GIS-based landslide susceptibility assessment in Seoul, South Korea, applying the radius of influence to frequency ratio analysis

The objective of this paper is to map landslide susceptibility using a statistical analysis model and the radius of influence within a geographic information systems environment. The statistical analysis included triggering factors (e.g., topography, land cover, forest, and soil properties) of adjacent areas, in addition to the landslide sites themselves. To estimate the probability of landslide occurrence using the radius of influence, and to produce a landslide susceptibility index (LSI), we performed frequency radio (FR) analysis by applying the radius of influence to the domain of specific training sites. Landslide susceptibility maps were generated for each radius of influence, ranging from 0 to 300 m in 30 m increments. We observed enhanced FR index values corresponding to reduced exaggeration of statistical anomalies within the proper radius of influence. It is referred that by adopting the radius of influence the classes that not only affect the landslide occurrence from the adjacent areas but also make anomaly errors can be taken into account in FR analysis. Moreover, comparing the FR values between adopting the optimum radius of influence or not, we inferred that the greater the gap, the bigger influence of adjacent areas the classes have. In the validation stage, we identified the optimum radius of influence by measuring the area beneath the relative operating characteristics curve. We found that the optimum radius of influence in the study area is 240 m, for which the LSI map is 5.95 % points more accurate than when not considering the radius of influence.