| 摘要: | 崩塌是一個高破壞性的事件，經常在連續的豪雨或大地震之後，發生在山區、郊外、或半都市開發地區。臺灣島的特徵又是多崎嶇陡峭的山地，且位於歐亞板塊與呂宋島弧的碰撞帶。由於氣候屬於亞熱帶與副熱帶，每年四到六月的梅雨季及七到九月的颱風季節，總是受到豪雨的侵襲。甚且，921集集大地震重創了中臺灣，造成數十萬個大大小小的崩塌地。本文崩塌敏感圖模式以南臺灣的旗山溪流域為研究區，測試區在荖濃溪流域，每年都受到颱風豪雨的摧殘，尤其是去年超大的莫拉克颱風，此區域受創最大，小林村整個消失。 本文企圖對台灣崩塌敏感圖建置有一個深入的研究，並以旗山溪為例。崩塌敏感圖模式建置主要有四個元素，包括崩塌引發因子、判別方法、崩塌判別臨界值(Demonstrator)及崩塌地資料。資料使用上包括21幅SPOT與福衛影像(產製崩塌地與NDVI)、地表高層資料(DTM)、地質圖、土地利用圖、水系圖、雨量、強地動記錄。模式判別方法是不安定指數法。本模式之10個崩塌相關因子包括：坡度、坡向、地質地層、道路、河流、斷層、時間序列及加權之地震強度、時間序列及加權之累積雨量、時間序列及加權之時雨量強度、及植生指數(NDVI)，驗證與測試結果顯示靈敏度(sensitivity)與明確性(specificity)都良好，不受時間與地域之影響。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. |
