LANDSLIDE SUSCEPTIBILITY EVALUATION BASED ON DEEP LEARNING ALONG KANGDING-LITANG SECTION OF CZ RAILWAY

The Kangding to Litang section of CZ Railway is located in the eastern edge of Qinghai-Tibet Plateau. The region is characterized by varied landforms,complicated geological structures and widely developed landslide disasters,which causes a serious threat to the planning,construction and future safe operation of the Kangding to Litang section of CZ Railway. Therefore,12 impact factors were chosen to be the evaluation indices. They include elevation,aspect,plane curvature,profile curvature,topographic relief,surface cutting degree,topographic wetness index,normalized difference vegetation index,stratum lithology,distance to fault,distance to river and distance to road. The landslide spatial database was constructed,and the deep learning convolutional neural network(CNN)model was used to evaluate the landslide susceptibility. According to the susceptibility index,the study area was classified into the following five grades: landslide extremely high-prone area(13.76%),landslide high-prone area(14.00%),landslide moderate-prone area(15.86%),landslide low-prone area(18.17%) and landslide extremely low-prone area(38.21%). The prediction performance was compared with the artificial neural network(ANN)model. The results show that the AUC value of the area under the ROC curve of the CNN model is 0.87,which is better than 0.84 of the ANN model,and the frequency ratio of the extremely high-prone areas is higher than the ANN model,so the CNN model has a higher predictive ability in this study area. The landslide extremely high-prone area and high-prone area are mainly distributed in the areas with relatively developed river,and the zones are distributed in the 2 km range along both sides of the Yalong River and other rivers. The results of landslide susceptibility well reflect the development and distribution of landslide hazards in the study area,which can provide a scientific basis for the construction of CZ railway and the work of disaster prevention and mitigation in the future safe operation.