Landslide susceptibility mapping with GIS in high mountain area of Nepal: a comparison of four methods

Abstract

Landslide susceptibility mapping (LSM) assists identifying and targeting landslide preventive measures, thereby minimizing potential losses. Multiple approaches are employed for LSM in various physiographic regions; however, their applicability has differed across studies, with limited understanding on the most suitable approach for LSM in high mountain areas. Hence, we conducted LSM in the Indrawati watershed, a high mountain area of Central Nepal, employing four approaches: frequency ratio, logistic regression, artificial neural network, and support vector machine. Nine landslide causal factors (slope, aspect, elevation, geological formation, proximity to river, proximity to road, land cover, soil type, and curvature) were considered for LSM. Rainfall-induced landslides were mapped by the on-screen digitization of satellite images and field observations. The landslides were randomly split into a ratio of 80:20 for training and validating the susceptibility maps. The LSMs obtained by four methods were then validated and compared using area under curve (AUC), kappa index, and statistical inferences (sensitivity, specificity, positive predictive value, negative predictive value, and accuracy). Our study showed that Eutric Cambisols, a class of soil type, has a strong association with landslide occurrence among the 52 classes of the nine causal factors. We found that the artificial neural network approach possessed the best prediction capability (AUC value = 86.9%) among the four methods, followed by logistic regression (85.6%), support vector machine (81.2%), and frequency ratio (80.1%) approaches. However, Kappa index and other statistical inferences suggested the support vector machine approach to be the second-best method. Overall, we found that the artificial neural network yields more accurate and reliable results and hence considered as a promising approach for susceptibility mapping in high mountainous region of Hindu-Kush Himalaya. The findings of this study might be useful for landslide analysts, development planners and decision-makers in conducting LSM and development planning in high mountain regions.