Report No FR0438
FURTHER DEVELOPMENTS OF THE
STOCHASTIC RAINFALL GENERATOR
To further develop the accuracy and range of synthetic rainfall
inputs for urban hydraulic and quality modelling applications including a scoping study
into the feasibility of developing a spatial variability rainfall model.
The existing SRG model suffers substantial loss of accuracy, in
respect of certain significant parameters, for events of return periods greater than one
in two years. Practical application procedures are likely to demand consideration of
return periods of up to at least one in ten years. Hence the model needs to be developed
to meet these requirements.
Spatial variability is a major characteristic of real rainfall events that is not
adequately considered in existing design procedures. A practical and effective way of
incorporating spatial variability into present design practice will remove a source of
uncertainty and inaccuracy.
During the technical development of an improved SRG the following
conclusions were drawn:
- Modification of the Neyman-Scott Rectangular Pulses (NSRP) model (which is the basis of
the SRG), in accord with meteorological observations, by allowing short duration
high-intensity raincells and long-duration low-intensity raincells to occur within the
same storm, provided a superior fit to historical extreme events when compared with the
original NSRP model.
- Validation of the modified model showed a remarkably good fit to annual maxima at the 1
and 2 hour levels of aggregation, but under-estimated the annual maximum at the 6 and 24
hour levels of aggregation. A more complex regionalisation procedure would inevitably lead
to an improved fit to the extreme values, however a more extensive data set would be
required representing different geographical regions in the UK.
- Overall the fit to the annual maxima was good given that the regionalisation procedure
was fairly simple and that the parameters of the modified NSRP model relate to physical
characteristics in the precipitation field and not directly to annual maximum rainfalls.
Independent testing of the Modified SRG (MSRG) was carried out by comparing the output
with historical rainfall series at four sites in the UK. The main conclusion from this
- The synthetic annual mean rainfall was accurate to within -18 and +9%.
- The percentage error between event depths over a range of return periods (RPs) was
generally between +/-10%. The exception to this was for the 2 year RP storms and for all
storms from the Exeter site, where errors ranged from -26 to +14%. Analysis of mean
intensity values indicated a tendency for over-prediction. Maximum intensity values were
usually well represented by the MSRG.
- Extreme storms appeared to be simulated well by the MSRG.
- When the rainfall events were run through sewer models, the MSRG showed a range of both
under- and over-estimated storage requirements (for specified spill frequency standards)
than the historical series.
- Percentage error in spill performance was considerably reduced when daily rainfall data
(rather than annual mean rainfall) were used in the model regionalisation procedure.
It is now considered that the MSRG can be used for a range of applications, for
example, pollution management and detention tank design. The following recommendations are
made to increase the level of confidence that the user can have when using any rainfall
- Long complete hourly rainfall historical data series, if available, will provide maximum
accuracy for any modelling study. The necessary length of this series will depend on the
user requirements. For pollution management studies at least 10 years sho uld be used, but
if specific extreme events are required, for example up to a 10 year RP, series in excess
of 20 years are necessary.
- Accuracy in the MSRG model can be improved if daily rainfall data are used to help
regionalise the model. Availability of this data is relatively wide spread and can be
obtained from the Meteorological Office. At least 20 years of daily rainfall data should
be obtained to ensure the data are representative of the site.
- The longer the series generated by the MSRG, the less likely that extreme years will
bias the results. To some extent this is the same for historical data. Twenty years are
recommended, but no less than 10 should be considered for most applications.
- The MSRG is not currently available in a supportable software package. However, it is
planned it will be incorporated into the commercially available STORMPAC(6) package to
allow its practical application by water industry users. The original SRG model, which is
currently part of the STORMPAC package, can still be used for applications requiring
rainfall with RPs of less than a 2 year RP (rather than up to a 10 year RP in the MSRG).
V RESUME OF CONTENTS
The report is divided into 3 parts:
Part 1 - Technical Development of the SRG
Part 1 (Sections 2 to 6) describes the technical development of
the SRG model. The modified model is defined in Section 2 and in Section 3 the model is
fitted to historical hourly data taken from one site. The model is regionalised in Section
4 and validation is described in Section 5. In Section 6 the Extreme values generated by
the model are compared to other representative values. Conclusions from Part 1 are given
in Section 7.
Part 2 - Additional Testing of the MSRG
In Part 2 (Sections 8 to 11) the MSRG is tested by assessing
selected characteristics of the synthetic rainfall and comparing them to equivalent
historical series; and by considering the accuracy of spill/storage predictions when MSRG
rainfall is used in simplified sewer flow models. Section 8 introduces the testing
programme and Section 9 describes the methodology adopted. The results are given in
Section 10 and the conclusions that can be drawn from the Part 2 testing are given in
Part 3 - Recommendations and Additional Developments
Part 3 contains the recommendations from the developmental and
testing work (Section 12) and provides a synopsis of the scoping study looking into the
viability of a spatial variability rainfall model (Section 13).
Copies of the report are available from FWR, price £35.00, less 20% to FWR Members.