. 51
Figure 4.9: RE between actual and simulated design rainfall intensities for O.R. Tambo
using the design rainfall from the DRESA software ..................................... 54
Figure 4.10: Average relative error for each standard time step (ARE_t) .......................... 55
Figure 4.11: The general rainfall distribution process ........................................................ 57
Figure 4.12: Typical adjusted six-hour DC extracted from a 24-hour DC ........................... 58
_______________________________________________________________________
xii
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Figure 4.13: D-hour to 24-hour ratio comparison with SCS-SA curves and DRESA
software design rainfall (1 of 3) .................................................................... 62
Figure 4.14: D-hour to 24-hour ratio comparison with SCS-SA curves and DRESA
software design rainfall (2 of 3) .................................................................... 63
Figure 4.15: D-hour to 24-hour ratio comparison with SCS-SA curves and DRESA
software design rainfall (3 of 3) .................................................................... 64
Figure 4.16: Average intermediate curve type for 22 stations computed using DRESA
design rainfall ............................................................................................... 65
Figure 4.17: D-hour to 24-hour ratio comparison for SCS-SA curves and at-site design
rainfall (1 of 3) .............................................................................................. 66
Figure 4.18: D-hour to 24-hour ratio comparison for SCS-SA curves and at-site design
rainfall (2 of 3) .............................................................................................. 67
Figure 4.19: D-hour to 24-hour ratio comparison for SCS-SA curves and at-site design
rainfall (3 of 3) .............................................................................................. 68
Figure 4.20: Average intermediate curve type for 22 stations using at-site design
rainfall .......................................................................................................... 69
Figure 4.21: Example of an intermediate SCS-SA curve ................................................... 70
Figure 4.22: Maximum intermediate curve type for the 5 to 30-min duration range ........... 70
Figure 4.23 Maximum interpolated intermediate curve type for the 5-min to 30-min
duration range using the IDW method ......................................................... 71
Figure 4.24: D-hour to 24-hour ratios of SCS curves compared to SCS-SA curves .......... 73
Figure 4.25: IC types associated with the SCS curves ...................................................... 73
Figure 4.26: Standardised mass curves (Huff curves) for O.R. Tambo ............................. 74
Figure 4.27: Standard time step ratios for the 1st quartile standardised mass curves
(90%) for the O.R Tambo station relative to the standard SCS-SA curves
..................................................................................................................... 75
Figure 5.1: Typical GOF between the shape of an observed individual storm event at
O.R Tambo on 29 October 1994, and a synthetic storm event .................... 79
Figure 5.2: MARE_S between observed storm events and synthetic design storms at
the five best stations in Gauteng .................................................................. 80
Figure 5.3: Synthetic design storms modified with peak earlier during an event................ 81
Figure 5.4: Typical GOF between the average intensities of an observed individual
storm event at O.R Tambo on 29 October 1994, and a synthetic storm
event ............................................................................................................ 83

_______________________________________________________________________
xiii
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Figure 5.5: MARE_I between observed storm events and synthetic design storms at
the five best stations in Gauteng .................................................................. 84
Figure 5.6: MARE_I frequency of occurrence for methods base on entire IDF curve ........ 84
Figure 5.7: Variation in Recurrence Interval (RI) for the five best stations in Gauteng ...... 86
Figure 5.8: The catchment used for the SWMM modelling ................................................ 86
Figure 5.9: Average RE of the peak discharge at node O1 between continuous
simulation and single event-based modelling using the REC and TRI
methods ....................................................................................................... 89
Figure 5.10: Average RE of the peak discharge at node O1 between continuous
simulation and single event-based modelling using a 2-hour storm event
..................................................................................................................... 90
Figure 5.11: Average RE of the peak discharge at node O1 between continuous
simulation and single event-based modelling using a 24-hour storm
event ............................................................................................................ 90
Figure 5.12: Three-day time series containing the event that resulted in the peak
discharge at O.R. Tambo in the year 2020 .................................................. 92
Figure 5.13: Extracted event that resulted in the peak discharge at O.R. Tambo in the
year 2020 ..................................................................................................... 92
Figure 5.14: Annual peak discharge and associated runoff volume obtained from a
continuous simulation for O.R. Tambo ......................................................... 92
Figure 5.15: Annual peak discharge and corresponding runoff volume for O.R Tambo .... 93
Figure 5.16: Interpolation of runoff volume for O.R Tambo ................................................ 93
Figure 5.17: Percentage RE of the runoff volume between continuous simulation and
single event-based modelling ...................................................................... 94
Figure 5.18: Critical storm duration for O.R Tambo according to the REC method ........... 95
Figure 5.19: Example of determining the minimum storm duration for the TRI method ..... 95
Figure 5.20: Result of the minimum storm durations for the five best stations in Gauteng
..................................................................................................................... 96
Figure 5.21: Design rainfall ratios in relation to 2-hour rainfall ........................................... 96
Figure 5.22: Intermediate curve type determined from flow results ................................... 97

_______________________________________________________________________
xiv
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
LIST OF TABLES
Page
Table 1.1. Project objectives ............................................................................................... 2
Table 2.1: Regression constants for the four southern African synthetic rainfall
distributions (Schmidt and Schulze, 1987) ................................................... 16
Table 3.1: SAWS short duration rainfall stations in Gauteng ............................................. 28
Table 3.2: Typical short duration rainfall data set .............................................................. 29
Table 3.3: Data quality classification criteria ...................................................................... 33
Table 3.4: Missing data analysis summary ........................................................................ 33
Table 4.1: Regression coefficients determined from the design rainfall obtained from
the DRESA software .................................................................................... 53
Table 4.2: Example regression coefficients for incremental intensities .............................. 56
Table 4.3: Six-hour rainfall distribution extracted from a 24-hour rainfall distribution ......... 59
Table 4.4: Sub-daily and sub-hourly ratios for the four SCS-SA curves ............................ 60
Table 4.5: D-hour to 24-hour ratios for the four SCS curves .............................................. 72
Table 5.1: CDS Regression coefficients for the SCS-SA curves ....................................... 80
Table 5.2: Sub-catchment SWMM characteristics ............................................................. 87
Table 5.3: Annual peak discharge at node O1 using the observed rainfall data at the
five best stations in Gauteng ........................................................................ 88
Table 5.4: Estimated peak discharge at node O1 using the GEV PD for the five best
stations in Gauteng ...................................................................................... 88

_______________________________________________________________________
xv
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
ACRONYMS AND ABBREVIATIONS

Acronym Description
AMS Annual maximum series
CDS Chicago design storm
DC Distribution Curve
DC5 1:5 year Distribution Curve
DC10 1:10 year Distribution Curve
DC20 1:20 year Distribution Curve

DDF Depth Duration Frequency

DRESA Design Rainfall Estimation in South Africa

EPA Environmental Protection Agency

GEV General Extreme Value
GOF Goodness-of-Fit
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
ILLUDAS Illinois Urban Drainage Area Simulator
MAP Mean annual precipitation (mm)
MARE Mean Absolute Relative Error
MDP Maximum Dry Period (min)
A Windows-based software program that
MIDUSS generates hydrographs for complex drainage
networks from single event storms.
MRD Minimum Rainfall Depth (mm)
MRI Minimum Rainfall Intensity (mm/hour)

_______________________________________________________________________
xvi
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
NOAA National Oceanic and Atmospheric Administration
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
REC Rectangular hyetograph method
RI Recurrence Interval (1:year)
R2 Coefficient of determination
SA(T2) SCS-SA Type 2
SA(T3) SCS-SA Type 3
SAWS South African Weather Services
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)
TRI Triangular hyetograph method

_______________________________________________________________________
xvii
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

BACKGROUND

INTRODUCTION

Engineers have for centuries concerned themselves with redirecting and managing
stormwater runoff to address the persistent increase in demands associated with civilisation.
As a result, flood estimation methods were developed which has evolved into sophisticated
computer-aided stormwater simulation modelling. The hydrological and hydraulic behaviour
of an urban stormwater drainage network is frequently simulated using software such as the
Stormwater Management Model (SWMM), developed by the United States Environmental
Protection Agency (EPA). Simulations can either run on a continuous basis using observed
rainfall data, or based on a single-event model simulation, using a hypothetical rainstorm
event (synthetic design storm) as input. Many urban stormwater and infrastructure designs
are based on single-event based modelling using synthetic design storms. However,
because of the abundance of methods to generate synthetic design storms, engineers
frequently base their method choice on familiarity with a method and preference rather than
sound evidence of the appropriateness of the selected method.

The focus of this pilot study in the Gauteng Province is to assess the performance of
methods used to generate synthetic design storms for single-event based modelling,
applicable to small urban catchments. Each of the considered methods presents some
degree of insufficiency and has received criticism in previous studies from which the need
to derive typical synthetic design storms for small urban catchments was identified. This
document describes the methodology followed and results obtained from considering
existing methods, and (1) comparing them against recently measured sub-hourly rainfall
data and (2) applying them to a typical urban stormwater network.

PROJECT AIMS

The aim of this study was to test the performance of the existing synthetic design storm
generation methods, and to identify the method most suited for conditions in small
catchments in Gauteng, using the 5-min interval rainfall records obtained from the South
African Weather Services (SAWS). The specific objectives to meet the aim are summarised
in Table 1.1.

_______________________________________________________________________
1
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Table 1.1. Project objectives
No. Objectives
1 To identify and assess the performance of currently available methods to generate
synthetic design storms used as input for single-event modelling in the selected
pilot study area.
2 To propose an improved procedure to generate synthetic design storms applicable
to small catchment areas in the study area.
3 To disseminate the information to managers, designers and technicians involved in
urban stormwater planning and design.

The complete list of tasks required to meet the project objectives include:
(a) Literature review – a literature review was conducted on the existing methods in terms
of their development, application, and limitations.
(b) Data collection – the rainfall data for all the stations with sub-daily data in the Gauteng
province were collected from the South African Weather Service (SAWS).
(c) Data verification – the data from each station was evaluated in terms of the length of
the record, and the missing periods were identified.
(d) Storm event identification – the data selected for the evaluation was analysed to
identify individual storm events in accordance with the single event-based.
(e) Storm analysis – the general storm parameters required to accurately apply certain
methods were determined from the individual storm events.
(f) Synthetic design storm evaluation – the Goodness-of-Fit (GOF) was calculated in
terms of the storm shape and peak intensities between the actual and the synthetic
design storms.
(g) SWMM Modelling – a hypothetical stormwater network was created; catchment
characteristics defined, and continuous simulation modelling conducted. A statistical
analysis was conducted on the peak discharge and runoff volume and compared to
results from single event modelling using different synthetic design storms.
(h) Report writing – the report detailing all analysis and results as well as knowledge
dissemination activities was completed.

_______________________________________________________________________
2
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
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 Gauteng. The
existing methods considered in the literature review include the following methods:
(a) Rectangular hyetograph (Mulvaney, 1851);
(b) Chicago design storm (Keifer and Chu, 1957);
(c) The Huff curves (Huff, 1967);
(d) Unites States Department of Agriculture’s Soil Conservation Services (SCS) synthetic
storm distribution curves (SCS, 1973);
(e) The expanded SCS version adapted for Southern African conditions (SCS-SA)
synthetic storm distribution curves (Schulze, 1984);
(f) Triangular hyetograph (Yen and Chow, 1980);
(g) Daily rainfall disaggregation model for South Africa (Knoesen, 2005); and
(h) HRU 1/72 time distribution for intermediate durations.

_______________________________________________________________________
3
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
LITERATURE REVIEW

BACKGROUND

A natural hydrological event, in its simplest form, starts with a hyetograph and ends with a
hydrograph, each having many unique characteristics that are quantifiable, such as total
volume, total duration, intensities and skewness (Adams and Howard, 1986). The link, being
the catchment with its various characteristics and antecedent moisture conditions,
determines the relationship between the hyetograph and hydrograph. For example, two
different hyetographs with two identical catchments and antecedent conditions can result in
differently shaped hydrographs, but with the same peak discharge. Conversely, two identical
hyetographs with two different catchments and antecedent conditions can result in two
different hydrographs with different peak discharges (Adams and Howard, 1986). Despite
the uncertainties about this transitional stage that rainfall needs to pass through before
runoff is generated, it is nevertheless assumed that the frequency of the hyetograph is
identical to the frequency of the hydrograph. This unrealistic assumption is well documented
by Adams and Howard (1986) who, as a result, states it should be used only in the strictest
of circumstances due to the large potential for error associated with the synthetic design
storm concept.

Despite the criticism synthetic design storms have received, these, together with Intensity-
Duration-Frequency (IDF) curves, are used extensively internationally in many urban
stormwater designs and studies (Balbastre et al. 2019). This is because of the complex
hydrological and hydraulic behaviour pertaining to an urban stormwater network which calls
for the use of sophisticated computer-aided rainfall-runoff simulation modelling. The inherent
advantages of synthetic design storms are highlighted by Balbastre et al. (2019) which is
summarised as follows:
(a) It guarantees a uniform level regarding quality and operation standards.
(b) It reduces and simplifies calculations, compared with continuous simulation
modelling.
(c) It provides a way of overcoming the problem of scarcity of short-duration rainfall
data.
(d) It can be regionalised which enhance its practicality.

To contextualise the advantages, the use of synthetic design storms for single event
modelling is done with the intention that it will provide flow results that have the same
frequency of exceedance then the statistical analysis of observed flow data. This is because
observed flow data is hardly ever available and in the best of circumstances, will only be
available at selective locations. In the case of a new stormwater network, observed flow data
does not exist, and observed short duration rainfall data will provide the next best information
_______________________________________________________________________
4
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
in the form of a continuous simulation. With the scarcity of short-duration rainfall data, the
complexity of stochastic generated historic rainfall data, and the excessive time consumption
of continuous simulation modelling, synthetic design storms have an important role to play
in the design and assessment of urban stormwater networks.

It is important to simulate a storm event with not only the critical duration but also the
hyetograph shape that will most likely be representative of the natural rainfall event, since
storm shape may have a significant impact on the peak discharge and total flood volume. It
becomes more noticeable at catchments with a high percentage of areas that saturate
quickly. This is due to the rapid decline of the infiltration capacity during the early part of a
rainstorm, which then tends towards an approximately constant value after a few hours for
the remainder of the event (Horton, 1933). If the peak intensity of a synthetic design storm
is applied earlier during the storm event, the infiltration capacity might be higher, which will
result in a lower peak flow result. Conversely, a storm event peaking later during the event
is likely to result in higher flow results. Although the existing synthetic design storms were
developed using the best data, technology, and engineering judgement available at the time,
they do present some degree of insufficiency. However, as identified previously, synthetic
design storms have an important role to play in the design and assessment of urban
stormwater networks, and, therefore, a review of each of the synthetic design storms is
provided in this chapter. Emphasis is placed on their development and the criteria and
assumptions on which it was based, which will provide the basis for the methodology that
will be followed in this study to assess the applicability of the methods.

RECTANGULAR HYETOGRAPH

The rectangular hyetograph method is commonly associated with the Rational Method,
which is the most employed formula in engineering hydrology (Gomez and Sanchez, 2014),
and it is used worldwide for flood protection design (Cordery and Pilgrim, 1993). This method
assumes that the peak discharge occurs when the duration of the rainfall event is equal to
the time of concentration of the catchment; that the rainfall intensity does not vary; and is
distributed uniformly over the catchment (Smithers, 2012). The first principle of the Rational
Method, the runoff coefficient, was concluded through the gradual development of flood
estimation methods by researchers in the British Isles before the year 1850. Thomas
Mulvaney presented the second principle, the time of concentration, and the method of
estimating the peak discharge in 1851, which became known as the Rational Method
(Gomez and Sanchez, 2014). With the development of the St. Venant equations for
modelling surface flow towards the end of the 19th century (Boussinesq and Flamant, 1871)
and the soil infiltration models, like the Green-Ampt model during the early 20th century
(Green and Ampt, 1911), methods like the Unit Hydrograph were developed. This method
in its simple form also assumes a uniform rainfall intensity distribution with a duration equal
to or greater than the longest time of concentration.

_______________________________________________________________________
5
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
THE CHICAGO DESIGN STORM (CDS)

The CDS method was designed for the City of Chicago by Keifer and Chu (1957) because
of the rapid increase in urbanization that followed the end of World War II. As depicted in
Figure 2.1 the method encompasses three important characteristics, namely:

(a) average intensity within the maximum storm duration,

(b) antecedent rainfall before the maximum duration, and
(c) location of peak intensity (Keifer and Chu, 1957).

Rainfall Hyetograph
Rainfall rate in millimetre per hour

(a) Average intensity during maximum period

(Taken from the selected Rate-Duration-Frequency
(c) Location of peak

(b)
Antecedent
Rainfall
Max period
or duration

Time from beginning of rainfall

Figure 2.1: A rainfall hyetograph showing the three most important characteristics
affecting the peak rate of runoff (Keifer and Chu, 1957)

The average rainfall intensities with associated probability or frequency of exceedance, are
determined for particular storm durations using statistical analysis of historical rainfall data.
These intensities concerning duration, expressed in terms of frequency of exceedance, are
commonly known as Intensity-Duration-Frequency (IDF) curves. The IDF curves have a
sigmoidal shape which can be related to a mathematical function in the form of Equation 2.1
and Equation 2.2 as follows:

a
iav = [ 2.1 ]
tb +c
and:
a
iav = [ 2.2 ]
(b + t)c
where:
iav = average rainfall intensity for a particular storm duration (mm/hour),
a = site and recurrence interval specific constant,
b,c = site specific constants, and
t = storm duration (min).

_______________________________________________________________________
6
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Keifer and Chu (1957) addressed the first characteristic of the average intensity within the
maximum storm duration, with the intensity distribution function which was derived from the
IDF curve, whereas the second and third characteristics of the CDS were represented by
the storm advancement coefficient (r). The IDF curve, presented by Equation 2.1 was used
to develop the intensity distribution (Keifer and Chu, 1957). However, the development was
somewhat transformed using a formula, presented by Equation 2.2 also known as the
Sherman formula (Sherman, 1931). Preceded by various substitutions and differentiation,
the intensity distribution of an advanced storm, which is a storm with the peak intensity
located at the beginning of the event, was expressed in terms of Equation 2.3 as follows:

a[(1 − c)t + b]
i= [ 2.3 ]
(b + t)c+1

The time before the peak intensity (tb) and the time after the peak intensity (ta) were
expressed in terms of storm advancement coefficient in the form of Equations 2.4 and 2.5
as follows:

tb
=r
td
tb
⸫ td = [ 2.4 ]
r
and:
ta
=1−r
td
ta
⸫ td = [ 2.5 ]
1−r

Substituting the storm duration in Equation 2.3 with Equations 2.4 and 2.5, yielded the
intensity distributions before and after the peak intensity in the form of Equation 2.6 and 2.7
as follows:

tb
a �(1 − c) + b�
ib = r
c+1 [ 2.6 ]
t
� b + b�
r
and:
ta
a �(1 − c) + b�
ia = 1−r
c+1 [ 2.7 ]
t
� a + b�
1−r

_______________________________________________________________________
7
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
where:
ta = specific time interval after the peak (min),
tb = specific time interval before the peak (min),
ia = specific intensity after the peak intensity (mm/hour),
ib = specific intensity before the peak intensity (mm/hour),
a,b,c = as defined in Equations 2.1 and 2.2, and
r = storm advancement coefficient (Smith, 2004; Watson, 1981).

The synthetic hyetograph for a completely advanced storm expressed by Equation 2.3 was
illustrated graphically by Keifer and Chu (1957) which is reproduced here as Figure 2.2.
However, Watson (1981) went further and illustrated the intensity distributions before and
after the peak intensity, which were derived from the IDF curve and expressed in terms of
Equations 2.6 and 2.7. The synthetic hyetograph illustrated by Watson (1981) with the peak
intensity positioned closer to the middle of the storm duration, is depicted in Figure 2.3. The
purpose of both these illustrations was to demonstrate that for a storm duration, the rainfall
volume obtained from the IDF curve, is equal to the cumulative rainfall volume obtained from
the hyetograph of the synthetic design storm.

