Comparative study on landslide susceptibility assessment of different models: a case study of alpine mountainous region in Xinjiang

Abstract

The complex engineering geological environment and unique climatic conditions in the alpine mountains of Xinjiang breed a large number of landslide geological hazards, and the accurate landslide susceptibility assessment (LSA) is of great significance to disaster prevention and mitigation. In this paper, based on historical landslide data and field geological survey, 4262 landslides were collected and analyzed, and 12 conditioning factors such as elevation, slope angle, slope aspect, curvature, topographic relief, lithology, road network kernel density, fault kernel density, land use type, vegetation cover, and snow cover were selected and through the independence test. 70% of the landslides were randomly selected as training samples, and the susceptibility of landslides in the alpine mountainous region was evaluated and compared using single model (Normalized Frequency Ratio (NFR), Information (I), Certainty Factor (CF)) and coupled model (Normalized Frequency Ratio-Logistic Regression (NFR-LR), Information-Logistic Regression (I-LR), Certainty Factor-Logistic Regression (CF-LR)), respectively. The remaining landslides were used as test samples to evaluate the accuracy. The main results show that the frequency ratio of landslide susceptibility level increases significantly from low susceptibility zone to very high susceptibility zone. The accuracy of the coupling model is greater than that of the single model, and that of the I-LR coupling model is the highest. The mean value of the coupled model was smaller than that of the single model, while the opposite standard deviation indicated that the prediction ability of landslide susceptibility was more vital. The six models have successfully evaluated the landslide susceptibility in alpine mountainous regions.