阿里山森林鐵路橫跨平地至高山，位於台灣中南部的嘉義地區，不僅是山區對外聯絡的交通要道，亦為觀光旅遊之勝地。由於近年來氣候變遷下導致極端氣候現象頻傳，使得阿里山森林鐵路邊坡崩塌災害日益嚴重，其中民國98年8月的莫拉克颱風更重創了森林鐵路。因此，本研究以鐵路沿線上、下邊坡1200公尺為研究對象，利用莫拉克颱風事件建立山崩潛感模式，以確保鐵路用路人之安全。 首先蒐集中央地質調查所製作的莫拉克颱風事件山崩目錄，並配合事件前後之航照圖進行檢核，以確立山崩潛感分析所需之山崩目錄。研究之分析單元為20公尺的網格，利用地理資訊系統萃取岩性、坡向、坡度、地形粗糙度、坡度粗糙度、事件前NDVI、事件前NDWI、水系距、道路距、鐵路距及高程等山崩潛在因子，以及選用莫拉克颱風事件之最大時雨量和總雨量做為山崩之促崩因子。 各因子分為山崩組與非山崩組並做標準化之處理，由於山崩組與非山崩組網格數量差異大，非山崩組以隨機抽樣之方式選取與山崩組相同數目的網格數，利用多變量統計中的羅吉斯迴歸法，建置山崩潛感模型，經由分類誤差矩陣總體正確率達69%，以及AUC曲線下之面積高達0.768後，繪製莫拉克颱風事件山崩潛感圖。最後分別將不同重現期距之時雨量及三日雨量輸入模式中(10、25、50及100年)，繪製不同雨量頻率年之山崩潛感圖。由本研究繪製之莫拉克颱風事件山崩潛感圖，及不同雨量頻率年之山崩潛感圖，可作為阿里山森林鐵路防災預警之災考。 The Alishan Forest Railway was suffered from the landslide disaster during typhoon Morakot. The landslide event of typhoon Morakot was selected as the research event. The buffer range of 1,200 meters of the the Alishan Forest Railway was used in the present study. The inventories of landslide during typhoon Morakot by Central Geological Survey was collected. The landslide inventories were checked by aerial photographs before and after Morakot typhoon event. The landslide factors were extracted from raster cell data of GIS by using the digital terrain model with 20m�20m resolution. The causative factors including lithology, slope aspect, slope, terrain roughness, slope roughness, NDVI before Morakot event, NDWI before Morakot event, river distance, fault distance, road distance, railway distance and elevation were used in this study. The trigger factors was maximum rainfall intensity and total rainfall of the Morakot event. The raster cell data extracted from every factor by GIS were divided into landslide group and non-landslide group. The landslide group and non-landslide group data were randomly sampled and the data numbers of two groups were equal. Those data were analyzed by a logistic model. Then the model was used to calculate a landslide susceptibility index for each cell. The results show that AUCs of the success rate curves for logistic regression is 0.768 and the overall accuracy in this event are 69%. Both method of verification have showed good performance. Finally, a landslide susceptibility map with susceptibility index for the research area is proposed for engineering and disaster prevention consideration.