Advancing rainfall-induced landslide hazard mapping: comparative study of semi-quantitative, statistical, hybrid, and deterministic models

Landslide hazard mapping is crucial for risk assessment, land-use planning, and disaster mitigation, especially in regions prone to rainfall-induced slope failures. This study compares five landslide hazard mapping approaches: Analytic Hierarchy Process (AHP), Frequency Ratio (FR), Logistic Regression (LR), AHP + FR + Penalty LR hybrid, and TRIGRS methods, applied to the Ardi Watershed, Thailand. Model performance was assessed based on methodology, data requirements, and predictive accuracy, using AUC, TPR, FPR, precision, and F1. TRIGRS showed the highest performance (AUC = 88.97%, TPR = 61%, FPR = 11%, precision = 67%, F1 = 64%), effectively modelling rainfall infiltration and slope instability. The AHP + FR + Penalty LR hybrid also performed well (AUC = 81.53%, TPR = 67%, FPR = 29%, precision = 23%, F1 = 34%), combining expert judgment with statistic. Standalone, the AHP, FR, and LR yielded lower accuracy (AUC = 68.03%–70.57%), and are suitable for preliminary or data-limited assessments. The study emphasizes the importance of rainfall classification and proper zonation techniques for producing a Landslide Hazard Map. Overall, findings underscore the need to choose hazard mapping methods tailored to local data availability and triggering factors. The study concludes with method-specific recommendations and the importance of modelling frameworks, particularly under climate change.