Optimizing runoff prediction in plains river network areas: a hybrid XAJ-BP modelling framework with basin structure division

The plain water network area is characterized by its flat terrain and a dense crisscrossing of rivers. The runoff process in this region is significantly influenced by human activities, resulting in a complex relationship between water storage and discharge. The application of basin hydrological models and modelling methods has significant limitations in plain water network areas. In this study, the Hekou Basin, located in the Taihu Lake region, was selected as the study area, and the water system was extracted using remote sensing images. A series of algorithms such as multi-flow-direction grid calculation and flow direction definition of special hydrological elements were used to realize the division of sub-basins in the plain water network area, and the topological relationship of hydrological spatial elements was established. The XAJ-BP model was constructed by coupling Xinanjiang (XAJ) model and Back Propagation neural network (BP) model. The rainfall-runoff process in the Hekou Basin from 2019 to 2021 was simulated, and the simulation results were compared with those obtained using the standalone XAJ and BP models. The results show that the basin structure division method based on remote sensing imagery effectively captures and preserves the complex water network structure and topological relationships in plain river network areas. The XAJ-BP model demonstrated strong performance in simulating the rainfall-runoff process in this region (NSE = 0.89|0.85), outperforming both the standalone XAJ model (NSE = 0.67|0.71) and the BP model (NSE = 0.82|0.80). This study indicates that the improved basin structure division method is effective, and the coupled model can improve the simulation accuracy of the runoff process in these areas, which is of great significance for the treatment of water problems in plain water network areas.