TRRL Rainfall PDF
TRRL Rainfall PDF
RESEARCH LABORATORY
by
The layout of this report is such that anyone who wishes to remove the
Appendix from the main body of the report, for use as a design manual,
can do so by removing the three centre double pages, less the final page
of the text, without damaging the report.
Environment Division
Transport Systems Department
Transport and Road Research Laboratory
Crowthorne, Berkshire
1974
CONTENTS
Page
Abstract 1
1. Introduction 1
4.2.1 7
4.2.2 7
4.5 Conclusions 12
5.1.4 A t u m a t a k network 18
° Conclusions 21
7. Acknowledgements 22
8. References 22
© ceowlvcoermaHr 1974
Extracts may be quoted except for commercial purposes,
provided the source is acknowledged
Ownership of the Transport Research
Laboratory was transferred from the
Department of Transport to a subsidiary of
the Transport Research Foundation on I st
April 1996.
n •
ABSTRACT
A simple method for predicting the characteristics o f storms for the design
of drainage structures in East Africa is described. The variation o f 2 year
daily point rainfall, and the 10:2 year ratio for daily rainfall, over East
Africa are given in map form. Using these, daily point rainfall for any return
frequency can be calculated. To arrive at the design storm the daily point
rainfall is adjusted using a generalised depth-duration equation and a
graphical representation of the variation of mean rainfall with area.
1. INTRODUCTION
Before the hydraulic and structural designs for a road bridge or culvert can be started, an estimate must be made
of the peak flow that the structure must safely pass. If flow measurements have been made for a number of
years on the river, or on a similar but adjacent river, this involves only a statistical analysis of recorded peaks
to arrive at a design flood of an appropriate return frequency. In East Africa, particularly on the smaller rivers,
such data rarely exist and use must be made of the much more common rainfall measurements published by the
East African Meteorological Department. From rainfall records a design rainstorm is constructed and routed
through a suitable catchment model to give thedesign flood.
Although an impressive amount of rainfall data exists, it has not been published in a form that can be
readily used by highway engineers for bridge and culvert design. The purpose of this report is to extract relevant
storm data from the published records and, combining these with certain unpublished data, to produce a simple
method for preparing design storms for flood estimation. The method involves first estimating the 2 year, 24 hour
point rainfall from a storm rainfall map o f East Africa. Three adjustments are then made:
(a) Using a generalised relationship between rainfall of any return frequency, and the two year values the
24 hour point rainfall for the design return frequency is calculated.
(b) A depth-duration rainfall equation is used to calculate the point rainfall for the appropriate time o f
concentration of the catchment.
(c) An areal reduction is read off a graph to convert this to the mean rainfall depth which is the required
rainfall input for the catchment model.
There are about 3,000 rainfall stations in East Africa which submit daily records to the Meteorological Department
for subsequent publication. The distribution of these is, however, far from uniform and m a n y have been installed
only in recent years.
The advice Often given to engineers requiring a design storm is to select a suitable rainfall station, on, or
close to, the catchment and to analyse the records for this station. For much of East Africa a station on or close
to the catchment cannot be found or, if present, it has often been recording for such a short period o f years
that it can give only unreliable estimates of flood producing rainfall. It was therefore decided to analyse all
available published records and use these to produce maps of storm rainfall from which values for individual
catchments could be interpolated.
Using records in published form this would have been a m a m m o t h task, but fortunately early in the
investigation the East African Meteorological Department transferred all their reliable daily rainfall records for
the years 1957-68 and selected stations for 1926-56 to magnetic tape and gave the Laboratory permission to
make a c o p y of the tapes.
F r o m the first tape 867 stations which had at least 10 complete data years and from the combined tapes
99 stations with about 40 complete data years were selected. The first set of data was used to map the variation
o f storm rainfall over East Africa and the second to establish a means of adjusting values read off the map for
alternative frequencies.
For each station selected for the first set o f data the highest 24 hour fall during each calendar year was
read off. These were ranked and given return frequencies using the expression:
n+l
Y -
m
If the rainfall depth is plotted against the assumed return frequency a non-linear relationship becomes
apparent. Many methods are available to linearise this relationship which, so long as extrapolation beyond the
period o f record is not attempted, give very similar results. The most commonly accepted method is the Gumbel
method (1) which is of the form
Y = a+cloglogT T1
Gumbel equations for all of the stations were produced. Because of the short period of record and the
known variability year to year of rainfall in East Africa, such equations are bound to be only very rough
estimates of storm rainfall at return frequencies approaching the period of record. The most accurate value is
likely to be near to the median value which has a return frequency of 2.3 years. For this reason a 2 year storm
was selected for mapping. '-
The 2 year dally storm rainfall estimates and the coordinates for each station were entered into the
Laboratory's 47/0 computer and a map was produced using the Calcomp General Purpose Contouring Program
(GPCP). As the surface is specified by random data, the data are gridcled; that is, the values of the function at the
mesh points of a rectangular array are generated by a procedure which analytically constructs a smooth surface
passing through every data point. The isohyets are produced by interpolation from the generated mode values
o f the mesh using a third order function. Among the parameters to be specified by the user of GPCP are the mesh
size, and the number of data points required in the vicinity of a node in order to determine the node value. The
effect o f assigning various values to these parameters was investigated, and as a result suitable values were chosen.
