A COMPARATIVE STUDY OF EMPIRICAL, STATISTICAL, AND ANALYTICAL MODELS FOR LANDSLIDE SUSCEPTIBILITY IN KULLU DISTRICT, HIMACHAL PRADESH

Abstract

Kullu district of Himachal Pradesh, India, is highly susceptible to landslides due to its rugged terrain, complex geological conditions, and heavy seasonal rainfall. This study evaluates and compares three different modeling approaches for landslide susceptibility mapping—empirical (Frequency Ratio), statistical (Shannon Entropy), and analytical (Analytical Hierarchy Process)—to determine the most effective technique for predicting landslide-prone areas. A comprehensive landslide inventory comprising 428 landslide events and ten Landslide Conditioning Factors (LCFs), including slope, elevation, aspect, lithology, and proximity to streams, was used to develop susceptibility maps. The Frequency Ratio (FR) model demonstrated the highest predictive accuracy with an AUC value of 0.738, followed closely by the Shannon Entropy (SE) model (AUC = 0.735). The AHP model (AUC = 0.635) exhibited lower predictive performance, suggesting limitations in its weighting scheme for this region. Validation techniques, including AUC-ROC analysis and Success-Prediction Rate curves, confirmed the reliability and generalizability of the models. The findings emphasize the effectiveness of statistical and empirical models over analytical methods for landslide susceptibility assessment in mountainous terrains. The generated susceptibility maps are a valuable tool for disaster risk management, infrastructure planning, and sustainable development in the Kullu district. This research improves landslide prediction methodologies and supports targeted mitigation strategies for high-risk regions.