Impacts of building configurations on urban stormwater management at a block scale using XGBoost

Abstract

Urban pluvial flooding has become a threatening hazard to ecosystem and human lives in recent years. Identifying its driving factors is essential for stormwater management. A significant underestimation of flood risk is caused by ignoring the coercive effects of building configuration on urban flooding. Data-driven machine learning algorithms provide insights into how building configurations affect urban flooding. This study utilized a cascade modelling chain comprising XGBoost, SHAP and PDP algorithms to determine the components of flooding risk and its impacts on local neighborhoods. It also provided practical guidance for city planners when planning from a building metrics perspective to mitigate pluvial flood risks. The result showed that building configuration plays an influential role in urban hydrology. The influence of Building Congestion Degree (BCD) on peak runoff became evident in an extreme rainstorm scenario, and especially in situations with low BCD. In addition, the Standard Deviation of Building Volume (SDBV) and Floor Area Ratio (FAR) had significant impacts on peak runoff. There was a significant inhibitory effect of SDBV on surface runoff when it is below 2.5 * 104 m3, as well as a suppression of flooding when the FAR is below the threshold of 2.8. The improvement strategy on building configuration for stormwater management can be carried out during urban renewal process or new district planning.