本研究利用斜坡單元當作基本測繪單元進行山崩潛感分析與製圖。斜坡單元的劃分具有地形上的考量，但以往應用在山崩潛感分析的斜坡單元劃設多使用集水區重疊法劃設，進一步的編修及合併等工作通常須以人工為之，相當耗費時間，且有較多人為主觀因素在內。本研究沿用簡瑋延（2011）所使用的物件導向式分析方法，自動化產製出斜坡單元，進一步的修改劃分流程並以統計方法檢核斜坡單元的劃分優劣。研究中將石門水庫上游集水區先進行地形分類，將沖積層、台地及階地等平坦區域先區分出來，再以子集水區為斜坡單元劃設前的限制範圍，最後以坡向為主要影像圖層匯入物件導向分析軟體Definiens之中，給定圖層權重及各項異質性指標進行影像分割，劃分斜坡單元。統計檢定則進行斜坡單元內部坡向及法向量的標準差計算，以研究區內兩個子集水區進行討論比較，結果顯示以本研究方法劃分斜坡單元得到較好的劃分成果。 接著利用斜坡單元進行山崩潛感分析，採用高程標準差、邊坡坡度比、濕度指數、坡度粗糙度、艾利颱風最大時雨量與總雨量進行羅吉斯迴歸。由於分析時斜坡單元屬於山崩或是非山崩群組的判定方式並無定論，本研究將採用不同崩壞比門檻值來判定山崩與非山崩組，找出最佳的山崩與非山崩組判定之崩壞比門檻值。分析結果顯示崩壞比門檻值定義為斜坡單元中山崩面積比為5%時得到的AUC為0.786是最高。 This study uses slope unit as mapping unit for landslide susceptibility assessment. The delineation of slope unit must consider topographic meaning. In the past, most of slope units applied to landslide susceptibility assessment were delineated by combining the watershed by DEM and the watershed by reverse DEM. Further editing and merging work is usually required and it needs much manual work. The process is times consuming, and containing too much subjective comment. This study follows Chien(2011) using an object-oriented analysis method to automatically generate slope unit. I have modified the processes of slope unit delineation and checked the slope unit quality by statistical methods. In the study area, Shihmen Reservoir catchment area, we first classified the flat area, and using sub-catchment for the former limits. Finally, I use slope aspect as an important component in the object-oriented analysis software, Definiens. According to the weight and the heterogeneity index of image layer, slope units are delineated. Then we calculated standard deviation of slope aspects and normal vectors in each slope unit. I choose two of sub-catchment in the study area for comparison and discussion. The result shows that the delineation methods of our study can get a better result. Then we use the slope unit for landslide susceptibility assessment. Standard deviation of elevation, steep slope ratio, wetness index, slope roughness, the maximum rainfall intensity, and total rainfall of Aere typhoon were adopted in a logistic regression analysis. Because slope units belonging to landslide group or non-landslide group had no conclusive judgment previously, so I will choose different landslide area ratio of slope unit to determine the slope units belong to which group. It reveals that a 5% landslide area ratio of slope unit is good for the threshold. The result shows success rate of the model is fairly good with an AUC of 0.786.
