Landslide susceptibility mapping and dynamic response along the Sichuan-Tibet transportation corridor using deep learning algorithms

Abstract

Landslides are one of the most serious natural hazards along the Sichuan-Tibet transportation corridor, which crosses the most complicated region in the world in terms of topography and geology. Landslide susceptibility mapping (LSM) is in high demand for risk assessment and disaster reduction in this mountainous region. A new model, namely Convolutional-Squeeze and Excitation-long short-term memory network (Conv-SE-LSTM), is proposed to map landslide susceptibility along the Sichuan-Tibet transportation corridor. Compared with conventional deep learning models, the proposed Conv-SE-LSTM adaptively emphasizes the contributing features of the conditioning factors by Squeeze and Excitation network (SE), and elaborately arranges the input order of the conditioning factors to utilize their dependence by long short-term memory network (LSTM). Considering the complex geological conditions and the wide range of the study area, the generalization and robustness of the proposed model are demonstrated from the perspective of global and sub-regions. Our proposed model yielded the best Area Under Curve (AUC) value of 0.8813, which is about 3%, 4% and 8% higher than that obtained by three traditional methods, respectively. An annual scale landslide susceptibility changes analysis method is also presented with an accuracy rate of 93.33%. The dynamic response relationship between landslide susceptibility and conditioning factors is revealed.