Synthetic hyetograph having the same average rate of rainfall

as given by the Rate-Duration-Frequency curve (Completely
Rainfall intensity (mm/hour)

advanced pattern and with no antecedent rainfall)

Rate-Duration-
Frequency Curve
a
iav =
tb + c
iav

td
Duration of maximum period (minutes)

Figure 2.2: Development of a synthetic storm pattern from the Intensity-Duration-

Frequency curve (Keifer and Chu, 1957)

_______________________________________________________________________
8
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

(a) Intensity-Duration-Frequency curve

Average Rainfall intensity

a
I=
(t + b)c

t Duration

(b) Design storm hyetograph

tb
a [(1-c) +b]
1-r
ia =
Rainfall intensity

t
tb (1-rb +b) 1+c
a [(1-c) +b]
r
ib = t
( rb +b)
1+c

Area A = Area B

t tb ta Time
t

Figure 2.3: Chicago Design Storm (Watson, 1981)

_______________________________________________________________________
9
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
By integrating Equation 2.6 and 2.7 from the beginning of the hyetograph to a time interval
before and after the peak intensity, Silveira (2016) formulated two cumulative rainfall
equations in the form of Equation 2.8 and Equation 2.9 as follows:

a�Tp − T�
Pb = r ∙ P −
Tp − T c [ 2.8 ]
�b + r �
and:
a�T − Tp �
Pa = r ∙ P +
T − Tp c [ 2.9 ]
�b + 1 − r �

where:
P = total rainfall depth (mm),
Pa = cumulative rainfall depth after the peak intensity (mm),
Pb = cumulative rainfall depth before the peak intensity (mm),
r = storm advancement coefficient,
T = time interval from the start of the event (min),
Tp = time when peak intensity occurs (min), and
a,b,c = as defined in Equations 2.1 and 2.2.

To address the second and third characteristic, Keifer and Chu (1957) calculated the storm
advancement coefficient for different durations and weighted in proportion to the antecedent
rainfall volume. The average antecedent rainfall volume was calculated for the 15, 30, 60
and 120-min storm durations. Keifer and Chu (1957) considered the rainfall volume before
the peak intensity, for which the antecedent rainfall volume was expressed in terms of the
storm advancement coefficient (r). The storm advancement coefficient for the specific storm
duration, using the average antecedent rainfall volume, was then calculated. The storm
advancement coefficient was then weighted proportionally to the average antecedent rainfall
volume for specific storm durations and then averaged for all durations. Following this
approach and assuming a total storm duration of 180-min for the City of Chicago, Keifer and
Chu (1957) determined the advancement coefficient for the 15, 30, 60 and 120-min storm
durations for which the weighted average was 0.386.

Watson (1981) determined the storm advancement coefficient for Norwood, Johannesburg
from 28 significant storms also following this approach. For total storm durations of two and
three hours, the storm advancement coefficient was 0.28 and 0.22, respectively. A graphical
comparison was conducted between the CDS and two real storms, and an Illinois Urban
Drainage Area Simulator (ILLUDAS) stormwater model was used to determine the peak
discharges. Based on this limited extent of testing, it was concluded that the CDS is an
adequate technique for predicting peak discharge.

_______________________________________________________________________
10
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The second approach which was adopted for calculating the storm advancement coefficient
was to ignore the second characteristic, which is the antecedent rainfall before the maximum
duration. Since the rainfall data was recorded in 5-min intervals, it was assumed that the
peak intensity is located exactly in the middle of the peak 5-min interval (Keifer and Chu,
1957). Keifer and Chu (1957) then determined the weighted average storm advancement
coefficient for all durations considering only the location of the peak intensity and weighted
proportionally to the time of duration. The weighted average was 0.375. Weesakul et al.
(2017) conducted a similar investigation on rainfall data recorded at the meteorological
station at the Asian Institute of Technology in Bangkok, Thailand, on data covering 21 years.
The storm advancement coefficient was between 0.20 and 0.49.

HUFF CURVES

According to Bonta (1997), cited by Bonta and Shahalam (2003), the curves presented by
Huff (1967) were developed by separating independent storms and then non-
dimensionalising each mass curve in terms of the total rainfall volume and total storm
duration. The dimensionless mass curves were then superimposed graphically showing the
breakpoints at 0.02 intervals along the horizontal axis with the fraction of the total rainfall
depth along the vertical axis, followed by the construction of curves with probability values
from 10% to 90% (Bonta and Shahalam, 2003). This methodology of developing the Huff
curves was, however, according to Bonta (2004), never documented which led to the
formulated methodology presented by Bonta (1997). In terms of identifying independent
rainstorms, Huff (1967) used a criterion of 6-hours as the minimum dry period to separate
consecutive rainstorms (Huff, 1990), whereas Bonta (1997) determined a minimum dry
period following the method of identification of independent rainstorms developed by
Restrepo and Eagleson (1982). The minimum threshold criterion for individual rainstorms
was 25 mm (Huff, 1990).

Typical dimensionless mass curves, also known as isopleths, which are lines connecting
intersecting points with equal probabilities of dimensionless storm depths, developed by
Bonta and Shahalam (2003) are depicted in Figure 2.4. Huff (1967) investigated the time
distributions from 261 storm events recorded in East-Central Illinois from 49 recording rain
gauges over 12 years from 1955 to 1966. The rain gauges were distributed over 1,036 km².
He divided the rainfall distributions between four quarters based on whether the heaviest
rainfall within each storm event occurred in the first, second, third or fourth quarter of the
total storm duration. The curves, with probability values from 10% to 90%, developed by
Huff are depicted in Figure 2.5. Huff (1990) further suggested that the first and second
quartile distributions be used for storm durations of less than 12-hours, the third quartile for
storm durations between 12 and 24-hours, and the fourth quartile for storm durations of
more than 24-hours. It is suggested that the 50% percentile is likely applicable for most
purposes.

_______________________________________________________________________
11
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 2.4: Dimensionless storm mass curve intersections with isopleths

connecting equal probabilities of dimensionless storm depths (i.e., Huff curves) for
a sample size of 322 May and June storms at Invercargill (base/smooth curves)
(Bonta and Shahalam, 2003)

Figure 2.5: Huff curves (Huff, 1990)

_______________________________________________________________________
12
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
SCS CURVES

Since the initial publication of the standard SCS temporal distribution curves, several
derivatives of the curves were developed (e.g. SCS, 1973; Schulze and Arnold, 1979; El-
Sayed, 2017; NRCS, 2019;). The adaptations included in this review are the South African
version, called the SCS-SA, the NOAA Atlas 14 curves that replaces the standard SCS
curves (NRCS, 2019). The development of each of these curves are discussed in detail in
this section.

2.5.1 Standard SCS temporal distribution curves

Standard SCS temporal distribution curves, Types I and II, were first published in 1973 by
the Soil Conservation Service (SCS), which later became the Natural Resources
Conservation Services (NRCS). These two curves were developed from the generalised
rainfall depth-duration relationships obtained from the US Weather Bureau technical papers
(TP-42) published in 1961 (US Weather Bureau, 1961), of which Type II covers most of the
USA. Types IA and III were later developed in the same way and were subsequently
published by the NRCS in 1986 (NRCS, 1986). The approximate geographic boundaries for
the four SCS rainfall distributions are depicted in Figure 2.6. However, according to the
National Engineering Handbook, little documentation is available that describes the
development of Type II and other legacy rainfall distributions (NRCS, 2019). From the
available information, the depth ratios relative to the 24-hour rainfall depth were plotted
against the duration for several locations in each of the four regions and then a curve was
selected with the best fit (SCS, 1973).

Figure 2.6: Approximate geographic boundaries for SCS rainfall distributions

(NRCS, 1986)

_______________________________________________________________________
13
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The curves were then developed for each of the four regions by positioning the greatest 30-
min rainfall depth at the 10-hour and 8-hour points for the Types I and IA respectively, and
the 12-hour point for both the Types II and III curves. It is important to note that the
positioning of the greatest 30-min rainfall depths at the 8-, 10- and 12-hour points were not
based on any meteorological factors but rather on design considerations (SCS, 1973). The
second-largest 30-min depth was positioned 30-min later, and the third-largest 30-min depth
was positioned at the preceding 30-min. The alternation of 30-min depths, which decrease
in magnitude, was repeated until the smallest 30-min depths were located at the beginning
and end of the 24-hour (SCS, 1973). The four SCS curves are depicted in Figure 2.7.

Figure 2.7: SCS Types I, IA, II and III curves (NRCS, 1986)

However, according to NRCS (2019), it was concluded that the use of the SCS curves be
discontinued and replaced by the National Oceanic and Atmospheric Administration (NOAA)
Atlas 14 curves. Locations in the USA that are already covered by the NOAA Atlas 14 data
are depicted in Figure 2.8.

_______________________________________________________________________
14
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 2.8: Map of States with updated synthetic rainfall distributions as of January
2016 (NRCS, 2019).

2.5.2 SCS-SA curves

The standard SCS Types I and II curves were originally adopted for use in South Africa
(Schulze and Arnold, 1979), but the need for revised synthesised storm distributions for
South Africa was identified after an analysis of digitised data for Natal (Schulze, 1984). The
development of the SCS-SA synthetic rainfall distributions was based on the selection of
four D-hour to one-day rainfall ratio range classes for various durations from 5-min to 24-
hours, which became known as the SCS-SA synthetic rainfall distributions Type 1 to 4
(Weddepohl, 1988). The D-hour to one-day rainfall ratios for 40 autographic rainfall stations
in South Africa were determined and the appropriate ratio range class was assigned to each
station (Schmidt and Schulze, 1987). Based on this analysis a map was drawn that
represents the regionalisation of the four ratio range classes, depicted in Figure 2.9. The D-
hour to one-day rainfall ratios for the four distributions were represented by Equation 2.10
as follows:

a∙D
R= [ 2.10 ]
(b + D)c

where:
R = ratio of D-hour to one-day design rainfall depth,
D = storm duration (hour), and
a,b,c = regression constants summarised in Table 2.1 (Schmidt and Schulze, 1987).

_______________________________________________________________________
15
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The rainfall distributions were then derived by positioning the middle of the peak 5-min ratio
at the 12-hour point, and distributing the ratios of increasing durations, equally on either side
of the peak intensity. The cumulative rainfall distributions depicted in Figure 2.10, therefore
represent the increase in intensity between consecutive durations from the start of the 24-
hour duration up to the 12-hour point, followed by the decrease in intensity up to the end of
the 24-hour duration. For example, the difference between the ratios of the 30-min before
and 30-min after the peak intensity is equal to the ratio of the 1-hour ratio determined with
Equation 2.15. More recently Smithers and Schulze (2002) demarcated the City of Tshwane
Metropolitan Municipality (CTMM) into four distinct regions with similar distributions of short
and long duration extreme rainfall. Males et al. (2004) then developed an integrated
catchment management plan for the City of Tshwane Metropolitan Municipality (CTMM) by
compiling a VisualSWMM model. It was found that the four regions conform to the SCS-SA
Type 2 rainfall distributions by considering the ratio range classes defined by Weddepohl
(1988).

Figure 2.9: Regionalisation of synthetic rainfall distributions in southern Africa

(Weddepohl, 1988)

Table 2.1: Regression constants for the four southern African synthetic rainfall
distributions (Schmidt and Schulze, 1987)
Distribution Type a b c
1 0.29935 0.059 0.62
2 0.45321 0.100 0.75
3 0.73402 0.230 0.90
4 1.01330 0.320 1.00
_______________________________________________________________________
16
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 2.10: Time distributions of accumulated rainfall depth divided by total rainfall
depths (Schmidt and Schulze, 1987)

2.5.3 NOAA Atlas 14 curves

The NOAA Atlas 14 consists of a series of volumes that contains the estimates of the design
rainfall for standard duration time steps and associated frequencies, together with 90%
confidence intervals for the USA, similar to the design rainfall estimates developed by
Smithers and Schulze (2000). Curves were also developed for four storm duration classes
(6, 12, 24, and 96-hours), separated depending on which quartile the greatest percentage
of the total rainfall occurred. According to Bonnin et al. (2011) and Perica et al. (2018), the
NOAA Atlas 14 curves were developed in the same way as the ones developed by Huff
(1967), except that a storm event was defined in terms of a fixed duration. In other words,
events always started with rainfall, but the end of the storm event was located after 6, 12,
24, and 96-hours respectively, irrespective of an event ending sooner (Bonnin et al. 2011).
Therefore, many storm events ended sooner than the duration class which lead to events
that were more front-loaded, compared to events selected based on the single event
approach like Huff (1967). Typical 6-hour NOAA Atlas 14 storm distributions for the Interior
Highlands region of the USA are depicted in Figure 2.11. Another approach of utilising the
NOAA Atlas 14 design rainfall, is by developing a 24-hour rainfall distribution from the 5-min
through to 24-hour design rainfall values. The procedure involves the calculation of ratios of
short durations to the 24-hour rainfall, that is distributed equally either side of 12-hours.
Equations were developed to interpolate the design rainfall for durations in between the
standard time steps (NRCS, 2019).

_______________________________________________________________________
17
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 2.11: Typical 6-Hour curves for the Interior Highlands region (Perica et al.,
2018)

TRIANGULAR HYETOGRAPH

The triangular hyetograph method was formulated by Yen and Chow (1980). The first-
moment arm was calculated for the recorded hyetograph concerning the beginning of the
rainstorm. It was then related to a triangular representation of the hyetograph with an equal
total rainfall volume and total storm duration. Each triangular hyetograph was then non-
dimensionalised in terms of the maximum intensity as well as the time to the maximum
intensity, relative to the total storm duration. Yen and Chow (1980) considered the typically
recorded hyetograph depicted in Figure 2.12, and defined the first-moment arm concerning
the beginning of the rainstorm in terms of Equation 2.11 as follows:

∆t�∑nj=1(j − 0.5)dj �
t̅ = [ 2.11 ]
∑nj=1 dj

where:
t = first-moment arm of the hyetograph (min),
dj = depth for the j-th time interval (mm),
∆t = equal time interval (min), and
n = number of time intervals for the rainstorm (Yen and Chow, 1980).
_______________________________________________________________________
18
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Actual
Rainfall

Rainfall Depth per Time Interval

Block
Diagram

Rainfall intensity

Δt

0 td
Time, t
Figure 2.12: Example Hyetograph (Yen and Chow, 1980)

Yen and Chow (1980) then considered the triangular hyetograph depicted in Figure 2.13,
which is defined in terms of the total storm duration (Td) and the maximum rainfall intensity
(h). The first-moment arm of the triangular hyetograph concerning the beginning of the
rainstorm was expressed in terms of Equation 2.12 as follows:

Td + a
t̅ = [ 2.12 ]
3

Solving for the time to the peak intensity (a) of the triangular hyetograph yielded Equation
2.13 as follows:

a = 3t̅ − Td [ 2.13 ]

Equation 2.13 was then solved using the first-moment arm and the total storm duration of
the recorded hyetograph. The maximum intensity of the triangular hyetograph was
expressed in terms of Equation 2.14 as follows:

2D
h= [ 2.14 ]
Td

where:
a = the time to peak intensity (min),
D = total rainfall volume (mm),
h = maximum rainfall intensity (mm/min), and
Td = total storm duration (min) (Yen and Chow, 1980).

_______________________________________________________________________
19
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
a b

Rainfall intensity, I

0 td
Time, t
Figure 2.13: Triangular representation of the hyetograph (Yen and Chow, 1980)

To generalise the geometric parameters of each hyetograph for all storm events, Yen and
Chow (1980) non-dimensionalised the time to the peak intensity (a) and the maximum
rainfall intensity (h) using Equation 2.15 and Equation 2.16 respectively:

a
ao = [ 2.15 ]
Td

h
ho = =2 [ 2.16 ]
D
� �
Td

where:
ao = dimensionless time to the peak intensity, and
ho = dimensionless maximum rainfall intensity.

Asquith et al. (2007) developed a method to estimate the rainstorm parameters using the
dimensionless cumulative hyetograph ordinates and the fraction of storm duration. For
fractions smaller and equal to the time to peak intensity, the quantile function Q(F) was
expressed in terms of Equation 2.17, and for fractions larger than the time to peak intensity
was expressed in terms of Equation 2.18 as follow:

1h 2
Q1 (0 ≤ F ≤ ao ) = F [ 2.17 ]
2a

_______________________________________________________________________
20
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
1 1h
Q 2 (ao < F ≤ 1) = ho ao + ho (F − ao ) − (F − ao )2 [ 2.18 ]
2 2b

By integration of the quantile functions and substitution of the dimensionless peak intensity,
the mean of the dimensionless cumulative hyetograph ordinates was expressed in terms of
Equation 2.19 as follows:

2 − ao
μ= [ 2.19 ]
3

where:
ao = dimensionless time to peak intensity,
F = fraction of storm duration,
ho = dimensionless peak intensity,
Q1 = dimensionless cumulative hyetograph ordinate for fractions of storm durations
smaller and equal than ao,
Q2 = dimensionless cumulative hyetograph ordinate for fractions of storm durations
larger than ao, and
µ = the mean of the dimensionless cumulative hyetograph ordinates (Asquith et
al. 2007).

Yen and Chow (1980) conducted a statistical analysis after the determination of the
dimensionless parameters. The effect of different total storm durations, total rainfall volumes
and different seasons, on the parameters were investigated by applying the triangular
hyetograph method to rainstorms which were defined as periods of nonzero rainfall.
Altogether 7 484 rainstorms were analysed for three weather stations located in the USA:
Boston, Urbana and Elizabeth City. The recorded hyetographs consisted of hourly recorded
data. The mean values of the time to peak intensity considering different total storm
durations, total rainfall volumes and seasons are depicted in Figure 2.14.

Yen and Chow (1980) also determined that the triangular hyetograph method produces
acceptably accurate design hyetographs and applied the method to the experimental data
of Izzard (1946) as well as Yu and McNown (1964), which consists of measured runoff
hydrographs from artificial catchments. Ellouze et al. (2009) conducted a similar
investigation by analysing hourly recorded rainfall data, recorded between 1974 and 1997,
at 10 rainfall stations located within a 7.74 km² catchment in central Tunisia. Altogether,
2,799 rainstorms were analysed. According to Ellouze et al (2009), the observed
hyetographs conformed adequately with the synthetic hyetographs. Asquith et al. (2007)
used a database of 1 659 storms for 91 small catchments with streamflow gauging stations
in Texas. Using the model developed by Asquith et al. (2007), the storm parameters were
estimated for two storm duration ranges, 0 to 24 hours, and 24 to 72 hours. For each duration
range, storms were classified into a sequence of integer depth intervals. The weighted
averages of the mean of the dimensionless hyetographs were 0.59 and 0.55 respectively.
_______________________________________________________________________
21
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The weighted averages were based on the number of storms in each depth interval.
Applying Equation 2.16, the time to peak intensity was 0.23 and 0.35 respectively. The
dimensionless cumulative hyetograph depicted in Figure 2.15, was calculated using the
quantile functions before and after the peak intensity expressed in terms of Equations 2.17
and 2.18.

Figure 2.14: Mean values of ao for nondimensional triangular hyetographs for

Boston, Massachusetts (Yen and Chow, 1980)

Figure 2.15: Dimensionless cumulative rainfall hyetographs for runoff producing

storms having 0 to 24-hour and 24 to 72-hour durations computed by the triangular
hyetograph model for Texas (Asquith et al, 2007)
_______________________________________________________________________
22
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
DAILY RAINFALL DISAGGREGATION MODEL FOR SOUTH AFRICA

Based on the work done by Boughton (2000), Knoesen (2005) developed and assessed a
model to disaggregate daily rainfall into hourly rainfall for South Africa. A total of 157 stations,
all of which had more than 10 years of hourly data were used to develop this model and to
regionalise the distribution of the maximum hourly fraction. According to Boughton (2000),
as cited by Knoesen (2005), the model treats each day as an independent event. Therefore,
only data of independent events of which all 24 hours from 00:00 to 00:00 with at least 1
mm rainfall recorded were used.

