Key points are not available for this paper at this time.
• Developed an interpretable spatial machine learning framework. • Investigated the nonlinear correlation between variables and social resilience. • Described the spatial heterogeneity of variables and social resilience. With the frequent occurrence of extreme rainstorms in global cities, understanding differences in social resilience is crucial for building climate-adaptive communities. However, quantitatively analyzing the compound effects and interactions of social resilience determinants remains challenging. Here, we developed an advanced interpretable spatial machine learning framework to analyze social resilience across 2,221 blocks in Zhengzhou City, China, from 2005 to 2022. The framework integrates the Geographically Weighted Random Forest model with the SHapley Additive exPlanations model to address non-linearity, spatial heterogeneity, and interpretability simultaneously. Our findings reveal a 64.75% increase in social resilience in response to rainstorm disasters, alongside a 13.19% widening in disparities between blocks. We also observed a pattern of high resilience in the city center and lower resilience in peripheral areas. The probability of blocks maintaining their resilience levels was 66.93% for low (Ⅰ)-resilience blocks and 52.60% for high (Ⅴ)-resilience blocks. Key factors—such as economic vitality, population size, government services, and rainfall intensity—significantly influenced social resilience, with nonlinear associations and local threshold effects. The impact of factors like land use diversity and facility supply varied spatially. This research deepens the understanding of the compound effects of social resilience determinants and highlights the importance of tailoring flood intervention strategies to local conditions.
Gu et al. (Fri,) studied this question.