76 o f the rainfall stations had records in b o t h the 10 and 40 year tapes. These were used toassess the
2
probable errors in the individual two year estimates using the shorter period of record. The results are plotted
in Fig. 2 where it will be seen that there is a considerable scatter. This is to be expected with so short a period
of record. A line of best fit through the points was calculated and is shown as a full line. It will be seen that
there is a tendancy for the results from the 10 year tape to over estimate the 2 year storm particularly at values
of 80 mm and over but the scatter of the points is such that no systematic adjustment to the map is justified.
The errors are on the safe side in design and for about 80 per cent of the area are well within the plotting accuracy.
The same analysis was made for the records for each region. All were consistent with the overall results
except the arid zone in North and East Kenya. Here the 10 year tape results consistently over estimated the 2
year values by about 10 mm.
The smoothing of isohyets achieved by the contouring programme should remove most of the random
errors in the estimates. At this time it is not possible to check how far this has been successfully accomplished
but this will become apparent if the same exercise is undertaken in 10 years time when further data are available.
Because of the uncertainty of estimation of rainfall in the arid zone of North East Kenya it is not
recommended at this time that Fig. 1 be adjusted to allow for the apparent over estimate from using only
10 years of record.
Records from 8 gauges within an area close to Nairobi, for which 35 years of reliable data were available,
were analysed as a further check on the probable errors in the use of short period records and to check on
possible cyclic behaviour in storm rainfall in East Africa.
Running means and standard deviations for annual maxima were calculated for each gauge. From these,
95 per cent confidence bands for the means were calculated and plotted against the period o f the running mean.
These are shown in Table 1 and Fig. 3. (Note: the mean annual maximum rainfall has a return frequency o f
approximately 2.3 years).
If any significant cyclic effects were present these would show up as peaks in Fig. 3. As can be seen a
smooth curve has resulted. It may therefore be concluded that cyclic effects are not significant.
From the figure the relative improvement in estimate of 2 year storms with increasing record length c a n
be seen. Below 10 year periods the confidence limits diverge rapidly but for longer periods the improvement is
less dramatic. An increase from 10 to 20 year period reduces the confidence limits by almost half which indicates
the order of the improvement in estimate in areas of sparse gauge coverage which can be expected if Fig. 1 is
replotted in about 10 years time when more data are available.
In section 2, Gumbel Analysis was used to establish 2 year values for storm rainfall. The Gumbel equation has
the form
y = a+cx
where y is the storm rainfall for the appropriate reduced variable x, and a and c are constants.
To use the 2 year rainfall map, Fig. 1, to predict storms of larger return frequency it is necessary tO b~
able to predict suitable values for a and c. To attempt to do this a selection o f the Gumbel regression lines from
the 40 year tape were superimposed to see if there was any pattern to them. It was noticed that certain regional
characteristics were evident. For example, in the arid areas and the coastal strip, slopes (c) tended to be higher
than in Uganda. The gauges were therefore split up into regional groupings. Earlier analysis o f data from Kenya
and Uganda suggested four regional groups; the coastal strip, the arid area of north and east Kenya, the central
Highlands and west of the Rift Valley. These regions were therefore used.
Linear regressions of Gumbel slope on the 2 year value (Y2) were made for each region. In two zones
3
TABLE 1
Gauge
913604 913606 913618 913620 913624 913628 913629 913630 Average
No.
Period
(yrs)
5 2.31 3.50 4.80 5.72 3.36 3.32 6.15 6.66 4.48
significant correlations were obtained. In the other two, the coastal and the arid zones, the correlations were not
significant but they contained only 8 and 5 gauges respectively and in the case of the coastal zone all the Y2
values were very similar.
4
For each of the two regression lines there was found to be no significant difference between the slope
of the line as calculated and the slope of a line through the origin (o, o) and (Y2' c). It was therefore concluded
that the slope of the Gumbel regression line could be replaced by a simple ratio of the values of 2 points on the
regression, the 10 year and 2 year ratio (Yl0:Y2) being most appropriate.
Yl0 - 1.885c +1
Y2 Y2
For each zone the average value for slope (c) and Y2 were calculated and the appropriate Yl0:Y 2 ratio
derived. These are shown below in Table 2.
TABLE 2
(c) (Y2)
Because of the scarcity of gauges on the 40 year tape it was necessary to use the records from the 10
year tape to establish
(b) the best value for the 10:2 year ratio particularly in the coastal and arid zones where records are
very scarce, and for most of Tanzania for which no data on the 40 year tape were available.
With only 10 years of record the estimates of the 10 year storms are bound to be inaccurate, but if the
scatter is random, and sufficient records are available, sufficiently accurate estimates for the appropriate value
for a zone are possible. To check if using the short period of records would introduce any bias a comparison
was made of the 10:2 year ratio for the 76 stations common to both tapes. The results are shown in Table 3.
With the possible exception of the arid zone, the results of which are very variable, it can be seen that
no bias is likely to be introduced by using the 10 year records and a much larger number of gauges will make
defining boundaries between zones easier.
