The Estimation of Design Rainfall for

Durban Unicity

Draft Report to

eThekwini Municipality

ACRUcons Report

November 2002

J C Smithers (Pr Eng, PhD)

School of Bioresources Engineering and Environmental Hydrology
University of Natal
Pietermaritzburg
South Africa

Disclaimer

While every reasonable effort has been made by the authors to obtain objective and
realistic results in this study, neither the authors, the School of Bioresources Engineering
and Environmental Hydrology, nor the University of Natal, nor any of their employees,
make any warranty, express or implied, or assume any legal liability or responsibility for
the accuracy, completeness or usefulness of any information, product or process
disclosed by this report.
 TABLE OF CONTENTS

Page

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

1 PROJECT OBJECTIVES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 APPROACHES TO DESIGN RAINFALL ESTIMATION . . . . . . . . . . . . . . . . . . . . . 1

3 THE RLMA&SI METHODOLOGY FOR DESIGN RAINFALL ESTIMATION . . . . 4

4 STUDY AREA AND AVAILABILITY OF OBSERVED RAINFALL . . . . . . . . . . . . 8

5 ESTIMATION OF THE MEAN OF THE AMS FOR DURATIONS < 24 HOURS . 10

6 EVALUATION OF RLMA&SI PROCEDURES IN STUDY AREA . . . . . . . . . . . . . 12

7 RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

8 DISCUSSION AND CONCLUSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

9 REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

APPENDIX A

Examples of the Occurrence of Errors in Rainfall Data Files . . . . . . . . . . . . . . . . . . . 21

APPENDIX B

Description of Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

ii
 LIST OF FIGURES
Page

Figure 1 Distribution of 15 clusters of relatively homogeneous extreme short duration (#

24 h) rainfall in South Africa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Figure 2 Distribution of 78 clusters of relatively homogeneous extreme daily rainfall in
South Africa (Smithers and Schulze, 2000b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Figure 3 Location of recording and daily raingauges within and surrounding the study
area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Figure 4 D:24 h means of AMS for Cluster 8 derived by Smithers and Schulze (2002) . . 10
Figure 5 D:24 h means of AMS for Cluster 8 revised in this study . . . . . . . . . . . . . . . . . . 11
Figure 6 Comparison of design rainfall estimated from observed data and by the
RLMA&SI procedures at the Chats1 site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Figure 7 Comparison of design rainfall estimated from observed data and by the
RLMA&SI procedures at the Crabtree site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Figure 8 Comparison of design rainfall estimated from observed data and by the
RLMA&SI procedures at the Dunkeld site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Figure 9 Examples of design rainfall and 90% confidence limits estimated using the
RLMA&SI procedures at the Chats 1 site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Figure 10 Examples of 15 minute design rainfall in the study area . . . . . . . . . . . . . . . . . . . 16
Figure 11 Examples of 1 h design rainfall in the study area . . . . . . . . . . . . . . . . . . . . . . . . . 17
Figure 12 Examples of 24 h design rainfall in the study area . . . . . . . . . . . . . . . . . . . . . . . . 18

iii
 LIST OF TABLES

Page

Table 1 Stations and length of record of rainfall data supplied to the study . . . . . . . . . . . . 9
Table 2 Data excluded from the analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

iv
1 PROJECT OBJECTIVES

The objectives of this study are to estimate design rainfall depths for durations ranging from 5
minutes to 24 hours and for return periods ranging from 2 to 200 years for the eThekwini
metropolitan municipality's area of jurisdiction. The output should indicate the spatial variation
in design rainfall depths for a given return period and duration within the study area. Estimates
of design rainfall are to be produced on a grid with a resolution 1' x 1' of a degree
latitude/longitude within the study area.

2 APPROACHES TO DESIGN RAINFALL ESTIMATION

One of the requirements for undertaking frequency analyses is the collection of long periods of
records. Given that the data at a site of interest will seldom be sufficient or available for
frequency analysis, it is necessary to use data from similar and nearby locations (Stedinger et al.,
1993). This approach is known as regional frequency analysis and utilises data from several sites
to estimate the frequency distribution of observed data at each site (Hosking and Wallis, 1987;
Hosking and Wallis, 1997). Thus, the concept of regional analysis is to supplement the time
limited sampling record by the incorporation of spatial randomness using data from different
sites in a region (Schaefer, 1990; Nandakumar, 1995).

Regional frequency analysis assumes that the standardised variate has the same distribution at
every site in the selected region and that data from a region can thus be combined to produce a
single regional rainfall, or flood, frequency curve that is applicable anywhere in that region with
appropriate site-specific scaling (Cunnane, 1989; Gabriele and Arnell, 1991; Hosking and
Wallis, 1997). This approach can then also be used to estimate events at ungauged sites where
no rainfall or runoff data exists at the site (Pilon and Adamowski, 1992).

