Multiple hydrological models comparison and an improved Bayesian model averaging approach for ensemble prediction over semi-humid regions

Abstract

In semi-humid regions, accurate prediction of flood processes is challenging. The goal of this study is to gain more insights into the runoff generation mechanism in semi-humid regions using multiple-model comparison method and explore the Bayesian model averaging (BMA) approach to improve flood prediction. This study compares seven runoff generation models for three semi-humid catchments in northern China. Flood events were classified into three categories, low-flow, medium-flow, and high-flow, according to flood peak flow in order to quantify the performance of each model and identify the dominant runoff generation mechanism for semi-humid catchments. Based on the performances of seven runoff generation models, three BMA schemes were used to integrate these models to compare the advantages of different combination methods. For the purpose of improving the performance of BMA over semi-humid regions, a physically based BMA approach, Green-Ampt-BMA approach (G-BMA), was proposed. In the G-BMA approach, an infiltration-excess flow module was added with the surface runoff calculated using the Green-Ampt equation. Considering the heterogeneity of precipitation and underlying surface characteristics, a distribution curve of infiltration capacity was introduced to simulate runoff processes. The results show that models with saturation-excess mechanism perform well for semi-humid catchments. The saturation-excess and infiltration-excess runoff exist simultaneously in a flood process over different catchments with different ratios of infiltration-excess to saturation-excess runoff. We found that the BMA approach effectively takes advantage of each model to provide more accurate forecasts. The physically based G-BMA approach performs better than the BMA approach for semi-humid regions with high ratio of infiltration-excess surface flow, especially in reducing flood peak error and forecast uncertainty.