Landslide mapping in Kerala is often conducted on spatially aggregated areas, limiting understanding of how landslide vary across different terrain types. This study addresses this gap by performing a comparative analysis of landslide susceptibility across five physiographically distinct catchments in Kerala, namely, Muthirapuzha, Pooyamkutty, Meenachil, Kuttiyadi, and Chaliyar using two different modeling approaches namely Random Forest (RF), representing machine learning techniques, and the Frequency Ratio (FR) method, representing conventional statistical approaches. Landslide susceptibility maps were developed by categorizing catchment areas into five classes, from Very Low to Very High . The conventional statistical approach, indicates that high and very high susceptibility zones together occupy 38.19% of Muthirapuzha, 20.21% of Meenachil, 22.12% of Kuttiyadi, 28.43% of Pooyamkutty, and 23.54% of Chaliyar. In contrast, the RF model produces a more differentiated spatial pattern, with high and very high susceptibility classes covering 31.49%, 25.42%, 38.58%, 32.99%, and 33.61% of the respective catchments. Model performance evaluation demonstrates the robustness of the RF approach, with AUC–ROC values of 0.87 (Chaliyar), 0.83 (Pooyamkutty), 0.93 (Muthirapuzha), 0.95 (Meenachil), and 0.96 (Kuttiyadi), and corresponding classification accuracies of 0.82, 0.76, 0.87, 0.90, and 0.88. Comparison with observed landslide inventories shows that actual landslide occurrences are comparable with the results obtained from RF method. These findings emphasise the influence of geomorphological, geological, and land-use characteristics on landslide occurrence in Kerala’s monsoon-dominated environment. The study highlights the capability of machine learning to capture complex, non-linear interactions among conditioning factors, offering improved tools for landslide hazard mapping and regional disaster risk management.
Akhila et al. (Mon,) studied this question.