In nearly all practical situations a regional method has been found to be more efficient than the
application of an at-site analysis (Potter, 1987). This view is also shared by both Lettenmaier
(1985; cited by Cunnane, 1989), who expressed the opinion that “regionalisation is the most
viable way of improving flood quantile estimation”, and by Hosking and Wallis (1997) who,
after a review of recent literature, advocate the use of regional frequency analysis based on the
belief that a “well conducted regional frequency analysis will yield quantile estimates accurate
enough to be useful in many realistic applications”. Where slight statistical heterogeneity exists
within a region, regional analysis yields more accurate design estimates than at-site analysis
(Lettenmaier and Potter, 1985; Cunnane, 1989; Hosking and Wallis, 1997). Even in
heterogenous regions, regional frequency analysis may still be advantageous for the estimation
of extreme quantiles (Cunnane, 1989; Hosking and Wallis, 1997).

Regional approaches are not new in frequency analysis and many different techniques are
available. The development of a regional index-flood type approach to frequency analysis based
on L-moments (Hosking and Wallis, 1993; 1997), termed the Regional L-Moment Algorithm
(RLMA), has many reported benefits and has been successfully used by Smithers and Schulze
(2000a; 2000b) to estimate short (# 24 h) and long (1 to 7 day) duration design rainfall in South
Africa.
Smithers and Schulze (2000a) utilised digitised rainfall data from 172 recording raingauges in
South Africa, which each had at least 10 years of record, and identified 15 relatively

1
homogeneous clusters of extreme short duration (# 24 h) rainfall in South Africa. Data from
1806 rainfall stations in South Africa which have at least 40 years of quality controlled daily
record were utilised by Smithers and Schulze (2000b) to identify 78 relatively homogeneous
clusters of extreme daily rainfall in South Africa. The location of the short and long duration
clusters are shown in Figures 1 and 2.

Figure 1 Distribution of 15 clusters of relatively homogeneous extreme short duration

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

2
Figure 2 Distribution of 78 clusters of relatively homogeneous extreme daily rainfall in
South Africa (Smithers and Schulze, 2000b)

For each cluster and for all durations (5 minutes to 7 days) and return periods (2 to 200 years)
considered, a growth curve relating the scaled design rainfall depth to return period was
developed. The mean of the Annual Maximum Series (AMS) was utilised as the index storm.
Hence, in order to estimate design rainfall depths at an ungauged location, it is necessary to
estimate the mean of the AMS for the required duration at the desired location.

The General Extreme Value (GEV) distribution fitted by L-moments was found by Smithers
(1996) and Smithers and Schulze (2000) to be the most appropriate probability distribution for
the estimation of short duration design rainfall in South Africa. The GEV fitted by L-moments
was also selected by Smithers and Schulze (2000b) for the estimation of long duration design
rainfalls in South Africa.

As illustrated by Smithers (1993) and Smithers and Schulze (2000a), much of the digitised
rainfall data in South Africa is viewed as unreliable, as many errors in the digitisation process
were evident in the data. In addition, comparisons between the 24 h rainfall totals, computed
from the digitised rainfall data, and daily rainfall, as measured at 08:00 every day using standard
non-recording raingauges at the same site, indicated numerous significant discrepancies in the
two values. The rationale adopted by Smithers and Schulze (2002) to overcome the deficiencies
in the digitised rainfall data was to utilise both the scaling characteristics of extreme rainfall and,

3
as far as possible, the more abundant and reliable daily rainfall data which generally have longer
record lengths than the digitised rainfall data.

The 24 h AMS are extracted from the digitised rainfall database using a continuously moving
24 h “window” to identify the maximum 24 h event. The 1 day AMS contains the maximum
events extracted from the daily rainfall database, which contains 24 h rainfall totals recorded at
fixed time intervals.

Smithers and Schulze (2002) developed regionalised regression equations to estimate the mean
of the 1 day AMS at any location in South Africa. Thus, using the regionalsied 24 h : 1day
rainfall ratios developed by Smithers and Schulze (2000a), the mean of the 24 h AMS can be
estimated at any location in South Africa. The estimation of the mean of the AMS for durations
< 24 h has been refined by Smithers and Schulze (2002). For each of the 15 extreme short
duration rainfall clusters and for durations ranging from 5 minutes to 24 h, regression equations
relating the D h to 24 h mean of the AMS have been developed. Thus, given the estimated mean
of the 1 day AMS at the required site, the mean of the AMS for durations #24 h can be
estimated. Similarly, regionalised relationships between the D day (2 # D #7) and 1 day AMS
were developed by Smithers and Schulze (2002). Thus, the mean of the AMS for durations
ranging from 5 minutes to 7 days can be estimated at any location in South Africa.