The gauge numbering adopted by the East African Meteorological Department is according to the degree
square in which the gauge lies. Gauges were therefore easily grouped, and for each degree square the mean and
standard deviation for the 10:2 year ratios were calculated. The means for adjacent squares were then compared
to see if the differences were significant. In this way the rough boundary to the zones was established. To get a
more accurate plot the gauges adjacent to the boundary were located on large scale maps and the boundary
5
TABLE 3
West zone
40 year 34 1.509 0.042
10 year 34 1.516 0.058
Central Highlands
40 year 29 1.597 0.046
10 year 29 1.585 0.057
Arid zone
40 year 5 2.006 0.264
10 year 5 1.864 0.317
Coastal strip
40 year 8 1.728 0.098
10 year 8 1.651 0.178
fixed by inspection. Once the line had been fixed it was superimposed on mean annual rainfall and topographical
maps to see if there was any physical explanation for zonal differences.
The results are shown in Table 4 and Fig. 4. I n order to obtain the appropriate storm rainfall value for
any given return frequency, read o f f from Fig. 5 the N:2 year ratio corresponding to the known 10:2 year ratio
and multiply by the 2 year daily rainfall, read from Fig. 1.
For most catchments, the rain falling in periods of less than one day are required. These can be estimated using
daily values and a suitable depth -- duration relationship. Two models were tested
a
(a) I -
Tn
(b) I - (T+b) n
where I = intensity in m m / h r
T = duration in hours
6
TABLE 4
Two sets of data were used. For 23 stations intensities for several durations from 15 minutes upwards were
used to select the best model. For a further 18 stations only 1 hour and 24 hour values were available. These
were used to assist model calibration. Details are given in Table 5.
4.2.1 I = a
Tn
This is a model that has been suggested by Mc Callum (3) as being appropriate for East Africa to model
intensities from 15 minutes to 24 hours. Mc Callum used data from 6 stations in Kenya and Tanzania. The
relationship was fitted to the highest intensity measured at each station. The period of record varied between 8
and 25 years. Because of the uncertainty about an appropriate return frequency to apply to Mc Callum's data
direct comparison between his results and those given below is difficult.
The Group I data, for which durations of from 15 m - 24 hours were available, were fitted to this model
for a 2 year return frequency and the results are given in Table 6.
4.2.2 I- a
(T + b) n
This model will be seen to be a general equation which reduces to the much quoted I _ T a+ b if the
7
TABLE 5
GROUP I
BUSIA
KASESE •
WADELA I
15 m -- 24hr
15 m -- 2 hr
15m- 2hr
) Supplied by Water Development Dept.
Uganda
MUGUGA 15 m - 24 hr
Extracted from records of East African
ATUMATAK 15 m - 24 hr
Agriculture and Forestry Research
SAMBRET 15 m - 24 hr
Organisation
SAOSA 15 m - 24 hr
EQUATOR 15m- lhr
KABATE 15m- 6hr
KISUMU 15m- 6hr
KITALE 15m- 3hr
MOMBASA 15m- 6hr
NANYUKI 15m- 1 hr
VOI 15m- lhr
D A R ES SALAAM 15m- 6hr
?- From Taylor & Lawes (2)
DODOMA 15m- 3hr
KIGOMA 15m- 3hr
MBEYA 15m-- 3hr
TABOCA 15m- 6hr
ZANZIBAR 15m-- 6hr
ENTEBBE 15m- 6hr
GULU 15m- 3hr
KAMPALA 15m- 3hr
\
G R O U P II
KITALE
MOLO
LAMU
!
LODWAR
GARISSA
NAKURU
KISUMU
MOMBASA Extracted from records of East African
NANYUKI lhrand24hr
Met. Dept.
VOI
JINJA
MBARARA
TORORO
KAMPALA
GULU
ENTEBBE
FORT PORTAL i
For each duration the largest rainfall value in each calendar year was ranked to form an annual series for the
station. Estimates of rainfall depth corresponding to recurrence intervals of 2, 5 and 10 years were then made using
the G u m b e l method (1). The period o f record for the group 1 stations was between 8-35 years, and the group 2
index n = 1. Often the s~mpler form is used with different values for the constants for different ranges o f
duration. Where, as in East Africa very little data other than daily totals exist, a relationship containing a
discontinuity is difficult to fit and the general expression, even if slightly more difficult to apply in practice, is
to be preferred.
The Group I data were fitted to the model for a number o f alternative values o f the c o n s t a n t ' b ' between
0.2 and 1.0 hours. The optimum value varied between stations but as no regional pattern to this variation could
be found it was assumed to be due to random errors in the data and an average value o f b = 0.33 h r s w a s
selected for further modelling. The results of fitting this model to the 5 stations, for which complete d a t a were
available, were very much superior to the previous model. It was therefore a d o p t e d and used with all group I
stations. The derived relationships with b = 0.33 hrs are given in Table 7.
Hourly and daily estimates o f rainfall with 2, 5 and 10 year recurrence intervals were available for 18
stations covering all the climatic zones of Kenya and Uganda. These were fitted to the intensity - duration
model with the constant being equal to 0.33. The results are shown in Table 8.
Tables 7 and 8 show that a constant value for " n " cannot apply to the whole o f East Africa. The area was
therefore once again split up into zones. Four zones were considered:--
(b) Arid
It will be seen that for most o f the stations in the Inland and Arid zones the value o f "n' is a p p r o x i m a t e l y
1.0 but that lower values are typical for the Coastal and Central Highland areas.
For the Inland zone, only Entebbe gave a value for ' n ' well under 1.0. A possible explanation for this is
that the period o f record included one very large storm which it has been estimated approached the probable
maximum precipitation (4). This would result in an under estimation o f the time value o f ' n ' . There is therefore
no firm evidence for excluding Entebbe from the general inland model.
