Advanced Machine Learning Models for Flood Susceptibility Mapping: A Case Study in Thai Nguyen Province (Old), Vietnam

Abstract

Floods are considered one of the most dangerous natural disasters in the world. Due to its wideranging impact, floods cause significant damage to people, property, livelihoods, habitats, infrastructure, and economic development. To date, flood susceptibility maps are still considered an effective tool in flood damage management and prevention. Thus, this study proposed the use of three advanced machine learning models, including CatBoost, LightGBM, and NGBoost, to generate flood susceptibility maps based on the historical flood locations and 11 influencing factors. The ROC-AUC results were utilized to compare and evaluate the forecasting accuracy of these ML models. The LightGBM model demonstrated superior forecast performance and was selected to build the flood susceptibility map. This map indicates that the low-lying districts and concentrated river systems in Thai Nguyen province, such as Thai Nguyen City, Song Cong, Phu Binh, Dong Hy, and Pho Yen districts, fall in high and very high flood susceptibility zones. Mountainous districts situated on the edge of Thai Nguyen province, such as Vo Nhai, Phu Luong, and Dinh Hoa, fall in medium and very low flood susceptibility zones. The obtained map provides a visual view of future flood-prone areas, assisting local authorities in land-use planning and implementing effective priority investment strategies.