Flood susceptibility mapping using machine learning models and remote sensing datasets has emerged as an effective approach for identifying flood-prone areas. The main objective of this study was to evaluate flood susceptibility using five ML algorithms: Extreme Gradient Boosting (XGBoost), Decision Tree (DT), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and Generalized Linear Model (GLM), as well as to assess the performance of their combination through an ensemble voting model (integrating RF, XGBoost, LightGBM, DT, and GLM). Flood extent data from 2000 to 2018 were obtained from the Global Flood Database (GFD), while ancillary spatial data related to climate, topography, hydrological, and land cover were collected from multiple sources. The individual models exhibited varying predictive performances, with XGB (AUC = 0.985), RF (AUC = 0.984), and LightGBM (AUC = 0.982) showing strong and statistically robust results. The DT model achieved moderate accuracy (AUC = 0.972), while GLM performed the least effectively (AUC = 0.879). Subsequently, the ensemble voting model outperformed all individual algorithms (AUC = 0.994), improving mapping accuracy and increasing reliability in identifying high- susceptibility areas. Overall, the results indicate that advanced ML techniques, particularly ensemble frameworks, are highly effective tools for spatial flood susceptibility analysis and risk management.
Rahimi et al. (2026) studied this question.