An XGBoost-SHAP approach to quantifying morphological impact on urban flooding susceptibility

Abstract

Urban flooding risks, often overlooked by conventional methods, can be profoundly affected by city configurations. However, explainable Artificial Intelligence could provide insights into how urban configurations affected urban flooding. This study, taking entered on Shenzhen City, deploys an XGBoost, integrating SHapley Additive exPlanation and Partial Dependency Plots, to assess how urban morphology influences urban flooding susceptibility. The models and strategies presented in this study aimed to adapt to extreme storms from the perspective of city spatial configuration planning. The findings underscore the varying impact of disaster variables on urban flooding, with morphological attributes becoming highly significant during severe inundations. In the analysis, mean building volume emerged as a pivotal parameter, with a mean SHAP value of 0.0107 m and a contribution ratio of 9.70 %. The study indicates that building configurations should be optimized to minimize urban flooding risks. It is recommended that the Mean Building Volume (MBV) be maintained within the range of 1.25 km3 to 2.5 km3, and the Standard Deviation of Building Volume (SDBV) be kept below 2.814 km3. By harnessing explainable Artificial Intelligence algorithms, this study offers insights into the intricate relationship between urban forms and flood risk, thereby informing the development of effective urban flooding adaptation strategies.

Keywords

• Urban flooding
• Building configuration
• XGBoost
• SHapley Additive exPlanation
• Explainable artificial intelligence
• Adaptive spatial planning