Smithers and Schulze (2002) investigated the sampling variability and the effect of unreliable
digitised rainfall data on the development of growth curves. They concluded that the growth
curves were scale invariant and that the growth curve for the 1 day duration, which was deemed
to be the most reliable, could thus be applied to durations ranging from 5 minutes to 7 days. This
regional approach to design rainfall estimation developed by Smithers and Schulze (2002), which
is based on the Regional L-Moment Algorithm (Hosking and Wallis, 1993; Hosking and Wallis,
1997), in conjunction with the Scale Invariance of the growth curves (RLMA&SI) has been
shown by Smithers and Schulze (2002) to result in reliable and consistent design values and was
thus utilised in this study to estimate design rainfalls. The RLMA&SI procedure is detailed in
the following chapter.

3 THE RLMA&SI METHODOLOGY FOR DESIGN RAINFALL ESTIMATION

An index storm approach, based on L-moments, has been developed for design rainfall
estimation in South Africa (Smithers and Schulze, 2002). Growth curves which relate design
rainfall, scaled by the mean of the AMS, to duration are utilised in conjunction with an estimate
of the mean of the AMS at the required location to compute the rainfall depth for the specified
duration and return period. Smithers and Schulze ( 2002) concluded that the growth curves are
scale invariant and that the most reliable estimate of the growth curves are those derived from
the daily rainfall database for the 1 day duration. Thus, the growth curves for the 1 day duration
are applied to all durations ranging from 5 minutes to 7 days. Smithers and Schulze (2002)
utilised 90% error bounds for the growth curves, which are used in conjunction with the 90%
prediction intervals for the mean of the AMS, to estimate 90% error bounds for the design
rainfall.

The mean of the 1 day AMS is estimated using regression equations for 7 regions in South
Africa, and corrected using a residual error surface. The mean for the 24 h AMS, as would be

4
estimated from continuously recorded rainfall data, is estimated as shown in Equation 2 from the
1 day mean of the AMS and the 24 h :1 day ratios developed by Smithers and Schulze (2000a).

L _124h = L _11day × Ratio24h : 1d ...2

where
L_124h = mean of the 24 h AMS, extracted from digitised rainfall data,
L_11day = mean of the 1 day AMS, extracted from daily rainfall data, and
Ratio24h : 1d = ratio to convert the mean of the 1 day AMS to 24 h AMS.

In order to estimate the means of the AMS for durations of 2 to 7 days, Equation 3 is used:

 
L _1D = φ D +  α D × L _11day  ...3

where
L_1D = mean of the AMS for duration = D days,
ND = regression constant for duration = D days, and
"D = regression coefficient for duration = D days.

Now, the regression constant and coefficient can be estimated from the following equations:

α D = θ + τ × Dσ ...4

where
2 = regression constant ,
J = regression coefficient, and
F = transformation exponent for duration = D days,
and
φD = ν + κ × Dρ ...5

where
< = regression constant,
6 = regression coefficient, and
D = transformation exponent for duration = D days.

Thus, for a given duration (D), Equations 4 and 5 can be used to estimate the parameters for
Equation 3 and hence L_1D (2 #D #7 days) can be estimated using Equation 3.

However, the regressions for different durations may intersect when L_11 day values are used
which are outside of the range of values used in the regression analysis. Therefore, it is necessary
to estimate the slope of the relationship between the mean of the AMS (L_1) and duration. This
is achieved by using L_1 values for durations of 1, 3 and 7 days derived from Equation 3 as
shown in Equations 6, 7, 8 and 9.

For durations (D): 1 day < D # 3 day

5
 log( L _13day ) − log( L _11day )
Slope(1day − 3day ) = ...6
log(3) − log(1)
L _1D = 10[ ]
log( L _ 11day ) + Slope( 1day − 3 day ) ) × [ log( D ) − log(1) ]
...7

For durations (D): 3 day < D # 7d day

log( L _17 day ) − log( L _13day )

Slope( 3day − 7 day ) = ...8
log(3) − log(1)
L _1D = 10[ ]
log( L _ 13 day ) + Slope( 3 day − 7 day ) × [ log( D ) − log( 3) ]
...9

To estimate the mean of the AMS for durations < 24 h Equation 10 is used:

L _1k = ( L _124 h × XCOEFFi ,k ) + CONSTi ,k ...10

where
L_1k = mean of AMS for duration = k, minutes
XCOEFFi,k = regression coefficient for cluster = i and duration = k minutes in
Table ?, and
CONSTi,k = regression intercept for cluster = i and duration = k minutes in
Table ?.

Thus L_1k could be estimated for any available duration = k minutes. However, the regressions
for different durations may intersect when L_124 h values are used which are outside of the range
of values used in the regression analysis. Thus, it is necessary to estimate the slope of the
relationship between L_1 and duration. This is achieved by using L_1 values for durations of 5,
15 and 120 minutes derived from Equation 10 and the 24 h value calculated using Equation 2.

