Assessing stage-dependent impacts of environmental factors on public-perceived flood risk: Insights from a GeoXAI framework

Abstract

Flooding is among the most destructive natural hazards, posing a major threat to sustainable development. Accurate assessment of public-perceived flood risk (PPFR) is essential for improving flood impact evaluation and informing management strategies. Taking the “7.20 Extreme Rainfall” event in Zhengzhou in 2021 as a case study, this study quantifies PPFR and its evolution across the pre-flood, during-flood, and post-flood stages using social media data. By incorporating a Geospatial Explainable Artificial Intelligence (GeoXAI) framework, this study assesses the nonlinear effects and spatial heterogeneity of natural, built, and socioeconomic environmental factors on PPFR. The results show that: (a) Precipitation, road density, building density, population density, and night light index are the key determinants of PPFR; (b) These five key factors exhibit highly similar threshold effects across the three stages. However, the magnitudes of their positive and negative effects show significant stage-specific variations, peaking in the during-flood stage and remaining weakest in the pre-flood stage; (c) High-value positive areas for natural and built environmental factors exhibit a relatively stable spatial pattern, whereas those of socioeconomic environmental factors shift significantly across stages. Based on these findings, a stage-specific and dynamically adjustable prioritization mechanism is recommended to enhance the precision and efficiency of urban flood management.