Landslide susceptibility mapping using Frequency Ratio, Random Forest, and XGBoost Models along NH-7 between Nandprayag and Vishnuprayag

Abstract

Landslides are a significant natural hazard in mountainous regions, causing substantial environmental and socio-economic damage. This study aims to evaluate landslide susceptibility using Frequency Ratio (FR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) models in the study area. A comprehensive landslide inventory and fourteen conditioning factors representing topographic, environmental, and anthropogenic characteristics were used as input variables. The FR model was employed to analyze the relationship between individual conditioning factors and landslide occurrence, while RF and XGBoost models were applied to capture complex nonlinear interactions among variables. Model performance was evaluated using ROC–AUC, success rate curves, and spatial validation approaches. The results indicate that land use/land cover (LULC), elevation, slope, and proximity-based factors such as distance to streams and roads play a crucial role in controlling landslide occurrence. Among the models, the RF model demonstrated the highest predictive performance (ROC–AUC = 0.880), followed closely by XGBoost (ROC–AUC = 0.867), whereas the FR model showed comparatively moderate accuracy (ROC–AUC = 0.761). The susceptibility maps effectively delineate high-risk zones, where a large proportion of landslides are concentrated within smaller spatial extents. Overall, the findings highlight the effectiveness of machine learning approaches in improving landslide susceptibility assessment and emphasize their potential application in hazard mitigation, land-use planning, and disaster risk management.