Flood susceptibility prediction using tree-based machine learning models in the GBA

Abstract

The Guangdong–Hong Kong–Macau Greater Bay Area (GBA) frequently suffered from floods accompanied with typhoons. This study developed a framework for evaluating flood susceptibility in the GBA using tree-based machine learning (ML) and geographical information system techniques. Based on the flood inventory, tree-based models, namely random forest, gradient boost decision tree, extreme gradient boosting, and categorical boosting considering topography, exposure, and vulnerability as influential factors, were used to train and test ML models, and the trained models were then used to predict flood susceptibility. All tree-based ML models achieved good performance, with accuracy values greater than 0.79. The categorical boosting model performed the best than other models to predict flood susceptibility. The flood susceptibility maps showed that more than 16% of the areas of the GBA were classified as having high flood susceptibility, and almost 70% of the historical floods were located in areas with high flood susceptibility. The model interpretation of the summary of Shapley additive explanation values indicated that the influential factors of elevation, population density, and typhoon intensity had a strong influence on flood susceptibility. The obtained spatial flood susceptibilities provide suggestions for flood disaster mitigation in the GBA.