Ensemble flood simulation for the typical catchment in humid climatic zone by using multiple hydrological models

Abstract

Abstract. A good performance of hydrological model for flood simulation is of critical importance for flood forecasting. Taking Yandu River catchment, as the study area, three hydrological models (i.e. Xin'anjiang model, TOPMODEL, artificial neural network model) and a multi-model ensemble simulation method (i.e. entropy-based method) were applied to simulate the hydrological processes of 30 flood events occurring in 1981–1987. The performance of the ensemble members and multi-model ensemble simulation method was evaluated by comparing indicators of Nash-Efficiency coefficient, errors in root mean square, peak occurrence time, and relative errors of flood peak discharge, event runoff depth. Results show that the three hydrological models perform well for hydrological simulation of all 30 storm floods with Nash and Sutcliffe Efficiency coefficient of above 0.75 and relative error of less than 10 %. However, different model exhibits a difference in simulation errors of peak discharge and peak occurrence time. For example, BP model has the smallest error of 3.78 % for peak discharge simulation while that of Xin'anjiang model and TOPMODEL are 20.9 % and 24.7 % respectively. The entropy-based ensemble simulation method improved flood simulation accuracy to some extent for all evaluation criteria comparing to the three hydrological models. It is feasible to apply entropy-based ensemble approach for improving accuracy of flood forecasting in humid regions of China.