山崩潛感分析中的許多潛感因子是直接或間接取自地形資料，例如：坡度、粗糙度、曲率等地形因子，或取自與距離尺度有關的各種區位因子，或結合地形資料去做內插的雨量分布或震度分布等促崩因子。用不同尺度下的地形資料運算出來的因子會有不同的輸出值，且並非高解析度地形資料就可以得到高效度的因子；它對一個特定區域的特定目的而言，會有一定的最佳尺度。本計畫將在石門水庫上游集水區及曾文水庫上游集水區兩個不同地區及採用各四個代表性颱風誘發山崩資料，分析各山崩潛感因子的尺度效應及探討各因子能解釋山崩分布的最佳尺度。因子解釋山崩的效度將以成功率曲線及曲線下面積(AUC)來表示，比較最佳尺度相對於原始資料尺度的AUC增長情形。 本計畫並選取高效度的獨立因子進一步建立兩個地區各四個事件的山崩潛感模型，比較由最佳尺度因子建立的模型相對於原始資料尺度建立的模型成功率之增長情形，及做交叉驗證以了解預測率之增長情形。 ;Many susceptibility factors for landslide susceptibility analysis are directly or indirectly derived from topographic data. These include topographic factors, like slope gradient, topographic roughness, curvature, etc., location factors, like distance to road which may be counted from different distance scale, and triggering factors, like rainfall value and earthquake intensity which are interpolated by cokriging with topographic data. A factor calculated from topographic data of different scale will output a different value, and high resolution data is not always outputting an effective factor for interpreting landslides. It may exist an optimum scale which can best interpret the landslide distribution at a specific region. This study will select the Shihmen Reservoir catchment area and the Tsengwen Reservoir catchment area as two study areas and select four Typhoon-triggered landslide inventories at each catchment to analyze the scale effect and to determine the optimum scale of each susceptibility factor. The effectiveness of a factor in interpreting the landslide distribution will be analyzed by using the success rate curve method and calculating their area under curve (AUC). Four success rate curves will be plotted and four AUCs will be calculated for each optimum factor at a catchment area for the four landslide inventories, and the increase of AUCs from the original scale to the optimum scale of a factor will.be discussed. This study will also select effective independent factors to build four landslide susceptibility models for each catchment. Four success rate curves will be plotted and four AUCs will be calculated for each model using the four landslide inventories at a catchment area, and the increase of AUCs from the original scale to the optimum scale of factors will be discussed. A cross validation between the models will also be done so that the increase of prediction rate of a model can be known.
