An XGBoost-SHAP framework for identifying key drivers of urban flooding and developing targeted mitigation strategies

Abstract

Urban flooding is a multifaceted and severe issue, exacerbated by global climate change and urban expansion. Hence, it is imperative to investigate effective strategies for mitigating the urban flooding risk. This research proposes an innovative framework to recognize the driving factors and evaluate the impact of land cover changes on urban flooding, using a machine learning simulation model in the Guangdong-Hong Kong-Macao Greater Bay Area (GBA). The results indicated that approximately 19.8% of the investigated areas are exposed to high-risk, predominantly in the densely populated urban center of the GBA. In terms of identifying driving factors, impervious surface percentage (ISP) and fractional vegetation cover (FVC) are the primary driving factors of urban flooding. Converting impervious surface (IS) into green space (GS), the areas exposed at medium- and high-risk of flooding in the urban–rural fringe were significantly reduced, while no significant alterations were observed in the areas at very high-risk induced by urban flooding in the central city. By contrast, converting GS into IS significantly increased the areas exposed to very high-risk in the central city. In addition, we proposed optimization strategies for effectively mitigating urban flooding, based on the distribution and regional characteristics of flooding in the central city, rural–urban fringe and rural areas, respectively. This study provides urban planners and designers with valuable insights into how land cover management and green infrastructure can be leveraged to address urban flooding, thus offering practical guidance for sustainable urban development in rapidly urbanizing regions.