Landslide susceptibility mapping based on GIS and support vector machine models for the Qianyang County, China

• Published in: Environmental earth sciences Vol. 75; no. 6; pp. 1 - 13
• Main Authors: Chen, Wei, Chai, Huichan, Zhao, Zhou, Wang, Qiqing, Hong, Haoyuan
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.03.2016; Springer Nature B.V
• Subjects: Aerial photography; Biogeosciences; Earth and Environmental Science; Earth Sciences; Environmental Science and Engineering; Geochemistry; Geographic information systems; Geology; Geomorphology; Hydrology/Water Resources; Landslides; Landslides & mudslides; Lithology; Mapping; Original Article; Terrestrial Pollution; Vectors (Biology); China; China, People's Rep., Shaanxi Prov; China; Geographic information system (GIS); Landslide susceptibility mapping support vector machine (SVM)
• ISSN: 1866-6280; 1866-6299
• DOI: 10.1007/s12665-015-5093-0

The main purpose of this study is to produce reliable susceptibility maps using GIS-based support vector machine (SVM) models and compare their performances for the Qianyang County of Baoji City, Shaanxi Province, China. In this paper, with kernel classifiers of linear, polynomial, radial basis function and sigmoid, the four various types were applied in landslide susceptibility mapping. The important input parameters for the landslide susceptibility assessment were acquired from different sources. Firstly, 81 landslide sites were obtained by aerial photographs, earlier reports and field surveys. Then, the landslide inventory was randomly classified into two datasets: 70 % (56 landslides) for training the models and 30 % (25 landslides) for validation purpose. Secondly, 15 landslide conditioning factors were selected (i.e., slope angle, slope aspect, altitude, plan curvature, profile curvature, distance to faults, distance to rivers, distance to roads, NDVI, STI, SPI, TWI, geomorphology, rainfall, and lithology). Subsequently, with four types of kernel function classifiers based on landslide conditioning factors, landslide susceptibility parameters were obtained using SVM models. Finally, the rationality of landslide susceptibility maps was verified using the receiver operating characteristics with both success rate curve and prediction rate curve. The validation results showed that success rates for the four SVM models were 83.15 % (RBF-SVM), 82.72 % (PL-SVM), 81.77 % (LN-SVM), and 79.99 % (SIG-SVM). The prediction rates for the four SVM models were 77.98 % (RBF-SVM), 77.50 % (PL-SVM), 77.07 % (LN-SVM), and 76.08 % (SIG-SVM), respectively. The results showed that the RBF-SVM model had the highest overall performance.