The Machine Learning-Based Mapping of Urban Pluvial Flood Susceptibility in Seoul Integrating Flood Conditioning Factors and Drainage-Related Data

In the last two decades, South Korea has seen an increase in extreme rainfall coinciding with the proliferation of impermeable surfaces due to urban development. When underground drainage systems are overwhelmed, pluvial flooding can occur. Therefore, recognizing drainage systems as key flood-conditioning factors is vital for identifying flood-prone areas and developing predictive models in highly urbanized regions. This study evaluates and maps urban pluvial flood susceptibility in Seoul, South Korea using the machine learning techniques such as logistic regression (LR), random forest (RF), and support vector machines (SVM), and integrating traditional flood conditioning factors and drainage-related data. Together with known flooding points from 2010 to 2022, sixteen flood conditioning factors were selected, including the drainage-related parameters sewer pipe density (SPD) and distance to a storm drain (DSD). The RF model performed best (accuracy: 0.837, an area under the receiver operating characteristic curve (AUC): 0.902), and indicated that 32.65% of the study area has a high susceptibility to flooding. The accuracy and AUC were improved by 7.58% and 3.80%, respectively, after including the two drainage-related variables in the model. This research provides valuable insights for urban flood management, highlighting the primary causes of flooding in Seoul and identifying areas with heightened flood susceptibility, particularly relating to drainage infrastructure.