Mapping of rainfall-induced landslide susceptibility in Wencheng, China, using support vector machine

Abstract

This study presents a framework for mapping rainfall-induced landslide susceptibility in the Wencheng area of Zhejiang Province, China, using support vector machine (SVM). Seven conditioning factors of elevation, slope, aspect, distance to roads, lithology, land use, and normalized difference vegetation index selected by correlation analyses, and two triggering factors of daily and cumulative rainfall data were employed as input data in the SVM modeling. The training dataset was constructed using 354 landslide inventories identified from field surveys between 1981 and 2010, and 354 points not prone to slides selected as positive and negative samples, respectively, based on a statistical method of factors. A fivefold cross-validation and receiver operation characteristic were applied to evaluate the performances of the SVM model. The summarized area under the curve result was 0.96, indicating that SVM demonstrated good and stable performance in mapping landslide susceptibility. Two practical cases were investigated: the Trami typhoon of August 22, 2013, and the Kong-Re typhoon of August 26, 2013, which brought heavy rainfall of 300 and 111 mm, respectively, to the Wencheng area. The resulting maps showed that SVM could provide a map with high probability values over small areas, demonstrating good properties of generalization. In addition, mapping of landslide susceptibility could be converted easily into landslide warning by replacing the triggering factor with weather forecasting data. Landslide susceptibility maps based on the results of this study could be used to assist governments and planners, and help reduce the economic and social costs of rainfall-induced landslides.