- 學位論文
以羅吉斯迴歸探討累積水平地震加速度對崩塌潛勢影響之研究
The Effect of Cumulative Horizontal Seismic Acceleration on Landslide susceptibility with Logistic Regression Model
摘要
臺灣位於西太平洋颱風帶上，颱風豪雨常造成不同規模之崩塌，因此，降雨量等水文因子和台灣特殊之地理、地質因子常被視為崩塌因子，雖然如此，台灣每年約千次之大小地震是否會加重崩塌發生之機率和規模，卻較少人探討。因此，本研究是假設地震對邊坡之破壞具累加性，利用地震累積水平加速度值來代表地震之累積能量大小，也就是假設每一次地震即會影響地質結構，造成大小不一的結構破壞，是考慮地震的潛在破壞到底會不會增加崩塌之機率，而非考慮其外在破壞，來探討地震對崩塌潛勢之影響，以及增加地震因子可否提昇崩塌潛勢模型之準確度。 為瞭解地震因子對於崩塌潛勢模型的影響大小，本研究蒐集2004年艾利颱風後之全臺灣崩塌地區位資料，做為崩塌與否的基準，以及1991年~2004年臺灣地區16場災害之累積水平地表最大加速度資料，與2004年之數值高程模型、2000~2004年臺灣地區年平均雨量，共11個崩塌影響因子，運用地理資訊系統和多變量羅吉斯迴歸，得到包含地震因子及不包含地震因子的兩個崩塌潛勢模型。同時，利用分類誤差矩陣、概似比檢驗、類R2指標、簡單統計指標及ROC curve共5種檢驗方法判別地震因子對於崩塌潛勢模型之影響權重。再以2003年及2006年之崩塌地資料，與2008年石門水庫集水區在經歷辛樂克颱風後之崩塌區位做驗證，以評估崩塌潛勢模型之準確率，並對其結果做卡方檢定，檢定此兩模型在判釋高、中、低崩塌潛勢是否有顯著差異。最後再針對臺灣地區崩塌地的空間分布進行分析，並與羅吉斯迴歸結果做比較，探討這11個崩塌因子可能影響崩塌之機制。 結果顯示，加入地震因子的崩塌潛勢模型對預測崩塌發生之準確性較佳，顯示在地震頻繁區域，使用歷史資料評估崩塌潛勢時，應該納入地震因子，以免錯誤估計該區域之崩塌潛勢。
並列摘要
Taiwan is located in the Western Pacific typhoon zone, and typhoons and its inducing heavy rains often result in different scale of landslides. Hence, rainfall and other hydrologic factors, as well as special geographic, geologic factors in Taiwan are selected as landslide factors. Even though, the topic about thousands of earthquakes a year might increase the probability and scale of landslides is seldom cared about by researchers. Therefore, this study assumes that every earthquake will affect the geological structure, causes structural damages and uses cumulative horizontal seismic acceleration to represent the cumulative energy of earthquakes. That is to consider the potential earthquake damage Not destruction of the external to explore the effect of earthquakes on landslides. Is it possible that seismic factor can increase the accuracy of landslide susceptibility model? This research will discuss about it. To understand how the seismic factor influences the landslide susceptibility model, we collected all landslides area after Typhoon Aere in 2004, which is used as a baseline of landslide. We also collected Cumulate Horizontal Peak Ground Acceleration of sixteen events of seismic disasters from 1991 to 2004 in Taiwan, the DEM from SWCB in 2004 and average annual rainfall from 2000 to 2004. In this study, we investigate eleven factors of landslides. And, GIS and Logistic regression are used to get two landslide susceptibility models, which are carried out with and without seismic factor. Meanwhile, five kinds of test methods such as classification error matrix, likelihood ratio test, analogous R2 index, simple statistical indicators and the ROC curve are calculated to determine how the seismic factor affect the landslide susceptibility model. Then, the landslide data in 2003 and 2006, and Shihmen Reservoir landslide locations after Typhoon Sinlaku in 2008 are collected to evaluate the accuracy of landslide susceptibility model. Finally, the spatial distribution of landslide in Taiwan are plotted and compared with the result of Logistic regression, and exploration of the relation between landslide and landslide factors. The result showed that the landslide susceptibility model which adds seismic factor is better on predicting the occurrence of landslide. So it showed that the assessment of the landslide susceptibility with historical data must consider the effect of the seismic factor to avoid misjudge the occurrence of landslide in earthquake-prone area.