A novel framework incorporating machine learning into GIS for flood susceptibility prediction of urban metro systems

Abstract

Floods have become increasingly destructive with climate change, resulting in the inundation of urban metro systems. This study complied with global data on flooded metro lines in recent decades. Based on these data, a framework incorporating machine learning (ML) with geographic information system (GIS) was developed to predict flood susceptibility in urban metro systems. To address the scarcity of subway flooding data, this study proposed a novel approach to generate a database for training and testing using ML and GIS. The 7.20 flood event in Zhengzhou, China, was analyzed as a case study. The optimal ML model was selected by comparing predicted flood states with recorded flooded metro stations. Flood susceptibility for the Zhengzhou metro system under future extreme rainfall scenarios was then predicted. Results demonstrated that the number of flooded stations and their flood susceptibility increased with rainfall intensity. These findings highlight the scale and vulnerability of metro systems, providing critical insights for developing resilient underground infrastructure.