The model is primarily based on the hourly fractions of the daily rainfall; the frequency of the
maximum hourly fraction; the clustering of the other 23 hourly fractions; and the arrangement
of the clusters into random daily temporal patterns. The average maximum fraction for each
independent event was collated into 20 ranges. The average highest 2-hour, 3-hour, 6-hour
and 12-hour factions for each range of maximum hourly fraction were calculated. The
second-highest hourly fraction added to the highest fraction was then found that best
approximates the 2-hour fraction, and the third-highest to approximates the 3-hour fraction
and so on, which then formed the clustered sequence (Knoesen, 2005). With the
regionalisation of the model, the maximum hourly fraction of all the stations were collated
into four revised ranges, and the average distribution of each range was calculated. Using
inverse distance weighting, a regionalised map was developed based on the mean value of
the maximum hourly fraction. The map, depicted in Figure 2.16, is then used to find the
appropriate range for the site of interest.

Figure 2.16: Regionalised map of the mean maximum hourly fraction (Knoesen,
2005)
_______________________________________________________________________
23
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The advantage of this model is that it can be applied to a site where only daily recorded
rainfall is available since short duration rainfall data are scarce compared to daily rainfall
data. However, the disadvantage of this model is that the distribution of the highest fraction,
considering the 20 ranges, and the 24 sequencings of the clustered fractions selected by
Boughton (2000), results in 480 different temporal patterns (Knoesen, 2005), whereas the
focus of this study is on single event modelling. Furthermore, this model disaggregates daily
rainfall into hourly rainfall, whereas sub-hourly disaggregation would be more appropriate
for this study. Based on this, this model was not considered.

SUMMARY AND CRITICAL EVALUATION OF METHODS

As identified in the literature review, the existing synthetic design storms have a strong
scientific basis. It guides the classification of existing synthetic design storms into three
categories, which is an adaptation of the four categories defined by Veneziano and Villani
(1999), as follows (Figure 2.17):
(a) Methods that are derived from the Intensity Duration Frequency (IDF) curves:
i. Methods based on a simple geometrical shape using a single point on the IDF
curve. This category includes the rectangular and triangular methods;
ii. Methods that use the entire IDF curve, which includes the CDS method and to
some degree the SCS and SCS-SA curves;
(b) Standardised mass curves generated directly from rainfall records, which includes
the Huff and NOAA Atlas 14 curves; and
(c) The simulation from a stochastic rainfall model, which includes the daily rainfall
disaggregation model for South Africa.

_______________________________________________________________________
24
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

SYNTHETIC DESIGN STORM METHODS

Rectangular method
Single point
(Category 1) on IDF curve
Triangular method
IDF based

Entire IDF Chicago design storm

curve
Rainfall distribution curves

SCS (I,Ia,II,III) SCS-SA (1,2,3,4) Site specific

(Category 2) Huff curves Dimensionless durations

Mass-curve
HRU 1/72 Fixed durations (2 to 24 h)
based

NOAA (14) Fixed durations (6 to 96 h)

(Category 3)
Knoesen’s semi-stochastic model
Stochastically
based
Considerable number of models not
covered in this study

Figure 2.17: Categorization of synthetic design storm methods covered in the

literature review

Although the existing synthetic design storms were developed using the best data,
technology, and engineering judgement available at the time, they do present some degree
of insufficiency. For example, the total precipitation volume is systematically underestimated
by the rectangular hyetograph (Arnell, 1982), and it was also realised that it gives a wrong
picture of a hyetograph (Niemczynowicz, 1982). In terms of the triangular hyetograph
method, Veneziano and Villani (1999) have noticed that, although it is quite simple and
intuitive, it does not have a strong conceptual basis and may produce biased flow estimates.
The CDS method, on the other hand, utilises point rainfall data that applies to a particular
site and therefore the variability and unique character of the rainfall patterns will
automatically be embedded in the CDS method.

_______________________________________________________________________
25
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
However, the CDS overestimates the peak discharge comparing with results from a
continuous simulation modelling analysis (Malik and James, 2007). Veneziano and Viallani
(1999) state that it tends to overestimate the peak discharge because it produces an
unrealistic single event.

Methods that utilise the entire IDF curve to generate a synthetic design storm hold much
potential since the design rainfall has been regionalised by Smithers and Schulze (2000).
Some of the variability and unique character of the rainfall pattern will therefore automatically
be embedded in these methods. In terms of Huff curves, Adamson (1981) suggested that it
may be considered representative of regions in Southern Africa with similar rainfall climate
and topography to the Mid-West of the United States of America (USA). However, to
determine the similarity, variables like latitude, longitude, altitude, distance from the sea,
mean annual precipitation (MAP) need all to be considered. Sufficient data also exists which
provide for an opportunity to develop similar type curves for the Gauteng province. The
standard SCS Types I and II curves, on the other hand, were developed from data in the
USA during the 1960 and 1970s. These curves are, however, in the process of being
replaced by the NOAA Atlas 14 curves, which are regionalised curves for the USA,
developed similarly to the Huff curves (NRCS, 2019), using the latest rainfall data. The
standard SCS Types I and II curves were originally adopted for use in South Africa (Schulze
and Arnold, 1979), but were further developed into four revised curves by Weddepohl (1988)
as cited by Ramlall (2020). The four curves were regionalised, resulting in the SCS-SA Type
1, 2, 3 and 4 curves (Ramlall, 2020). Following the regionalisation map of the four SCS-SA
curves, Type 3 applies to the Gauteng province, as well as the North-West, Northern Cape,
and parts of the Eastern Cape, Mpumalanga, and Limpopo provinces. However, from
practical experience, it appears that the Type 3 curve yields peak discharges that seem too
high for specific sites in Gauteng when applied to a SWMM model (Males et al., 2004). This
is partly due to too high ratios for the D-hour to one-day rainfall compared to the ratios
calculated using the design rainfall of Smithers and Schulze (2000).

_______________________________________________________________________
26
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
DATA COLLECTION AND STORM EVENT
IDENTIFICATION

The first aim of this study is to identify and assess the performance of currently available
methods to estimate synthetic design storms used as input for single-event modelling in the
selected pilot study area. In order to achieve this aim, short-duration rainfall data for the
study area had to be collated and reviewed, and applicable observed storm events identified.
This chapter describes the data collection, quality assessment and storm event identification
processes applied in this study.

SAWS RAINFALL DATA SOURCE

Rainfall data, recorded in 5-min intervals, were obtained from the SAWS for 35 stations
situated in Gauteng. The stations consist of four Weather Office Stations (WO), 13
Automatic Weather Stations (AWS) and 17 Automatic Rain Stations (ARS). The locations of
the stations are depicted in Figure 3.1 and their general details summarised in Table 3.1.

Figure 3.1: SAWS stations with short duration rainfall data in Gauteng

_______________________________________________________________________
27
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Rainfall is recorded at all the stations using a tipping bucket rain gauge, which consists of a
funnel and a seesaw-like container that is divided in the middle into two individual
compartments. The funnel channels the rainfall into the bucket, one compartment at a time,
with a rainfall resolution of 0.2 mm. When the bucket tips an electronic pulse is sent to a
data logger which records the total number the bucket has tipped in 5-min and translate it to
a rainfall depth. The data logger records a zero value for each 5-min interval during periods
of no rainfall. If the power supply is interrupted, the data logger will not record which will
result in a gap in the dataset. The data recordings are described in detail in the next section.

Table 3.1: SAWS short duration rainfall stations in Gauteng

Data
ID Name Number Latitude Longitude Type Period
(Years)
1 Jhb OR Tambo 0476399_0 26.1430 28.2346 WO 26.2
2 Lanseria 0512746_7 25.9436 27.9188 WO 10.7
3 Irene 0513385A2 25.9105 28.2106 WO 26.2
4 Bolepi House 0513439A1 25.8094 28.2564 WO 17.6
5 Vereeniging 0438784_3 26.5699 27.9582 AWS 26.2
6 Zuurbekom 0475528B7 26.3008 27.8136 AWS 6.3
7 Kloofendal 0475637_1 26.1308 27.8799 AWS 3.9
8 Jhb Bot Gardens 0475879_0 26.1566 27.9991 AWS 26.1
9 Springs 0476764_2 26.2395 28.4351 AWS 3.9
10 Bronkhorstspruit 0514408AX 25.8087 28.7386 AWS 11.7
11 Waterkloof AFB 0513379_8 25.8277 28.2235 AWS 7.9
12 Pretoria Unisa 0513346_0 25.7663 28.2005 AWS 26.2
13 Pretoria Proefplaas 0513435A4 25.7520 28.2585 AWS 9.9
14 Pretoria Pur 0513284_8 25.7351 28.1760 AWS 2.5
15 Pretoria TUT 0513253B3 25.7301 28.1627 AWS 1.9
16 Wonderboom RWY11 0513339B8 25.6516 28.2115 AWS 0.4
17 Wonderboom RWY24 0513369C4 25.6525 28.2272 AWS 0.4
18 Fochville Police 0474899_9 26.4877 27.4941 ARS 9.3
19 Westonaria Kloof 0475174_4 26.3944 27.5979 ARS 9.3
20 Goudkoppies 0475736B8 26.2715 27.9265 ARS 11.0
21 Dube 0475674_5 26.2466 27.8947 ARS 9.2
22 Dobsonville 0475613_X 26.2232 27.8626 ARS 9.2
23 Magaliesburg Police 0512090_5 25.9980 27.5386 ARS 9.2
24 Sterkfontein 0475361_8 26.0174 27.7317 ARS 11.2
25 Jhb Sandton 0476096_0 26.1026 28.0690 ARS 10.8
26 Alexandra Depot 0476156_X 26.1088 28.1130 ARS 9.2
27 Ivory Park 0513359_3 25.9906 28.2019 ARS 9.2
28 Diepsloot 0513025_X 25.9239 28.0218 ARS 9.2
29 Pta presidency 0513404_6 25.7394 28.2325 ARS 10.9
30 Pta Rietondale 0513404A0 25.7292 28.2361 ARS 9.7
31 Pta Mountain View 0513222_0 25.6994 28.1567 ARS 9.0
32 Baviaanspoort 0513611_4 25.6908 28.3633 ARS 9.3
33 Kameeldrift 0513550_3 25.6603 28.3144 ARS 9.8
34 Shosanguve 0550115_8 25.5289 28.1137 ARS 11.4
35 Wonderboom 0513369_0 25.6631 28.2166 AWS 10.6

_______________________________________________________________________
28
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
DATA COLLATION

Each rain gauge is equipped with a solar panel and a General Packet Radio Service (GPRS)
so that data can be remotely downloaded. Each AWS and ARS is assigned to a WO from
where the data for the previous day’s recordings are downloaded daily. The data then goes
through several quality checks before it gets uploaded into the SAWS database. If any
irregularities are noticed in the data during the quality checks which cannot be
meteorologically explained, the rainfall recordings are then manually deleted. Therefore, an
interval will be present in the data set, but the rainfall column will be left blank. Typical
deleted rainfall and missing intervals in a data set are depicted in Table 3.2.

Table 3.2: Typical short duration rainfall data set

Number Name Lat Long year month day hour minute Rain
∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙
0476399_0 JHB INT WO -26,14 28,23 1994 12 31 1 35 0,0
0476399_0 JHB INT WO -26,14 28,23 1994 12 31 1 40 0,0
0476399_0 JHB INT WO -26,14 28,23 1994 12 31 1 45 0,0
0476399_0 JHB INT WO -26,14 28,23 1994 12 31 1 50 0,0
0476399_0 JHB INT WO -26,14 28,23 1994 12 31 1 55 0,0
0476399_0 JHB INT WO -26,14 28,23 1994 12 31 2 0 0,0
0476399_0 JHB INT WO Missing
-26,14 28,23 1995 1 1 2 5 0,0
0476399_0 JHB INT WO intervals
-26,14 28,23 1995 1 1 2 10 0,0
0476399_0 JHB INT WO -26,14 28,23 1995 1 1 2 15 0,0
0476399_0 JHB INT WO -26,14 28,23 1995 1 1 2 20 0,0
0476399_0 JHB INT WO -26,14 28,23 1995 1 1 2 25 0,0
∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 17 15 0,2
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 17 20 0,0
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 17 25 0,0
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 17 30 0,0
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 17 35
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 Deleted
17 40
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 rainfall
17 45
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 17 50
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 17 55
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 0
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 5
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 10
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 15
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 20
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 25
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 30
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 35 0,0
0476399_0 JHB INT WO -26,14 28,23 2000 10 20 18 40 0,0
∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙ ∙∙∙∙∙∙

_______________________________________________________________________
29
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
DATA PROCESSING

The short duration rainfall data sets were processed using a Java application called the
Rainfall Processor (Rain-Pro), which was developed using IntelliJ IDEA Community Edition
2019.1.3 and JDK 1.8 with JRE 1.8 (Munro, 2021, personal communication, 23 August).
Subsequent analysis of the data was done using Microsoft (MS) Excel. The order in which
the data was processed is summarised in the flow chart depicted in Figure 3.2.

The process began with the short duration rainfall data of a station (RAW DATA SET.txt) as
input for the Rain-Pro software, and the output consists of three files. The first being the
PROCESSED.txt file, which consisted of the rainfall data, except for all intervals with zero
rainfall. This file was used as input to the ANNUAL MAXIMUM SERIES.xlsx, which provided
the input for the STATISTICAL ANALYSIS.xlsx file. This file provided the design rainfall
estimations as described in Section 4.2. The PROCESSED.txt file was also used as input
to the MS Excel file, STORM FILTER.xlsx, which applied the maximum dry period and the
minimum rainfall depth criteria, described in Sections 3.5.1 and 3.5.2, respectively. After
applying the maximum dry period and minimum rainfall depth criteria, the STORM
FILTER.xlsx file provided the start and end dates of all individual events, which was used to
create the STRIP DATES.txt file. The second output file of the Rain-Pro software, the
MISSING_INTERVAL.txt file, contained the missing data which was subsequently used as
input to the MISSING ANALYSER.xlsx file as described in Section 3.4. The third output file
of the Rain-Pro software, the STATS.txt file provided a summary of the data set. This
summary consisted of the total number of lines processed, total number of lines with zero
rainfall as well as lines with rainfall larger than zero, the start and end dates of the data set,
the data period, and the total duration of missing data.

The Rain-Pro software was executed for a second time, which extracted the individual
events from the RAW DATA SET.txt file according to the STRIP DATES.txt file. The output
from the Rain-Pro software, after the second run, consisted of the storm event files,
0M_EVENTS.txt, 15M_EVENTS.txt, etc., which contained the individual events according
to the maximum dry period and minimum rainfall depth criteria. These files were used as
input to the RI FILTER.xlsx file, which applied the minimum rainfall intensity criterion
described in Section 3.5.3. The RI FILTER.xlsx file provided the start and end dates of the
significant events, which was used to update the STRIP DATES.txt file.

The Rain-Pro software was executed for a third time, and the storm event files were
subsequently updated. These files contained the rainfall data of only the significant events
which were used as input for the STORM PARAMETERS.xlsx file. This file determined the
storm parameters described in Sections 4.1.2 and 4.1.3, respectively. The storm event files
were also used as input for the SHAPE ASSESSMENT.xlsx and INTENSITY
ASSESSMENT.xlsx files, which were used for the storm shape and average intensity
assessments described in Sections 5.1 and 5.3.

_______________________________________________________________________
30
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

RAW DATA SET.txt

STRIP DATES.txt
CONFIG FILE.txt

RAIN-PRO.jar
STORM FILTER.xlsx

0M_EVENTS.txt

Read Strip Dates “NO”

PROCESSED.txt

Read Strip Dates “YES”

15M_EVENTS.txt

MISSING_INTERVAL.txt
30M_EVENTS.txt

STATS.txt 60M_EVENTS.txt

MISSING ANALYSER.xlsx RESULTS 120M_EVENTS.txt

RI Filter Complete “NO”

RI FILTER.xlsx

RI Filter Complete “YES”

ANNUAL MAXIMUM SERIES.xlsx
DESIGN RAINFALL RATIOS.xlsx RESULTS
STATISTICAL ANALYSIS.xlsx

DESIGN STORMS.xlsx STORM PARAMETERS.xlsx

SWMM MODEL.inp SHAPE ASSESSMENT.xlsx INTENSITY ASSESSMENT.xlsx

RESULTS RESULTS

LEGEND

Java application (.jar) Microsoft Excel files (.xlsx)

Results on which the findings of

Java application input file (.txt)
this study are based
Output files of the Java
applications (.txt) EPA SWMM software file (.inp)

Figure 3.2: Data processing flow chart

_______________________________________________________________________
31
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
MISSING DATA ANALYSIS

The missing data analysis was done by identifying missing periods of data, consisting of
missing 5-min intervals, as well as intervals that had no rainfall value assigned, as depicted
in Table 3.2. The information of all the missing periods of each station was written to a
separate text file (MISSING_INTERVAL.txt) which was used for further processing in MS
Excel. The time of the year when missing data occurs was taken into consideration. The
average monthly rainfall, depicted in Figure 3.3, for the three stations situated at O.R. Tambo
International Airport, Irene, and Johannesburg Botanical Gardens, were used to guide this
process. Also indicated in Figure 3.3 are the minimum and maximum values of the three
stations, which indicates the variances in the average rainfall. Based on this analysis, the
months of May to September were considered to be ‘dry’ months and as a result, missing
data occurring during these months were assumed to be zero rainfall. Only missing data
occurring during the ‘wet’ months were used for the classification of the data quality.

Figure 3.3: Average monthly rainfall for O.R. Tambo, Irene and Jhb Bot Gardens

The data quality of each station was characterised according to the criteria summarised in
Table 3.3 and the results are summarised in Table 3.4. Stations with data periods of more
than 20 years and periods of missing data during ‘wet’ months of less than 5% were
classified as stations with good data sets. These stations were used to determine the storm
parameters and design rainfall estimation described in Chapter 4. They were also used for
the shape and intensity assessments, as well as the continuous simulation modelling
described in Chapter 5. Data period of less than 20 years but more than 10 years, and
missing data less than 20%, were classified as Average data sets. These stations were
considered to be less reliable and were only used for the design rainfall estimation. Data
period of less than 10 years, and missing data of more than 20%, were classified as poor
sets. These stations were omitted from this study due to the data sets being incomplete. In
total 35 stations were assessed, with five having good data sets, 17 average, and 13 poor,
as depicted in Figure 3.4.

_______________________________________________________________________
32
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Table 3.3: Data quality classification criteria
Data Period
Description
years ≥ 20 20 > years ≥ 10 10 > years
Missing data ≤ 5% Good Average Poor
during wet ≤ 20% Average Poor Poor
months > 20% Poor Poor Poor

Table 3.4: Missing data analysis summary

Data Missing Data (%) Data
ID Name Period
All year round Wet months Quality
(Years)
1 Jhb OR Tambo 26.2 1.9 1.0 Good
2 Lanseria 10.7 1.1 1.0 Average
3 Irene 26.2 1.2 0.7 Good
4 Bolepi House 17.6 16.3 10.6 Average
5 Vereeniging 26.2 6.9 5.2 Good
6 Zuurbekom 6.3 1.4 0.6 Poor
7 Kloofendal 3.9 17.9 17.0 Poor
8 Jhb Bot Tuine 26.1 4.1 2.8 Good
9 Springs 3.9 55.7 31.7 Poor
10 Bronkhorstspruit 11.7 3.5 3.0 Average
11 Waterkloof AFB 7.9 7.6 7.0 Poor
12 Pretoria Unisa 26.2 8.6 4.7 Good
13 Pta Proefplaas 9.9 1.4 1.1 Average
14 Pretoria Pur 2.5 18.3 3.3 Poor
15 Pretoria TUT 1.9 23.5 19.6 Poor
16 Wonderb. RWY11 0.4 17.0 16.3 Poor
17 Wonderb. RWY24 0.4 10.1 8.8 Poor
18 Fochville Police 9.3 2.2 2.0 Average
19 Westonaria Kloof 9.3 13.4 7.8 Average
20 Goudkoppies 11.0 6.5 4.2 Average
21 Dube 9.2 9.2 5.8 Average
22 Dobsonville 9.2 25.5 17.2 Average
23 Magalies Police 9.2 40.0 18.8 Average
24 Sterkfontein 11.2 11.8 6.8 Average
25 Jhb Sandton 10.8 8.2 7.1 Average
26 Alexandra Depot 9.2 7.3 6.1 Average
27 Ivory Park 9.2 20.4 12.4 Average
28 Diepsloot 9.2 3.9 2.6 Average
29 Pta presidency 10.9 35.1 23.2 Poor
30 Pta Rietondale 9.7 51.5 31.3 Poor
31 Pta Mount. View 9.0 42.1 23.0 Poor
32 Baviaanspoort 9.3 38.0 22.6 Poor
33 Kameeldrift 9.8 52.7 31.6 Poor
34 Shosanguve 11.4 13.2 9.3 Average
35 Wonderboom 10.6 4.5 3.5 Average

_______________________________________________________________________
33
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 3.4: Total number of stations according to quality classification criteria

A further assessment of the data quality was conducted by comparing the daily rainfall data
measured by a standard non-recording raingauge at 08:00 every day with the 5-min data
accumulated for the same 24 h period. The daily rainfall at each WO is recorded daily at
08:00 am, using a standard non-recording rain gauge that is independent from the automatic
rain gauge. To verify the accuracy of the 5-min data, the total rainfall was determined
between 08:00 am of each day and the previous day. The total rainfall for each day was
then compared with the recorded daily rainfall. For example, the difference at the OR Tambo
station between the two totals, on a daily scale, is depicted in Figure 3.5. This was compiled
by subtracting the total rainfall from the 5-min data between 08:00 and 08:00, from the daily
rainfall. It was observed that the daily rainfall was frequently assigned to the incorrect day.
For example, the total rainfall from the 5-min data on 29 April 1995, was 23.4 mm. The same
value was recorded in the daily rainfall, but on the day before. The same applies to all
intervals where two identical amounts are depicted but in opposite directions, for example,
26 January 2015.

