Investigating the influence of nonlinear spatial heterogeneity in urban flooding factors using geographic explainable artificial intelligence

Abstract

Urban pluvial flooding is one of the most significant environmental challenges impacting human society. Understanding the mechanisms through which geographical elements affect flooding is essential for developing effective flood mitigation strategies. However, due to limitations in current research methods, the nonlinear spatial heterogeneity of urban flooding factors remains underexplored. This study aims to design a novel framework based on geographic explainable artificial intelligence (GeoXAI) to investigate the nonlinear spatial heterogeneity of urban flooding factors in a case study of Guangzhou, China. In the attribution analysis of urban flooding susceptibility (UFS), a spatial statistical method and a conventional explainable artificial intelligence method were used for comparative evaluation with the GeoXAI method. The results reveal that: (a) flooding factors exert varying influences across different regions, although they generally increase UFS in the central-southern, western, and southeastern sectors of Guangzhou; (b) kernel normalized difference vegetation index and impervious surface density are dominant factors in urban flooding, with optimal thresholds for effectively mitigating flooding at above 0.25 and below 0.2, respectively; (c) GeoXAI demonstrates superior performance over traditional methods, offering enhanced model accuracy, more reliable interpretability, and better consideration of geospatial variables and spatial effects. These findings provide significant guidance for flood management in Guangzhou and underscore the broader potential of GeoXAI for disaster management in various regions.