9
TABLE 7
I - - -
n
2 yr correlation coefficient is shown only for 5 stations for comparison with TABLE 6
10
TABLE 8
Model I - a
( T + ~_)n
The 'n' values for the Central Highlands are very variable. The western stations (Equator and Nakuru) give
very similar results to Inland stations whereas stations around Nairobi give much lower values. The explanation
for this must be differences in synoptic weather processes. This can be checked by looking at the diurnal
variations in rainfall occurrence which have been studied by Thompson (5). The conventional "continental"
rainfall model gives convective thunderstorms in late afternoon. Much of the inland zone does have a rainfall
maximum at this time as does the northern and western parts of the Central Highlands zone, but in the
Nairobi area heavy rain occurs in the evening, spreading through to the early morning in the "short rains"
(November). Thompson claims that a large part of this rain results from the spread o f storms from the Highlands
close by after about 5.00 pm. This would explain the longer duration and lower intensity (small 'n') of Nairobi
rainfall. A similar pattern would be expected on all windward slopes of the Kenya - Aberdare range and the
Kilimanjaro area. The Central Highlands zone has therefore been divided into two halves (by the dotted line in
Fig. 4) to show the area similar to Nairobi and the area similar to the inland stations.
Molo is the one station that does not fit this pattern. It is on the eastern facing slope of the Mau plateau
and at an altitude o f 2,500 m, but on the evidence of just one station it is not possible to draw any firm conclusions.
11
Voi does not fit the arid model. Thompson (5) shows that although much of the rain at Voi occurs in the
afternoon, there is a significant amount of morning rainfall due possibly to the effect of the adjacent Taita hills.
Voi's position (Fig. 4) on the border between the arid and highland zones is appropriate and Voi is therefore not
included in computing average values for'n'below.
At the coast thunderstorms are not very common and heavy rainfall is more frequently the result of a
disturbance or discontinuity in the monsoon (13). A different model for the coast is therefore appropriate.
TABLE 9
in the equation I -
Recurrence Interval
3. Eastern slopes of
0.82 0.85 0.85
Kenya-Aberdare Range
It is proposed that in practice an engineer will estimate the daily rainfall for the appropriate recurrence
interval using Figs. 1, 4 and 5. He will then enter this into the relevant intensity - duration model above, to
obtain the design rainfall he requires. The form of the above models were therefore adjusted to simplify this
operation.
a
I -
(T + b) n
aT
Rainfall in time T (RT)
(T + b) n
24 a
The daffy total (RD)
(24 + b) n
T (24+b'~ n
Eliminating 'a' R T
24 \T + b]
with b = 0.33, a unique set of curves can be developed for converting daffy rainfall to rainfall of any given
duration. These are shown in Fig. 6.
4.5 Conclusions
It was concluded that Fig. 6 is the best means at present available for estimating depth - duration
ratios for rainfall in East Africa. All inland areas other than Eastern and South Eastern facing slopes of the
Aberdare - Kenya ranges are satisfactorily modelled using the average inland curves. It is possible that in
very wet mountainous areas elsewhere curves similar to the Nairobi curve are appropriate but these areas will
12
be very limited in extent and with present data impossible to predict. In these areas use of the average inland
curve is probably conservative.
5. A R E A REDUCTION FACTORS
In the previous sections a method has been developed for predicting point rainfall for any duration and recurrence
interval. Over a catchment the point rainfall will vary and it is necessary to be able to predict this variation to
estimate the volumetric rainfall input to the catchment. The most convenient way is by means of areal reduction
factors (ARF). These are factors by which the appropriate estimates of point rainfall are multiplied to give the
average depth of rainfall over the catchment.
No factors have been published for East African data. In this section data from four dense networks of
raingauges in East Africa are analysed to derive areal reduction factors and from them a general equation for
East Africa is developed. This is then compared to published equations for other parts o f the world.
5.1.1 T h e K a k i r a N e t w o r k Sixteen years of record from 29 standard daily read raingauges were available
from a sugar estate on the northern shore of Lake Victoria, 12 miles east of Jinja, Uganda. The estate is
approximately 82 km 2 in area and undulating. The gauge network is shown in Fig. 7.
The method adopted for deriving the areal reduction factors in this and the following network studies
was to derive depth-frequency equations for point rainfall for each gauge and to compare these with similar
equations for the average rainfall over the catchment.
Depth-frequency equations for each of the 29 gauges were obtained using the Gumbel method (1). Annual
series were formed by ranking the maximum 1 day rainfall for each calendar year, for the 16 years. These were
plotted on Gumbel extreme value paper to provide a visual check on the assumption of linearity. All gave
reasonable straight line plots. The Gumbel regression lines were then calculated. Goodness of fit was checked by
calculating the correlation coefficients as described by Nash (6) and these are given in Table 10.
Over the network the point data relationships were a.veraged to give the best estimate of the depth-frequency
characteristics of the area. To do this one must assume that the area is homogeneous and that differences between
gauges can be reasonably expected to be due to chance. This was checkbd by using the Langbein homogeneity
test (7). The equation for point rainfall for the whole network is also given in Table 10.
The network was divided up into six areas labelled A - F in Fig. 7. The mean areal rainfall was calculated
for the following combinations of area:
The mean rainfall for each area was calculated for each day of heavy rainfall using the Thiessen method
(14). Annual series were formed for these and Gumbel regression equations computed as before. These are
shown in Table 11.
