A GIS-based System for Automated Extraction of Landslide Hazards

Abstract

This study developed a landslide auto-detection system coupled with GIS technique, vegetation index, image-differencing method and iterative algorithm for automatically landslides extraction. The auto-detection system, including rough and precise classification algorithms, can be used to effectively assess landslide change and extract the optimal accuracy of landslide sites from multi-temporal SPOT images. The rough classification could rapidly but roughly extract the landslide sites using selection of the landslide seed points coupled with GIS techniques of spatial analysis and display. In order to estimate the accurate landslides, the optimized kappa method was developed in this study to calculate the optimal change threshold for landslide extraction. A case with earthquake-induced massive landslides at Jou-Jou Mountain area in central Taiwan on September 21, 1999 was chosen. The optimal accuracy analysis for landslides indicates that the landslide area on September 27, 1999 was 823.0 ha with 0.83 of optimal kappa accuracy. Over about 4 years of vegetation restoration, the landslide area on July 20, 2003 decreased to 309.8 ha with 0.72 of optimal kappa accuracy, giving an overall reduction of 62.35%. The analyzed results are useful for decision making and policy planning in the landslide area.