Data Mining For Knowledge-Based Approach for Landslide Susceptibility Mapping

Abstract

Statistical approaches to landslide susceptibility mapping can be an effective tool for determining areas of high risk, but have also had problems with portability, reliability, stability, and impracticality as they often require large amounts of data in order to be effective. The knowledge based approach has been proven to address these issues, but the necessary preparation required for obtaining domain knowledge on landslide susceptibility and predisposing factors demands time and effort from the analyst. This paper proposes a knowledge based data mining approach: data mining techniques for defining the knowledge on the relationship between landslide susceptibility and predisposing factors and the knowledge based approach for mapping landslide susceptibility. A set of environmental clusters were generated through fuzzy classification of key environmental layers describing spatial variation of predisposing factors. These clusters were then hardened and later ranked according to density of landslides. The cluster centroid values are ordered in the order of landslide density for hardened classes and used as control points for the construction of fuzzy membership functions on knowledge of relationships between landslide susceptibility and predisposing factors. The fuzzy membership functions are then used through a knowledge-based approach to map landslide susceptibility. The accuracy of the so generated map in the Kaixian area is comparable with that from a logistic regression model. However, the generated fuzzy membership functions are more portable than those in the logistic model. It was also found that the combination layer was correlated strongly with landslide instances, and could serve as a useful input for future studies. The results of this study provide a useful illustration for potential in the construction of fuzzy membership maps from field observation data.