Spatial prediction of landslide susceptibility using integrated frequency ratio with entropy and support vector machines by different kernel functions

• Published in: Environmental earth sciences 2016-10, Vol.75 (20), p.1, Article 1344
• Main Authors: Chen, Wei, Wang, Jiale, Xie, Xiaoshen, Hong, Haoyuan, Van Trung, Nguyen, Bui, Dieu Tien, Wang, Gang, Li, Xinrui
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Biogeosciences; Earth and Environmental Science; Earth Sciences; Entropy; Environmental risk; Environmental Science and Engineering; Geochemistry; Geology; Hydrology/Water Resources; Land use; Landslides; Landslides & mudslides; Lithology; Original Article; Risk assessment; Spatial analysis; Terrestrial Pollution
• ISSN: 1866-6280; 1866-6299
• DOI: 10.1007/s12665-016-6162-8

The main purpose of the present study is to compare the prediction capability of frequency ratio (FR), index of entropy (IOE), and support vector machines with four kernel functions (LN-SVM, PL-SVM, RBF-SVM, and Sig-SVM) for landslide susceptibility mapping at Long County, China. For this purpose, a total of 171 landslide locations were collected from historical landslide reports, interpretation of satellite images, and field survey data. These landslides were separated into two parts (70/30): 120 landslides were randomly selected for training the models, and the remaining 51 landslides were used for validation purpose. Eleven landslide-related parameters were selected to produce landslide susceptibility maps, including slope aspect, slope angle, plan curvature, profile curvature, altitude, NDVI, land use, distance to faults, distance to roads, distance to rivers, and lithology. The landslide susceptibility maps were produced by FR, IOE, and SVM models, and these maps were validated and compared using area under the curve method. The results show that the RBF-SVM model has the best performance for this study area, while the success rate is 82.51 % and prediction rate is 77.83 %. For the other models, the results are as follows: the PL-SVM model (success rate is 82.44 %; prediction rate is 75.71 %), the FR model (success rate is 79.79 %; prediction rate 75.42 %), the LN-SVM model (success rate is 79.76 %; prediction rate is 74.76 %), the IOE model (success rate is 78.29 %; prediction rate is 74.01 %), and the Sig-SVM model (success rate is 75.22 %; prediction rate is 73.75 %). The results of this study are useful for land-use decision makers, landslide risk assessment and management study in this region, and other similar areas.