應用邊坡單元預測大規模崩塌

The Application of Slope-unit in Prediction of Large-scale Landslide

嚴文彬(Wun-Bin Yan)

指導教授 ： 林美聆

國立臺灣大學/工學院/土木工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201602783

摘要

本研究區位於高雄市那瑪夏區楠梓仙溪流域，2009年莫拉克颱風期間在多處發生淺層崩塌、土石流以及57處大規模崩塌。大規模崩塌既可能造成鉅額生命財產損失，亦可能發生二次災害，事先預測大規模崩塌發生的可能性是重要的課題。
研究目的在探討大規模崩塌在發生前的地形、地質與水文特性是否具有特徵，以及是否與非大規模崩塌的特性有所不同。以研究區內在莫拉克颱風期間發生的大規模崩塌為對象，坡單元為統計單元，透過統計方法預測環境是否具有大規模崩塌發生的潛在可能，以及在何種降雨條件下發生大規模崩塌。
研究流程為在ArcGIS以何種流程可以劃設適合描述大規模崩塌的坡單元；隨後萃取災前地形、地質、集水區以及降雨因子，在SPSS進行假設檢定，找出與大規模崩塌關係顯著的因子，再由區別分析預測在該環境條件下發生崩塌的潛勢值；以及探討降雨對崩塌的關係，最後同時討論諸多環境、降雨的條件與大規模崩塌之間的關係。
研究發現以集水區四坡向單元比較適合描述大規模崩塌的範圍，崩塌面積30%以上視為崩塌單元的統計效果較好。坡度種類、上半部坡度標準差、地質種類、距構造線距離、集水區寬度、單元水系密度等六項因子，其崩塌與非崩塌群體分布具有顯著差異，彼此的相關性低，並且各自具有物理意義，當作預測分析的使用因子，具有不錯的預測效力。區別分析隨著因子數量增加，有正判率上升、崩塌與非崩塌群體之間分離、群體之內集中的現象，在考量到6項因子時，區別分析的擬合與預測正判率分別為70.3%、75.8%。
降雨因子分析中，觀察到大規模崩塌與長時間降雨因子的關係較明顯，而且長時間的累積雨量與地形高程有正相關的分布，配合環境潛勢值Z>0.15，而以48小時累積雨量超過1,000毫米可作為大規模崩塌發生的警戒值。

關鍵字

大規模崩塌 ； 預測 ； 坡單元 ； 假設檢定 ； 區別分析 ； 降雨分析

並列摘要

Large-scale landslides triggered by rainfall or earthquake often caused much loss of lives and property, thus landslide susceptibility assessment plays an important role in hazard mitigation. The study focused on the geo-environmental characteristics before the large-scale landslide occurrences including morphological, geological and hydrological features and the differences between landslide and non-landslide distributions. The Qishan River basin in Namashia District, Kaohsiung County, Taiwan, with 57 events of large-scale landslides during Typhoon Morakot in 2009 were used as study area. With application of slope-units, environmental landslide potential has been estimated by statistical methods and rainfall condition has also been discussed.
Slope-units mapped by ArcGIS tools were used to extract the morphological, geological, watershed and rainfall factors. By conducting the hypothesis test and discriminant analysis in SPSS, significant factors correlated to landslides were found and selected to form the potential analysis model. It was found that slope-units mapped according to watershed and 4 reclassified aspects, fitted the range of large-scale landslides better. Under the principal of landslide area ratio more than 30% as a landslide unit, 6 factors including: slope type, standard deviation of slope degree above half ramp, types of formations, distance to geological structural line, watershed width and unit drainage density, all display significant differences between landslide and non-landslide groups with physical meanings and low correlation to one another. The overall accuracies of discriminant analysis model for fitting and prediction were 70.3% and 75.8%. Gradual addition of these factors helps to separate groups and discriminant function to concentrate within each group.
Concerning rainfall factors, the relation and the correlation with environmental potential to landslides were analyzed. It was discovered that rainfall increased with higher elevation. Long-term rainfall had better correlation with large-scale landslides. If the discriminant function Z score is higher than 0.15, it was classified as a potential landslide unit. It was suggested that the accumulative rainfall in 48 hours with 1,000 mm may be used as an alert rainfall threshold to large-scale landslides.

並列關鍵字

large-scale landslide ； prediction ； slope-unit ； hypothesis test ； discriminant analysis ； rainfall analysis

參考文獻

何岱杰、張維恕、林慶偉、劉守恆 (2014)，應用數值地形及光學影像於潛在大規模崩塌地形特徵判釋，航測與遙測學刊，第十八卷，第2期，第109-127頁。

謝秉寰 (2013)，旗山溪流域莫拉克颱風觸發大規模崩塌之地形特徵研究，國立成功大學地球科學系碩士論文，96頁。

王建方 (2014)，坡向坡單元劃設及其應用於大規模崩塌潛勢預測，國立臺灣大學土木工程學研究所碩士論文，131頁。

任庭暘 (2013)，大規模崩塌的潛感分析：以高屏溪流域為例，國立成功大學地球科學系碩士論文，88頁。

李明熹 (2006)，土石流發生降雨警戒分析及其應用，國立成功大學水利及海洋工程學系博士論文，229頁。