Flood dynamic monitoring and XGBoost-SHAP based risk assessment: A case study of the 23·7 extreme rainstorm in BTH region, China

Amid intensifying climate change, extreme rainfall has made urban flooding a major threat to city safety. Dynamic monitoring and risk assessment are vital for disaster management and recovery. Centered on the 23·7 Extreme Rainstorm in the BTH core region, we mapped inundation dynamics using a terrain-constrained MLC method with SAR images. We coupled XGBoost with the explainable AI algorithm SHAP to quantify spatial heterogeneity in flood risk and explore nonlinear drivers. The results show that: (1) The inundation area in the experimental region expanded greatly and peaked at 220.85 km 2 on 17 August, severely affecting 40% of roads, 67% of cropland, and several critical facilities; (2) The flood risk presented the spatial distribution characteristics of “multinuclear agglomeration - peripheral expansion”, with 8.3% of the very high risk area concentrated in the urban centers such as Beijing and Tianjin; (3) XGBoost-SHAP analysis identified RFI(SHAP=+0.709) and RVD(SHAP=+0.567) as major positive drivers, while ROD(SHAP=-0.318) had a significant risk suppressive effect. These findings inform targeted flood-risk prevention and enhance urban resilience across the BTH urban agglomeration. • Terrain-constrained MLC flood extraction with SAR tracked flood dynamics in BTH’s 23·7 Extreme Rainstorm. • Dongdian Flood Detention Basin inundation peaked on Aug 17 at 220.85 km 2 . • XGBoost–SHAP quantified spatial heterogeneity and nonlinear drivers of flood risk. • BTH flood risk showed multi-nuclear clusters with peripheral expansion and 8.3% was very high risk.