GIS-based support vector machine modeling of earthquake-triggered landslide susceptibility in the Jianjiang River watershed, China

摘要: Abstract Support vector machine (SVM) modeling is based on statistical learning theory. It involves a training phase with associated input and target output values. In recent years, the method has become increasingly popular. The main purpose of this study to evaluate mapping power SVM in earthquake triggered landslide-susceptibility for section Jianjiang River watershed using Geographic Information System (GIS) software. river was affected by Wenchuan May 12, 2008. Visual interpretation colored aerial photographs 1-m resolution extensive field surveys provided detailed landslide inventory map containing 3147 landslides related 2008 earthquake. Elevation, slope angle, aspect, distance from seismogenic faults, drainages, lithology were used as controlling parameters. For modeling, three groups positive negative samples concert four different kernel functions. Positive include centroids 500 large landslides, those all 5000 randomly selected points polygons. Negative 500, 3147, slopes that remained stable during functions are linear, polynomial, radial basis, sigmoid. total, 12 cases susceptibility mapped. Comparative analyses probability area relation curves show both polynomial basis suitably classified data either or though function more successful. generated maps compared known centroid locations polygons verify success rate predictive accuracy each model. results further validated area-under-curve analysis. Group 3 polygons, along gave best 79.20% 79.13% under function. Of results, sigmoid least skillful when samples, regions (success rate = 54.95%; accuracy = 61.85%). This paper also provides suggestions reference selecting appropriate types modeling. Predictive could be useful hazard mitigation helping planners understand regions.