以坡單元為分析單元之霧社水庫集水區未開發地區大規模崩塌評估模式建立

Abstract

台灣地處板塊交界處，地層長期受到板塊運動影響，導致地質脆弱、破碎，地質條件的不穩定再加上台灣位於季風氣候帶，降雨量豐富且集中，山崩、地滑等土砂災害容易發生，造成嚴重的生命財產損失。因此，大規模崩塌的分析是一個重要課題，本研究區位於南投縣仁愛鄉東側之霧社水庫集水區。 本研究以集水區四坡向坡單元為分析單元，以各項圖資為參考依據，編修集水區四坡向坡單元與大規模崩塌圈繪之邊界，提高分析單元及崩塌圈繪的準確性，以降低後續分析時產生的誤差。 使用ArcGIS軟體計算地形、地質、水文因子，再利用假設檢定以及皮爾森相關係數分析篩選出對大規模崩塌較顯著的6項低相關影響因子，包含高程標準差、水系密度、平均坡度、岩層類別、坡單元高度面積積分和地質構造線密度。利用影響因子進行區別分析，以區別函數建構本研究區的大規模崩塌潛勢分析模型，探討各影響因子對大規模崩塌之影響，發現坡度和高程標準差屬於較重要之因子，以坡度和高程標準差進行分層抽樣，依坡度分層抽樣樣本的區別分析有較高的正判率，以此區別函數建置研究區之大規模崩塌潛勢模型，以ROC曲線決定大規模崩塌潛勢值的臨界值為0.25，以此探討評估結果並與大規模崩塌圈繪進行比較，發現大規模崩塌圈繪涵蓋之坡單元多為中潛勢以上之區域，評估模型有較良好的準確性。

Taiwan is located at the junction of tectonic plates. The strata have been affected by plate movements and resulting in fragile, broken, and unstable geological conditions. When typhoons or earthquakes struck, it oftan caused landslide disasters, resulting in serious loss of lives and properties. This research aims to construct large scale landslide susceptibility model based on discriminant analysis method using larges cale landslide data and slope unit. This reasearch takes the Wushe reservoir basin as the study area. This research uses the watershed slope unit as the analysis unit, and refers to various maps to edit the boundaries of the slope unit and large-scale landslide, which reduce errors in subsequent analysis. ArcGIS software was used to derive topographic, geological and hydrological factors. The hypothesis testing and Pearson correlation coefficient analysis were performed to screen out significant six low-correlation factors, including standard deviation of elevation, river density, average slope angle, geological category, hypsometric integral, and geological structure density. With significant factors, the discrimination functions was constructed for a large-scale landslide susceptibility model. It was found that slope angle and standard deviation of elevation are the more important factors. Stratified sampling is carried out based on the standard deviation of elevation and slope angle. The discrimination analysis of stratified samples based on slope angle has a higher accuracy. This discrimination function is used to construct large-scale landslide susceptibility model in the research area, and the ROC curve is used to determine the landslide susceptibility. We can found taha the critical value is 0.25. Consistency between the eatimated results and the actual large-scale landslide scars distribution are reasonably good.