Comparison of probabilistic and expert-based models in landslide susceptibility zonation mapping in part of Nilgiri District, Tamil Nadu, India

Abstract

In the present research work, the landslide susceptibility zonation (LSZ) mapping was carried out for the landslide prone area Nilgiri hills, Tamil Nadu, India. The LSZ mapping was carried out using ten landslide influencing factors along with extensive field investigation. The geospatial database was prepared through integrated remote sensing, geographical information systems, and GPS technologies. The methods adopted for the present study are frequency ratio (FR) which is probabilistic and analytical hierarchical process (AHP) which is subjective and objective based model. The FR values were evaluated through evaluating relationship between causative factors and past landslide (training) locations. The FR values were considered as the base for assigning the weights in AHP method along with the subjective knowledge. The final LSZ map were derived through the spatial integration of all causative factors and classified as different susceptibility classes viz. very low, low, moderate, high, and very high. The prediction accuracy of final LSZ map were validated using past landslide (validation) locations using area under curve (AUC) method. The FR model shown the highest prediction accuracy with AUC value of 0.6279, while the AHP model shown the AUC value of 0.5620.