Urban pluvial flooding has emerged as a serious threat to environmental conditions and human lives. Identifying its key drivers is crucial for the mitigation of flood risks. Although previous studies have demonstrated that pluvial flooding is caused by both natural (e.g., topography) and anthropogenic factors (e.g., land cover condition), much less effort has been devoted to investigating the potential influence of three-dimensional building configuration on pluvial flooding. To shed some light on this topic, we first analyzed the linear relationship between the density of flooding hotspots and different potential drivers in a highly-urbanized city using Pearson correlation analysis. Next, we designed two random forest-based models to quantify the importance of various building metrics. The first model considers only common drivers, while the second one also includes different types of building metrics. Results indicate that the density of buildings, building congestion degree, and building coverage ratio have exerted considerable influence on the occurrence of pluvial flooding. For example, the root relative squared error of our enhanced model (28.36%) is lower than that of the baseline model (32.58%). Our findings are expected to provide practical guidance for the mitigation of pluvial flood risks from the perspective of three-dimensional urban planning. Moreover, this methodological framework can be further applied to the analysis of flooding in many other regions.
Keywords: Random forest; Three-dimensional building configuration; Urban pluvial flooding.
Copyright © 2020 Elsevier Inc. All rights reserved.