For durations (D minutes): 120 < D < 1440

log( L _11440 ) − log( L _1120 )
Slope(1440−120) = ...11
log(1440) − log(120)
L _1D = 10[ ]
log( L _124 h ) − Slope( 1440 −120 ) × [ log(1440 ) − log( D ) ]
...12

For durations (D minutes): 15 < D < 120

log( L _1120 ) − log( L _115 )
Slope(120−15) = ...13
log(120) − log(15)
L _1D = 10[ ]
log( L _1120 ) − Slope( 120 −15 ) × [ log(120 ) − log( D ) ]
...14

For durations (D minutes): 5 < D < 15

6
 log( L _115 ) − log( L _15 )
Slope(15− 5) = ...15
log(15) − log(5)
L _1D = 10[ ]
log( L _ 115 ) − Slope( 15−5 ) × [ log(15) − log( D ) ]
...16

Design rainfall depths are calculated using Equation 17.

DREi, j = GCi, j × L_1i ...17

where
DREi,j = design rainfall estimate for duration = i and return period = j,
GCi,,j = growth curve for duration = i and return period = j, and
L_1i = mean of AMS for duration = i estimated using the above procedures.

However, as shown by Smithers and Schulze (2002), the best estimate of the growth curve for
all durations are the values derived from the daily rainfall database for the 1 day duration. Hence
Equation 17 may be re-written as:

DRE i, j = GC1 day, j × L_1i ...18

The 90 % prediction interval for DREi,j is estimated by:

U 90 DRE i, j = U 90 GCi,1day × U 90 L_1i

...19
L90 DRE i, j = L90 GCi,1day × L90 L_1i

where
U90DREi,j = upper 90% error bound of design rainfall estimated for duration = i and
return period = j,
U90GCi,j = upper 90% error bound of the growth curve for duration = i and return
period = j,
U90L_1i = upper 90% error bound of the estimated mean of annual maximum series
for duration=i,
L90DREi,j = lower 90% error bound of design rainfall estimated for duration = i and
return period = j,
L90GCi,j = lower 90% error bound of the growth curve for duration = i and return
period = j, and
L90L_1i = lower 90% error bound of the estimated mean of annual maximum series
for duration=i.

The prediction interval for the mean of the AMS is symmetrical about the estimated value and
hence may be represented as

7
 Pi ...20
PI ( L_1i ) = L_1i × (1 ± )
100

where
PI(L_1i) = prediction interval for the mean of the annual maximum series for
duration = i, and

U 90 L_1i - L_1i ...21

Pi = × 100
L_1i

Therefore

Pi
U 90 DRE i, j = U 90 GC1day, j × L_1i × (1 + )
100 ...22
Pi
L90 DRE i, j = L90 GC1day, j × L_1i × (1 − )
100

The above procedures, which are based on the Regional L-Moment Algorithm (RLMA) and the
Scale Invariance (SI) properties of extreme rainfall, have been termed the RLMA&SI
procedures. The data used in the study and the performance of the RLMA&SI procedures are
assessed in the following sections.

4 STUDY AREA AND AVAILABILITY OF OBSERVED RAINFALL

Rainfall data recorded by tipping bucket raingauges and data loggers, programmed to log at 5
minute intervals during a rainfall event, were supplied to the study by Durban Metro (DM). The
years of record of these data are summarised in Table 1.Stations with less than 10 years of record
were not included in the analyses performed in this study. The location of both recording and
daily raingauges within and surrounding the study area are shown in Figure 3. The autographic
and daily raingauges shown in Figure 3 have at least 10 and 40 years of record respectively.

These data were reformatted and checked for errors using routines developed by Smithers and
Schulze (2000a). This analysis revealed a large number of errors in the data, examples of which
are summarised in the Appendix A. The majority of the errors were negative time steps in the
data and repeated data points.

Table 1 Stations and length of record of rainfall data supplied to the study

8
 Site Start End Years of Record in Annual Maximum
Year Year Series (AMS)
Bluff3 1987 2001 14
Chats1 1987 2001 14
Chats4 1987 2001 13
Cityeng 1987 2001 14
Crabtree 1987 2001 14
Dbnnhi 1987 2001 14
Dunkeld 1987 2001 14
Firwood 1987 2001 14
Islandvw 1987 2001 8
Kennedy 1987 2001 14
Newlnds3 1987 2001 14
Phoenix1 1987 2001 14
Phoenix4 1992 2001 8
Ridgeend 1987 2001 13
Ridgevw 1987 2001 13
Sandpump 1987 2001 8
Sherwd3 1990 2001 6
Stthom 1997 2001 4
Wentwth 1997 2001 4
Woodlnd2 1997 2001 4

In addition to the errors noted in Appendix A, the periods of data listed in Table 2 were excluded
from the analysis as the rainfall recorded during these periods are extremely high. It is suspected
that the rainfall logged during these periods are incorrect.

