Using neural network classifier and the GIS technique for automatic landslide hazard assessment

The catastrophic earthquake, 7.3 on the Richter scale, occurred on September 21, 1999 in central Taiwan. Much of standing vegetation on slopes was eliminated and massive, scattered landslides were induced at the Jou-Jou Mountain area of the Wu-Chi basin in Nantou County. This paper proposed unsupervised neural network classifier coupled with pre- and post-quake SPOT satellite images to extract the landslide, and combined multiple criteria decision making methods with the GIS technique to assess the priority of landslide treatment site. The analyzed results indicate that there were 849.20 ha of the landslide area extracted in the initial earthquake stage. According to the calculations of the AHP model, the weights of evaluative factor for landslide scale, number of building, road density, and river density in a watershed are 0.08, 0.48, 0.30 and 0.14, respectively. The factors, number of building and road density, are more critical. The top priority sub-watersheds for the landslide treatment are Nos. 19, 18 and 20, with outranking flow 13.4485, 3.6925 and 2.2887, calculated from the PROMETHEE algorithms. Those sub-watersheds were located near Wu-Chi River with gently sloped terrain so that there were higher density of buildings and roads. A GIS-based system to extract the landslides and assess the landslide treatment site was also developed in this study. The analyzed results are useful for decision making and policy planning in the landslide area.