A quick method of early landslide identification based on dynamic susceptibility analysis using M-SVM method: a case study

Landslides are irreversible and highly destructive disasters, so the early identification of landslides is necessary to minimize damage and the loss of life. Landslides are especially common in the mountainous areas of southwest China, where the extreme conditions make the exploration and early identification of landslides difficult. Therefore, this paper proposes a new early landslide identification method based on dynamic landslide susceptibility analysis, extracting accumulative deformation of the land over the past 5 years by SBAS-InSAR (Small Baseline Subset InSAR) technology, divided into 15 stages based on time and modeled using these 15 dynamic conditioning factors at different periods and conventional conditioning factors to get 15 landslide susceptibility maps using a machine learning algorithm. From this, the gradient change index of landslide susceptibility of each grid is calculated, and a gradient change level map of landslide susceptibility is established. In the map, the grids are divided into five susceptibility levels, i.e., very high (VH), high (H), moderate (M), low (L), and very low (VL). The VH and H levels were used for the early identification of landslide prone areas. Thirteen areas exhibiting early signs of landslides were identified, and eight were confirmed through field surveys. This procedure is of great significance as a new method that fully utilizes the change rate of susceptibility values to identify landslides.