Environ Sci Pollut Res Int. 2026 Jul 27. doi: 10.1007/s11356-026-38077-0. Online ahead of print.
ABSTRACT
Floods are among the most devastating disasters, posing significant risks to communities and infrastructure, particularly in tropical regions where hazard mapping is often constrained by limited data availability. This study applies an integrated geospatial and machine learning (ML) approach to improve flood hazard assessment in a mountainous tropical region of Aceh Jaya, Indonesia, with the aim of evaluating model performance and identifying dominant causative factors linked to spatially targeted mitigation strategies. A set of ten flood causative factors, together with historical flood inventory data, was analyzed using four ML algorithms: Random Forest (RF), Support Vector Machine (SVM), Boosted Regression Tree (BRT), and Generalized Linear Model (GLM). Flood hazard maps were classified into five levels, ranging from very low to very high susceptibility. High to very high hazard zones cover approximately 14-22% of the study area and are primarily concentrated in low-elevation downstream areas. Model evaluation using Area Under the Curve (AUC), True Skill Statistics (TSS), correlation, and deviance indicates that RF achieves the highest predictive performance (AUC = 0.983; TSS = 0.92). The results consistently identify elevation as the dominant controlling factor, underscoring the influence of terrain-driven hydrodynamic processes. The coherence between model outputs, underlying physical mechanisms, and observed spatial patterns strengthens the basis for delineating flood hazard zones and informing mitigation priorities. This integrated approach enhances the applicability of the study by supporting evidence-based decision-making and enabling more targeted flood risk management in data-limited tropical regions, with a transferable framework that can be readily applied to similar mountainous settings.
PMID:42509508 | DOI:10.1007/s11356-026-38077-0
