Climate-driven flood hazard assessment in data-scarce mountainous basins using a GIS-based machine learning and...

Floods pose increasing risks in mountainous regions such as the Swat River Basin, where climatic variability, glacial influences, and data limitations hinder conventional flood risk assessment. This study introduces a hybrid framework that integrates explainable SHapley Additive exPlanations (SHAP)-based XGBoost for Global Climate Model (GCM) ranking, Random Forest (RF) ensemble modeling, and coupled hydrologic-hydraulic simulations (HEC-HMS-HEC-RAS) for multi-scenario flood hazard mapping. The approach provides an interpretable, data-driven, and physically based method for assessing climate-induced flood hazards in data-scarce basins. Daily precipitation, maximum temperature (Tmax), and minimum temperature (Tmin) data from eleven CMIP6 (Coupled Model Intercomparison Project Phase 6) Global Climate Models (GCMs) were bias-corrected using the linear scaling approach. These GCMs were ranked using XGBoost regression with SHAP interpretation, achieving high predictive accuracy (R²: 0.934/0.926 for precipitation, 0.953/0.949 for Tmax, and 0.947/0.943 for Tmin). A Multi-Model Ensemble built with RF regression further improved performance (R²: 0.74/0.71 for precipitation; 0.97/0.963 for Tmax; 0.965/0.958 for Tmin). These datasets were used to drive the HEC-HMS model, calibrated (1993-2013) and validated (2014-2019) with satisfactory results (NSE: 0.612/0.603; PBIAS: + 3.96%/- 6.75%). Flood frequency analysis using distribution fitting and the VIKOR method identified Log-Logistic (historical), Gumbel (SSP245), and GEV (SSP585) as optimal models. Simulated hydrographs for different return periods were input into a 2D HEC-RAS model to estimate flood depth, extent, and velocity. Flood hazard was quantified using a composite index of depth and velocity. Under SSP585, high to very high hazard zones expanded to 78% of the floodplain for the 100-year event, compared to 69% under historical conditions. This integrated, explainable, and scalable framework enhances climate-driven flood hazard prediction in complex mountainous terrains.