Figure 3.5: Rainfall difference between daily and 5-min data on a daily scale at O.R.
Tambo station
_______________________________________________________________________
34
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Due to the discrepancy between the dates of the daily and 5-min rainfall, the verification
could not be conducted daily. The verification was therefore conducted by comparing the
annual maximum daily rainfalls from the daily and the 5-min data as depicted in Figure 3.6.
The highest discrepancies are observed for the years 2001, 2006, 2009 and 2020. However,
according to Figure 3.7, the missing data for O.R. Tambo occurred mostly between 1999
and 2003. This provided the basis to conclude that there were substantial discrepancies
between the daily and the 5-min rainfall data and that more investigation is needed. It was,
however, assumed for this study that the 5-min data is sufficiently accurate to conduct this
study.

Figure 3.6: Annual maximum daily rainfall from the 5-min and daily rainfall data for
the O.R. Tambo station

Figure 3.7: Annual missing data for the O.R. Tambo station

_______________________________________________________________________
35
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
STORM EVENT IDENTIFICATION

A single event-based methodology was adopted for this study to identify independent storm
events. Generally, this methodology consists of defining an allowable Maximum Dry Period
(MDP) between rainfall spells, as depicted in Figure 3.8, as well as a Minimum Rainfall Depth
(MRD) threshold. This methodology was adapted by the addition of a third threshold criterion
that is related to the Minimum Rainfall Intensity (MRI) of the storm event. The three criteria
and the results are described in this section.

3.5.1 Maximum Dry Period (MDP)

The MDP criterion refers to the threshold period of no rainfall that occurs between two rainfall
spells. If the dry period exceeds the MDP threshold, then the two spells are considered as
two separate events. Conversely, if the dry period is less than the MDP, then the two spells
are considered as one event. This concept is depicted in Figure 3.8. Different methods to
identify dry periods are documented in the literature. For example, Huff (1967) used a
maximum of 6-hours to separate independent rainfall events. Ramlall (2020) initially used
6-hours as the MDP but found that events frequently included extensive periods of low to
zero rainfall. Subsequently, the MDP was reduced to 1-hour which resulted in an improved
coefficient of determination (R²) between total rainfall depth and total storm duration.

The MDP can also be determined by considering the statistical independence of events
using an empirical relation for estimating the minimum time between independent events
(Restrepo and Eagleson, 1980). This process involves the computation of the mean and
standard deviation of dry periods in a continuous rainfall data set. In an iterative process,
the smallest dry periods are omitted until the mean and standard deviation is equal (Bonta,
2004). However, for this study, the MDP was systematically increased from 0 to 120-min to
identify idealised independent events on which the assessments were conducted. In other
words, each data set was assessed from start to finish using a zero MDP criterion. Then a
second assessment was conducted on the same data set but with a 15-min MDP criterion,
then the 30-min MDP, and so on. The total rainfall depth and total storm duration of a
particular storm event could therefore differ depending on the MDP used to distinguish
between idealised events.

_______________________________________________________________________
36
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 3.8: Idealised independent events (after Restrepo and Eagleson, 1982)

3.5.2 Minimum Rainfall Depth (MRD)

Following the separation of the continuous rainfall data into individual storm events, the total
rainfall of each event was considered. A storm event was ignored if the total rainfall was less
than 10 mm. The minimum rainfall of 10 mm was selected considering the design rainfall
obtained from the computer software, Design Rainfall Estimation in South Africa (DRESA)
(Smithers and Schulze, 2003), which implements the procedures to estimate design rainfall
in South Africa per Smithers and Schulze (2000). For example, the five and 10-min design
rainfall for the O.R. Tambo station with a recurrence interval of two years depicted in Figure
3.9 are 8.8 and 12.7 mm, respectively. This implies that if the total rainfall is less than 10
mm, the total duration of the event was between five and 10 min, which would have been
recorded in only one or two five-min intervals. The location of the peak intensity of such an
event will therefore be either at the beginning of, or the end, of the event whereas it is
expected for the peak to be somewhere within the event.

The second argument is that if the total rainfall of 10 mm was achieved over a longer period,
the intensities will be lower than the 5-min design rainfall with a two-year recurrence interval.
_______________________________________________________________________
37
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Therefore, limiting the minimum rainfall to 10 mm to select events ensured that insignificant
storm events were eliminated early during the storm event identification process.

Figure 3.9: Typical design rainfall estimation results for O.R. Tambo International
Airport obtained from the DRESA software

3.5.3 Minimum Rainfall Intensity (MRI)

The minimum intensity was the final criterion that was applied before an event could be
classified as a significant event. The recurrence interval of each storm event was estimated
using the DRESA software. The ninety per cent upper (U) and lower (L) bounds for the
design rainfall values were not considered for this study. The depth-frequency relationship
for the design rainfall was then approximated using an exponential function. The 5-min
design rainfall for the O.R. Tambo International Airport station, for example, is depicted in
Figure 3.10 as well as the fitted exponential regression line. If the maximum recurrence
interval of the event was less than two years, the storm event was ignored. This was done
to eliminate storm events with insignificant intensities considering the 5-min to 24-hour
design rainfall for each station.

A typical application of this criterion is depicted in Figure 3.11, which was applied to the
storm event that occurred at the O.R. Tambo station on 13 March 2011. The total storm
duration for this event was 12 hours. A moving window was used to extract the maximum
rainfall for each of the standard time steps from 5 minutes to 12 hours. As depicted in Figure
3.11 the 5-min to 6-hour time steps were less than the specified threshold of two years,
however, the 8, 10 and 12-hour time steps exceeded the threshold which therefore qualified
this storm event.

_______________________________________________________________________
38
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 3.10: Typical depth-frequency relationship for O.R. Tambo International

Airport

Figure 3.11: Typical maximum recurrence intervals per standard time step for the
storm event that occurred on 13 March 2011 at the O.R. Tambo International Airport

3.5.4 Identification of single storm events

The Rain-Pro software was programmed to extract all data records with rainfall values more
than zero rainfall from the raw data set. A new data file was then written to a separate text
file (PROCESSED.txt). This process simplified the data file in terms of the number of line
items, because the data loggers of the SAWS automatic rain gauges continuously records
data in 5-min intervals. In any raw data file, more than 95% of the line items have zero
rainfall. Because all these items have the same value, namely zero, these items are,
therefore, not unique.

_______________________________________________________________________
39
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
All further processing of the new data set could, therefore, assume that all intermediate
intervals of time have zero rainfall. The processing of the PROCESSED.txt file was done in
MS Excel. For a MDP of 0-min, which means that a storm event ended immediately when
precipitation stopped, the total number of storm events identified from the 22 stations with
good to average data sets, was 4 052 events. The number of events consistently increased
as the dry period increased. This is because of the combined precipitation of short spells,
forming a single event, that exceeds the threshold of 10 mm, whereas individually the spells
do not meet the criterion. As a result, the total number of events for a MDP of 120-min was
5 726. The total number of storm events for the selected stations are depicted in Figure 3.12.

It is also important to note that a large portion of the total number of events consisted of
events for which the recurrence interval was less than two years, which are insignificant
storm events. Over 80% of the events had a recurrence interval of less than two years. The
storm events with recurrence intervals exceeding two years were considered significant, and
the analyses were conducted on these events. The frequency of events for the 22 stations
with good to average data sets, considering different MDP’s, is depicted in Figure 3.13. It
can be seen that 80 to 85% of all events are insignificant.

Figure 3.12: Total number of storm events and insignificant events identified based
on different MDP’s

_______________________________________________________________________
40
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 3.13: Frequency of events based on different MDP’s

_______________________________________________________________________
41
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
CHAPTER SUMMARY

This chapter has provided details about the data source, the collation of the rainfall data,
and an overview of the data processing. The missing data analysis was based on the total
number of missing periods during rainy months. These months, consisting of January to
April and October to December, were identified by considering the average monthly rainfall
of the O.R Tambo, Irene and Johannesburg Botanical Gardens stations.

The missing data as well as the data period of each station were used to classify the data
quality of each rainfall station. A station was classified as either, good, average or poor,
depending on the selected classification criteria. The stations with poor data sets were
omitted from the study, whereas the good and average stations are used in Chapter 4 in
subsequent investigations. The criteria used for the identification of individual events were
also described. These criteria consisted of an MDP ranging from 0 to 120-min, an MRD of
10 mm, and an MRI of 1:2 years. In Chapter 4 a single MDP will be selected and subsequent
analyses in the chapter, such as the advancement coefficient and dimensionless time to
peak, are based on this MDP.

_______________________________________________________________________
42
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
METHODOLOGY AND DATA ANALYSIS

This chapter consists of seven sections that describe several aspects relating to the
generation of synthetic design storms which provides the foundation for Chapter 5. The first
section of this chapter describes the extraction of parameters relating to the CDS method,
and the dimensionless time to peak relating to the Triangular (TRI) method. The significant
storm events identified in the previous chapter were used to determine these storm
parameters. The second section describes the design rainfall estimation of stations with
good and average data sets as defined in the previous chapter, using Probability Distribution
(PD) analysis. The third section describes the methodology that was developed to determine
the regression coefficients associated with the CDS method. The fourth section describes
the methodology that was adopted and adapted for the distribution of the design rainfall to
create a 24-hour synthetic design storm. It also describes the process of extracting storm
events from a 24-hour DC which is necessary for comparing the shape of synthetic design
storms and significant storm events in the next chapter. The fifth section describes the
comparison of the design rainfall ratios with the ratios of the SCS-SA curves and the
recommendation with regards to the implementation of intermediate curve types. Due to this
comprehensive comparison, the sixth section describes the comparison of the SCS curves
with the SCS-SA curves and the conclusion with regards to the use of the SCS curves. In
section seven standardised mass curves are developed using the significant storm events
identified in the previous chapter.

STORM PARAMETERS

In this section, an appropriate MPD, as defined in Chapter 3.5, is investigated and selected.
Based on this finding, the storm parameters that were then considered include the storm
advancement coefficient associated with the CDS method, and the time to the peak intensity
associated with the TRI method. The adaptations of the procedures to determine these
parameters and the results are described below.

4.1.1 Selection of an appropriate MDP

As noted in Chapter 2 (literature review), Ramlall (2020) based the decision to use an MDP
of 1-hour instead of 6-hour, on the improved correlation between total rainfall depth and total
storm duration. A similar investigation was initially conducted whereby the correlation was
determined using six different MDP criteria, namely, 0-min, 15-min, 30-min, 60-min, 120-
min, and 360-min. This investigation was based on the significant storm events of five
stations that were pooled together, namely: O.R Tambo, Irene, Vereeniging, Jhb Bot
Gardens and Unisa. The results are summarised in Figure 4.1. This indicates that the
correlation improves consistently as the MDP increases, but decreases dramatically with an
MDP of 360-min.
_______________________________________________________________________
43
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
This finding is consistent with the finding of Ramlall (2020), who found an increase in
correlation by decreasing the MDP from 6-hour to 1-hour. However, an optimum correlation
should exist with an MDP between 2-hour and 6-hour.

Figure 4.1 Correlation between total rainfall and duration for different MDP criterion

_______________________________________________________________________
44
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Subsequent to the above initial approach, an alternative approach was taken with this study
which based the selection of the MDP on the response time of the catchment. It is argued
that if the MDP is longer than the reaction time of a catchment, then any rainfall after a dry
period would not impact the peak discharge from the first spell. The second spell is then
rather seen as the start of the next event. The Antecedent Moisture Content (AMC) during
the second spell is significantly impacted by the first event. The focus of this study is,
however, not on the AMC but rather the shape assessment of synthetic design storms.
Therefore, the alternative approach is warranted. The typical reaction time of small urban
catchments are approximately 15-min, and therefore, the 15-min MDP was selected for this
study.

4.1.2 Storm advancement coefficient – CDS method

As detailed in Section 2.3, the storm advancement coefficient was determined in this study
using Keifer and Chu’s (1957) second approach. Firstly, according to Keifer and Chu’s
(1957) first approach, the antecedent rainfall was determined by first identifying the
maximum rainfall within a 15, 30, 60 and 120-min period within each significant storm event.
Each significant storm event was then assumed to be 180 min in total, and the 180 min
design rainfall for the 1:5 year storm event was divided on either side of the maximum
intensity using the storm advancement coefficient. Since the significant storm events
identified for this study were shown to vary significantly in terms of recurrence interval, a
single design rainfall would therefore be inappropriate. Secondly, there was an insignificant
difference between the coefficients determined by Keifer and Chu (1957), which were 0.387
and 0.375 for the first and second approaches, respectively. Therefore, the first approach
was not considered for this study. The procedural steps that were followed to determine the
advancement coefficient in this investigation, adopted from Keifer and Chu’s (1957) second
approach, are summarised as follows:

(a) The maximum rainfall within a 15, 30, 60 and 120-min duration within each significant
storm event, was identified.
(b) For each of these durations, the ordinal position of the 5-min interval with the
maximum rainfall was identified.
(c) The location of the peak intensity for each period was determined in accordance with
Equation 4.1 This concept is illustrated in Figure 4.2, which depicts a typical storm
event for which the location of the peak intensity for a 30-min period was calculated.
(d) The weighted average for each significant event was determined, using the location
of peak intensity and weighted proportionally to 15, 30, 60 and 120-min in accordance
with Equation 4.2.
(e) The average for all significant events for each of the five best stations was determined
in accordance with Equation 4.3.

_______________________________________________________________________
45
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
ti
rd = (po − 0.5) [ 4.1 ]
dt
where:

rd = location of peak intensity,

po = ordinal position of interval with maximum rainfall,
ti = duration of interval (5) (min), and
dt = total duration (15, 30, 60 and 120) (min).

∑N
t=1 rd ∙ dt
d
rs = Nd [ 4.2 ]
∑t=1 dt
where:

rs = advancement coefficient of significant storm event s, and

Nd = Number of storm durations (4).

Ns
1
ri = � rs [ 4.3 ]
Ns
s=1
where:

ri = advancement coefficient of station s, and

Ns = Number of significant storm events of station s.

30-min

r ∙ 30 5-min intervals with the

Location of the highest combined rainfall
peak intensity for a 30-min period
Rainfall

Ordinate (3rd) with

highest rainfall
Location of peak intensity:
5
𝑟𝑟 = (3 − 0.5) = 0.417
30

Time

Figure 4.2: Typical location of the peak intensity within a 30-min duration

_______________________________________________________________________
46
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The storm advancement coefficient was calculated from the significant storm events
identified from the five stations with good data sets in Gauteng, namely, O.R Tambo, Irene,
Vereeniging, Jhb Bot Gardens and Unisa. The results, depicted in Figure 4.3, indicates an
advancement coefficient that is very similar to the coefficient proposed by Keifer and Chu
(1957) of 0.375. Therefore, an average value of 0.380 was used in this study. It is, however,
recommended that the impact of the advancement coefficient has on the peak discharge,
be investigated in a future study by conducting a sensitivity analysis.

Figure 4.3 Storm advancement coefficients, based on Keifer and Chu’s (1957)
second approach

4.1.3 Dimensionless time to peak - TRI method

The hyetograph’s first-moment arm for each significant storm event was calculated and
related to a triangular representation of the hyetograph with an equal total rainfall volume
and total storm duration, as described by Yen and Chow (1980). However, for some events
the first-moment arm was small which resulted in an obtuse type of triangle, meaning that
one angle is larger than 90 degrees. This is, however, not possible, and therefore, in such
instances, a right angle was assumed. An example of such an event was observed at O.R
Tambo on 4 January 1997, is depicted in Figure 4.4. The first-moment arm was calculated
using Equation 2.11, which resulted in 43.2 minutes, and using Equation 2.13 to calculate
the time to peak intensity, yielded negative 40.3 minutes. The negative time to peak was
adjusted to zero minutes as depicted in Figure 4.4. Approximately 20% of the significant
single event-based events required this adjustment. The effect of different total storm
durations, total rainfall volumes and different seasons were not considered for this study, as
proposed by Yen and Chow (1980).
_______________________________________________________________________
47
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.4: Correction of an obtuse triangle such as for the storm event at O.R.
Tambo on 4 January 1997 at 00:50

Yen and Chow (1980) has determined the statistical mean values for the storm events sorted
into various duration groups. The first group contained all events, the second group the 2-
hour events, the third group the 3-hour events, etc. However, an insignificant difference was
reported between the groups. Therefore, all events were grouped together for this study, for
which the average time to peak ratio was computed at each of the five best stations, as
shown in Figure 4.5. From the results the average time to peak intensity for the five stations
were 0.20.

Figure 4.5 Time to peak intensity using method proposed by Yen and Chow (1980)

_______________________________________________________________________
48
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
DEPTH-DURATION-FREQUENCY (DDF) CURVES

The development of short duration DDF curves for the rainfall stations in Gauteng is
described in this section. The DDF curves were then subsequently used for further
investigations in this study, namely:
(a) comparison with the DRESA design rainfall,
(b) generating rainfall distribution curves, and
(c) determining of the D-hour to 24-hour design rainfall ratios, which were then compared
with the ratios of the SCS-SA curves.

Municipal stormwater infrastructure associated with minor stormwater drainage networks is

generally designed to accommodate storm events with a return period of 1:5 up to 1:20
years. The DDF curves were, therefore, developed for stations that have a minimum data
period of approximately 10 years. With a data period of at least 10 years the estimation of
design events with a return period of up to 1:20 years is allowed, as estimation of return
periods greater than twice the record length are generally not recommended (Smithers and
Schulze, 2000). In terms of missing data, Smithers and Schulze (2000) found that there was
no significant effect on design rainfall estimates if up to 20% of the years in the Annual
Maximum Series (AMS) do not contain their true value. Even though this finding does not
explicitly correlate with the missing data in the context of this study, 22 stations were
identified to comply with this criterion which is located as depicted in Figure 4.6.

Figure 4.6: Rainfall stations in Gauteng that were used to develop IDF curves
_______________________________________________________________________
49
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The AMS was extracted for 16 time steps ranging from 5-min to 24-hours (standard time
steps), and a PD analysis was conducted using the General Extreme Value (GEV)
distribution. Based on Goodness-of-fit (GOF) tests and L-moments to fit the distributions to
the data, Smithers (1996) recommended the GEV and the 3 parameter Log-Normal (LN3)
distributions for short duration rainfall PD analysis in South Africa. In a further study,
Smithers and Schulze (2000) concluded that if a single PD was to be adopted for all regions
in South Africa, the GEV would be the most appropriate PD to use. Therefore the GEV
distribution was used to estimate design rainfall in this study.

