Author(s): Lakwon Choi; Ryotaro Shimizu; Makiko Iguchi; Masayasu Irie
Linked Author(s):
Keywords: RRI model; Optimization; Multiple evaluation functions; SCE-UA; Parameter optimization
Abstract: Climate change shifts a precipitation trend toward more torrential characteristics, which would increase the risk of flooding. In particular, a more accurate prediction of river discharge is needed for forecasting floods in rivers in a near real-time manner. This study used the Shuffled Complex Evolution method developed at The University of Arizona (SCE-UA) method to evaluate the parameters of the Rainfall-Runoff-Inundation (RRI) model, which is one of the distributed runoff models for torrential rainfall. In general, distributed runoff models have many parameters, and it is difficult to optimize parameters for various rainfall patterns. This study aims to propose procedures for optimizing parameters for multiple intense rainfall events with relatively short flood arrival times. First, the parameter optimization of each evaluation function was performed as using single evaluation function optimization; the model results with optimized parameters showed a fair agreement with the observed data. The use of a single evaluation function tended to fit to one of the rainfall patterns, but not to multiple rainfall wave patterns, because the characteristics of evaluation functions differ from each other and trade-off relations among peak shape, volume, and overall goodness of fit. Therefore, we introduced combinations of evaluation functions to optimize the parameters for multiple rainfalls. Overall, the combination of evaluation functions improved the accuracy of river flow forecasting more than the use of a single evaluation function. Among the combinations of evaluation functions, the combination of peak error, squared error, and correlation coefficient showed better performance of the peak flow rate, while the combination of RMSE and the peak error produced a better overall wave shape.
Year: 2024