Geospatial Analysis of Landslide Susceptibility Using  Weight of Evidence, Frequency Ratio and Information Value Methods in the Thotne Khola Watershed of Okhaldhunga District, Nepal

Abstract

Landslides constitute a significant destructive threat in Nepal's mountainous regions, frequently resulting in fatalities, damage to structures and infrastructure, and disruption of the livelihoods of local inhabitants. This study aims to develop and validate a landslide susceptibility map of the Thotne Khola watershed in the Okhaldhunga District, Nepal. This study used Frequency Ratio (FR), Weights of Evidence (WoE), and Information Value (IV) models to prepare landslide susceptibility maps (LSMs). An inventory map of 234 landslides was created by the interpretation of Google Earth images and fieldwork, thereafter divided into 80% training data and 20% testing data. Fifteen conditioning factors contributing to landslides, including aspect, slope, plan curvature, profile curvature, geology, dominant soil, land cover, rainfall, distance to stream, distance to road, NDVI, relative relief, Sediment Transport Index (STI), Stream Power Index (SPI), and Topographic Wetness Index (TWI), were combined with a training dataset using GIS tools to generate the Landslide Susceptibility Maps (LSMs) for the study area. The area was then classified into five zones of landslide susceptibility: very low, low, moderate, high, and very high. The proportions of the entire area categorized as having very low to very high susceptibility for FR were 21.06%, 20.43%, 20.55%, 19.98%, and 17.99%, respectively; for WoE, they were 13.81%, 25.10%, 29.44%, 24.40%, and 7.25%, respectively; and for IV, they were 11.15%, 22.73%, 29.25%, 28.00%, and 8.87%, respectively. The resulting maps were further confirmed using the area under the curve (AUC) and landslide density index methodologies. The AUC values obtained from three models are 0.909 for FR, 0.902 for IV, and 0.871 for WoE.  The result shows that all the models performed strong prediction rates; however, the weight of evidence model has an AUC value less than 0.9.  The increased LDI values in the high and very high susceptibility categories further confirm the model's accuracy. The results of this study can assist local government in land use management, disaster preparedness, and mitigation strategies.