Table 2 Data excluded from the analysis

Station Period of Record

Removed

Cityeng 01/07/1991 - 30/09/1991

01/11/1999 - 31/12/1999

Dunkeld 01/11/1992 - 30/11/1992

Bluff3 01/12/11989 - 31/12/1989

Dbnhi 07/09/1994 - 09/09/1994

Kennedy 01/10/1998 - 06/10/1998

Ridgeview 20/05/1997 - 24/05/1997

Sandpump 25/11/1991 - 15/11/1991

9
Figure 3 Location of recording and daily raingauges within and surrounding the study
area

The Durban Metro data supplied to the study were utilised revise the D to 24 h regression
relationships used to estimate the mean of the AMS for durations < 24 h.

5 ESTIMATION OF THE MEAN OF THE AMS FOR DURATIONS < 24 HOURS

As indicated in Figure 1, the study area falls largely within short duration Cluster 8. The D : 24 h
relationship derived by Smithers and Schulze (2002) for Cluster 8 is depicted in Figure 4. Using
data supplied by Durban Metro, these relationships were revised and are depicted in Figure 5.
Clearly the relationships depicted in Figure 5 can be used with more confidence than those
shown in Figure 4.

10
Figure 4 D:24 h means of AMS for Cluster 8 derived by Smithers and
Schulze (2002)

Figure 5 D:24 h means of AMS for Cluster 8 revised in this study

11
6 EVALUATION OF RLMA&SI PROCEDURES IN STUDY AREA

The General Extreme Value (GEV) distribution, which has been determined as a suitable
distribution for design rainfall estimation in South Africa by Smithers and Schulze (2000a), was
fitted to the AMS by L-moments and used to estimate design rainfall depths in this study. A
comparison between design rainfall estimated using the observed Durban Metro rainfall data,
the RLMA&SI procedures and from daily raingauges was performed at selected sites.

The results of these comparisons at selected sites and durations are contained in Figures 6 to 8.
The daily raingauge closest to the site in question was used to estimate design rainfall for
durations > 1 day. A clear distinction is made between design rainfall estimated from rainfall
recorded continuously and rainfall recorded at fixed 24 h intervals, such as daily rainfall.
Inconsistencies are evident in the design values computed from the Durban Metro data,
particularly for longer return periods. These results indicate that:

• Design rainfall estimated with the RLMA&SI procedures compare well with the design
values computed from the long records of observed daily rainfall.
• Design rainfall for durations < 24 estimated with the RLMA&SI procedures are similar
to values computed from the observed rainfall data for return period #20 years.
• Design rainfall estimated with the RLMA&SI procedures are consistent over the
durations considered, and thus overcome the inconsistencies in design values computed
from the observed data.
• Design rainfall estimated with the RLMA&SI procedures reflect the differences in
design values computed from continuously recorded (24 h) and rainfall recorded at
fixed time intervals (1 day).

Based on the above observations, it is postulated that the RLMA&SI procedures, with revised
D to 24 h mean of AMS relationships developed in this study, can be used with confidence to
estimate design rainfalls in the study area.

12
 Figure 6 Comparison of design rainfall estimated from observed
data and by the RLMA&SI procedures at the Chats1
site

Figure 7 Comparison of design rainfall estimated from observed data and by

the RLMA&SI procedures at the Crabtree site

13
 Figure 8 Comparison of design rainfall estimated from observed data and by
the RLMA&SI procedures at the Dunkeld site

7 RESULTS

The RLMA&SI procedures were applied on a 1' x 1' of a degree latitude and longitude grid in
the study area. Typical results, with 90% confidence limits for the design values, are contained
in Figure 9.

14
 Figure 9 Examples of design rainfall and 90% confidence limits estimated
using the RLMA&SI procedures at the Chats 1 site

The above analyses were performed for durations ranging from 5 minutes to 7 days and for
return periods of 2 to 200 years. Typical results from applying the RLMA&SI procedures to
estimate design rainfalls in the study area with a spatial resolution of 1' x 1' of a degree latitude
and longitude are shown in Figures 10 to 12 for 15 minute, 1 h and 24 h durations respectively.
From these examples it is evident that there is some variation in the design values within the
study area. The design rainfalls at each 1' x 1' latitude and longitude grid point covering the study
area are supplied in electronic format on the accompanying CD. The format of the database file
(dbf) contained on CD is listed in Appendix B.

15
Figure 10 Examples of 15 minute design rainfall in the study area

16
Figure 11 Examples of 1 h design rainfall in the study area

17
Figure 12 Examples of 24 h design rainfall in the study area
18
8 DISCUSSION AND CONCLUSIONS

The regional approach as developed by Smithers and Schulze (2000a), and further refined and
termed the RLMA&SI procedures by Smithers and Schulze (2002), has been utilised and
refined to estimate short duration design rainfalls for the study area which encompasses the
eThekwini Metropolitan Municipality's area of jurisdiction. This approach differs from the
single site approach to design rainfall estimation used by, for example, Midgley and Pitman
(1978), Alexander (1978; 1990), Adamson (1981), Schulze (1984) and Weddepohl (1988), and
improves the reliability of the design values.

