莫拉克颱風於民國98年8月間,在高屏溪流域降下了百年來最高的雨量,誘發了大量山崩。為了解極端降雨下之山崩特性及山崩潛感分析可能遭遇的問題,本研究選取莫拉克颱風前後解析度2公尺的福衛二號影像,以人工判釋方法分別圈繪颱風前後之山崩,並做檢核及前後之比對,建立誘發山崩目錄。以此品質較佳之山崩目錄分析山崩分布特性並做為訓練資料,以流域地形、地質及區位為潛感因子,以莫拉克雨量為促崩因子,並以羅吉斯迴歸分析建立新的莫拉克山崩潛感模型。由於此極端降雨誘發了大量的深層滑動型山崩,分析時必須謹慎地將其分辨出來而不納入分析,同時因為山崩多集中於中等坡度(22°~38°),而陡坡及峭壁上的落石型山崩顯然較少,因其崩壞機制與淺層滑動型山崩不同,分析時也須將落石潛勢區分開處理而不納入淺層山崩之潛感分析。本研究比較了五種不同組合的山崩潛感模型(A1、A2、A3、B、C)解釋莫拉克颱風誘發山崩與預測海棠颱風事件誘發山崩之良窳,期能瞭解在極端降雨之山崩特性下以往山崩潛感模型是否有可改進之處。結果以傳統之模型A1較為適當,其餘模型改善有限。模型A1解釋莫拉克誘發山崩,淺山區AUC達0.830,高山區AUC達0.721;預測海棠誘發山崩,淺山區AUC達0.790,高山區AUC達0.729。此模型除了可解釋極端事件,於驗證海棠的結果也可接受。The Typhoon Morakot brought extreme rainfall and induced numerous landslides in Gaoping catchment in August, 2009. To realize the feature of landslides and the potential problems involved in the landslide susceptibility analysis under an extreme rainfall event, this study uses landslide inventories interpreted from FORMOSA-2 images before and after the Morakot typhoon event. The landslide inventories were checked by examining rectified aerial photographs, high resolution topographic maps, so as to establish an event-based landslide inventory. This study uses rainfall data of the Typhoon Morakot as trigger factor, and uses topographic factors and geological factors as causative factors, and then uses logistic regression as analytical method to establish a susceptibility model. Because this extreme rainfall event induces a large number of deep slides, to identify them carefully and to exclude them from the landslide inventory are necessary. The event-based landslide inventory shows that most landslides locate at moderate slopes(22°~38°) instead of scarp slope. Because the differences of mechanism between shallow landslides and rockfalls, we exclude rockfall region from the shallow landslide susceptibility analysis.To clarify the issue of landslide susceptibility analysis under an extreme rainfall event, this study compares five different models (model A1, model A2, model A3, model B, and model C). The results show that modle A1 established with traditional method is still suitable, the other models have not shown any advantages. Model A1 shows that AUCs of the success rate curves for hill terrain and for mountainous terrain are 0.830 and 0.721, respectively. It also shows that AUCs of the validation rate curves for hill terrain and for mountainous terrain are 0.790 and 0.729, respectively in the Haitang event. The results are all satisfactory.