Assessment of susceptibility to rainfall-induced landslides using improved self-organizing linear output map, support vector machine, and logistic regression

Quantitative landslide susceptibility assessment is necessary for mitigating casualties, property damage, and economic loss. Identification of landslides and preparation of landslide susceptibility maps are crucial steps in landslide susceptibility assessment. Therefore, an optimal landslide susceptibility model is presented that is capable of producing accurate landslide susceptibility maps and assessing landslide susceptibility. To construct the optimal landslide susceptibility model, the effectiveness of the improved self-organizing linear output map (ISOLO), support vector machines (SVM) with four kernel functions (LN-SVM, PL-SVM, RBF-SVM, and SIG-SVM) and logistic regression (LR) was compared. Twelve landslide causative factors (namely, slope, slope aspect, elevation, curvature, profile curvature, plan curvature, slope length, topographic wetness index, distance to river, distance to road, distance to fault and annual maximum 24- and 48-h rainfalls) were used in this landslide susceptibility analysis. These models were applied to the Kaoping River basin in Southwestern Taiwan to assess its performance. Landslide inventory maps from 2008 to 2011 were collected. Data from the first three-year period were used for training and the remaining data was used for testing. The performance of the models was compared using accuracy and the area under the receiver operating characteristic curve as criteria. The results show that the RBF-SVM model outperformed the logistic regression in the study area. Using the RBF-SVM model, the landslide susceptibility under the annual 48-h maximum rainfall of various return periods were analyzed to assist local administrations and decision makers in disaster planning.