Active landslide identification with a combined method of D-InSAR and random forest model

Early identification of disaster is a key technical problem in disaster prevention and mitigation. In this study, Zhouqu County, Gansu Province was taken as an example. Based on Sentinel-1A radar satellite orbit landing data from January 2018 to January 2019 and Sentinel-2 optical remote sensing image data from May 2021, D-InSAR technology was used to obtain surface deformation information in the study area, and Random Forest model was used to identify potential landslides. The results show that using the existing landslide data set, the random forest model can identify the potential landslide well. The distribution locations of potential landslide are all located in areas with large surface shape variables. The overall deformation occurred along the east-west direction, mainly distributed in the northeast and southwest directions of Zhouqu County, and overlapped with the potential landslide. The identified potential landslide point (Beishan landslide in Lijie Township) has an annual variable of 0.12 m, and the landslide occurred on January 18, 2021. This typical landslide case also confirms the effectiveness of the proposed method.