Evaluation of Landslide Susceptibility of the Ya'an-Linzhi Section of the Sichuan-Tibet Railway based on Deep Learning
Abstract The Qinghai-Tibet Plateau is one area with the most frequent landslide hazards due to its unique geology, topography, and climate conditions, posing severe threats to engineering construction and human settlements. The Sichuan-Tibet Railway that is currently under construction crosses the Qinghai-Tibet Plateau; there are frequent landslide disasters along the line, which seriously threaten the construction of the railway. This paper applied two deep learning (DL) algorithms, the convolutional neural network (CNN) and deep neural network (DNN), to landslide susceptibility mapping of the Ya’an-Linzhi section of the Sichuan-Tibet Railway. A geospatial database was generated based on 587 landslide hazards determined by Interferometric Synthetic Aperture Radar (InSAR) Stacking technology, field geological hazard surveys, and 18 landslide influencing factors were selected. The landslides were randomly divided into training data (70%) and validation data (30%) for the modeling training and testing. The Pearson correlation coefficient and information gain method were used to perform the correlation analysis and feature selection of 18 influencing factors. Both models were evaluated and compared using the receiver operating characteristic (ROC) curve and confusion matrix. The results show that better performance in both the training and testing phases was provided by the CNN algorithm (AUC = 0.88) compared to the DNN algorithm (AUC = 0.84). Slope, elevation, and rainfall are the main factors affecting the occurrence of landslides, and the high and very high landslide susceptibilities were primarily distributed in the Jinsha, Lancang, and Nujiang River Basins along the railway. The research results provide a scientific basis for the construction of the Ya'an-Linzhi section of the Sichuan-Tibet Railway within the region, as well as the disaster prevention and mitigation work during future safe operations.