Incorporating landslide failure mechanisms into susceptibility mapping under climate extremes: A hybrid NFR-RF approach in the Western Ghats, India

Increasingly erratic monsoon behaviour has intensified extreme rainfall events in the Western Ghats, resulting in diverse landslide mechanisms that challenge conventional susceptibility modelling. Existing susceptibility models overlook landslide failure mechanisms and rely mostly on multi-decadal cumulative inventories that do not adequately represent slope behaviour under extreme climatic stressors. Addressing this gap, the present study developed a typology-integrated, hybrid framework to improve susceptibility prediction under climate-driven extremes. The objectives were to: (ⅰ) identify key conditioning factors and quantify their relative importance for different landslide failure mechanisms under extreme rainfall conditions, (ⅱ) compare type-specific and integrated models developed using a Normalized Frequency Ratio–Random Forest (NFR–RF) hybrid approach, and (ⅲ) quantify uncertainty and spatial patterns in the resulting susceptibility maps. An extreme-rainfall-triggered inventory comprising 943 debris flows, 1, 219 shallow slides, and 102 rockfalls was used. Ten conditioning factors, elevation, slope angle, slope aspect, profile curvature, Topographic Wetness Index (TWI), distance to streams, Normalized Difference Vegetation Index (NDVI), geology, distance to lineaments, and accumulated rainfall, were recoded using type-specific NFR values prior to RF modelling with a 70:30 training–testing split. Additionally, three integrated models were developed, Max, Mean, and All-landslide, assigning the highest value, averaging susceptibility, and combining all landslides without type distinction, respectively. Results show that slope angle is the dominant factor across all failure mechanisms, followed by elevation for debris flows, geology for shallow slides, and distance to lineaments for rockfalls. The shallow slide model shows balanced performance (AUC = 84.67%; FR = 95.37%), while the all-landslide model achieves the highest AUC (85.64%). Rockfalls exhibited the greatest uncertainty (SD = 1.338; mean SI = 4.502). FR accuracy for the Max, Mean, and All-landslide models reach 93.91%, 94.44%, and 93.39%, respectively. Spatially, 60.96% of the basin is classified as high–very-high susceptibility for debris flows, 48.64% for rockfalls, and 42.30% for shallow slides. The research demonstrates that typology-based hybrid modelling enhances interpretability and risk-informed land-use planning. The findings from this study could aid decision-makers in land-use planning and tailored risk-mitigation strategies for climate extremes in the Western Ghats and similar mountainous regions.