By comparing these regression equations with the mean equations for point rainfall, areal reduction
factors were calculated. These are given in Table 12, and shown also in Figs. 8 to 10.
There is no evidence for a change in areal reduction factor with recurrence interval, bearing in mind the
width of the confidence band. The regressions are likely to be most accurate at a recurrence interval of just over
two years. The two year values were therefore taken as the best estimate of areal reduction factor for all
recurrence intervals.
13
TABLE 10
Kakira Network
14
TABLE 11
Kakira N e t w o r k
TABLE 12
Kakira N e t w o r k
Areal R e d u c t i o n F a c t o r
Area
(km 2 ) 2 yr 5 yr 10 y r
15 0.904 0.892 0.886
40 0.856 0.842 0.835
80 0.797 0.797 0.797
Fig. 11 shows the network o f gauges and the outline o f the areas over which m e a n rainfall was calculated.
The areas are:
15
T A B L E 13
Nairobi Network
Total
Network Y = 56.20 + 21.77 X 0.91 64.2 88.9 105.2
16
The area C was arranged to include the whole of the built up area of the city of Nairobi (shown in Fig. 11
by a full line) so that by comparison with area B any effect on areal reduction factors due to the modifications
of the climate locally by urbanisation would be shown up. As will be seen below no effect was observed.
A major difficulty in calculating rainfall for such a network is that the number of gauges in operation
varies from storm to storm. Manual calculation of the Thiessen weightings for over 100 storms is very tedious.
A computer program was therefore prepared which, given the co-ordinates and catch for each gauge in operation
for a particular storm, calculates the appropriate Thiessen weightings and average depth of rainfall for any area. (8).
A second difficulty is that as continuous records are not available for all gauges, point rainfall relationships
can only be calculated for the few gauges for which continuous records are available. For the Nairobi network
18 gauges were available with continuous records for the 20 year study period 1937-56. This period was chosen
because prior to 1937 relatively few gauges were installed and for the years 1957-60 only a selection of gauge
records were published.
The Gumbel regression equations for the index gauges are given in Table 13. A Langbein homogeneity
test on the data showed that the area could be considered as homogeneous. The data were therefore combined
to produce a Gumbel regression equation for the whole area. This is also given in Table.13.
Using the Thiessen Polygon program the average rainfall for each area was calculated for all large storms.
Annual series were then prepared and Gumbel regression equations for areal rainfall calculated. These are shown
in Table 14.
TABLE 14
Nairobi Network
TABLE 15
Nairobi Network
As with the Kakira network the evidence for a variation in areal reduction factor with recurrence interval
is inconclusive. The 2 year values are therefore taken as the best estimate for all recurrence intervals.
The analysis using the same methods as described for the Kakira network resulted in the following Gumbel
regression equations:
TABLE 16
Predicted Rainfall
Return period Areal Reduction
(yrs) Point Areal Factor
(mm) (mm)
Gumbel regression equations for point and areal rainfall were calculated as before and are given in Tables
17 and 18. The areal reduction factors from these are given in Table 19.
18
TABLE 17
Atumatak Network
E s t i m a t e d S t o r m Rainfall
Period Regression
Regression equation
(hrs) coefficient 2 yr 5 yr I0 yr
(mm) (mm) (mm)
¼ Y = 19.14 + 4.61 X 0.86 20.83 26.06 29.51
Y = 26.69 + 5.38 X 0.86 28.66 34.76 38.80
1 Y = 32.10 + 7.01 X 0.73 34.67 42.62 47.87
2 Y = 35.39 + 6.47 X 0.71 37.76 45.10 49.95
8 Y = 40.22 + 8.08 X 0.80 43.19 52.34 58.40
24 Y = 42.01 + 8.63 X 0.71 45.18 54.96 61.43
TABLE 18
Atumatak Network
E s t i m a t e d S t o r m Rainfall
Period Regression
Regression equation
(hrs) coefficient 2 yr 5 yr 10 y r
(mm) (mm) (mm)
¼ Y = 9.95 + 3.62 X 0.95 11.28 15.38 18.10
1/i Y = 16.03 + 5.49 X 0.95 18.04 24.27 28.38
1 Y = 23.07 + 6.11 X 0.93 25.31 32.24 36.82
2 Y = 27.15 + 5.65 X 0.93 29.22 35.63 39.86
8 Y = 32.87 + 6.23 X 0.98 35.16 42.22 46.89
24 Y = 35.46 + 8.33 X 0.98 38.52 47.96 54.22
19
TABLE 19
Atumatak Network
With data from only four networks it is not possible to arrive at a number of models for areal rainfall and
an objective plot of the boundaries of the zones appropriate to each model. All that can be done at this stage is
to develop a single model and to apply this in all cases where it is not obviously inapproPriate.
Further data will become available when analysis is complete on three dense networks of autographic
gauges over Nairobi, Kampala and Dar es Salaam and the networks.of the Kenya and Uganda Rural Catchment
programme (9). At that time an improved model will be possible.
A plot of 24 hour areal reduction factors against area (Fig. 17) shows that the two Uganda networks
(Kakira and Atumatak) give smaller values than the two Kenyan networks. This is consistent with the observations
of Johnson (15) who divided East Africa up into four zones.
a. Highland regions o f Kenya and Southern Tanzania where rain tends to be widespread.
