以隨機森林法評估淺層崩塌潛勢

Abstract

在莫拉克颱風侵襲後，在高雄市那瑪夏鄉引發了大規模的淺層崩塌，造成了在當地生命財產上的慘痛損失。為了降低之後再發生類似的災害損失，對於過往的崩塌建立模型對其分析，以期了解崩塌發生之原因。 本研究針對2013年水土保持局公開之那瑪夏鄉崩塌圈繪資料。整理過往崩塌圈繪，衛星圖像和現地觀測，建立起崩塌目錄。本研究主是使用隨機森林模型以不同的尺度來構建那瑪夏鄉的崩塌潛勢圖。尺度分別為5m×5m，10m×10m，20m×20m，25m×25m和40m×40m。以9個崩塌環境因子進行分析，分別是坡度、地質、坡向、曲率、地形粗糙指數、斷層距離、道路距離和河流距離。使用混淆矩陣、接收者操作特徵曲線以及特徵曲線下面積（AUC）來評估在各尺度下模型結果的結果。 產出的崩塌潛勢圖能在災害事情發生前預先針對山坡地區域進行災害發生潛勢及可能造成風險進行評估。以期在未來進行規劃做為參考，也能事先針對高風險區進行補強，以達到防災、減災及避災的效果

After Typhoon Morakot, a large number of shallow slope failures occured in the Namasia of Kaohsiung, causing widespread damage to property and infrastructure, injuries and losses of life. Assessing and modeling the landslide events provides information for mitigating the impact or minimizing similar disaster in the future. In our study, landslide information and disaster events in the Namasia area were collected, based on the landslide in Taiwan published by Soil and water Conservation Bureau (2013). The landslide inventory was compiled using data from existing catalogues, satellite imagery and field observations. The main objective of the study is to apply a Random Forest model for construction of landslide susceptibility map in the Namasia. Nine landslide variables were analyzed, namely: slope, lithology, aspect, terrain ruggedness index, curvature, distance to fault, distance to road and distance to stream. The another purpose of the study is to produce landslide susceptibility maps using different scale. Five scale, respectively 5m×5m,10m×10m、20m×20m、25m×25m and 40m×40m are used, and their results are compared for landslides susceptibility mapping .The comparison and validation of the outcome of model were achieved using confusion matrix, the receiving operating characteristic curve, and the area under the receiving operating characteristic curve . The output susceptibility map has the potential to be useful for general planned development activities in the future and to prioritize areas where hazard mitigation measures are required.