Landslide hazard analysis based on SBAS-InSAR and MCE-CNN model: a case study of Kongtong, Pingliang

A new multi-channel expanded convolutional neural network (MCE-CNN) model was proposed for landslide hazard analysis based on 'dynamic and static' + 'internal and external' factors. Firstly, 102 landslide samples were collected in Kongtong area, of the total samples, 75% were utilized for training, and the remaining 25% were applied for validation. At the same time, 14 landslide evaluation indicators including dynamic surface deformation rate were collected. Then, the result was validated using overall accuracy (OA) and area under curve (AUC) measurements. In order to further prove the effectiveness of the proposed method, comparative experiments were designed from two aspects, different models (MCE-CNN, SVM, RF) and different factors ('dynamic and static' + 'internal and external' and only on the static 'internal and external'). The results show that the AUC results of three models (MCE-CNN, SVM, RF) based on 'dynamic and static' + 'internal and external' factors were 95.4%, 83.7%, 84.1% respectively. The AUC results of three models (MCE-CNN, RF, SVM) based only on the static 'internal and external' factors were 92.3%, 82.9% and 81.8% respectively. Therefore, the results of hazard analysis by the method proposed in this paper are more reasonable and reliable, and the proposed method has great potential in practical landslide hazard analysis.