土砂災害潛勢區風險評估模式之建置

Risk Assessment for the Sediment Disaster Areas

倪柏寧(Po-Ning Ni)

指導教授 ： 陳怡睿

長榮大學/管理學院/土地管理與開發研究所/碩士(2010年)

摘要

台灣擁有高密度的山坡地，整體陸地中高達四分之三屬於山坡地；近年來由於科技與經濟的高度發展，原有的平原區已逐漸飽和，導致人為開發不斷向山坡地帶延伸，但由於地勢陡峻，每遇颱風暴雨或地震，常有大規模的土砂災害發生，如山崩、土石流等，使得當地居民之生命財產安全備受威脅。因此，建立一套風險評估模式以評估土砂災害事件發生時，保全對象可能暴露於潛在災害環境中之受害機率，提供政府防災規劃時之參考實屬必要。
本研究運用基因演算法自動演化類神經網路技術應用於高解析度衛星影像之判釋分類，藉以獲取地表資訊與災害記錄資料，並利用Logistic迴歸建置土砂災害潛感模式。本研究考量之土砂災害潛勢因子包括「坡度」、「坡向」、「地質」、「高程」、「土壤厚度」、「距水系距離」、「地形粗糙度」、「坡度粗糙度」、「有效累積雨量」、「降雨強度」與「開發情形」等11項因子，並透過主成份分析將其萃取為9個成份，並以萃取後之成份作為Logistic迴歸之自變數，而因變數源自颱風事件(丹瑞颱風與龍王颱風)前後之衛星影像，以影像相減法以及常態化植生指數獲取災害資訊，並採用6：4方式擇取有發生災害與無發生災害之樣本，進行災害潛感模式之建置，其土砂災害潛感模式之推估結果精確性可達到93.5%。
本研究藉由實地探勘以及與里長進行訪談，透過相關文獻回顧後，設定出保全對象之受害機率條件，並推算出研究區域內之空間衝擊、時間衝擊、易致災性與耐災程度之機率，配合土砂災害潛感模式所推算之土砂災害發生機率，統合發展土砂災害風險評估模式。研究中亦結合地理資訊系統繪製區域風險圖，探討區域內因災害來臨時保全對象暴露於潛在災害環境之風險。由風險圖配合實地探勘結果得知，研究區域內之高風險地帶多集中於鳳義里與山興里，且部份河段存有大量土石淤積之情形，已嚴重影響民眾生命財產安全，因此管理單位應立即針對高風險地帶與河道土石淤積等情形著手，進行相關措施之擬定，以達到防災減災之目的。
本研究藉由所建置之風險評估模式，結合地理資訊系統繪製區域風險圖，將可提供政府針對高風險地帶擬定出適當之因應對策及方針，期可減少傷亡人數外，也可大幅減低災後所需耗費之社會成本。

關鍵字

土砂災害 ； 風險評估 ； 基因演算法 ； 類神經網路

並列摘要

Up to three-quarters of land in Taiwan are hillsides. In recent years, the original plain area has been highly developed due to the development of technology and economy. Therefore, the artificial development is extended from plain areas to hillside areas. After typhoons, rainstorms or earthquakes, large-scale sediment disasters such as landslides, debris flow, etc., threaten lives and properties of local residents. Therefore, the establishment of a risk assessment model for disaster prevention is essential.
In order to obtain the records of surface information and disaster information, this study uses genetic algorithms to automatically evolve artificial neural networks technology in high-resolution satellite image interpretation. This study uses Logistic regression model to build the model of sediment disasters potential. In this study, the sediment disaster potential factors includes slope, aspect, geology, elevation, thicknesses of soil, the distance from the river, terrain roughness, slope roughness, effective accumulated rainfall, the intensity of rainfall and development condition. The important components will be extracted from sediment disaster potential factors by principal component analysis and used as independent variables of Logistic regression. The dependent variable was derived from satellite images of pre- and post-typhoon events (Typhoon Damrey and Longwang) by images subtraction and normalized difference vegetation index for disaster information. In this study, a ratio of 6 to 4 was used for randomly chose specimen with/without sediment disaster to construct the sediment disaster potential model. The accuracy of prediction by the model of sediment disaster potential is 93.5%.
Through field survey, interview with the head of a subdivision of the district and literature review, the damage probability of the object to assure safety was determined and the probabilities of spatiotemporal impact, vulnerability and disaster resistance in research area were also calculated. With probability of sediment potential by the model of sediment disasters potential, a comprehensive risk assessment of sediment disaster model was established.
This study also combined geographic information system to plot regional risk maps and to explore the risk of the object to assure safety in research area. Risk maps in cooperated with field investigation reveal that the high-risk zones are located on Fongyi and Shanhsing and mass sediment in some parts of the rivers threaten resident’s lives and properties. Therefore, the risk assessment model can be used by the authority to immediately propose disaster prevention plan for high risk area as a countermeasure against sediment disaster to reduce the number of casualties and social costs of post-disaster.

並列關鍵字

sediment disasters ； risk assessment ； genetic algorithms ； artificial neural networks

參考文獻

周天穎、雷祖強、陳駿賢(2003)。遙測及景觀分析技術於崩塌地判釋與變遷之研究。中華水土保持學報，34(4)，353-368。

林妙雀、酈芃羽(2004)。影響消費者對零售通路商店形象因素之研究－以基因演算法最佳化類神經網路模式加以驗證。運籌研究集刊，6，80-97。

張崑宗、劉進金(2006)。利用類神經網路方法於高解析衛星影像及地形資料之崩塌地辨識－以九份二山為例。航測與遙測學刊，11(2)，161-174。

楊明德、蘇東青、楊曄芬(2005b)。草嶺地區土石流潛勢調查與評估。中華水土保持學報，36(3)，301-312。

劉守恆、林慶偉(2004)。SPOT衛星影像之崩塌地自動分類研究。航測與遙測學刊，9(2)，9-22。