Runoff simulation and flood risk assessment of heavy rainfall-flood events in the Juhe River Basin, Beijing

Abstract

Heavy rainfall in mountainous regions typically triggers extreme flood events, and accurate simulation of the spatiotemporal evolution of floods is crucial for flood management and damage mitigation at the watershed scale. This study takes the Juhe River basin in the Beijing region as the study area. Rainfall-runoff processes were simulated using a CNN-BiLSTM-Attention model, and the spatiotemporal evolution of flood events was further modeled using the integrated flood modeling system (IFMS). On this basis, the flood risks of different heavy rainfall-flood events (HRFEs) under land use and land cover (LULC) scenarios from different periods were analyzed. Optimization of the key parameters of the CNN-BiLSTM-Attention model using a population-based algorithm significantly improved simulation accuracy, with NSE values exceeding 0.75 and relative errors in peak discharge remaining below 10%. The IFMS was able to effectively simulate the inundation extent and water depth of the “25.7” HRFE, with errors in simulated water depth remaining less than 0.40 m, thereby demonstrating strong applicability in the study area. Land use structure and the evolution of HRFEs jointly determine the spatial distribution and temporal dynamics of flood risk. In particular, the encroachment of urbanization on cropland has increased the exposure of residential areas in low-lying floodplains, thereby making these areas more susceptible to inundation during HRFEs.