Report No FR0217

STOCHASTIC GENERATION OF RAINFALL TIME SERIES

FR0217

Dec 1991

SUMMARY

I OBJECTIVE

To enhance existing techniques for the temporal modelling of rainfall inputs to combined and storm sewer systems and, where necessary, to develop new procedures.

II REASONS

Rainfall time series provides an effective means of assessing the hydraulic performance and pollution impact of sewer systems by quantifying frequency, rates, volumes and durations of discharges. The methods currently available to the engineer do not meet all of these demands. For example, design rainfalls only represent extreme conditions and so do not provide an indication of the day to day performance of an existing sewerage system. The currently used one-year Time Series Rainfall enables the assessment of storm discharges which occur more frequently than once per year. However, the available one-year series does not contain extreme events nor does it accurately represent all geographical areas. Other recent research has made use of long local historical records of rainfall for simulation of spill volumes and has, therefore, overcome the problems of regionality. However, these techniques are labour intensive and rely on the availability of a local source of rainfall data.

III CONCLUSIONS

The Neyman-Scott Rectangular Pulses rainfall model has been identified as particularly suitable for the proposed application. The model has been successfully fitted to hourly and daily rainfall data for many UK sites and validated by comparing historical and simulated rainfall statistics not used in the fitting procedure.

The model parameters have been regionalised to enable hourly rainfall data to be generated at sites lacking historical data. A disaggregation procedure has been developed and tested to enable the generated hourly data to be disaggregated to five minute data.

IV RECOMMENDATIONS

The Stochastic Rainfall Generator (SRG) model as developed by the University of Newcastle upon Tyne has been shown to provide a satisfactory statistical fit to historical rainfall data. However, in order to further enhance the confidence of the typical water industry user WRc recommend that further work be undertaken to use the model to generate long chronological records at selected additional locations where further historic rainfall data are available. The statistics of the simulated and historic data may then be compared to fully demonstrate the model's capabilities. WRc propose to carry out this work during the early part of 1992 under FWR project F-1901. Furthermore, the practical performance of the model needs to be assessed by comparing the responses obtained when historic and simulated data sets are routed through selected flow simulation models of sewer systems known to exhibit pollution and flooding deficiencies. Subject to satisfactory results from the foregoing, the stochastic rainfall generator model should be developed into an easily usable software package. The function of this software would be to allow the production of a rainfall time series by non-specialists which would be statistically equivalent to a historical time series for a specified location. Parameter estimation should be achieved by input of the following variables:

The above information could be supplemented by the mean rainfall for each calendar month at the site, or more detailed rainfall records, if available.

Any software development should also include the disaggregation technique to convert the generated hourly rainfall into five minute values for convenient input to flow simulation models.

In order to achieve greater computational efficiency in sewer system simulation, further procedures should be developed to extract the critical events for a particular application from the generated time series. In addition, more work should be carried out on the use of simplified simulation models to allow the full use of a chronological long-term rainfall time series.

V RESUME OF CONTENTS

This report describes the development of a regionalised stochastic rainfall time series model for the UK. It is primarily intended for use with models which simulate the flow and quality regimes of sewer systems. The report describes the model development and is not a User Manual for the model. Further work is necessary to achieve a user friendly software product.

The reasons for the choice of model are given in Section 2. A detailed description of the model is given in Section 3, together with a tested procedure for fitting it to hourly data.

Section 4 describes the method for fitting the model to daily data, and Section 5 contains an extreme value analysis of the model. In Section 6 a regionalisation procedure is developed So allow the model parameters to be estimated at sites lacking data. Section 7 introduces the disaggregation approach.

Conclusions and recommendations arising from the work are presented in Sections 8 and 9. Details are provided of the historic rainfall database used for model development, together with test results for the model fitting procedure in the Appendix.

Copies of the report are available from FWR, price 35.00, less 20% to FWR Members.