A Support Vector Machine for Landslide Susceptibility Mapping in Gangwon Province, Korea

Abstract

Abstract:In this study, the support vector machine (SVM) was applied and validated by using the geographic information system (GIS) in order to map landslide susceptibility. In order to test the usefulness and effectiveness of the SVM, two study areas were carefully selected: the PyeongChang and Inje areas of Gangwon Province, Korea. This is because, not only did many landslides (2098 in PyeongChang and 2580 in Inje) occur in 2006 as a result of heavy rainfall, but the 2018 Winter Olympics will be held in these areas. A variety of spatial data, including landslides, geology, topography, forest, soil, and land cover, were identified and collected in the study areas. Following this, the spatial data were compiled in a GIS-based database through the use of aerial photographs. Using this database, 18 factors relating to topography, geology, soil, forest and land use, were extracted and applied to the SVM. Next, the detected landslide data were randomly divided into two sets; one for training and the other for validation of the model. Furthermore, a SVM, specifically a type of data-mining classification model, was applied by using radial basis function kernels. Finally, the estimated landslide susceptibility maps were validated. In order to validate the maps, sensitivity analyses were carried out through area-under-the-curve analysis. The achieved accuracies from the SVM were approximately 81.36% and 77.49% in the PyeongChang and Inje areas, respectively. Moreover, a sensitivity assessment of the factors was performed. It was found that all of the factors, except for soil topography, soil drainage, soil material, soil texture, timber diameter, timber age, and timber density for the PyeongChang area, and timber diameter, timber age, and timber density for the Inje area, had relatively positive effects on the landslide susceptibility maps. These results indicate that SVMs can be useful and effective for landslide susceptibility analysis.

Land Use/Land Cover Dynamics and Modeling of Urban Land Expansion by the Integration of Cellular Automata and Markov Chain

“…SVMs can be generally grouped into four kernel functions-namely, linear, polynomial, radial basic function, and sigmoid [78]. The mathematical representations of each kernel are mentioned in the help section of ENVI version 5.3; whereas, the same mathematical representations are presented by Lee et al [76] using ENVI version 4.4.…”

“…SVM is a supervised, non-linear, non-parametric classification technique that is widely used in the remote sensing field due to its specific capability to draw conclusions even with limited training samples [75]. Previously, Lee [76] evaluated the SVM algorithms for landslide vulnerability mapping in the Gangwon province of Korea and the results suggested that SVM was quite suitable for a wide range of classification problems, even if the problems were of high dimension and were non-linear separable. Moreover, the approach was applied by Mahmoud et al [71] to monitor urbanization in Abuja, Nigeria.…”

“…Nevertheless, SVMs formulations are not completely free from drawbacks. The major problem concerns the selection of kernels [76,78]. Despite this limitation, SVMs are still more popular and produce more accurate classification results than the traditional methods [76,78,79].…”

“…The major problem concerns the selection of kernels [76,78]. Despite this limitation, SVMs are still more popular and produce more accurate classification results than the traditional methods [76,78,79]. SVMs can be generally grouped into four kernel functions-namely, linear, polynomial, radial basic function, and sigmoid [78].…”

Developing an Accessible Landslide Susceptibility Model Using Open-Source Resources

“…Landslides-defined as the displacement of soil and rocks on slopes-are one of the most common natural hazards in many mountainous areas and greatly affect the social sustainability of human beings [1][2][3]. They can have natural causes, such as heavy rainfall and earthquakes, but also human causes, including urban encroachments and increased surface impermeability to water infiltration.…”