Application of a GIS-based slope unit method for landslide susceptibility mapping along the rapidly uplifting section of the upper Jinsha River, South-Western China

Abstract

The objective of this study was to produce a landslide susceptibility map along the rapidly uplifting section of the upper Jinsha River. Firstly, a total of 40 landslides were identified in the study area from the interpretation of remote sensing (RS) and field survey data. Following landslide identification, ten variables including slope angle, slope aspect, curvature, land use, normalised difference vegetation index (NDVI), rainfall, lithology, distance to river, distance to fault, and Strahler’s integral value were selected as the influencing factors in landslide susceptibility mapping. All of the influencing factors were extracted by the slope unit. The Strahler’s integral value was used to represent the relationship between the rate of uplift and rate of denudation in each slope unit. Furthermore, three methods, including logistic regression, a support vector machine, and an artificial neural network, were applied to landslide susceptibility modelling. Five-fold cross validation, a statistical analysis method, and the area under the receiver operating characteristic curve (AUC) were used to compare the evaluation results of the three models. Finally, the variance-based method was used to find the key factors associated with landslides in the study area. The results show that the mean prediction accuracies of the logistic regression model, artificial neural network model, and support vector machine model were 80.47%, 87.30%, and 83.94% in the training stage, respectively, and 81.08%, 82.16%, and 83.51% in the validating stage, respectively. The mean AUCs of the three models were 88.16%, 93.96%, and 89.68% in the training stage, respectively, and 87.68%, 92.60%, and 89.88% in the validating stage, respectively. These results show that the artificial neural network model is the best model for evaluating landslide susceptibility in this study. The landslide susceptibility map produced by the artificial neural network model was divided into five classes, including very low, low, moderate, high, and very high, and the percentages of the areas of the five susceptibility classes were 17.23%, 28.32%, 22.73%, 16.73%, and 15.00%, respectively. Furthermore, the distance to river, slope aspect, lithology, and distance to fault are the most important influencing factors for landslide susceptibility mapping in the study area. Consequently, this study will be a useful guide for landslide prevention, mitigation, and future land planning in the study area.