Landslide susceptibility modelling using GIS-based machine learning techniques for Chongren County, Jiangxi Province, China

• Published in: The Science of the total environment Vol. 626; pp. 1121 - 1135
• Main Authors: Chen, Wei, Peng, Jianbing, Hong, Haoyuan, Shahabi, Himan, Pradhan, Biswajeet, Liu, Junzhi, Zhu, A-Xing, Pei, Xiangjun, Duan, Zhao
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01-06-2018
• Subjects: Bayes' net; China; Landslide susceptibility; Logistic model tree; Radical basis function classifier; Random forest; Random forest; Landslide susceptibility; China; Bayes' net; Radical basis function classifier; Logistic model tree
• ISSN: 0048-9697; 1879-1026
• DOI: 10.1016/j.scitotenv.2018.01.124

•Bayes' net, RBF classifier, logistic model tree and random forest models were applied for landslide susceptibility modelling.•Information gain method was used to evaluate the relationship between landslide and conditioning factors.•Random forest model shows the better result in landslide prediction.
[Display omitted]
The preparation of a landslide susceptibility map is considered to be the first step for landslide hazard mitigation and risk assessment. However, these maps are accepted as end products that can be used for land use planning. The main goal of this study is to assess and compare four advanced machine learning techniques, namely the Bayes' net (BN), radical basis function (RBF) classifier, logistic model tree (LMT), and random forest (RF) models, for landslide susceptibility modelling in Chongren County, China. A total of 222 landslide locations were identified in the study area using historical reports, interpretation of aerial photographs, and extensive field surveys. The landslide inventory data was randomly split into two groups with a ratio of 70/30 for training and validation purposes. Fifteen landslide conditioning factors were prepared for landslide susceptibility modelling. The spatial correlation between landslides and conditioning factors was analyzed using the information gain (IG) method. The BN, RBF classifier, LMT, and RF models were constructed using the training dataset. Finally, the receiver operating characteristic (ROC) and statistical measures, including sensitivity, specificity, and accuracy, were employed to validate and compare the predictive capabilities of the models. Out of the tested models, the RF model had the highest sensitivity, specificity, and accuracy values of 0.787, 0.716, and 0.752, respectively, for the training dataset. Overall, the RF model produced an optimized balance for the training and validation datasets in terms of AUC values and statistical measures. The results of this study also demonstrate the benefit of selecting optimal machine learning techniques with proper conditioning selection methods for landslide susceptibility modelling.