Rainstorm-induced flood risk assessment in developed urban area using a data-driven approach with watershed units

Abstract

Rainstorm flooding in developed urban areas has become a global focus. This study proposes a data-driven approach to urban rainstorm flood risk assessment. In contrast to the existing research, this study focuses on terrain watersheds as an assessment unit. Using Changsha as the study area, an inventory of 238 historical rainstorm flood locations was produced using automatic web crawling and literature data mining. Subsequently, an assessment model was developed based on a Bayesian algorithm and 16 influencing factors, and its accuracy was verified using a receiver operating characteristic curve. Because underground infrastructure is prone to backflow at its entrances and exits during rainstorms, the developed model was used to assess the backflow risk of two typical underground structures subjected to three rainstorm return periods: 5 (scenario 1), 10 (scenario 2), and 100 years (scenario 3). The conclusions are as follows: (1) The proposed method has a prediction accuracy of 88 % for flood risk. The most influential factors were H11 (proportion of impervious surface), H4 (mean elevation), and H1 (rainfall), contributing 52 %, 14.3 %, and 11.9 %, respectively. (2) Watersheds are classified into “Very Low,” “Low,” “High,” and “Very High” based on the degree of flooding impact, accounting for 83.6 %, 11.9 %, 3.9 %, and 0.7 %, respectively. Watersheds classified as “Very High” are mainly distributed in the central region. (3) A total of 48 subway stations (7.9 % of the total) and 148 underground parking lots (6.5 % of the total) in the study area are located in “Very High” risk areas. (4) Compared to that in scenario 1, the proportion of underground entrances and exits with a “Very high” protection level in scenario 3 increased by approximately 10 %. In conclusion, this framework can assist urban planners in understanding the risks of urban flooding and mitigating potential flooding impacts.