Following the PD analysis of the 22 stations, the Relative Error (RE) in the design rainfall
obtained from the DRESA was computed for the 1:5, 1:10 and 1:20 year return periods,
respectively. The RE was determined following Equation 4.4, and the average for all stations
(ARE_t) was determined using Equation 4.5. The result of this analysis is depicted in Figure
4.7. This indicates that the ARE_t of the 1:20 year, 5-min to 15-min time steps, for example,
range between negative 9% and negative 18%. These differences could be attributed to: (i)
the inaccuracies in the 5-min rainfall data used in this study, (ii) the data period of the rainfall
records used in this study, (iii) the GOF of the GEV distribution for short duration design
rainfall, and (iv) the use of a regional approach in the DRESA study compared to the at-site
approach used in this study. However, the exact cause of the errors are yet to be
determined. The ARE_t for the 1:5 year, 30-min to 1.5-hour time steps, for example, range
between 10% and 16%. However, the ARE_t of all standard time steps are still within the
RE associated with the 90% upper and lower bounds given by the DRESA software as
depicted in Figure 4.7. Therefore, the estimation of the at-site design rainfall following this
PD analysis was deemed to be acceptable.

PPD − PSS
RE = ∙ 100 [ 4.4 ]
PSS
where:
RE = RE of time step t, and return period, T (%),
PPD = design rainfall obtained from the PD analysis (mm), and
PSS = mean design rainfall obtained from the design rainfall software (mm).

N𝑡𝑡 NS
1
ARE_t = ∙ � � REt,T [ 4.5 ]
Nt ∙ NS
t=1 T=1
where:
ARE_t = RE of time step t, and return period, T (%), and
NS = number of stations (22), and
N𝑡𝑡 = number of time steps (16).

_______________________________________________________________________
50
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
N𝑡𝑡
1
ARE_S = ∙ � REt [ 4.6 ]
Nt
t=1
where:
ARE_S = average RE of station, S, and return period, T (%), and
Nt = number of time steps (16).

Figure 4.7: Average RE for each standard time step (ARE_t)

Figure 4.8: Average RE for each rainfall station (ARE_S)

_______________________________________________________________________
51
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
INTENSITY-DURATION-FREQUENCY (IDF) REGRESSION COEFFICIENTS

A methodology of using a nonlinear algorithm to determine IDF regression coefficients as

well as the RE as a measure of their accuracy are presented in this section. IDF regression
coefficients (a, b, and c) are needed to generate a synthetic storm event using the CDS
method. These coefficients are associated with Equation 2.2, which is reproduced here as
Equation 4.7 for convenience, as follows:

a
ia = [ 4.7 ]
(b + t)c

where:
ia = Average rainfall intensity for a particular storm duration (mm/hour),
a,b,c = Site specific regression constants, and
t = Storm duration (min).

The design rainfall depths obtained from the DRESA were used to develop the regression
coefficients. These were determined using an optimisation algorithm that adjusts the
coefficients until the simulated intensity distribution is approximately equal to the actual
intensities. In this case, the actual intensities are the design rainfall obtained from the
DRESA software, divided by their respective durations and the simulated intensities are the
intensities that were simulated using Equation 4.7. The coefficients that yielded the best
design intensities were found using the Generalised Reduced Gradient (GRG) nonlinear
algorithm (Lasdon et al, 1976). The GOF was determined using the Root Mean Square Error
(RMSE) between the actual and simulated design intensities using Equation 4.8. The
coefficients were adjusted until the RMSE was optimised by reaching a minimum. The built-
in solver function of MS Excel, using the GRG nonlinear algorithm as the solving method
and the calculated RMSE value as the objective value was used for this procedure.

2
∑N
j=1 �IDFA𝑗𝑗 − IDFS𝑗𝑗 �
t
[ 4.8 ]
RMSE = �
Nt
where:
RMSE = root mean square error,
IDFA = actual design rainfall intensities (mm/hour),
IDFS = simulated design rainfall intensities (mm/hour), and
Nt = number of time steps (16).

_______________________________________________________________________
52
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Table 4.1: Regression coefficients determined from the design rainfall obtained from
the DRESA software
a
ID Name b c
1:5 1:10 1:20

1 Jhb OR Tambo 732 885 1047 4.269 0.726

2 Lanseria 974 1158 1347 5.869 0.763

3 Irene 745 901 1067 4.528 0.735

4 Bolepi House 988 1195 1414 5.621 0.756

5 Vereeniging 781 913 1040 5.086 0.752

6 Zuurbekom 740 896 1060 4.430 0.730

8 Jhb Bot Tuine 723 874 1035 4.101 0.720

10 Bronkhorstspruit 991 1198 1418 5.693 0.757

11 Waterkloof AFB 965 1166 1380 5.956 0.767

12 Pretoria Unisa 992 1200 1420 5.710 0.757

13 Pretoria Proefplaas 979 1183 1400 5.757 0.760

18 Fochville Police 742 867 988 4.434 0.731

19 Westonaria Kloof 738 862 982 4.339 0.729

20 Goudkoppies 741 895 1059 4.423 0.730

21 Dube 741 896 1060 4.430 0.731

22 Dobsonville 719 868 1028 4.006 0.718

23 Magaliesburg Police 981 1168 1358 5.776 0.760

24 Sterkfontein 715 850 988 3.926 0.715

25 Jhb Sandton 724 875 1035 4.119 0.721

26 Alexandra Depot 740 896 1060 4.437 0.731

27 Ivory Park 760 919 1087 4.780 0.742

28 Diepsloot 968 1170 1386 5.954 0.766

29 Pta presidency 979 1183 1400 5.757 0.760

32 Baviaanspoort 977 1182 1398 5.817 0.762

34 Shosanguve 988 1195 1414 5.686 0.757

_______________________________________________________________________
53
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The RE of the simulated intensities, using the optimal regression coefficients presented in
Table 4.1, was calculated in accordance with Equation 4.9 as follows:

IDFS − IDFA
RE = ∙ 100 [ 4.9 ]
IDFA
where:
RE = relative error (%),
IDFA = actual design rainfall intensities (mm/hour) estimated from the DRESA
software, and
IDFS = simulated design rainfall intensities (mm/hour) from Equation 4.7.

For example, the optimal coefficients from Table 4.1 for the 1:5-year design intensities for
O.R. Tambo were 732, 4.269 and 0.726, respectively. Applying these coefficients, the
simulated intensities were calculated for all standard time steps, and the RE concerning the
actual intensities were determined. The same applies to the 1:10 and 1:20 year regression
coefficients and the results are depicted in Figure 4.9.

Figure 4.9: RE between actual and simulated design rainfall intensities for O.R.
Tambo using the design rainfall from the DRESA software

The Average Relative Error (ARE_t) of each standard time step for the 1:5, 1:10 and 1:20
year return periods, respectively, of all stations, were determined following Equation 4.10
and the results are depicted in Figure 4.10. This indicates that the results are similar for all
recurrence intervals, and the example results of O.R Tambo depicted in Figure 4.9.

NS
1
ARE_t = ∙ � REt,T [ 4.10 ]
NS
j=1
where:

_______________________________________________________________________
54
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
ARE_t = relative error of time step t, and return period, T (%), and
NS = number of stations (22).

Figure 4.10: Average relative error for each standard time step (ARE_t)

RAINFALL DISTRIBUTION CURVE (DC)

The methodology of developing a DC is described in this section which is divided into two
parts. The first part describes the development of a 24-hour DC using the DDF curves
described in Section 4.2. Each DC was then subsequently used for the comparison between
single event-based and continuous simulation modelling. The second part describes the
methodology of developing a DC for a duration < 24-hours. This methodology was used for
the comparison between observed and simulated storm events.

4.4.1 Development of a 24-hour DC

The development of a 24-hour DC entails the embedment of the entire short duration DDF
curve in a single synthetic design storm. The peak intensity is located at the centre of the 24
hours and the design rainfall of shorter durations are then divided equally on either side of
the peak intensity. This process entails the interpolation of intermediate design rainfall for
durations between the standard time steps (5-, 10-, 15-, 30-, 45-min, etc.). For example, the
design rainfall for durations in 5-min intervals between 15-min and 30-min is interpolated
using a power regression function fitted through the incremental intensities of the standard
time steps. This concept was adopted, and adapted, from the National Engineering
Handbook (NRCS, 2019). This process is demonstrated in Table 4.2 and the procedural
steps that were followed to complete Table 4.2 are summarised as follows:

_______________________________________________________________________
55
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
(a) Populate Column 1 and 2 with the standard time steps and design rainfall,
respectively.
(b) Calculated the rainfall ratios in Column 3 by dividing each rainfall value by the 24-
hour rainfall value.
(c) Calculate the non-dimensionalised incremental intensity in Column 4 by dividing the
difference in ratio by the difference in duration. For example, the 0.5-hour non-
dimensionalised incremental intensity is (0.386 - 0.445) / (0.50 – 0.25) = 0.351.
(d) Calculate the logarithmic value of the duration and non-dimensionalised incremental
intensity in Columns 5 and 6.
(e) Determine the power regression coefficients, defined in Equation 4.11, in Columns 7
and 8 for each duration and the next duration. For example, the coefficients 0.175
and negative 0.965 for duration = 0.083 hours, is applicable for all durations from
0.083 to 0.167 hours. The LINEST function in MS Excel is suitable for this task, which
uses the least square statistical procedure to calculate the regression coefficients
that best fits the data.

Iincrement = a ∙ (Dintermediate )b [ 4.11 ]

where:
Iincrement = Dimensionless incremental intensity,
a ,b = Power regression coefficients, and
Dintermediate = Intermediate storm duration. (hour).

Table 4.2: Example regression coefficients for incremental intensities

1 2 3 4 5 6 7 8
Duration Design Rainfall log log
Inc.Int. a b
(hour) Rainfall Ratio (hour) (Inc.Int.)
0.083 12.1 0.161 1.928 -1.079 0.285 0.175 -0.965
0.167 18.3 0.243 0.988 -0.778 -0.005 0.177 -0.961
0.25 22.5 0.299 0.669 -0.602 -0.174 0.184 -0.933
0.5 29.1 0.386 0.351 -0.301 -0.455 0.175 -1.000
0.75 33.5 0.445 0.234 -0.125 -0.631 0.175 -1.000
1 36.8 0.489 0.175 0.000 -0.756 0.175 -0.890
1.5 41.4 0.550 0.122 0.176 -0.913 0.187 -1.051
2 44.8 0.595 0.090 0.301 -1.044 0.171 -0.918
4 52 0.691 0.048 0.602 -1.320 0.191 -1.000
6 56.8 0.754 0.032 0.778 -1.497 0.191 -1.000
8 60.4 0.802 0.024 0.903 -1.622 0.131 -0.817
10 63.4 0.842 0.020 1.000 -1.701 0.199 -1.000
12 65.9 0.875 0.017 1.079 -1.780 0.178 -0.954
16 69.7 0.926 0.013 1.204 -1.899 0.238 -1.059
20 72.7 0.965 0.010 1.301 -2.002 0.105 -0.785
24 75.3 1.000 0.009 1.380 -2.064
Inc.Int. = Non-dimensionalised Incremental Intensity
_______________________________________________________________________
56
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Following the determination of the power regression coefficients, the design rainfall ratios
are determined in 5-min intervals, for all durations from 5-min to 24-hour. The construction
of the DC is initiated by positioning the peak ratio, namely the 5-min ratio, at the 12-hour
mark. The difference between the 15-min and 5-min ratios is then divided equally before
and after the peak ratio so that the total is equal to the 15-min ratio. The difference between
the 25-min and 15-min ratios is again divided equally on either side of the 15-min ratios so
that the total is equal to the 25-min ratio, and so on until the 24-hour ratio is distributed. The
distribution of the first 55 minutes design rainfall ratios is demonstrated in Figure 4.11.
Finally, the DC is created by accumulating each 5-min ratio from the start to the end of the
24-hour period. The distribution described in this section was subsequently applied to the
construction of the DC5, DC10, and DC20 curves, as well as constructing the SCS-SA
curves in 5-min intervals.

Figure 4.11: The general rainfall distribution process

_______________________________________________________________________
57
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
4.4.2 Development of a DC for durations < 24-hours

The procedure used to develop a DC for a synthetic design storm with durations < 24-hours
was adopted from the National Engineering Handbook (NRCS, 2019). As an example, the
extraction of a 6-hour rainfall distribution using the SCS-SA Type 2 curve is described below.
Although a DC at 5-min intervals are more appropriate, this example uses intervals at 30-
min to shorten the example and still demonstrate the concept (NRCS, 2019). The
calculations to extract a 6-hour event from the 24-hour SCS-SA Type 2 curve is shown
schematically in Figure 4.12, summarised in Table 4.3, and the procedural steps are
summarised as follows:

(a) Because a 24-hour DC is centred around 12-hour, a 6-hour DC will, therefore,

start at 9-hour and end at 15-hour.
(b) The new cumulative rainfall ratio at 9-hour is zero. The unadjusted ratio at 9.5-
hour is equal to the difference between 9.5-hour and 9-hour, which is 0.0169.
(c) The unadjusted ratio at 10-hour is equal to the sum of the unadjusted 9.5-hour
ratio, and the difference between 10-hour and 9.5-hour, which is 0.0360.
(d) The remainder of intervals, up to 15-hour is calculated in the same way, which
has an unadjusted ratio of 0.6980.
(e) Lastly, the adjusted ratios are calculated by dividing each unadjusted ratio with
the unadjusted ratio of 15-hour of 0.6980, which results in the 6-hour DC, as
depicted in Figure 4.12.

Figure 4.12: Typical adjusted six-hour DC extracted from a 24-hour DC

_______________________________________________________________________
58
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Table 4.3: Six-hour rainfall distribution extracted from a 24-hour rainfall distribution
Unadjusted 6-Hour 6-Hour distribution
Time lapsed SCS-SA
cumulative 6-hour distribution time cumulative rainfall
(Hour) Type 2
rainfall ratio (hour) ratio
0 0.0000
0.5 0.0054
1 0.0110
1.5 0.0168
2 0.0230
2.5 0.0295
3 0.0360
3.5 0.0424
4 0.0490
4.5 0.0562
5 0.0640
5.5 0.0722
6 0.0810
6.5 0.0907
7 0.1010
7.5 0.1115
8 0.1230
8.5 0.1361
9 0.1510 0.0000 0 0.0000
9.5 0.1679 0.0169 0.5 0.0242
10 0.1870 0.0360 1 0.0516
10.5 0.2100 0.0590 1.5 0.0845
11 0.2420 0.0910 2 0.1304
11.5 0.2920 0.1410 2.5 0.2020
12 0.5000 0.3490 3 0.5000
12.5 0.7080 0.5570 3.5 0.7980
13 0.7580 0.6070 4 0.8696
13.5 0.7900 0.6390 4.5 0.9155
14 0.8130 0.6620 5 0.9484
14.5 0.8321 0.6811 5.5 0.9758
15 0.8490 0.6980 6 1.0000
15.5 0.8639
16 0.8770
16.5 0.8884
17 0.8990
17.5 0.9094
18 0.9190
18.5 0.9272
19 0.9350
19.5 0.9432
20 0.9510
20.5 0.9577
21 0.9640
21.5 0.9705
22 0.9770
22.5 0.9832
23 0.9890
23.5 0.9946
24 1.0000
_______________________________________________________________________
59
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
SCS-SA DESIGN RAINFAL RATIO COMPARISONS

In this section, the D-hour to 24-hour ratios of the design rainfall at each site were calculated
and compared with the ratios of the four SCS-SA curves. This was done to confirm the use
of the Type 3 curve for the Gauteng province depicted in Figure 2.9 as recommended by
Weddepohl (1988). The methodology that was used to achieve this was by comparing the
D-hour to 24-hour ratios of the design rainfall with the SCS-SA curves. Two sources of
design rainfall were used for this analysis. The first being the design rainfall obtained from
the DRESA software and the second is the design rainfall obtained from the at-site PD
analysis, as describe in Section 4.2. The ratios for the SCS-SA curves were calculated using
Equation 2.10 and the coefficients listed in Table 2.1. These ratios are summarised in Table
4.4 and form the baseline ratios against which the two sources of design rainfall were
compared with.

Table 4.4: Sub-daily and sub-hourly ratios for the four SCS-SA curves
Duration Type 1 Type 2 Type 3 Type 4
5m 0.084 0.135 0.174 0.209
10m 0.126 0.204 0.281 0.347
15m 0.155 0.249 0.355 0.444
30m 0.215 0.332 0.487 0.618
45m 0.256 0.384 0.561 0.710
1h 0.289 0.422 0.609 0.768
1.5h 0.341 0.478 0.672 0.835
2h 0.383 0.520 0.713 0.874
4h 0.502 0.629 0.802 0.938
6h 0.588 0.701 0.849 0.962
8h 0.657 0.755 0.881 0.974
10h 0.715 0.800 0.905 0.982
12h 0.767 0.838 0.925 0.987
16h 0.857 0.902 0.956 0.993
20h 0.933 0.955 0.980 0.997
24h 1.000 1.000 1.000 1.000

4.5.1 Comparison with DRESA design rainfall

The ratios for the design rainfall obtained from the DRESA software were calculated by
dividing the rainfall for each duration with the 24-hour rainfall according to Equation 4.12:

Pt,T
RDt,T = [ 4.12 ]
P24,T
where:
RDt,T = D-hour design rainfall ratio for duration, t and recurrence interval, T,
_______________________________________________________________________
60
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Pt,T = Design rainfall (at-site and DRESA) for duration, t and RI, T (mm), and
P24,T = 24-hour Design rainfall for recurrence interval, T (mm).

The position of the design rainfall ratios relative to the ratios of the four curves were linearly
interpolated and were subsequently called Intermediate Curve (IC) types. Following the
calculation of the D-hour design rainfall ratios, the IC for each duration and recurrence
interval was determined according to Equation 4.13. The result of this analysis is
summarised in Figure 4.13 to Figure 4.15. Thereafter, the average for each standard time
step (AIC_t) was determined according to Equation 4.14 and the results are depicted in
Figure 4.16. This indicates that the average intermediate curve type range between 1.96
and 2.18 with an average (AIC_G) of 2.03 using Equation 4.15. The AIC_G value, therefore,
represents the average of all standard time steps of the 22 stations in Gauteng. This
concludes that the average curve type applicable to the Gauteng province, using the design
rainfall from the DRESA software, is approximately equal to the Type 2 curve. However, the
variation in the IC values with regards to the standard time steps, summarised in Figure 4.13
to Figure 4.15, is notable.

RDt,T − RCi,t
ICt,T = + Ci [ 4.13 ]
RCi+1,t − RCi,t
where:
ICt,T = Intermediate curve type for duration t and recurrence interval T,
Ci = SCS-SA curve type i (1, 2, 3 or 4),
RCi,t = SCS-SA ratio for curve type i and duration t,
RCi+1,t = SCS-SA ratio for curve type i+1 and duration t,
RDt,T = D-hour design rainfall ratio for duration t and recurrence interval T,

NS NT
1 1
AIC_t = ∙ ∙ � � ICT,S [ 4.14 ]
NS NT
T=1 S=1
where:
AIC_t = Average intermediate curve type for duration t,
ICT = Intermediate curve type for recurrence interval T and station S,
NS = Number of stations (22),
NT = Number of recurrence intervals (3),

Nt
1
AIC_G = ∙ � AIC_t i [ 4.15 ]
Nt
i=1
where:
AIC_G = Average intermediate curve type for Gauteng,
ICTT = Intermediate curve type for recurrence interval T and station S,
Nt = Number of standard time steps (16),

_______________________________________________________________________
61
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.13: D-hour to 24-hour ratio comparison with SCS-SA curves and DRESA
software design rainfall (1 of 3)
_______________________________________________________________________
62
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.14: D-hour to 24-hour ratio comparison with SCS-SA curves and DRESA
software design rainfall (2 of 3)
_______________________________________________________________________
63
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.15: D-hour to 24-hour ratio comparison with SCS-SA curves and DRESA
software design rainfall (3 of 3)

_______________________________________________________________________
64
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.16: Average intermediate curve type for 22 stations computed using
DRESA design rainfall

4.5.2 Comparison with at-site design rainfall

The same comparison was done using the design rainfall obtained from the PD analysis
described in Section 4.2. The average for each standard time step (AIC_t) was again
determined according to Equation 4.14 and the results are depicted in Figure 4.17 to Figure
4.19. This shows that the intermediate curve type (IC) varies between a minimum of 1.67
and a maximum of 2.21, with an average (AIC_T) of 1.97 using Equation 4.15. Therefore, it
can be concluded that the average curve type applicable to the Gauteng province, using the
at-site design rainfall, is approximately equal to the Type 2 curve. However, the variation in
the IC values with regards to the standard time steps, summarised in Figure 4.17 to Figure
4.19, is much more noticeable compared to the results using the DRESA design rainfall.
Therefore, a recommended solution is presented in the following section.