C. Dry regions of N.E. Kenya and S.W. Tanzania which are intermediate between these two zones.
It is therefore concluded that the Sambret and Nairobi results can be combined to form an upper limit
curve which will apply to highland areas of Kenya and Tanzania. The same curve will probably not be too
conservative for all other areas except Uganda where the results from Kakira and Atumatak should be used as
a guide until such time as further data are available.
Factors for periods o f less than 24 hours are much smaller. This is particularly important for urban catch-
ments which is one of the main reasons for initiating the urban raingauge networks referred to above. For rural
areas the lag in runoff means that most storms are shorter than the time of concentration of the catchment so
that 24 hour values are appropriate.
Very few published data are available for tropical Africa and the equations published for other parts of
20
the world are of little use in interpreting African results because the rainfall elsewhere appears to be much more
extensive. For example the equation published by the US. Weather Bureau (12) as being appropriate to
continental U.S.A. is:
The factors predicted by this equation are much higher than those appropriate to East Africa. (The
1 000 km 24 hour value = 0.91).
ARF = 1 - C
This gives values as high or even higher than the U.S. Weather Bureau equation.
The only published figures for Tropical Africa known to the authors are those by Rodier and Auvray (11)
for West Africa. These are shown in the Table 20 below.
TABLE 20
0-25 1
26-50 0.95
51-100 0.90
101-150 0.85
151-200 0.80
These are similar to the East African upper limit curve. It is difficult however to make a direct comparison
as they are design recommendations and not experimental values and it is possible that some rounding up has
been allowed to simplify design techniques.
6. CON C LU S I ON S
It has been shown that daily point rainfall can be predicted for any catchment in East Africa using Figs. 1 , 4
and 5. Figs. 1 and 4 are in such a form that they can be updated when the East African Meteorological
Department's taped daily records have been extended. It is recommended that the mapping be repeated in
about 1978 when the short period tape will cover a 20 year period.
The depth-duration-frequency equations are adequate for design use for flood prediction in rural areas.
They are not adequate for use with urban flood models but this will be rectified when current research in East
Africa using high speed autographic recorders is complete. Improved models for short duration rainfall should
therefore be available by early 1975.
21
Further data are required to give a complete picture of the variation of areal reduction factor with
location and storm duration. These will be made available by the current rural catchment and urban rain gauge
programme, which it is anticipated will be reported upon early in 1975.
7. ACKNOWLEDGEMENTS
The work described in this report was carried out in the Environment Division of the Transport Systems Depart-
ment o f the Transport and Road Research Laboratory, and forms part of the programme of research undertaken
at the Laboratory on behalf of the Overseas Development Administration of the Foreign and Commonwealth
Office. The assistance o f the Director General o f the East African Meteorological Department, the Director
of the East African Agriculture and Forestry Research Organisation and the Manager of the Kakira sugar estate
in making data available is gratefully acknowledged.
8. R E F E R E N C E S
1. GUMBEL, E. J. Statistical Theory of Extreme Values and some Practical Applications. National Bureau
of Standards, Applied Mathematics series 33. U.S. Dept. of Commerce 1954.
3. McCALLUM, D. The Relationship between Maximum Rainfall Intensity and Time. East African
Meteorological Dept. Memoirs. Vol. III No. 7 1959.
5. THOMPSON, B. W. The Diurnal Variation of Precipitation in British East Africa. East Africa Meteorological
Dept., Technical Memorandum No. 8 1957.
6. River Engineering and Water Conservation Works, Ed Thorn R. B. Ch. 6 Applied Flood Hydrology Nash
J.E.
8. GRIGG; A. O. A program for calculating Thiessen average rainfall. Department of the Environment
T R R L Report LR 470, Crowthorne, 1972 (Transport and Road Research Laboratory).
9. FIDDES, D and J. A. FORSGATE. Representative rural catchments in Kenya and Uganda. Ministry of
Transport, RRL Report LR 318, Crowthorne, 1970 (Road Research Laboratory).
11. RODIER, J. A. and C. AUVRAY. Preliminary general studies of floods on experimental and representative
catchment areas in tropical Africa. 1ASH Symposium of Budapest 1965 Vol 1 pp 22-38.
12. H E R S H F I E L D , D. M. Rainfall Frequency Atlas for the United States. U.S. Weather Bureau. Technical
paper No. 40 1961.
13. SONSDIN, H. W. The Maximum Possible Rainfall in East Africa, East African Meteorological Dept.
Technical Memorandum No. 3 1953.
14. THIESSEN, A. H. Precipitation averages for large areas, Monthly Weather Review Vol. 39 pp 1082 - 84
1911.
15. JOHNSON, D. H. Rain in East Africa. Ch. 5 Royal Met. Soc. Vol. 88 Jan 1962 pp 1-19.
22
9. A P P E N D I X 1
The relevant figures and table are reproduced below. These examples were designed to act as a guide to the
design methods developed in the main report and also to show the range of variation in short period rainfall
over East Africa.
Example I
Calculate the design storm required to estimate the flood resulting from 25 year recurrence interval s t o r m
rainfall on a 20 km 2 catchment, grid reference 32°E I°N.
