A landslide expert system: image classification through integration of data mining approaches for multi-category analysis

Remote Sensing (RS) data can assist in the classification of landscapes to identify landslides. Recognizing the relationship between landform/landscape and landslide areas is, however, complex. Soil properties, geomorphological, and groundwater conditions govern the instability of slopes. Previous study of Wan (2009 Wan, S. 2009. A spatial decision support system for extracting the core factors and thresholds for landslide susceptibility map. Engineering Geology, 108: 237–251. [Crossref], [Web of Science ®] , [Google Scholar]; A spatial decision support system for extracting the core factors and thresholds for landslide susceptibility map. Engineering Geology, 108, 237–251) used the maximum-likelihood classifier to classify the multi-category landslide image data. Unfortunately, the classification does not consider the geomorphologic condition. Accordingly, a Landslide Expert System was developed to modify these problems. The system uses multi-date SPOT image data to develop the landslide database. The threshold slope which becomes vulnerable to landslides is obtained by the K-means method. Then, an innovative Data Mining technique – Discrete Rough Sets (DRS) – is applied to obtain the core variables and their relevant thresholds. Finally, the Expert Knowledge Translation Platform (EKTP) is used to create the rules for classification. This study used a new approach called ‘Rough Set Tree’ to demonstrate the performance of the approach. The classification of landslide vulnerable areas, bare land, rock, streams, and water-body is greatly improved.