Landslide detection based on deep learning and remote sensing imagery: A case study in Linzhi City

Abstract

Landslides result in serious damage to economic and land resources. Automated landslide detection over a wide area for the study and prevention of geohazards is important. Linzhi is located in the southeastern part of the Tibetan Plateau, one of the most landslide-prone regions in China. In this paper, we utilize a deep learning approach in combination with remote sensing images to detect landslides in Linzhi City. SHAP-based interpretability analysis and exponential Weighted Method and Technique for Order Preference by Similarity to Ideal Solution (EWM-TOPSIS) method are employed to investigate the catastrophic factors that affect landslides and results of landslide detection in Linzhi City. The obtained results indicate that the model is basically accurate in landslide detection in the Linzhi area, and most of the evaluation indexes of the model training are above 80%. Moreover, vegetation cover and rainfall are important causal factors triggering landslides in Linzhi City. Our research will provide a reference for landslide detection in similar areas.