Landslide Susceptibility Mapping Using Frequency Ratio and Weight of Evidence Models in Purchaudi Municipality, Baitadi District, Nepal

Landslides are prevalent in the Himalayas, resulting in annual fatalities and economic losses during the monsoon season. Various studies are conducted for landslide susceptibility mapping in the Himalayas however many rural settlements are under the threat of landslide. The main objective of this study is to prepare the landslide susceptibility map by using the bivariate frequency ratio (FR) and weight of evidence (WoE) models. For this study, the landslide inventory map was initially developed allocating 70% of the data for training and 30% for testing by using Landsat-8, Google Earth, and Sentinel imageries. After a thorough examination of existing literature and extensive field research, a total of twelve potential factors that can cause landslides were selected. The success and prediction rate of both models were obtained by using the area under the curve (AUC) method. The findings show that the landslide susceptibility maps prepared by the weight of evidence (WoE) and the frequency ratio (FR) models are more or less similar and have similar prediction and accuracy rates. Both the WoE and FR models exhibit a success rate of 0.703. Similarly, the accuracy rates of the FR and WoE models are 0.777 and 0.775, respectively. Both models have an accuracy and prediction rate that exceeds 70%, making them suitable for evaluating landslide susceptibility in the same terrains of Sudurpaschim province. However, the frequency ratio model shows slightly better predictive ability compared to WoE. The landslide densities derived from both models exhibit a gradual increase from low to very high susceptibility class. Nevertheless, there is no significant difference in the success and prediction rates between the two models. This result may be useful as a reference for future studies on similar terrain in Nepal.