GIS-Based Landslide Susceptibility Mapping in Gorkha Municipality: A Comparison of Frequency Ratio and Logistic Regression Models

Abstract

Gorkha Municipality, located in the hilly region of Nepal, has been experiencing an increasing number of landslides due to a combination of steep topography, high rainfall, and human activities such as road construction. These landslides damage the infrastructures resulting in economic losses. To address this issue, a landslide susceptibility map was developed to identify areas prone to future landslides. Thirteen causative factors were considered in the analysis, including Slope, Aspect, Curvature, Rock and Soil Types, Topographic Wetness Index, Stream Power Index, Distance to Streams, Normalized Difference Vegetation Index, Rainfall, Distance to Roads, Land Use Land Cover, and Distance to Faults. A total of 347 landslide and 347 non-landslide points were collected through field surveys and Google Earth. Two statistical GIS based models, Frequency Ratio (FR) and Logistic Regression (LR), were applied to assess landslide susceptibility. The results identified distance to roads as the most influential factor contributing to landslides in Gorkha Municipality. Pearson correlation and Variance Inflation Factor analyses and confirmed low multicollinearity among the causative factors. Model performance was evaluated using AUC-ROC, where the FR method achieved an accuracy of 0.819, while the LR method performed better with an accuracy of 0.909. This indicates that the multivariate approach (LR) is more effective than the bivariate method (FR) for this terrain. The results show that 15% of the area falls within the very high susceptibility zone, 19% in high, 19% in moderate, 20% in low, and 28% in very low susceptibility zones. These findings provide valuable insights for disaster risk reduction, land-use planning, and infrastructure development in Gorkha Municipality.