ABSTRACT The spatial prediction of landslide susceptibility (LS) is a very important tool for preventing losses caused by landslide disasters. Deep learning is a state‐of‐the‐art method that has been used to predict the spatial distribution of LS, but the influence of combining deep learning technology with a partition membership model and then applying the hybrid model to the spatial prediction of LS is still an important research question and challenge. The aim of this study is to design a hybrid model combining deep learning (DL) with partition membership (PM) for modeling LS. To validate our hybrid model method, Zixi County, which is located in Jiangxi Province, China, was selected as the experimental area. On the basis of the landslide inventory map, 233 landslide locations were identified, and fifteen environmental factors were analyzed on the basis of expert knowledge. To assess the effectiveness and superiority of the proposed PMDL model, the support vector machine (SVM), Hoeffding tree (HT), Naive Bayes (NB) and stochastic gradient descent (SGD) methods were selected as representative traditional models, and they were compared with the proposed PMDL model on the basis of the index of the AUC value. The results indicate that the designed PMDL hybrid model is more reliable and stable. The effectiveness of predicting landslide susceptibility is confirmed by the proposed hybrid model.
Lin et al. (Wed,) studied this question.