_______________________________________________________________________
65
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.17: D-hour to 24-hour ratio comparison for SCS-SA curves and at-site
design rainfall (1 of 3)
_______________________________________________________________________
66
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.18: D-hour to 24-hour ratio comparison for SCS-SA curves and at-site
design rainfall (2 of 3)
_______________________________________________________________________
67
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.19: D-hour to 24-hour ratio comparison for SCS-SA curves and at-site
design rainfall (3 of 3)

_______________________________________________________________________
68
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.20: Average intermediate curve type for 22 stations using at-site design
rainfall

4.5.3 Recommended intermediate curve types for Gauteng

The variation in intermediate curve types with duration and between sites in Gauteng is
significant as it can lead to an underestimation or overestimation of the peak discharge. For
example, if a stormwater network near O.R Tambo, with a reaction time of approximately
30-min, is simulated in a dynamic rainfall-runoff simulation model, using the SCS-SA Type
2 curve to generate a 1:20 year synthetic design storm, then the peak discharge will be
underestimated. This is because the intermediate curve type for the 30-min time step at O.R
Tambo is 2.64, as depicted in Figure 4.17. Therefore, the Type 3 curve would be more
appropriate of the fours SCS-SA distributions, or a linear interpolated curve between the
Type 2 and 3 curves that represents the intermediate curve of 2.64. This means that an
intermediate DC could be determined by either following the rainfall distribution process
described in Section 4.4 or by linear interpolation between the cumulative rainfall ratios,
using Equation 4.16 below. An example of the final IC type 2.64 is depicted in Figure 4.21.

IC − Ci
ICTt = ∙ RCi,t [ 4.16 ]
Ci+1 − Ci
where:
ICTt = Intermediate cumulative rainfall ratio for time step t,
Ci = SCS-SA curve type i (1, 2, 3 or 4),
Ci+1 = SCS-SA curve type plus 1 (1, 2, 3 or 4), and
RCi = Cumulative rainfall ratio of SCS-SA curve i and time step t.

_______________________________________________________________________
69
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.21: Example of an intermediate SCS-SA curve

To simplify the variation of the at-site intermediate curve types across Gauteng, the
maximum of the 5-min to 30-min, 1:5 to 1:20 year recurrence intervals at each site were
determined, as depicted in Figure 4.22. This means that, for example, the intermediate curve
type of 2.64 for O.R. Tambo would then be applicable for all duration from 5-min to 30-min
and recurrence intervals from 1:5 to 1:20 year. Although the use of the maximum value will
result in an overestimation for certain durations and return period, e.g. for the 1:20 year, 5-
min intermediate curve type of 1.96, the use of the maximum value will result in a
conservative (i.e. most intense) rainfall distribution. However, interpolating the 2.64
intermediate curve type is an improvement from using the recommended Type 3 curve for
Gauteng.

Figure 4.22: Maximum intermediate curve type for the 5 to 30-min duration range
_______________________________________________________________________
70
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
To envisage the variation of intermediate curve types, the maximums from Figure 4.22 was
interpolated with the Inverse Distance Weighting (IDW) method. Moeletsi et al. (2016)
evaluated the IDW method for patching daily rainfall over the Free State Province, South
Africa, and found the optimal power parameter as both 2.0 and 2.5. Although the application
of the IDW in this study is somewhat different, a power parameter of 2.5 was used, and the
IC types across Gauteng interpolated as depicted in Figure 4.23. The appropriate power
parameter for this application needs to be determined in a future study. Smaller duration
ranges for different recurrence intervals could also be considered. However, any future
refinements of the intermediate curve types must be preceded by the recommendations
made in Section 4.2.

Figure 4.23 Maximum interpolated intermediate curve type for the 5-min to 30-min
duration range using the IDW method

SCS CURVE COMPARISON

The use of the SCS curves (Brooker, 2021, personal communication, 22 February) is still
preferred by some engineers. Hence, the SCS curves were also investigated. A
comprehensive comparison between the D-hour to 24-hour ratios of the design rainfall and
the ratios of the four SCS-SA curves were conducted in Section 4.5. Therefore, the SCS
curves were compared to the SCS-SA curves rather than the at-site or DRESA estimated
design rainfall.

_______________________________________________________________________
71
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The development of the SCS curves is, however, not the same as the SCS-SA curves. The
approach used in the development of the SCS-SA curves follows the rainfall distribution
procedure described in Section 4.4, whereas the SCS curves were based on the peak 30-
min design rainfall for specific regions in the USA, as described in Section 2.5.1. The SCS
curves for 6-min intervals were downloaded from NRCS’s website (USDA, 2020) and the
maximum ratios were extracted from the data by using a moving window period. For
standard time steps which are not multiples of six minutes, the ratios were determined using
the incremental intensity concept as described in Section 4.4. The final ratios for the
standard time steps of the four SCS curves are summarised in Table 4.5. These ratios were
then compared with the ratios of the SCS-SA curves depicted in Figure 4.24. This indicates
that, for example, the SCS Type II is approximately equal to the SCS-SA Type 2. However,
the 30-min IC type is 2.31 as depicted in Figure 4.25, which is the maximum ratio that could
be achieved with the SCS curves. The ratio of the Type II curve decreases towards the 5-
min as well as towards the 20-hour ratio. The intermediate SCS-SA curve types that are
associated with the SCS Type III are all less than 2.00 for all durations less than 6-hours.
Therefore, Type III as well as the Type 1 and Type 1A, are not recommended. The use of
the Type II curve is, however, very limited and it must be used with caution.

Table 4.5: D-hour to 24-hour ratios for the four SCS curves
Duration Type I Type Ia Type II Type III
5m 0.061 0.020 0.112 0.070
10m 0.113 0.040 0.207 0.140
15m 0.150 0.060 0.274 0.197
30m 0.213 0.115 0.380 0.287
45m 0.252 0.147 0.423 0.359
1h 0.281 0.171 0.454 0.404
1.5h 0.330 0.215 0.501 0.458
2h 0.370 0.252 0.538 0.500
4h 0.492 0.371 0.639 0.622
6h 0.578 0.468 0.707 0.709
8h 0.647 0.548 0.760 0.772
10h 0.707 0.621 0.803 0.819
12h 0.761 0.687 0.841 0.856
16h 0.857 0.811 0.904 0.914
20h 0.935 0.917 0.955 0.961
24h 1.000 1.000 1.000 1.000

_______________________________________________________________________
72
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.24: D-hour to 24-hour ratios of SCS curves compared to SCS-SA curves

Figure 4.25: IC types associated with the SCS curves

STANDARDISED MASS CURVES

Standardised mass curves, or more commonly known as Huff curves, were developed by
non-dimensionalising all significant storm events with respect to the total duration and total
rainfall depth. The curves that were developed for O.R. Tambo, using the significant
independent storm events that were identified in Section 3.5, are depicted as examples in
Figure 4.26. The 10, 50 and 90% percentiles were determined at 0.05 dimensionless time
intervals along the horizontal axis. Tenth-degree polynomials were fitted through the
ordinates and coefficients in the form of Equation 4.17 were determined for each curve.

_______________________________________________________________________
73
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
P = a ∙ T10 + b ∙ T 9 + c ∙ T 8 + d ∙ T 7 + e ∙ T 6 + f ∙ T 5 + g ∙ T 4 + h ∙ T 3 + i ∙ T 2
[ 4.17 ]
+j∙T
where:
P = fraction of total rainfall,
a−j = regression coefficients, and
T = fraction of total duration.

The D-hour to 24-hour ratios of the first quartile, 90% standardised mass curve was then
compared with the ratios of the four SCS-SA curves as depicted in Figure 4.27. Only the
first quartile, 90% curve was considered because of its intensities being the highest. As
demonstrated in Section 4.5, the ratios for the O.R Tambo station varies between the SCS-
SA Types 2 and 3. It is seen in Figure 4.27 that the ratios of all standard time steps up to 1-
hour are less than 0.1 which is an underestimation of the real ratios. This is, however,
concerning because the peak discharge from catchments with rapid response times will be
underestimated. The effects of different synthetic design storms applied to an EPA SWMM
model are demonstrated in Chapter 5.

Figure 4.26: Standardised mass curves (Huff curves) for O.R. Tambo

_______________________________________________________________________
74
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 4.27: Standard time step ratios for the 1st quartile standardised mass curves
(90%) for the O.R Tambo station relative to the standard SCS-SA curves

CHAPTER SUMMARY

This chapter has provided the basis to generate synthetic design storms using the methods
that are considered for this study, and to assess the performance of the methods. For
example, the primary storm parameters consisting of the storm advancement coefficient and
the dimensionless time to peak were determined to be 0.38 and 0.20, respectively. It can be
seen that the two coefficients, although similar by definition, are not similar. This is because
of the difference in calculation procedure. Storm events, identified in the previous chapter,
with an MDP of 15-min, were used to determine these parameters. The use of this MDP was
based on the argument that if the MDP exceeds the time of concentration, any rainfall after
the dry period could no longer influence the peak discharge of the previous spell.

The reaction time of the EPA SWMM model considered in the next chapter, which is
representative of the catchment size targeted in this study, was found to have a reaction
time of 15-min, and therefore, the 15-min MDP were selected. The storm shape and average
intensity assessments in the next chapter will, therefore, also be based on the 15-min MDP.
A methodology of determining regression coefficients for the CDS method, using the GRG
non-linear solver was also developed. These coefficients, together with the storm
advancement coefficient, will be used for the CDS method in the next chapter. The simulated
design rainfall intensities, using the regression coefficients, were evaluated by determining
the RE of the actual design rainfall intensities obtained from the DRESA software. From this
analysis it was evident that the results obtained from the methodology, involving the GRG
non-linear solver, was sufficiently accurate to determine the regression coefficients required
by the CDS method.
_______________________________________________________________________
75
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Design rainfalls were determined using the at-site AMS for 16 standard durations for each
station and then compared with the design rainfall obtained from the DRESA software. It
was found that the average RE was within the 90% upper and lower bounds given by the
DRESA software. Short duration design rainfall of between 5- and 30-min was identified to
be substantially lower than the DRESA software and further investigation is recommended
to identify the cause of the error.

The procedure to distribute design rainfall equally around the peak intensity was adapted to
utilize the concept of incremental intensities for the interpolation of design rainfall for
intermediate time steps. This was then subsequently utilised to generate the DC5, DC10
and the DC20 curves from the design rainfall developed from the PD analysis. The
procedure to extract an event of less than 24-hour was also illustrated. This procedure will
be utilised in the next chapter to extract synthetic design storms from the SCS, SCS-SA,
DC5, DC10 and the DC20 curves, for durations equal to the significant events.

The D-hour to 24-hour ratios were also determined and their position relative to the four
standard SCS-SA curves were determined, using the design rainfall of the DRESA software
as well as the at-site design rainfall analysis. The results from both sources of design rainfall
indicated that, on average, Gauteng conforms more closely to the SCS-SA Type 2 curve,
rather than Type 3. However, a significant variation in the ratios for the various durations
were observed because the derived ratios plotted between standard SCS-SA type curves.
This led to the development of intermediate curve types whereby a curve is linearly
interpolated between two standard type curves. The method of interpolating between
standard type curves was also documented.

Standardised mass curves (Huff curves) were developed and based on the limited tests
conducted on these curves, the results indicate an underestimation of the short duration
design rainfall. Therefore, this method was not considered for the single event-based
modelling in the next chapter because it will lead to an underestimation of the peak
discharge.

In the next chapter, the flow results from a single event-based modelling, using the standard
type curves and the intermediate type curves, will be demonstrated.

_______________________________________________________________________
76
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
EVALUATION OF SYNTHETIC DESIGN STORMS

This chapter contains details of the evaluation of synthetic design storm evaluation by
comparison between mass curves and average intensities, as well as simulated peak
discharge and runoff volumes. The evaluation of the synthetic design storms was conducted
by considering two characteristics of an event, namely the shape of the mass curves, and
the average intensities of standard time steps. This was achieved by non-dimensionalising
each significant event in terms of rainfall and generating a synthetic storm event matching
the total duration. Comparison between the simulated peak discharge and runoff volume
generated with different synthetic design storms were also compared with results from
continuous simulation using observed rainfall data as input. The methods used to generate
synthetic design storms includes the following:

(a) the CDS method, using the regression coefficients from Table 4.1, and an average
advancement coefficient of 0.38,
(b) standard curves, including the SCS-SA Type 2 [SA(T2)] and 3 [SA(T3)],
(c) DC’s developed using the methodology described in Section 4.4, and the on-site
design rainfall described in Section 4.3,
(d) the Rectangular (REC) method, and
(e) the Triangular (TRI) method using a dimensionless time to peak of 0.20.

MASS CURVE COMPARISON

The evaluation of the shape of the mass curves was conducted by considering each
significant storm event with an MDP of 15-min. The GOF of the shape of the synthetic storm
event compared to the observed storm event at 5-min intervals were determined using the
Mean Absolute Relative Error (MARE) technique. The average of all significant events was
the final score given to the method used to generate the synthetic storm event. The general
MARE formula for the shape of the mass curve (MARE_S) is expressed in Equation 5.1.

Ns Nd
1 �Si,j − Oi,j �
MARE_S = ��� � [ 5.1 ]
Ns ∙ Nd j=1 Oi,j
i=1

where:
Si,j = synthetic depth for duration i and storm j (mm),
Oi,j = observed depth for duration i and storm j (mm),
Nd = average number of 5-min data points per storm, and
Ns = number of storms.

_______________________________________________________________________
77
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
A typical result of the GOF between an observed individual storm event and the synthetic
design storms is summarised in Figure 5.1. The storm event that occurred on 29 October
1994 at O.R. Tambo is used to illustrate the assessment process followed for all significant
events identified in Section 3.5. For example, this event had a total duration of 50-min and
the total rainfall was 17 mm. Considering the results presented in Figure 5.1, the DC5 best
represents the shape of the observed mass curve with a MARE_S of 0.311. The MARE_S
of the DC10, and DC20 curves performed slightly worse with values of 0.326 and 0.350,
respectively. The SA(T2) and SA(T3) also showed good performance with values of 0.366
and 0.337, respectively. The CDS method performed better than the REC and TRI, with
values of 0.431, 0.535, 0.494, respectively. Therefore, considering this single event and the
O.R Tambo station, the DC5 presented the best fit of all the methods, and the REC the
worst.

To determine the overall performance of the methods, the average and median MARE_S
values were determined for all the significant events from the five best stations in Gauteng,
namely O.R Tambo, Irene, Vereeniging, Johannesburg Botanical Gardens, and Unisa. The
significant events from the five stations were pooled together before the analysis was
conducted. The result of this analysis is depicted in Figure 5.2. It is seen that the TRI best
represents observed events with the lowest MARE_S values, followed by the REC. The
results of both TRI and REC outperformed CDS, DC5, DC10, DC20, SA(T2) and SA(T3). It
is important to note that this means the methods that use a single point on the IDF curve
according to Figure 2.17, performed better than the methods that use the entire IDF curve.
This is attributed to the position of the peak intensity relative to the total duration. For DC5,
DC10, DC20, SA(T2) and SA(T3) the peak intensity is in the middle of 24-hours, whereas
with CDS it is slightly earlier. Therefore, the effect of adjusting the position of the peak, was
investigated. This is described in the next section.

_______________________________________________________________________
78
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 5.1: Typical GOF between the shape of an observed individual storm event at
O.R Tambo on 29 October 1994, and a synthetic storm event

_______________________________________________________________________
79
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 5.2: MARE_S between observed storm events and synthetic design storms at
the five best stations in Gauteng

ADJUSTMENT OF THE PEAK INTENSITY’S POSITION

This chapter describes the development of a procedure to adjust the position of the SCS-
SA curves. The updated results of the mass curve comparison, described in Section 5.1,
using the modified curves are also presented below.

As described in Section 2.5.2 of the literature review, the SCS-SA curves were developed
with the peak intensity at the centre of 24-hours and the design rainfall of increasing
durations divided equally on either side of the peak intensity. Conversely, with the CDS
method the peak intensity can be positioned anywhere between the start and end of the
event, but with the average intensities of all durations still embedded in the event as depicted
in Figure 2.3. Therefore, the same methodology used to determine regression coefficients
described in Secton 4.3 was used to determine the coefficients of the SCS-SA curves. This
means that the SCS-SA curves were recreated using the CDS method, but with the peak
earlier during the event. The optimised regression coefficients of the SCS-SA curves are
summarised in Table 5.1.

Table 5.1: CDS Regression coefficients for the SCS-SA curves

SCS-SA curves
Description
Type 2 Type 3
a 10.079 27.517
b 6.470 11.686
c 0.755 0.891

_______________________________________________________________________
80
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The CDS method with IDF regression coefficients described in Section 4.3, were also re-
evaluated, but with an optimized advancement coefficient. The updated results for the mass
curve comparison are depicted in Figure 5.3. It is seen that the results for the CDS, SA(T2)
and SA(T3) improved significantly. An advancement coefficient of 0.01 was found to give
the best results with improved average GOF values. The same optimization procedure was,
however, not attempted for the DC5, DC10 and DC20 curves. This will entail the smoothing
of the DDF curves described in Section 4.2, which falls outside the scope of this study.
Therefore, only improved results for the CDS, SA(T2) and SA(T3) are depicted in Figure
5.3.

Figure 5.3: Synthetic design storms modified with peak earlier during an event

It can be concluded that the performance of the methods that use the entire IDF curve can
be altered and improved by modifying the position of the peak intensity relative to the total
duration of an event. However, any value of MARE_S above zero implies that the shape of
a synthetic design storm does represent observed hyetographs. Another factor that
contributes to an MARE_S value above zero is the variation of recurrence interval within
observed storm events. Conversely, synthetic design storms assume that intensities of
increasing durations, embedded in an event, have the same recurrence interval. This can
lead to an overcompensation by reducing the advancement coefficient to a value beyond
the true optimum point. This phenomenon is explained in detail in Section 5.3 which
describes the assessment of the average intensities between observed storm events and
synthetic design storms.

_______________________________________________________________________
81
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
AVERAGE INTENSITY COMPARISON

The evaluation of the intensities was conducted by considering the ratios of average
intensities of the standard durations (5, 10, 15, 30, 45-min, etc.) of each significant storm
event. The GOF of the ratios of the synthetic design storms compared to the observed storm
events was determined using the general MARE formula for the average intensity ratios
(MARE_I), expressed in Equation 5.2.

Ns
1 �Sj − Oj �
MARE_I = �� � [ 5.2 ]
Ns Oj
j=1
where:
Sj = synthetic intensity for storm j (mm/h),
Oj = observed intensity for storm j (mm/h), and
Ns = number of storms.

Similar to the mass curve assessment, the average intensity ratio assessment was
conducted considering the CDS, DC5, DC10, DC20, SA(T2), SA(T3), REC and TRI
methods. The same storm event, depicted in Figure 5.1, is used to illustrate the comparison
of the average intensity ratios, although all significant storm events were used for the
assessment. The results for the example event are depicted in Figure 5.4. In contrast to
MARE_S, the MARE_I result for REC is the worst. This is attributed to the uniform intensity
assumed by the REC method. The result for TRI is slightly better than REC because of the
minor increase in variation of intensities. This can be seen by considering Equation 2.14
which stipulates that the peak intensity of a triangular hyetograph is twice the total rainfall
depth of the event, divided by the total duration. This means the longer the duration of an
event, the lower the peak intensity becomes, which explains the poor MARE_I for the TRI
method. The results for the CDS, DC5, DC10, DC20, SA(T2) and SA(T3), depicted in , are
all very similar with MARE_I ranging from a minimum of 0.218 for SA(T2) to a maximum of
0.307 for DC20.

