Using SOM and DBSCAN-based models for landslide hazard and spatial correlations analysis: A case study in central Taiwan

Abstract

Typhoon Morakot, occurred on August 8, 2009, caused serious damages such as lots of villages and bridges destroyed by landslides, debris flow and flood hazards in south-central and eastern Taiwan. The Alisan Creek watershed in central Taiwan was one of serious hazard areas. This study proposed (1) two-layer SOM-based classifier coupled with pre- and post-typhoon SPOT satellite images to extract accurate landslide, (2) the concept of landslide slope unit (LSU) derived from automated watershed modeling delineation to calculate spatial characteristics of the landslide and (3) DBSCAN-based model to analyze the spatial correlations at landslides to assess the priority of landslide treatment sites. The evaluated result shows that the landslide area soon after the typhoon is 619.91 ha (Kappa coefficient 0.9766). The studied watershed was delineated as 596 LSU, using the threshold area of 10 ha, in which there are 29, 86 and 481 of high, medium and low potential zones, respectively, clustered based on the landslide scale. According to DBSCAN spatial analysis, high and medium potential zones could be clustered as four and five groups. However, the clustering effect of low potential zones was not apparent due to the scattered distribution. High and medium potential zones also show tight spatial correlations due to adjacent to each other and/or linked in the same sub-watershed. Both potential zones would increase the possibility of secondary disasters, such as debris flows and additional landslides, and directly threaten downstream neighboring villages during typhoon seasons. The landslides at the hazardous zones should be treated in high priority. The analyzed results are useful for decision making and policy planning watershed management in the landslide area.