Integrating machine learning and explainable AI for flood susceptibility and driver analysis in the Amazon basin

Despite increasing flood frequency and intensity in the Amazon River Basin, flood susceptibility mapping remains limited by its large scale and data constraints. This study evaluates the performance of machine learning models for flood susceptibility mapping and identifies the main flood conditioning factors. Random forest (RF) and extreme gradient boosting (XGBoost) were used to generate the flood susceptibility maps based on flood extent data and 14 conditioning factors from meteorological, hydrological, topographical, and geological features. Shapley additive explanations (SHAP) were used to assess the most significant flood conditioning factors. XGBoost slightly outperformed RF, with both models showing similar results that remained stable after feature reduction. SHAP analysis revealed that land use/land cover (LULC) and precipitation are the key factors influencing flood susceptibility patterns. These findings provide valuable insights for improving flood susceptibility mapping in complex regions such as the Amazon River Basin.