Using maximum entropy modeling for landslide susceptibility mapping with multiple geoenvironmental data sets

The potential of using maximum entropy modeling for landslide susceptibility mapping is investigated in this paper. Although the maximum entropy model has been applied widely to species distribution modeling in ecology, its applicability to other kinds of predictive modeling such as landslide susceptibility mapping has not yet been investigated fully. In the present case study of Boeun in Korea, multiple environmental factors including continuous and categorical data were used as inputs for maximum entropy modeling. From the optimal setting test based on cross-validation, the effective feature type for continuous data representation was found to be a hinge feature and its combination with categorical data showed the best predictive performance. Factor contribution analysis indicated that distances from lineaments and slope layers were the most influential factors. From interpretations on a response curve, steeply sloping and weathered areas that consisted of excessively drained granite residuum soils were very susceptible to landslides. Predictive performance of maximum entropy modeling was slightly better than that of a logistic regression model which has been used widely to assess landslide susceptibility. Therefore, maximum entropy modeling is shown to be an effective prediction model for landslide susceptibility mapping.