The RLMA&SI procedures developed by Smithers and Schulze (2002) to estimate design
rainfalls in South Africa have been applied in the study area on a 1' x 1' latitude and longitude
grid. These estimates have been compared to values estimated directly from the observed rainfall
data. In this study it has been shown that the RLMA&SI procedures generally result in consistent
and reliable estimates of design rainfall, which are similar to design values estimated directly
from observed rainfall data. Design rainfall estimated using the RLMA&SI procedures are not
affected by at-site anomalies in the data. Daily rainfall data, which have longer records than the
continuously recorded rainfall, are utilised in the analysis. In addition, the regional approach is
far more detailed and is based on an analysis of longer periods of records from more stations than
any other previous study on rainfall frequency analysis in South Africa. With the refinement of
the relationship between the D h and 24 h mean of the AMS using data from the study area, it
is postulated that the estimated design rainfalls may be used with confidence within the study
area.

A listing and description of files containing the results generated in this study are contained in
Appendix B.

9 REFERENCES

Adamson, P.T., 1981. Southern African storm rainfall. Technical Report No. TR 102.
Department of Water Affairs, Pretoria, RSA.
Alexander, W.J.R., 1978. Depth-area-duration-frequency properties of storm precipitation in
South Africa. Technical Report No. TR 83. Department of Water Affairs, Pretoria,
RSA.
Alexander, W.J.R., 1990. Flood Hydrology for Southern Africa. SANCOLD, Pretoria, RSA.
Alexander, W.J.R., 2001. Flood Risk Reduction Measures. University of Pretoria, Pretoria, RSA.
Cunnane, C., 1989. Statistical distributions for flood frequency analysis. WMO Report No. 718.
World Meteorological Organization, Geneva, Switzerland.
Dent, M.C., Lynch, S.D. and Schulze, R.E., 1987. Mapping mean annual and other rainfall
statistics over southern Africa. Report 109/1/89. Water Research Commission, Pretoria,
RSA. 198 plus Appendices pp.
Gabriele, S. and Arnell, N., 1991. A hierarchical approach to regional flood frequency analysis.
Water Resources Research, 27(6): 1281-1289.
Hosking, J.R.M. and Wallis, J.R., 1987. An index flood procedure for regional rainfall frequency
analysis. EOS, Transactions, American Geophysical Union, 68: 312.

19
Hosking, J.R.M. and Wallis, J.R., 1993. Some statistics useful in a regional frequency analysis.
Water Resources Research, 29(2): 271-281.
Hosking, J.R.M. and Wallis, J.R., 1997. Regional Frequency Analysis: An Approach Based on
L-Moments. Cambridge University Press, Cambridge, UK. 224 pp.
Lettenmaier, D.P., 1985. Regionalisation in flood frequency analysis - Is it the answer ? US-
China Bilateral Symposium on the Analysis of Extraordinary Flood Events. Nanjing,
China.
Lettenmaier, D.P. and Potter, K.W., 1985. Testing flood frequency estimation methods using a
regional flood generation model. Water Resources Research, 21: 1903-1914.
Midgley, D.C. and Pitman, W.V., 1978. A depth-duration-frequency diagram for point rainfall
in Southern Africa. HRU Report 2/78. University of Witwatersrand, Johannesburg,
RSA. 57 pp.
Nandakumar, N., 1995. Estimation of extreme rainfalls for Victoria - Application of the Forge
method. Working Document 95/7. Cooperative Research Centre for Catchment
Hydrology, Monash University, Clayton, Victoria, Australia.
Op Ten Noort, T.H., 1983. Flood peak estimation in South Africa. The Civil Engineer in South
Africa, October: 557-563.
Pilon, P.J. and Adamowski, K., 1992. The value of regional information to flood frequency
analysis using the method of L-moments. Canadian Journal of Civil Engineering, 19(1):
137-147.
Potter, K.W., 1987. Research on flood frequency analysis: 1983-1986. Review of Geophysics,
25(2): 113-118.
Schaefer, M.G., 1990. Regional analyses of precipitation annual maxima in Washington State.
Water Resources Research, 26(1): 119-131.
Schmidt, E.J. and Schulze, R.E., 1987. SCS-based design runoff. ACRU Report No. 24.
Department of Agricultural Engineering, University of Natal, Pietermaritzburg, RSA.
164 pp.
Schulze, R.E., 1984. Depth-duration-frequency studies in Natal based on digitised data. In: H.
Maaren (Editor). South African National Hydrology Symposium. Department of
Environment Affairs, Technical Report TR119. Pretoria, RSA. 214-235.
Smithers, J.C., 1993. The effect on design rainfall estimates of errors in the digitised rainfall
database. In: S.A. Lorentz, S.W. Kienzle and M.C. Dent (Editors). Proceedings of the
Sixth South African National Hydrological Symposium. Department of Agricultural
Engineering, University of Natal. Pietermaritzburg, RSA. 95-102.
Smithers, J.C., 1996. Short-duration rainfall frequency model selection in Southern Africa.
Water SA, 22(3): 211-217.
Smithers, J.C. and Schulze, R.E., 2000a. Development and evaluation of techniques for
estimating short duration design rainfall in South Africa. WRC Report No. 681/1/00.
Water Research Commission, Pretoria, RSA. 356 pp.
Smithers, J.C. and Schulze, R.E., 2000b. Long duration design rainfall estimates for South
Africa. WRC Report No. 811/1/00. Water Research Commission, Pretoria, RSA. 69 pp.
Smithers, J.C. and Schulze, R.E., 2002. Design rainfall and flood estimation in South Africa.
WRC Project No. K5/1060. Draft final report (Project K5/1060) to Water Research
Commission, Pretoria, RSA. 155 pp.
Stedinger, J.R., Vogel, R.M. and Foufoula-Georgiou, E., 1993. Frequency analysis of extreme
events. Handbook of Hydrology. McGraw-Hill, New York, USA.
Weddepohl, J.P., 1988. Design rainfall distributions for Southern Africa. Unpublished M.Sc.
Dissertation, Department of Agricultural Engineering, University of Natal,
Pietermaritzburg, RSA.