_______________________________________________________________________
82
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 5.4: Typical GOF between the average intensities of an observed individual
storm event at O.R Tambo on 29 October 1994, and a synthetic storm event

_______________________________________________________________________
83
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Following the analysis of individual storm events, the average and median values of the five
stations with the best data sets were determined and the results for significant events at
these sites are shown in Figure 5.5. The average and median values for the REC and TRI
methods are approximately equal, which indicates an approximately symmetrical frequency
distribution. According to this results the REC method has the worst representation of
average intensities, followed by the TRI method. The median values for the CDS, DC5,
DC10, DC20, SA(T2) and SA(T3) methods are similar and much lower than the REC and
TRI methods. This indicates a much better representation of average intensities. However,
a significant difference between the average and median values for these methods can be
observed. This shows that the results are skewed, which is confirmed by considering the
frequency of occurrences depicted in Figure 5.6.

Figure 5.5: MARE_I between observed storm events and synthetic design storms at
the five best stations in Gauteng

Figure 5.6: MARE_I frequency of occurrence for methods base on entire IDF curve

_______________________________________________________________________
84
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The results depicted in Figure 5.5 show that the CDS, SA(T2) and SA(T3) performed the
best of all the methods considering the MARE_I. However, the modification to the location
of the peak intensity applied in Chapter 5.2 will not improve the MARE_I value. This is
because of the embedment of all intensities with increasing duration in a single event
associated with the methods that utilize the entire IDF curve. This procedure, therefore, does
not alter the intensities of the synthetic design storm but merely the location of the peak.
The variation of the Recurrence Interval (RI) of average intensities during an observed event
was therefore further investigated in Section 5.4 to see the causes of the MARE_I value.

VARIATION IN RI OF AVERAGE INTENSITIES

To determine the variation in RI of the average intensities during a storm event, the MRI
criterion described in Section 3.5 was applied. This was done by determining the RI ratio
(RIR) of the standard durations relative to the Maximum RI (RIM) of each storm event with
the same maximum standard duration. The average RIR of the five best stations in Gauteng,
depicted in Figure 5.7, which was determined using Equation 5.3.

NS NSD
1 RIt,i,s
RIR t,D = �� [ 5.3 ]
Ns ∙ NSD RIMi
s=1 i=1
where:
RIR = recurrence interval ratio of time step t, and total duration D,
NSD = number of storm events with total duration D,
NS = number of stations (5),
RIt,i,s = recurrence interval of time step t, storm event i, and station s (year), and
RIMi = maximum recurrence interval of storm event i, (year).

The results depicted in Figure 5.7 show that, on average, the maximum standard durations
for 6, 8, 12 and 16-hour events have the maximum RI, and the smallest duration the
minimum RI. For example, storm events with a total duration of more than 12 but less than
16 hours, denoted by the 12h curve, have a 12-hour RIR of approximately 1.0, but a 5-min
RIR of approximately 0.1. This finding contradicts the assumption that the RI of all standard
durations have the same RI, which is associated with the methods that utilize the entire IDF
curve.

For storm events with total duration less than 6-hours, the RI of the maximum time steps
starts to decrease which indicates a tendency for the maximum RI to shift to the shorter time
steps. Considering the 1h and 2h curves, the RIR of the 1 and 2-hour time steps are lower
than the 30-min and 1-hour RIR, respectively. This effect is presented as the cause of an
MARE_I above zero, as described in Section 5.3.

_______________________________________________________________________
85
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 5.7: Variation in Recurrence Interval (RI) for the five best stations in Gauteng

EVENT-BASED AND CONTINUOUS SIMULATION

An assessment was conducted in terms of the peak discharge and runoff volume simulated
from a SWMM model of a hypothetical catchment area. This was achieved by comparing
the result of an event-based simulation with the results from a continuous simulation
application of the model. The hydrological urban catchment defined by Gironas et al. (2009)
as depicted in Figure 5.8 was used in this assessment. The catchment has a total area of
11.6 ha which was divided into seven sub-catchment areas. The characteristics of the sub-
catchments are summarised in Table 5.2.

Figure 5.8: The catchment used for the SWMM modelling

_______________________________________________________________________
86
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Table 5.2: Sub-catchment SWMM characteristics
Sub-catchment
Description
S1 S2 S3 S4 S5 S6 S7
Area (ha) 1.829 1.906 1.505 2.740 1.930 0.793 0.935
Width (m) 480.1 500.3 438.8 706.2 506.6 208.1 273.4
Slope (%) 2.5 2.5 4.5 4.0 2.5 2.5 3.8
Imperv (%) 56.8 63.0 39.5 49.9 87.7 95.0 0.0
N-Imperv 0.015 0.015 0.015 0.015 0.015 0.015 0.015
N-Perv 0.240 0.240 0.240 0.240 0.240 0.240 0.240
Dstore-Imperv (mm) 1.5 1.5 1.5 1.5 1.5 1.5 1.5
Dstore-Perv (mm) 7.5 7.5 7.5 7.5 7.5 7.5 7.5
Zero-Imperv (%) 25 25 25 25 25 25 25
Subarea Routing Outlet Outlet Outlet Outlet Outlet Outlet Outlet
Percent Routed (%) 100 100 100 100 100 100 100
Suction head (mm) 110 110 110 110 110 110 110
Conductivity (mm/h) 11 11 11 11 11 11 11
Initial deficit (m/m) Note 0.263 0.263 0.263 0.263 0.263 0.263 0.263
Note: Difference between porosity and field capacity

5.5.1 Peak discharge comparison

The catchment depicted in Figure 5.8 has a single outfall node, namely O1. This node was
selected to conduct the flow comparison between the continuous simulation and the event-
based model. The time step used in the simulations was 5-min. The observed rainfall data
of five stations with the best available data sets were used as input data for five different
continuous simulation models, namely O.R. Tambo, Irene, Vereeniging, Johannesburg
Botanical Gardens and Unisa. Small continuity errors existed following the simulations which
confirm the validity of the analysis results. The AMS of the flow rates, using the five different
rainfall data sets, were extracted, and are summarised in Table 5.3. This was followed by a
PD analysis using the GEV distribution to estimate deign peak discharges. The result of the
PD analysis is summarized in Table 5.4.

_______________________________________________________________________
87
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Table 5.3: Annual peak discharge at node O1 using the observed rainfall data at the
five best stations in Gauteng
PEAK FLOW RATE (m³/s)
YEAR
O.R Tambo Irene Vereeniging Jhb Bot Unisa
1994 1.282 1.251 4.466 1.042 2.770
1995 0.935 2.365 5.208 1.129 2.292
1996 0.954 1.561 1.455 2.102 4.317
1997 1.543 1.567 2.446 1.047 2.057
1998 0.958 1.050 2.264 1.433 0.676
1999 1.126 1.945 1.988 0.960 1.303
2000 1.257 2.368 1.334 1.818 1.132
2001 2.350 1.556 1.263 0.472 0.932
2002 1.496 0.853 0.741 1.010 1.811
2003 1.348 0.626 0.749 0.814 0.977
2004 1.146 1.020 0.790 1.109 1.916
2005 1.070 1.213 0.974 2.448 1.580
2006 1.211 1.213 0.920 0.800 1.476
2007 0.495 1.053 0.677 1.593 1.169
2008 0.872 1.007 1.059 1.015 2.069
2009 1.380 0.849 1.887 1.683 2.207
2010 1.611 1.362 1.182 3.198 1.337
2011 1.946 1.294 1.180 2.184 1.414
2012 1.365 0.958 2.299 1.100 1.431
2013 1.242 1.646 1.295 1.746 1.042
2014 1.291 2.591 1.804 1.241 2.894
2015 2.601 1.506 2.483 1.214 0.933
2016 4.160 1.268 1.588 1.225 1.362
2017 1.089 1.423 1.207 1.995 1.118
2018 0.882 0.765 1.563 1.085 1.247
2019 2.241 1.266 1.434 1.510 0.840
2020 1.245 2.159 1.317 3.097 1.771

Table 5.4: Estimated peak discharge at node O1 using the GEV PD for the five best
stations in Gauteng
RI FLOW RATE (m³/s)
(years) O.R Tambo Irene Vereeniging Jhb Bot Unisa
1:2 1.286 1.328 1.460 1.376 1.472
1:5 1.866 1.797 2.321 1.969 2.131
1:10 2.302 2.096 2.959 2.366 2.606
1:20 2.765 2.371 3.629 2.744 3.090
1:50 3.437 2.711 4.595 3.238 3.760

_______________________________________________________________________
88
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The same catchment depicted in Figure 5.8 was then simulated using a single event-based
approach, using synthetic design storms as input data. Synthetic design storms for the 1:5,
1:10 and 1:20-year RI’s were generated using the SA(T2), SA(T3), CDS, DC5, DC10 and
the DC20 curves. A storm duration of 2-hours was considered for these methods since it will
generally be adequate to exceed the longest time of concentration (Watson, 1981). Longer
durations will lead to higher levels of soil moisture content, and therefore, the 24-hours was
also considered to evaluate the effect on the peak discharge.

The REC and TRI methods were applied by finding the critical duration for the REC method
and determining the minimum duration for the TRI method. This process is explained in
detail in Section 5.5.3. Again, small continuity errors were obtained after each simulation
which confirms the validity of the analysis results. The peak discharge at node O1 from the
event-based simulation was obtained which was then compared to the peak discharge
obtained from the continuous simulation which used the observed rainfall from of each
rainfall station.

The average percentage RE between the results from the continuous simulation for each
station, and the results from the REC and TRI methods, are depicted in Figure 5.9. The
results using the CDS, SA(T2) and SA(T3) methods for the 2-hour and 24-hour single event-
based modelling, are depicted in Figure 5.10 and Figure 5.11, respectively.

Figure 5.9: Average RE of the peak discharge at node O1 between continuous

simulation and single event-based modelling using the REC and TRI methods

_______________________________________________________________________
89
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 5.10: Average RE of the peak discharge at node O1 between continuous

simulation and single event-based modelling using a 2-hour storm event

Figure 5.11: Average RE of the peak discharge at node O1 between continuous

simulation and single event-based modelling using a 24-hour storm event

The average percentage RE depicted in Figure 5.9 to Figure 5.11 represents the average
for the 1:5, 1:10 and 1:20 year results. This indicates that both the REC and TRI methods
underestimated the peak discharge at all five stations. Since the same parameters were
used in both the continuous simulation and single event-based modelling, the
underestimation is attributed to the incorrect initial deficit applicable to single event-based
modelling.

_______________________________________________________________________
90
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
In terms of the results depicted in Figure 5.9 to Figure 5.11, the following are observed:

(a) The difference between SA(T2) and SA(T3) for the 24-hour event is much greater
than the difference for the 2-hour event. This is attributed to the change in the ratio
characteristics which is described in detail in Section 5.5.4.
(b) The RE for the CDS method is much greater for the 24-hour event compared to the
error for the 2-hour event, which is expected because of the consistency provided by
the CDS method in terms of average intensities. In other words, the average
intensities of the 5, 10, 15-min, etc. remains constant, irrespective of the total duration
of the event. This means that if the average intensities remain consistent, the soil has
reached a higher level of soil moisture content during a 24-hour event compared to a
2-hour event.
(c) The results of the DC5, DC10 and the DC20 curves relative to SA(T2) and SA(T3) is
consistent with the SCS-SA design rainfall ratio comparison described in Section 4.5.
This consistency is described in detail in Section 5.5.5.
(d) The cause for the larger overestimation by the CDS, SA(T2), SA(T3), DC5, DC10
and DC20 at the Unisa station relative to the other stations is unknown and requires
further investigation.
(e) The results of the CDS method, using a 2-hour event, is consistent with the results
obtained from the continuous simulation. This can be seen with the small positive RE
at O.R Tambo, Irene, and Vereeniging. The RE at Unisa was approximately 20%
which is less than the overestimation of up to 60% by the SA(T2), SA(T3) and DC
methods.

5.5.2 Runoff volume

An assessment of runoff volume was done by identifying the events that resulted in the
annual peak discharge. The process involved the extraction of a three-day time series, one
day before and one day after the date of the maximum, as depicted in Figure 5.12. The start
and end of the event associated with the maximum discharge were determined graphically,
and the event was then extracted for which the cumulative runoff volume was determined
as depicted in Figure 5.13. This process was repeated for each annual maximum peak
discharge and the associated runoff volume was calculated. The process is illustrated using
the rainfall event of 9 February 2020 for O.R. Tambo in Figure 5.14. It is seen that the
magnitude of the annual maximum discharge and associated runoff volumes is
approximately related. However, the relationship is not directly related which can be seen,
for example, by comparing the results for the years 1994 and 1995. In 1995 the maximum
discharge is lower than the previous year, whereas the runoff volume was higher. Similar
occurrences of this inconsistency are observed in 1997/98, 1998/99, 2003/04, 2009/10.

_______________________________________________________________________
91
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 5.12: Three-day time series containing the event that resulted in the peak
discharge at O.R. Tambo in the year 2020

Figure 5.13: Extracted event that resulted in the peak discharge at O.R. Tambo in the
year 2020

Figure 5.14: Annual peak discharge and associated runoff volume obtained from a
continuous simulation for O.R. Tambo
_______________________________________________________________________
92
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The inconsistency in the runoff volume and peak discharge is observed by ranking the
events in descending order according to the peak discharge as depicted in Figure 5.15. For
example, the runoff volume in the years 2009, 1998 and 1995 should be ranked much
higher. The smoothing of the runoff volume was therefore necessary to improve the
consistancy, to prevent some runoff volumes with lower RI’s to be more than others with
higher RI’s. Therefore, the same recurrence interval of the peak discharge was assigned to
the runoff volume, and a logarithmic regression line for the runoff volume with the RI as the
dependent variable, was determined as depicted in Figure 5.16. The R² of 0.795 indicates
a reasonable fit to the data. The same process was followed for Irene and Johannesburg
Botanical Gardens and R² of 0.524 and 0.072 were obtained. Only these three stations was
used in this analysis, and future analyses could include additional sites.

Figure 5.15: Annual peak discharge and corresponding runoff volume for O.R
Tambo

Figure 5.16: Interpolation of runoff volume for O.R Tambo

_______________________________________________________________________
93
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The runoff volume for the 1:5, 1:10 and 1:20 year recurrence intervals for O.R Tambo was
calculated according to the logarithmic relationship depicted in Figure 5.16. These volumes
were then used for comparison with the runoff volume obtained from the event-based
simulations. The runoff volume was determined at the outfall node O1 depicted in Figure
5.8, using all the synthetic design storms as input data. The 2-hour rather than the 24-hour
event was used for the SA(T2), SA(T3), CDS and DC methods because the total discharge
from a 24-hour event would be much higher. The percentage RE between the runoff
volumes of the single event-based simulations is depicted in Figure 5.17. It is seen that the
peak discharge was underestimated by between 35% and 55% using the REC and TRI as
input data. The peak discharge obtained from the SA(T2), SA(T3), CDS and DC methods
are overestimated by up to 45%. The poor performance could be attributed to the poor R²
between the RI and runoff volume.

Figure 5.17: Percentage RE of the runoff volume between continuous simulation

and single event-based modelling

5.5.3 Critical and minimum durations

The critical and minimum durations required to apply the REC and TRI methods,
respectively, to a dynamic rainfall-runoff simulation model are described in this section. The
critical duration refers to the duration which allows sufficient time for discharge to reach the
outlet point. However, the duration must be as short as possible so that the discharge is not
underestimated. This involves the simulation of multiple synthetic design storms, starting at
5-min and gradually increasing the duration until the discharge has reached a maximum. If
the duration of the synthetic design storm is less than the critical duration, then the discharge
from the remote parts of the catchment has not yet reached the point of discharge. This is
according to Mulvaney (1851), as cited by Dooge (1974), who states that the peak discharge
from the catchment will occur when the discharge from every portion of the catchment
arrives simultaneously at the point of discharge. Conversely, if the duration exceeds the
critical duration, then the intensity of the synthetic design storm will be lower which will result
in a lower peak discharge.
_______________________________________________________________________
94
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The results of applying this concept to the catchment shown schematically in Figure 5.8,
and using the design rainfall estimation for O.R Tambo described in Section 4.2 as input
data, the critical storm duration was found to be 15-min, as shown in Figure 5.18.

Figure 5.18: Critical storm duration for O.R Tambo according to the REC method

The minimum duration applicable to the TRI method was determined by limiting the
maximum intensity of the triangular hyetograph expressed in Equation 2.14 to the maximum
intensity following the PD analysis described in Section 4.2. An example of this limitation is
depicted in Figure 5.19. From the IDF graph, the intensity of the 5-min duration is equal to
125 mm/h. However, the maximum intensity for a 5-min triangular hyetograph computed
using Equation 2.14, is 250 mm/h. Therefore, generating a 5-min triangular hyetograph is
incorrect. Limiting the triangular hyetograph to 30-min as shown in Figure 5.19, will therefore
ensure that the peak intensity is not exceeded. The results for the five best stations, following
this methodology, are depicted in Figure 5.20

Figure 5.19: Example of determining the minimum storm duration for the TRI
method
_______________________________________________________________________
95
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯

Figure 5.20: Result of the minimum storm durations for the five best stations in
Gauteng

5.5.4 SCS-SA ratios for 2-hour events

To create 2-hour events from the standard SCS-SA curves, the development of a DC for a
duration < 24-hours, as described in Chapter 4.4, was applied. This involved determining
the D-hour to 2-hour ratios as depicted in Figure 5.21. Considering the critical storm duration
of 15-min depicted in Figure 5.18, the ratios for the 2-hour, Types 2, 3 and 4 curves at 15-
min duration, as depicted in Figure 5.21, are approximately equal. The 15-min ratios for the
24-hour storm event are, however, very different. This explains the difference between the
peak discharge comparison for the SA(T2) and SA(T3) considering a 24-hour event
compared to a 2-hour event, as described in Section 5.5.

Figure 5.21: Design rainfall ratios in relation to 2-hour rainfall

_______________________________________________________________________
96
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
5.5.5 Consistency of DC

This section contains a comparison between the D-hour to 24-hour ratios, with the results
of the peak discharge comparison, as described in Section 5.5. Using the rainfall distribution
process to determine DC’s as described in Section 4.4, includes the embedment of the entire
IDF curve in a single event. These ratios were found to vary between the standard SCS-SA
curves as described in Section 4.5. By considering the critical storm duration of 15-min
depicted in Figure 5.18, the intermediate curve types for the five best stations were
determined as shown in Figure 5.22. The same method of determining intermediate curve
types as described in Section 4.5.3 was followed to determine the intermediate curve types
for the results of the DC method. As depicted in Figure 5.22, the comparison between the
intermediate curve types following the two methodologies is consistent. This confirms the
validity of the 15-min critical duration for the model depicted in Figure 5.8.

Figure 5.22: Intermediate curve type determined from flow results

CHAPTER SUMMARY

This chapter has provided details of the synthetic design storm evaluations. The shape and
average intensities of synthetic design storms were compared with observed storm events.
It was concluded that the shape and average intensities of the REC method least represents
observed events. It was, however, further demonstrated that the REC method can be used
to determine the response time of a catchment. This was based on the concept of the critical
duration which allows sufficient time for discharge from the entire catchment to reach the
point of discharge.

_______________________________________________________________________
97
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The methods that consider an entire IDF curve, which includes the CDS, SA(T2), SA(T3)
and DC methods, were also shown to be a poor representation of observed events. By
manipulating the position of the peak intensity, it was demonstrated that the GOF has
improved considerably. However, the variation of the RI relative to the average intensities
of the standard time steps, also contributes to the poor GOF. It was also demonstrated that
the maximum RI of 6-hour and longer events is likely to be associated with the maximum
standard duration. It was, however, concluded that the shape of synthetic design storms are
not similar to observed events.

