Probabilistic flood susceptibility mapping using explainable AI for the Western United States

Abstract Floods are among the most consequential environmental hazards, threatening lives, infrastructure, and economic stability. In the western United States, increasing variability in streamflow and shifting hydrologic regimes have heightened concerns about flood susceptibility and long-term water security. This study aims to predict flood susceptibility and analyze multi-decadal trends in extreme streamflow events across 1,088 United States Geological Survey flow-monitoring stations. We employ an explainable artificial intelligence (XAI) framework, integrating a random forest model with SHapley Additive exPlanations (SHAP), to identify and interpret the dominant drivers of flood susceptibility. The XAI model achieved a high predictive performance (Area Under Curve = 0.92), offering a reliable probabilistic tool for hazard mitigation and emergency planning. SHAP analysis showed that low channel elevation, high discharge variability, and flatter slopes were the strongest predictors of flood susceptibility, jointly accounting for 53% of model importance. Additionally, a statistically significant positive trend in annual mean discharge emerged as another important predictor (7% SHAP contribution). Seasonal analysis further revealed that streams with lower summer and spring flow thresholds were particularly susceptible, as reduced channel capacity and flow volatility heightened flood potential. Spatially, flood susceptibility hotspots (with predicted susceptibility >50%) emerged along the Pacific slope, extending from coastal Washington through northern California, where low elevations and flat terrain facilitate water accumulation. Additional clusters were identified in interior mountainous regions, including the Sierra Nevada and Cascades. To complement station-level findings, we developed a county-level composite risk index incorporating flood susceptibility, land cover, settlement patterns, population exposure, and social vulnerability. This probability-based XAI-driven framework provides a transparent and actionable tool for identifying at-risk areas, enabling targeted mitigation strategies and resilience-building efforts across the western United States.