Automatic Identification of Earthquake Landslides Based on Multi-Source Remote Sensing Data —A Case Study of the Jiuzhaigou Earthquake

Earthquake is a critical triggering factor for landslide development. Rapid and accurate identification of historical and potential landslides after an earthquake is essential for post-disaster risk reduction. To satisfy current technical requirements for post-earthquake disaster prevention, this study combines Interferometric Synthetic Aperture Radar (InSAR) technique, hotspot analysis, and deep learning technique, using the Jiuzhaigou earthquake as a case study, to establish an automatic identification system for post-earthquake landslide hazards. Results show that hotspot analysis accurately identifies high deformation areas detected by InSAR, with an extraction accuracy and omission rate of 67% and 9.7%, respectively. A U-Net deep learning model trained based on post-earthquake historical landslides achieves an accuracy, recall, and intersection over union (IoU) of 0.80,0.72, and 0.51, respectively. The proposed framework enables rapid identification of post-earthquake landslides, providing significant support for disaster prevention and mitigation in earthquake-affected regions.