| dc.description.abstract | Landslides are highly destructive events that typically occur in mountainous, rural and semi-urban areas after continuous torrential rain or earthquake. The island of Taiwan features very rough and steep terrain on the collision zone between the Asiatic continent and the Luzon Arc. Due to it locates at tropical and subtropical zone; it is always struck by storms during the annual April-June plum rain season and July-September typhoon season every year. Furthermore, on September 21, 1999 the Chichi Earthquake, with a seismic magnitude (ML) of 7.3, struck central Taiwan. The quake caused landslides and rock avalanches at several tens of thousands of sites in the disaster area. Chishan River Basin and Lao-Nong River Basin both located in the south Taiwan, the former is our study area, the latter test area; which always experience heavy rainfall every year, leading to further landslide events. Especially typhoon MORAKOT in 2009 caused devastating damages and claimed many human lives. The rainfall made a record high in 50 years over 3,000 mm in just three days. Some of the habitats and villages were totally destroyed. The landscapes were fiercely changed at extensive and substantial scales. Among the most destructed area is Xiaoliin village within Chishan River Basin.
With respect to the worsening scenario, this dissertation attempts an in-depth look at the model creation of landslide susceptibility index(LSI) map in Taiwan, especially focused on the study area of Chishan River Basin (753 km2), using multi-temporal SPOT series and FORMOSAT 2 satellite images acquired since 1999 until 2009. We propose a modified landslide susceptibility map model in this dissertation, using 10 landslide causative factors, including slope, aspect, NDVI, geological formation, distance to fault, distance to river, distance to road, time-weighted long-term cumulative rainfall, time-weighted maximum one hour rainfall intensity, and time-weighted Arias intensity (AI, Ia) of earthquake shaking; the analysis of each factor is by using instability index method (IIM). In this dissertation time-weighted long-term rainfall and earthquake are first used as environmental factors, not triggering factor in single event. The results of checking and verifying show that a model with these 10 factors have the best specificity in prediction, and the result in different area is also very good. It also proves that a good landslide susceptibility map model includes 4 elements: causative factors, discriminant method, risk classification, and landslide inventory; while selection of causative factors is the key; and high accuracy of LS mapping for large geographical areas is feasible to utilize multi-temporal high-resolution remote sensing images.
| en_US |