Spatiotemporal Dynamics and Multi-Scale Diagnosis of Urban Resilience to Typhoon Disaster Chains in Fujian, China

Abstract Coastal cities in southeastern China face increasing threats from typhoon-induced compound disasters (for example, torrential rainfall, urban waterlogging, and storm surges) that can cascade into interconnected disaster chains under climate change and rapid urbanization. However, dynamic multi-scale assessments of resilience to such compound disasters remain limited. This study develops an integrated framework that combines multi-scale geospatial analysis with explainable machine learning (XGBoost-SHAP). Using Fujian Province as a case study, we assess typhoon disaster chain urban resilience (TDCUR) in 2010, 2015, and 2020 across grid, administrative unit, and watershed scales, characterize spatiotemporal patterns, and apply XGBoost-SHAP as a post hoc diagnostic to summarize nonlinear indicator-TDCUR association patterns and their spatial concentration under the predefined TDCUR framework. The results indicate that: (1) Provincial TDCUR increased by 6.9% and regional disparities converged, yet major coastal cities experienced declining resilience despite strong economic development; (2) Resilience showed pronounced spatial polarization, with low-resilience cold spots expanding by 48% and clustering in the Xiamen-Quanzhou area; (3) Machine learning diagnostics indicate that typhoon-strong wind-storm surge sensitivity (B8), typhoon-rainfall-flood sensitivity (B7), and impervious surface proportion (A2) show the strongest model-based associations with the spatial variation of TDCUR and display significant interaction effects; and (4) SHAP-based spatial diagnosis identifies the Xiamen-Quanzhou-Fuzhou coastal belt and the Jinjiang Basin as priority areas with concentrated low TDCUR and high cumulative SHAP magnitudes. The proposed framework is transferable and can support spatial screening for targeted resilience actions in coastal regions, with implications for SDG 11.