Decoding urban flood resilience in the Henan section of the Yellow River Basin: Insights from an XGBoost-SHAP analysis.

Abstract

Urban flooding has become an escalating challenge to urban safety and sustainable development, especially in rapidly urbanizing and climate-sensitive regions. This study evaluates urban flood resilience across ten cities in the Henan section of the Yellow River Basin, China, by proposing a novel assessment framework that integrates the Resistance-Adaptability-Recovery (RAR) model and the Wuli-Shili-Renli (WSR) system, guided by the United Nations Sustainable Development Goals. To ensure objectivity in weight determination, EFAST is employed for global sensitivity-based weighting, followed by TOPSIS to derive composite resilience scores. Furthermore, an explainable machine learning approach combining XGBoost with SHAP is applied to identify and interpret the most influential resilience drivers, ensuring both predictive accuracy and interpretability. The results reveal a steady enhancement in regional resilience from 2010 to 2021, forming a spatial pattern of central radiation centered on Zhengzhou. Subsystem-level analysis indicates resistance dominance, adaptability convergence, and recovery lag. SHAP-based interpretation identifies insurance density, urban residents' per capita disposable income, radio population coverage, and infrastructure investment as the key determinants of resilience. These findings provide valuable insights for resilience-informed urban planning and targeted interventions in flood-prone regions.