20
 APPENDIX A

Examples of the Occurrence of Errors in Rainfall Data Files

STATION DATE TIME RAINFALL (mm) COMMENT

CHATS1 21/03/93 26:00 0.51 Changed time to 24:00
24/03/93 26:00 0.91 Changed time to 24:00
24/03/93 26:00 0.91 deleted duplicate
18/10/93 24:00 0.80 occured after 20/10/93
19/10/93 20:50 0.20 occured after 20/10/93
CHATS4 18/11/92 00:00 0.07 occurred after 04:45
CITYENG 30/09/93 14:10 0.60 occurred after 14:25
CRABTREE 27/12/92 03:57 0.00 occurred after 24:00
19/10/99 24:45 0.60 changed to 24:00
26/10/99 0.6:25 2.60 occurred after 24:00
DBNNHI 14/02/90 20:45 0.41 occurred after 20:48
10/12/92 00.00 0.00 deleted - duplicate
DUNKELD 28/09/91 19:01 0.80 occurred after 19:20
28/09/91 21:00 0.80 occurred after 21:45
29/09/91 03:01 0.26 deleted - duplicate
29/09/91 03:01 0.26 occurred after 03:15
22/11/92 24:00 0.00 deleted - duplicate (after 07/10/92)
23/11/92 24:00 0.00 deleted - duplicate (after 07/10/92)
21/01/95 24:00 0.00 deleted - duplicate
FIRWOOD 15/04/89 19:55 0.60 deleted - duplicate
01/10/93 12:10 0.20 occurred after 13:05
01/10/93 16:05 0.60 occurred after 16:40
02/10/93 22:20 2.60 occurred before 22:05
10/04/94 20:00 0.00 occurred after 20:25
01/01/98 11:74 0.25 assumed 11:74 is 11:14 - changed
NEWLND3 18/06/92 24:00 0.00 deleted - duplicate
19/06/92 24:00 0.00 deleted - duplicate
20/06/92 24:00 0.00 deleted - duplicate
PHOENIX1 08/03/94 17:30 0.20 occurred after 01:25
08/03/94 00:05 1.00 occurred after 02:55
08/03/94 00:10 2.72 occurred after 02:55
21/11/94 00:40 0.60 occurred after 16:20
23/10/99 18:15 0.40 occurred after 23:50
23/10/99 19:20 2.60 occurred after 23:50
23/10/99 19:30 0.20 occurred after 23:50
23/10/99 20:45 3.00 occurred after 23:50
23/10/99 21:05 0.20 occurred after 23:50
23/10/99 23:20 2.40 occurred after 23:50
26/10/99 05:35 0.60 occurred after 18:00
26/10/99 05:40 1.80 occurred after 11:35
26/10/99 05:10 1.80 occurred after 11:35
26/10/99 05:10 1.98 deleted - duplicate

