摘要Climate change has intensified extreme rainfall events,challenging progress toward SDG 11's urban resilience targets.Current assessment methods often neglect dynamic recovery processes and regional precipitation disparities.We propose a three-phase framework combining interpretable machine learning(ML)and factorial experiments,using the Prep_shock index that integrates standardized rainfall intensity,capital exposure,and historical probability,to evaluate the dynamic resilience of 220+Chinese cities from 2019 to 2022.Key findings reveal that:(1)The Prep_shock index effectively elimi-nates north-south precipitation biases,identifying Shandong coastal cities and Yangtze River Delta city clusters(36.2％)as high-resilience areas,in contrast to Henan Province.COVID-19 exacerbated systemic risks in megacities,undermining their capital protection capacities.(2)Spatial diagnostics classify 75.6％of the cities into Quadrant Ⅲ(the balanced resilience category),with recovery times decreasing from the west to the east.Super-large cities like Zhengzhou(2021)exhibited criti-cal recovery deficiencies(Quadrant Ⅳ).(3)Interpretable ML models(XGBoost/EBM)identify redundancy as the dominant resilience driver-robustness governs baseline resilience,while recovery relies on emergency support(for example,hospital beds density and fiscal inputs)and redundant infrastructure(for example,road network density).(4)Factorial experiments reveal optimization trade-offs:simultaneous enhancement of rapidity and redundancy diminishes their individual benefits,necessitating context-specific prioritization.The study advances dynamic resilience assessment methods and proposes quadrant-specific strategies for tailored urban adaptation.