A novel hybrid framework of high-resolution flood susceptibility mapping in ungauged mountainous regions

Flood susceptibility mapping in mountainous regions remains a critical tool for mitigating escalating flood risks in the context of climate change. However, its accuracy is constrained by data scarcity of flood inventory maps and uncertainties in flood susceptibility mapping. This study addresses these gaps by introducing a novel hybrid framework that integrates a physically based Topography-based Subsurface Storm Flow (Top-SSF) model, a terrain-processed Height Above Nearest Drainage (HAND) model, and Random Forest (RF) model to achieve high-resolution (15 m) flood susceptibility mapping. Trained and tested on data from 80 gauged catchments, the hybrid framework demonstrated excellent performance, with training and test AUC of 0.995 and 0.992, respectively. Its robustness and applicability were subsequently validated against two independent historical flood datasets spanning 2000-2024, yielding high AUC values of 0.895 and 0.969, respectively. Sensitivity analysis identified slope, Topographic Position Index (TPI), and Topographic Wetness Index (TWI) as key drivers, collectively contributing 80.46% to flood susceptibility. Finally, we used the hybrid framework to produce the first high-resolution flood susceptibility map for the Southwest Mountainous Region of China, with potential to significantly enhance early warning systems and improve flood management effectiveness. • A novel hybrid framework of high-resolution flood susceptibility mapping was proposed. • Slope, TPI, and TWI are identified as the major factors affecting flood susceptibility. • The first high-resolution flood susceptibility map for the Southwest Mountainous Region of China was made