Following the shape and average intensity assessments, the synthetic design storms were
used in a dynamic rainfall-runoff simulation model. A stormwater network in a catchment
was simulated in EPA SWMM. The rainfall data from the five best stations in Gauteng was
used as input data for the model. The AMS of the flows and volumes were determined
following a continuous simulation and a PD analysis was conducted. The results from the
analysis were used as baseline for the comparison with the results of single event-based
simulations. The REC, TRI, CDS, SA(T2), SA(T3), DC5, DC10 and DC20 methods were
used to generate synthetic design storms. The results of the SA(T2) and SA(T3) was
successfully related to the intermediate curve types, and it was demonstrated that the
characteristics of 2-hour events were very different to 24-hour events.

The peak discharge initial deficit associated with the infiltration parameters of the Green-
Ampt model method is questioned, and further research is recommended. The effect of the
location of peak intensity and the total duration of the storm event should also be further
investigated as it was shown that these two parameters can have a significant effect on the
peak discharge.

The next chapter contains a general discussion and conclusions drawn from the findings of
this study and includes areas recommended for future research.

_______________________________________________________________________
98
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
DISCUSSION, CONCLUSIONS AND
RECOMMENDATIONS

This chapter presents a discussion of the results from the study and conclusions that
emanated from this study. The proposed way forward is also described.

OBJECTIVE

The aim of this study was to test the performance of the existing synthetic design storm
generation methods, and to identify the method most suited for conditions in small
catchments in Gauteng, using the 5-min interval rainfall records obtained from the SAWS.
The specific objectives of the study to meet this aim were to: 1) identify and assess the
performance of currently available methods to estimate synthetic design storms used as
input for single-event modelling in the selected pilot study area; 2) to propose an improved
procedure to generate a synthetic design storm applicable to small catchment areas in the
study area; and 3) to disseminate the information to managers, designers and technicians
involved in urban stormwater planning and design.

DISCUSSION

The literature review found that existing synthetic design storms have a strong scientific
basis, with synthetic design storms being classified into three categories: a) methods that
are derived from the Intensity Duration Frequency (IDF) curves; b)standardised mass curves
generated directly from rainfall records, which includes the Huff and NOAA Atlas 14 curves;
and c) the simulation from a stochastic rainfall model, which includes the daily rainfall
disaggregation model for South Africa.

Although the existing synthetic design storms were developed using the best data,
technology, and engineering judgement available at the time, they do present some degree
of insufficiency. For example, the total precipitation volume is systematically underestimated
by the rectangular hyetograph (Arnell, 1982), and it was also realised that it gives a wrong
picture of a hyetograph (Niemczynowicz, 1982). In terms of the triangular hyetograph
method, Veneziano and Villani (1999) have noticed that, although it is quite simple and
intuitive, it does not have a strong conceptual basis and may produce biased flow estimates.
The CDS method, on the other hand, utilises point rainfall data that applies to a particular
site and therefore the variability and unique character of the rainfall patterns will
automatically be embedded in the CDS method. Methods that utilise the entire IDF curve to
generate a synthetic design storm hold much potential since the design rainfall has been
regionalised by Smithers and Schulze (2000). Some of the variability and unique character
of the rainfall pattern will therefore automatically be embedded in these methods.
_______________________________________________________________________
99
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The first aim of this study was to identify and assess the performance of currently available
methods to estimate synthetic design storms used as input for single-event modelling in the
selected pilot study area. In order to achieve this aim, short-duration observed rainfall data
recorded at 35 automatic rainfall stations situated in the Gauteng province, were collected,
and assessed in terms of completeness. The missing data as well as the data period of each
station were used to classify the data quality of each rainfall station. A station was classified
as either, good, average or poor, depending on the selected classification criteria. The
stations with poor data sets were omitted from the study, whereas the good and average
stations are used in subsequent investigations. Five stations have good data sets with each
having a data period of 26 years and less than 5% of missing data during the rainy months.
A further 17 stations were identified with average quality data sets. The criteria used for the
identification of individual events were also described. These criteria consisted of an MDP
ranging from 0 to 120-min, an MRD of 10 mm, and an MRI of 1:2 years. The correlation
between the total storm duration and total rainfall depth related to different MDPs was
consistent with the findings from Ramlall (2020). However, this study is based on an MDP
associated with the reaction time of a small urban catchment. Appropriate storm parameters
were determined which were 0.38 for the advancement coefficient and 0.20 for the
dimensionless time to peak. The determination of regression coefficients for any DDF curve,
and by re-generating the SCS-SA curves, using the GRG nonlinear algorithm to determine
appropriate coefficients, was demonstrated. Design rainfalls were determined using the at-
site AMS for 16 standard durations for each station and then compared with the design
rainfall obtained from the DRESA software. It was found that the average RE was within the
90% upper and lower bounds given by the DRESA software. Short duration design rainfall
of between 5- and 30-min was identified to be substantially lower than the DRESA software
and further investigation is recommended to identify the cause of the error. The procedure
to distribute design rainfall equally around the peak intensity was adapted to utilize the
concept of incremental intensities for the interpolation of design rainfall for intermediate time
steps. This was then subsequently utilised to generate the DC5, DC10 and the DC20 curves
from the design rainfall developed from the PD analysis. The procedure to extract an event
of less than 24-hour was also illustrated. The D-hour to 24-hour ratios were also determined
and their position relative to the four standard SCS-SA curves were determined, using the
design rainfall of the DRESA software as well as the at-site design rainfall analysis. The
results from both sources of design rainfall indicated that, on average, Gauteng conforms
more closely to the SCS-SA Type 2 curve, rather than Type 3. However, a significant
variation in the ratios for the various durations were observed because the derived ratios
plotted between standard SCS-SA type curves. This led to the development of intermediate
curve types whereby a curve is linearly interpolated between two standard type curves. The
method of interpolating between standard type curves was also documented. The design
rainfall ratios of the SCS curves were compared with the SCS-SA curves and concluded
that the application of the SCS curves is very limited and that it must be applied with caution.
A methodology of determining regression coefficients for the CDS method, using the GRG
non-linear solver was also developed. The simulated design rainfall intensities, using the

_______________________________________________________________________
100
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
regression coefficients, were evaluated by determining the RE of the actual design rainfall
intensities obtained from the DRESA software. From this analysis it was evident that the
results obtained from the methodology, involving the GRG non-linear solver, was sufficiently
accurate to determine the regression coefficients required by the CDS method.
Standardised mass curves (Huff curves) were developed and based on the limited tests
conducted on these curves, the results indicate an underestimation of the short duration
design rainfall. Therefore, this method was not considered for the single event-based
modelling in the next chapter because it will lead to an underestimation of the peak
discharge.

The synthetic design storm evaluation was conducted by comparing the synthetic design
storms with the observed rainfall events. Two aspects were evaluated, namely the shape of
the mass curves, as well as the average intensities embedded in each synthetic design
storm. It was concluded that the shape and average intensities of the REC method least
represents observed events. It was, however, further demonstrated that the REC method
can be used to determine the response time of a catchment. This was based on the concept
of the critical duration which allows sufficient time for discharge from the entire catchment
to reach the point of discharge. The methods that consider an entire IDF curve, which
includes the CDS, SA(T2), SA(T3) and DC methods, were also shown to be a poor
representation of observed events. By manipulating the position of the peak intensity, it was
demonstrated that the GOF has improved considerably. However, the variation of the RI
relative to the average intensities of the standard time steps, also contributes to the poor
GOF. It was also demonstrated that the maximum RI of 6-hour and longer events is likely to
be associated with the maximum standard duration. It was, however, concluded that the
shapes of synthetic design storms are not similar to observed events. This analysis provided
sound evidence that synthetic design storms do not exist in nature.

Following the shape and average intensity assessments, the synthetic design storms were
used in a dynamic rainfall-runoff simulation model. A stormwater network in a catchment
was simulated in EPA SWMM. The rainfall data from the five best stations in Gauteng was
used as input data for the model. The AMS of the flows and volumes were determined
following a continuous simulation and a PD analysis was conducted. The results from the
analysis were used as baseline for the comparison with the results of single event-based
simulations. The REC, TRI, CDS, SA(T2), SA(T3), DC5, DC10 and DC20 methods were
used to generate synthetic design storms. The results of the SA(T2) and SA(T3) was
successfully related to the intermediate curve types, and it was demonstrated that the
characteristics of 2-hour events were very different to 24-hour events. Various observations
were made from these results of which the most important was that the suggested values
for the initial deficit associated with the Green-Ampt infiltration parameters did not result in
good simulations. The effect of the location of peak intensity and the total duration of the
storm event should also be further investigated as it was shown that these two parameters
can have a significant effect on the peak discharge.

_______________________________________________________________________
101
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
The CDS method was also shown to provide consistent results compared to continuous
simulation. However, a sensitivity analysis must be conducted to determine the effect of the
advancement coefficient and total storm duration on the peak discharge. This method has
the following advantage:
(a) The IDF coefficients can be accurately determined from the rainfall obtained from the
DRESA software, which can them be used to generate a synthetic design storm.
(b) The location of the peak intensity concerning the total duration of the event can be
adjusted to be anywhere between the start and end of the event.

CONCLUSIONS

It is, in general, concluded that synthetic design storms, applied to a single event-based
model, provides the engineer with the ability to assess the complex hydrological and
hydraulic characteristics of an urban stormwater network. Despite its unrealistic
assumptions, shortcomings, and the criticism the synthetic design storm concept has
received, applying it to a single event-based model has resulted in good peak discharge and
runoff volume estimates. Three methods, to generate synthetic design storms, were
identified that could be applied to a single event-based model. They are the REC, SCS-SA
and CDS methods.

The REC method also provides a means of evaluating the response time of an urban
catchment, but the initial deficit defined as the difference between the porosity and field
capacity, associated with the Green-Ampt infiltration method, did not result in good
simulations. The design rainfall ratio comparisons from both the at-site and DRESA design
rainfall, provided the bases to conclude the inappropriateness of the SCS-SA Type 3 curve
for Gauteng. However, it was also concluded that an interpolation between the standard
type curves is needed for better single event-based simulation results. The methodology
that was used to determine the CDS regression coefficients from the DRESA design rainfall
was sufficient and resulted in good results when applied to a single event-based simulation.

It can, therefore, be concluded that all three project aims were achieved, namely:

(a) The performance of currently available methods to estimate synthetic design storms
was assessed and used as input for single-event modelling in the selected pilot study
area.
(b) An improved procedure to generate a synthetic design storm for the study area was
proposed. An improved procedure to generate a synthetic design storm applicable to
small catchment areas in the study area was proposed through the development of
intermediate SCS-SA curve types. The intermediate curves could be used in
combination with the interpolated map of Gauteng with lower values to the north of
the province and higher values to the south.
_______________________________________________________________________
102
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
(c) The results of this project were disseminated to managers, designers and technicians
involved in urban stormwater planning and design through a WRC workshop
presented in conjunction with the annual National Flood Studies Programme’s
workshop, as well as at the 2022 UP Flood Hydrology course.

RECOMMENDATIONS

Recommendations that emanated from this study are summarised as follows:

(a) The impact the advancement coefficient and total storm duration have on the peak
discharge must be investigated by conducting a sensitivity analysis. The parameters
that could be considered for the analysis includes the size of catchment, slope,
roughness, and soil types.
(b) The impact that missing data has on the design rainfall estimation and by implication
the design rainfall ratios, in the context of this study, should be investigated.
(c) The current DRESA software, be further developed to provide the user with an
opportunity to determine a synthetic design storm for a specific location.
(d) The discrepancy between the dates of the daily and 5-min data, and the annual
maximum daily rainfalls from the daily and the 5-min data.
(e) The appropriateness of the GEV PD for short duration rainfall should be re-confirmed.
(f) An appropriate power coefficient for the IDW interpolation technique should be
investigated.
(g) The suggested values for the initial deficit associated with the Green-Ampt infiltration
parameters should be investigated.
(h) The relevance of the CDS, SCS-SA and REC methods for generating synthetic
design storms applicable to single-event based modelling of small urban catchments
must be expanded on a national scale, with the possibility of following a regional
approach and ensemble modelling investigated.
.

_______________________________________________________________________
103
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
REFERENCES
Adamson, P. T. 1981. Southern African storm rainfall. Technical Report No TR 102,
Department of Environment Affairs, Pretoria, RSA. pp 19.

Arnell, V. 1982. Estimating runoff volumes from urban areas. Journal of the American Water
Resource Association, Vol 18, No 3, June, pp 383-387.

Asquith, W. H. Bumgarner, J. R. and Fahlquist, L. S. 2007. A triangular model of

dimensionless runoff producing rainfall hyetographs in Texas. Journal of the American
Water Resources Association, Vol 39, No 4, June, pp 911-921.

Bonnin, M. Martin, D. Lin, B. Parzybok, T. Yekta, M. and Riley, D. 2011. NOAA Atlas 14
Volume 1 Version 5, Precipitation-Frequency Atlas for the United States. NOAA,
National Weather Service, Silver Spring, MD, USA.

Bonta, J. V. 1997. Proposed use of Huff Curves for hyetograph characterization. pp 111-
124. In: C.W. Richardson et al. (ed.) Proceedings of the Workshop on Climate and
Weather Research. Denver, Colorado. July 17-19, 1995. USDA-Agricultural Research
Services, 1996-03, pp 223.

Bonta, J. V. 2004. Development and utility of Huff curves for disaggregating precipitation
amounts. Applied Engineering in Agriculture, Vol 20, No 5, pp 641-653.

Bonta, J. V. and Shahalam, A. 2003. Cumulative storm rainfall distributions: comparison of

Huff curves.Journal of Hydrology, Vol 42, No 1, June, pp 65-74.

Boughton, W. 2000. A model for disaggregating daily to hourly rainfall for design flood
estimation. Cooperative Research Centre for Catchment Hydrology, Monash
University, Clayton, Victoria, Australia. Report 00/15, 36 pp.

Boussinesq, J. and Flamant, A. 1886. Notice sur la vie et les travaux de Barré de Saint-
Venant. Annales des ponts et chaussées, Vol 12, pp 557-595.

Brooker, C. 2021. Personal communication, 22 February 2021, Mr Chris Brooker, CBA

Specialist Engineering, Gauteng, South Africa, 2055.

Cordery, I. and Pilgrim, D. 1993. Handbook of hydrology. McGraw-Hill, New York, USA.

Ellouze, E. Azri, C. and Abida, H. 2009. Spatial variability of monthly and annual rainfall data
over Southern Tunisia. Atmospheric Research, Vol 93, No 4, pp 832-839.

El-Sayed, E. A. H. 2017. Development of synthetic rainfall distribution curves for Sinai area.
Ain Shams Engineering Journal, Vol 9, No 4, December, pp 1949-1957.

_______________________________________________________________________
104
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Gomez, M. and Sanchez, X. 2014. Rational method application (Lecture notes).5 December
2020, https://ocw.camins.upc.edu/ocw/home.htm?execution=e1s6.

Green, W. H. and Ampt, G.A. 1911. Studies on Soil Physics. The Journal of Agricultural
Science, Vol 4, No 1, pp 1-24.

Horton, R. E. 1933. The role of infiltration in the hydrologic cycle. Transactions American
Geophysical Union, Vol 14, No 1, June, pp 446-460.

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

Huff, F. A. 1990. Time distributions of heavy rainstorms in Illinois. Department of energy and
natural resources, Champaign.

Izzard, C. F. 1946. Hydraulics of runoff from developed surfaces. Proceedings, highway

research board, Vol 26, pp 129-146.

Keifer, C. J. and Chu, H. H. 1957. Synthetic storm pattern for drainage design. ASCE Journal
of the hydraulics division, Vol 83, No 4, pp 1332.1-1332.25.

Knoesen, D.M. 2005. The development and assessment of techniques for daily rainfall
disaggregation in South Africa. University of KwaZulu-Natal, School of Bioresources
Engineering and Environmental Hydrology, Pietermaritzburg.

Males, R. Braune, M. van Bladeren, D. and Mahlangu, D.J. 2004. Regional hydrological
modelling of City of Tshwane municipality using Visual SWMM. Journal of Water
Managment Modelling, Vol 12, No R220-30, pp 627-649.

Malik, U. and James, W. 2007. Reliability of design storms used to size urban stormwater
system elements. Journal of water management modelling, pp 309-326.

Munro, G. 2021. Personal communication, 23 August 2021, Mr Gerhard Munro, Oracle Java
Certified SE Programmer, Gauteng, South Africa, South Africa, 1709.

Niemczynowicz, J. 1982. Areal intensity duration frequency curves for short term rainfall
events in Lund. Hydrology Research, Sweden.

NRCS. 2019. National Engineering Handbook. Part 630 Hydrology. Chapter 4: Storm rainfall
depth and distribution. Department of Agriculture, Washington.

NRCS, 2020. United States Department of Agriculture, September 2021,

https://www.nrcs.usda.gov/wps/portal/nrcs/detailfull/national/water/manage/hydrology
/?cid=stelprdb1044959.

_______________________________________________________________________
105
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Ramlall, R. 2020. Assessing the performance of techniques for disaggregating daily rainfall
for design flood estimation in South Africa. University of KwaZulu-Natal, School of
Agricultural Earth and Environmental Sciences, Pietermaritzburg.

Perica, S. Pavloci, S. St. Laurent, M. Trypaluk, C. Unruh, D. and Wilhite, O. 2018. NOAA
Atlas 14 Volume 11 Version 2, Precipitation-Frequency Atlas for the United States,
Texas. National Weather Service, Silver Spring, MD.

Restrepo, P. J. and Eagleson, P. S. 1982. Identification of independent rainstorms.Journal

of Hydrology, Vol 55, pp 303-319.

Schmidt, E.J. and Schulze, R. E. 1987. Design stormflow and peak discharge rates for small
catchments in Southern Africa. Water Research Commission. WRC Report No. TT
31/87. Pretoria. pp 65-70.

Schulze, R. E. 1984. Hydrological models for application to small rural catchments in

southern Africa: Refinements and development. University of Natal, Department of
Agricultural Engineering, Pietermaritzburg.

Schulze, R. E. 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,
Department of Agricultural Engineering, Pietermaritzburg.

SCS. 1973. A method for estimating volume and rate of runoff in small watersheds. US
Department of Agriculture, Soil Conservation Service, Washington, D.C.

Sherman, C. 1931. Frequency and intensity of excessive rainfall at Boston, Massachusetts.

American Society of Civil Engineers, Vol 95, pp 951-960.

Silveira, A. L. L. 2016. Cumulative equations for continuous time Chicago hyetograph

method. Brazilian Journal of Water Resources, Vol 21, No 3, pp 646-651.

Smith, A. A. 2004. MIDUSS Version 2, Reference Manual. Alan A. Smith Inc., Dundas,
Ontario, Canada.

Smithers, J. C. 2012. Methods for design flood estimation in South Africa. Water Research
Commission, Vol 38, No 4, July, pp 633-646.

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.

Smithers, J.C. and Schulze, R.E. 2002. The estimation of design rainfall for Tshwane –
Report to SRK Consulting. School of Bioresources Engineering and Environmental
Hydrology. Pietermaritzburg. ACRUcons Report 38.

_______________________________________________________________________
106
 Assessment and development of synthetic design storms: Gauteng pilot study
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Smithers, J. C. and Schulze, R. E. 2003. Design rainfall and flood estimation in South Africa.
Water Research Commission.WRC Report No 1060/01/03. Pretoria.

Veneziano, D. and Villani, P. 1999. Best linear unbiased design hyetograph. Water
Resources Research, Vol 35, No 9, September, 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,
Department of Agricultural Engineering, 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.

U.S. Weather Bureau. 1961. Generalised Estimates of Probable Maximum Precipitation and
Rainfall-Frequency Data for Puerto Rico and Virgin Islands. U.S. Department, Soil
Conservation Service, Technical paper, No 42.

Yen, B. C. and Chow, V. T. 1980. Design hyetographs for small drainage structures. ASCA
Journal of the hydraulics division, Vol 106, No HY6, pp 1055-76.

Yu, Y. S. and McNown, J. S. 1964. Runoff from impervious surfaces. Journal of hydraulic
research, Vol 2, No 1, pp 3-23.

_______________________________________________________________________
107