Deciphering flood vulnerability of Himalayan river basin using explainable machine learning and multivariate probabilistic analysis

Floods pose an escalating hazard in the Upper Ganga Basin (UGB), where rugged Himalayan terrain, intense monsoonal rainfall, and increasing human occupation amplify flood risk. This study presents an integrated assessment of flood susceptibility, vulnerability, and risk across the UGB using explainable machine-learning (ML) models and a multivariate exposure framework. Geomorphological, hydrological, land-surface, and socio-infrastructural conditioning factors were compiled and evaluated. Decision Tree-based ML classifiers, including Random Forest (RF), Extra Trees (ET), XGBoost (XGB), and LightGBM (LGBM), were trained and optimized to generate flood susceptibility maps, while model interpretability was assessed using Shapley Additive exPlanations (SHAP). The ML models demonstrated excellent predictive performance, with RF producing the most spatially coherent susceptibility patterns. Flood susceptibility is highest along valley bottoms and major river corridors and is primarily controlled by elevation, land-use/land-cover, topographic position, proximity to rivers, vegetation condition, and slope. Hydrological connectivity and antecedent moisture further reinforce susceptibility in narrow and deeply incised terrain. A multivariate vulnerability analysis integrating flood-affected areas, population exposure, settlement concentration, and road density reveals strong nonlinear and threshold behaviour in vulnerability escalation. The highest vulnerability occurs in the central and southeastern UGB, where dense settlements and limited connectivity coincide with extensive flood-prone areas. Overall, the results demonstrate that flood risk in the UGB arises from complex interactions between physical susceptibility and human exposure, highlighting the need for integrated flood risk management and sustainable land-use planning in Himalayan regions.