From Appendix 1 Fig. 3 for a 10 year: 2 year ratio of 1.49 and a recurrence
interval of 25 years the flood factor = 1.74 1.74
From Appendix 1 Fig. 4 read off the area reduction factor for a 20 k m 2 area = 0.9 0.9
The areal rainfall for the catchment is 122 x 0.9
I10 mm
= 109.8 (say 110)
• From Appendix 1 Table 1 select a suitable ' n ' value for an inland station (Zone 1) = 0.96 0.96
Using ' n ' = 0.95 in Appendix 1 Fig. 5 select rainfall ratios for 15 mins,
30 mins, 1 hour, 2 hours, 4 hours and multiply by 110 m m to obtain
R T for each period. These are then plotted as a symmetrical histogram,
R T b e i n g shown in units of (mm of rain in 15 mins)
23
E x a m p l e II
A s s u m i n g a symmetrical shape calculate the 10 year recurrence interval design storms for point rainfall appropriate
for
(a) Nairobi
(b) Kampala
(c) D a r e s Salaam
(a) Nairobi
Using ' n ' = 0.85 in A p p e n d i x 1 Fig. 5 calculate R T for 15, 30 mins, 1 hr, 2 hr, 4 hr and plot as a symmetrical
histogram in units o f m m / 1 5 mins.
(b) Kampala
24
Using 'n' = 0 . 9 5 in A p p e n d i x 1 Fig. 5 calculate R T for 15 min, 30 m i n , 1 , 2 , 4 hrs and p l o t as s y m m e t r i c a l
histogram in units o f r a m / 1 5 rains.
25
APPENDIX 1
TABLE 1
Recurrence Interval
Zone
2 year 5 year 10 year
26
30"E 35,E 40"E
" 5°N 5"h
i I
¢ f I
I
!
t \ o P / / ~ I I
I " K ~tt ~ n
I
o" _ ~ o , I
# ~ LAKEVICTORIA
1
,-"U
--%~. ; / I
•
,, ~
. "~ L_
I~' "* " -
!
:....._. , .. , ,
• • da ~ s S
I .mi-Arid
• " • 1.64 1
(. ~ ~ ( . J
':
I
Z
~'" ~. . ).. •
•
~
•
•
•o
.':..~,
.l
.~ • ,~
Arid
1-89
2
("
~''~
~ ' ~ ~. \" :'~',.."~'
~ i l, '
~".'.'-. IA ~ IB ~'.
i~. 6 tTanzania~
.~. In la|.49nd . 0o~o~o . "" ." ~:oast • '
~
;'~.:~. .. (
• -'~ ran,~
•
I
" '"%' "~ "~ - -
:
~"~° ~~ i ~ °
:oo:
1.74
10°S
I •":':':.'i:.-'i.;.:-i.:.-'.:.:~:!:~:i;.:.:-ii!~.".:~.'
~s.:i
0
I
F
50
I
100
I
150
I
200 k m
!
":J~
::~::!:. ,
fm %~'
l " '~~
//
3"0
O
+~
0J
¢N
.o
r-
2"0
C
r~
D
u
1-1
1"0
Io
I
III I I
1"0 1"4 1"8 2-2
10 : 2 year ratio.
O
O
O
o
6
I
0:
II 0
ii I-
o
<~ o
co i.i.
Z
0
l-
f,J
I,i.i
ne
E .,.I
,¢
I,,i.I
ne
,¢
¢.D
Z
,¢
ne
I.I.
,¢
l""
I.Ll
Q
Q
I,I..
X
°--
"o
¢-
e,L
o m
,¢
"-- co
o .~.
Z v ~ 00
o . @ <1
I I I I I
o I.D .e0
Q o cb o o
JO~.OgJ. u o ! ~ . : : ) n p a J e g J V
0.1 0.5 1-0 5-0 10-0 20"0
1.0 I I 1"0
I I I III
0.9 0.9
0-8 0.8
0"7 0-7
0
0-6 0-6
'E
CE 0.5 0-5
0-4 0-4
0-3 0-3
0-2 0-2
do
E
. m
6 ¢-
o O0
. m
co
I m
z
I I L. .E
~
I-
A r7 E •
0 r-
A
-0
v
l-
W
z
c-
I.I.I
I I I I i I I I I I O
0 0 0 0 0 0
I.L
°
T=.
X
b C
C e- _c O.
E
ra G,~
E E
'v'
I
e-
-E
I I-- ! I-
E
E E
O O
O')
A
m
..E
I I I I I I I I I I
Q 0 0 O O O
120
Y 2 ( 1 0 ) = Y 2 ( 4 0/) ~•/~'~/.
./ . " , ~
/
100
E 80
E /:/.
•. |
o~
>.
J •
O
,=¢ • I
I 60
CN
>- ~.:."
40
20
I I I I I I I
20 40 60 80 100 120 140
Y2 -- 10 year(ram)
A
E
E 3
"10
"¢-.
O
40
(13
..E
(J
t'-
"(3
O
¢J 2
03
I I I i
5 10 15 20
Period of record (years)
'W "
~,,,~
I
/ ,.. ./.!
"ii'~i
:~i~iii..'.
.iii,i~i,i.
"'iii~...i...
"""!ii-',~ ~i~
"
6
1.49
Inland
•
~o~o~a
.'"
g
/
l"anzaniar~
,~ o,o
~oast "
" 151~.
. oo
P
"1-64
I "
°•
Sol
"-:'°N/"° ~n~
~oa:
Fig. 4. 10 Y E A R : 2 Y E A R RATIO
4"0 m
3.0
2
=..
°..
c"
2"0
1.0
I I I I I !
