Author(s): Soon-Thiam Khu; Dawei Han; Edward Keedwell; Oliver Pollard
Linked Author(s): Dawei Han, Soon-Thiam Khu
Keywords: Genetic programming; Real-time flood forecasting; Updating; Rainfall-runoff; Direct; Iterative
Abstract: Data driven methods are highly flexible methods to extract information (mainly causal relationship) between available data and has been shown to be one of the effective methods in real-time rainfall-runoff forecasting. Moreover, these methods tend to be easy to implement, requiring “loose” coupling with existing rainfall-runoff models. Methods such as autoregressive (AR) or autoregressive integrated moving average (ARIMA) have been widely used but the main disadvantage of such approaches is the prior assumption of the form of error correlation. Genetic programming (GP), a relatively new evolutionary-based technique, can be used to generate a suitable expression linking the observations, simulation model results and the error in the simulation for the purpose of error correction. In this study, GP functions as an error correction scheme to complement a runoff forecasting model (PRTF) used by the UK Environment Agency (Southwest region) WRIP system. WRIP (Weather Radar Information Processor) is a real time flood forecasting system with spatial radar rainfall data as its primary input. PRTF (Physically Realizable Transfer Function) is basically a linear transfer function model with three controllable parameters. The proposed method is tested on several flashy catchment in Devon, UK. Hourly runoff forecasts of different updating intervals are performed for forecast horizons of up to six hours. The main objective of this study was to investigate the use of direct versus iterative updating. The results show that the proposed updating scheme is able to forecast the runoff quite accurately for all updating intervals considered and particularly for those updating intervals not exceeding the time of concentration of the catchment. These results form part of an ongoing study by the UK Environment Agency and the proposed method will be extended to longer forecasting horizon in the future.
Year: 2005