有許多廣泛使用的方法可以建立山崩潛感圖，例如地貌測繪、山崩目錄分析、統計方法、物理模型以及人工智慧等等。統計方法經常被用來分析山崩與山崩潛感因子之間的關係式，其中以羅吉斯回歸模型最為廣泛使用，因為其具有較高之準確性及穩定性。本篇論文之研究目的是以梅姬颱風誘發之山崩目錄訓練模型，建立台灣曾文水庫集水區之山崩潛感模型。我們需要分析數值高程模型、地質背景資料以及降雨量資料以取得山崩潛感因子，其中降雨量可以由雨量站量測資料統計內插取得。本研究使用兩種雨量內插方法，分別為普通克利金法(Ordinary Kriging, OK)以及回歸克利金(Regression Kriging, RK)，並依據不同降雨量內插方法及不同潛感因子數量分為四種山崩潛感模型，分析結果顯示誘發山崩與潛感因子之間的空間關係良好。潛感模型之間的差異顯示，使用回歸克利金法內插雨量所得到的模型比起普通克利金法較佳及穩定，有利於建立穩定的潛感模型。根據分析結果，最後選用RK_8山崩潛感模型建立曾文水庫集水區之山崩潛感圖，可用於預測未來在不同雨量事件下之淺層山崩。;There are many methods which are commonly used to construct a landslide susceptibility map, such as geomorphological mapping, analysis of inventories, statistical method, physically based models, and artificial intelligence method. The statistical method is widely used to fit the mathematical relationship between observed landslides and the factors related to the influence of slope failure. The logistic regression model is the most popular for its robusticity and high accuracy. The purpose of this study is to establish a landslide susceptibility model in Zengwen Reservoir Catchment, Taiwan using statistical modeling techniques. Megi typhoon triggered landslides was selected to train the susceptibility model. DEM, geological data, and rainfall data are analyzed to achieve causative factors, besides that, the rainfall triggers are able to interpolate by ordinary kriging (OK) method and by regression kriging (RK) method, respectively, to build four of different landslide susceptibility models. The results show the spatial relationship between landslide occurrences and the causative factors is good. The comparative differences between models indicate regression kriging is better and suitable for interpolation the rainfall triggers and it is the good point to build the stable model. The RK_8 model is used to produce landslide susceptibility map of the region and used for prediction of future shallow landslides under different rain events with good prediction rate.