21
 STATION DATE TIME RAINFALL (mm) COMMENT
KENNEDY 18/09/93 18:21 0.18 incorrectly placed - not in order
18/09/93 18:30 1.00 incorrectly placed - not in order
18/09/93 19:10 0.40 incorrectly placed - not in order
18/09/93 23:50 0.20 incorrectly placed - not in order
18/09/93 23:51 0.20 incorrectly placed - not in order
18/09/93 24:00 0.00 deleted - duplicate
30/09/93 03:00 0.60 occurred after 06:50
05/10/93 20:00 0.20 occurred after 22:50
06/10/93 00:51 1.60 occurred after 03:41
06/10/93 01:01 0.60 occurred after 03:41
06/10/93 01:21 0.18 occurred after 03:41
06/10/93 01:50 0.60 occurred after 03:41
06/10/93 10:20 0.38 deleted - duplicate
06/10/93 11:01 0.38 occurred after 11:21
06/10/93 10:20 2.60 occurred after 13:40
06/10/93 10:40 0.60 occurred after 13:40
06/10/93 10:41 0.20 occurred after 13:40
06/10/93 11:21 1.60 occurred after 13:40
06/10/93 20:00 0.40 occurred after 22:31
07/10/93 00:40 0.18 occurred after 03:40
07/10/93 01:51 0.18 occurred after 03:40
07/10/93 20:00 0.00 occurred after 23:21
PHOENIX4 24/10/99 20:55 5.00 occurred after 09:20
24/10/99 21:10 0.60 occurred after 09:20
24/10/99 24:00 0.60 occurred after 09:20
RIDGEEND 09/03/94 16:10 0.14 deleted - duplicate
09/03/94 16:10 2.20 occurred after 16:30
SANDPUMP 11/05/91 22:25 0.60 deleted - duplicate
11/05/91 24:00 0.00 deleted - 2 duplicates
13/05/91 24:00 0.40 deleted - duplicate
13/05/91 24:00 0.00 deleted - duplicate

22
 STATION DATE TIME RAINFALL (mm) COMMENT
STTHOM 18/07/97 04:10 1.80 deleted - duplicate
18/07/97 05:00 0.80 occurred after 05:35
18/07/97 05:05 1.54 occurred after 05:35
18/07/97 05:10 0.80 deleted - duplicate
18/07/97 06:10 1.54 deleted - duplicate
18/07/97 07:10 0.54 occurred after 07:35
18/07/97 09:00 2.20 occurred after 09:35
18/07/97 09:05 1.40 occurred after 09:35
18/07/97 09:10 0.80 occurred after 09:35
18/07/97 09:15 0.80 occurred after 09:35
18/07/97 10:05 0.40 deleted - duplicate
18/07/97 11:00 0.80 occurred after 11:35
18/07/97 11:05 0.40 occurred after 11:35
18/07/97 11:15 1.40 deleted - duplicate
18/07/97 12:05 0.80 deleted - duplicate
18/07/97 13:10 1.20 deleted - duplicate
18/07/97 13:15 0.40 occurred after 13:25
18/07/97 14:05 1.20 occurred after 14:35
18/07/97 14:15 0.80 occurred after 14:35
18/07/97 15:00 0.40 deleted - duplicate
18/07/97 15:10 0.54 occurred after 15:35
18/07/97 15:15 0.14 deleted - duplicate
18/07/97 16:00 1.00 occurred after 16:35
28/07/97 03:00 0.40 occurred after 03:35
28/07/97 04:00 0.14 occurred after 04:35
28/07/97 04:15 1.00 occurred after 04:35
WOODLND2 16/10/99 24:00 deleted
08/11/99 24:00 0.20 deleted - duplicate
18/09/00 20:00 0.40 occurred after 20:40
14/11/00 24:00 0.00 deleted - duplicate
15/11/00 24:00 0.00 deleted - duplicate
15/11/00 24:00 0.20 occurred after 24/11/00 - 24:00
01/01/01 24:00 0.00 deleted - duplicate
01/01/01 24:00 1.00 occurred after 31/01/01 - 24:00
01/01/01 23:45 0.60 occurred after 31/01/01 - 24:00
03/02/01 19:00 0.60 occurred after 19:45
03/02/01 19:11 0.20 occurred after 19:45
14/02/01 04:00 1.40 occurred after 04:35
18/02/01 00:15 0.40 occurred after 04:30
18/02/01 14:00 1.80 deleted - duplicate
04/04/01 20:00 0.40 occurred after 20:10

23
 APPENDIX B

Description of Files

dm_dre.dbf
Relevant Description Units
Fields o
LATDEG Latitude
LATMIN Latitude 'o
LONGDEG Longitude
LONGMIN Longitude '
SC Short duration cluster number
DC Daily cluster number
RP Return Period Years
CODE Flag M = design value
L = lower 90% confidence limit
U = upper 90% confidence limit
M5 Design rainfall for 5 minute duration event mm
M10 Design rainfall for 10 minute duration event mm
.....
M1440 Design rainfall for 1440 (24h) minute duration event mm
D1 Design rainfall for 1 day duration event mm
...
D7 Design rainfall for 1 day duration event mm
o
LATDEC Latitude decimal
o
LONGDEC Longitude decimal
POSM n/a

24