1-0 1"4 1.8 2"2
10 : 2 year ratio
0-9 0"9
/
//
0:8 K" ,,,'
0"8
0-7 0"7
0-6 0"6
0
0-5 0"5
oo
'E
n.-
0-4 0"4
0-3 0"3
0-2 0"2
2~ 2~
G 00
i ..'"
!
I 26
0
o u~
" " - o oo o,
~
• • °o
°°°
©,
9
,.o
;..
17
Estate boundary . . . .
Heights in feet
Contours 100ft vertical interval
0 I 2 3Km
"I I I I
120
110
100
90
E
E
>-
80
¢t3
r~ Point
70
Areal rainfall
60
50
Y
40
30 I I I I I
--1"0 0"0 1"0 2"0 3"0 4-0
Reduced variable
110
100
90
A
E
E
>-
¢o
"E
80
Point rainfall
S
I//
//
I /
~./II ///
//7///X/'"
C~
70
///
60
Areal rainfall
50
///
40 /
30 I I I I I
-1-0 0"0 1"0 2"0 3"0 4-0
Reduced variable
120
110
1 O0
90
A
E
E
>-
80
c-
Point rainfall / /
/
n--
70
4/ / //
60
/ / \ Areal rainfall
/" /
50
/ /
/ /
/ /
/ /
40
30 I / t t I
-1"0 0"0 1-0 2-0 3"0 4.0
Reduced variable
0 5 10 15 20 km
! i J i I
120
/
/
II0
,/
100
/i
i /
/
90
/
A
/
/I/
E
E Point rainfall
>-
80
tY"
70
/
Areal rainfall
60
50
/
/
40
/I
/
I
30 I /I !', I I
-- 1-0 0-0 1" 0 2-0 3"0 4"0
Reduced variable
120
110 m
O
100
f
90
•4
E / /
E
v
>-
80 Point rainfall
/ /
¢..
/ /
n-
/ /
70
i /
/
60 m
Areal rainfall
50
/
40
///
/
30 I I I I I
--1"0 0"0 1"0 2"0 3"0 4"0
Reduced variable
2 5 10 20
130
I I I I
120
110
100 ;//
/
90
/ ,,
/ /
A
E Point r a i n f a l l
E
>-
N=
80 / /
t"
n¢
/
70 /
Areal rainfall
60- m--
50
/j
//
/
/_
40
I.
/ I I I I
30
--1-0 0-0 1-0 2"0 3"0 4"0
Reduced variable
Fig. 14 C O M P A R I S O N OF A R E A l ' A N D P O I N T D A I L Y R A I N F A L L --
F R E Q U E N C Y R E L A T I O N S H I P S FOR AN A R E A OF 300 km 2
Return frequency
2 5 10 20
130 I I I I
120
110
100
j /
E
E
>-
t-
oM
90
80 -
Point rainfall // /
/
/
/
/
/
/
t~
/ /
70 - / /
I ./
,,///~/
60 -
/ / / / // ~ Areal rainfall
/ /
50 -
///i//
t /I/
40 -
//
/
30 I /I I I I
-1"0 0"0 1"0 2"0 3"0 4"0
Reduced variable
120
110
1O0
90
A
E
E
>-
80
/
E:
t-
Point rainfall
/
//
70
/ /
/ /
60 B / /
/
/ / Areal rainfall
50
/
/
40
"/ /
30 i; / (' = = I
-1-0 0"0 1"0 2"0 3"0 4.0
Reduced variable
O
u O
O
LO
¢,q
d~
m
O
I
H
LL
fr O
m O
GO n-
O
I--
L~
tL
i Z
O
I
i-
ffq
E a
ee
-I
w-
tw
z
cJ
i
tw
U.
I--
I
O
O ggl
.a
U.
O•
0
0
z x/ <~ oo
O •
I I I I I
Q O
r,,. (D LD ~" O
Jo:l.3eJuo!:l.:)npgJ
I~aJv
T h e predicting of storm rainfall in East Africa: D. FIDDES, B.Sc., M.Sc., C.Eng., M.I.C.E., DIC.,
J. A. FORSGATE, B.Sc., and A. O. GRIGG: Department of the Environment, T R R L Laboratory Report 623 :
Crowthorne, 1974 (Transport and Road Research Laboratory). A simple method for predicting the
characteristics of storms for the design o f drainage structures in East Africa is described. The variation of 2 year
daily point rainfall, and the 10:2 year ratio for daffy rainfall, over East Africa are given in map form. Using
these, daily point rainfall for any return frequency can be calculated. To arrive at the design storm the daffy
point rainfall is adjusted using a generalised depth-duration equation and a graphical representation of the
variation of mean rainfall with area.
ABSTRACT
The predicting of storm rainfall in East Africa: D. FIDDES, B.Sc., M.Sc., C.Eng., M.I.C.E., DIC.,
J. A. FORSGATE, B.Sc., and A. O. GRIGG: Department of the Environment, TRRL Laboratory Report 623:
Crowthorne, 1974 (Transport and Road Research Laboratory). A simple method for predicting the
characteristics of storms for the design of drainage structures in East Africa is described. The variation of 2 year
daily point rainfall, and the 10:2 year ratio for daily rainfall, over East Africa are given in map form. Using
these, daily point rainfall for any return frequency can be calculated. To arrive at the design storm the daffy
point rainfall is adjusted using a generalised depth-duration equation and a graphical representation of the
variation of mean rainfall with area.