| 摘要: | 台灣由於地理環境特殊，地狹人稠，將近七成的人口居住在面積不到兩成的都市地區，又我國為一高天然災害潛勢的國家，高度的人口聚集加上環境的過度開發造成都市居住環境安全性降低，同時也產生許多致災因子，使得都市遭受自然及人為災害侵害的機率大增，且每當災害來臨時往往造成極為嚴重之生命財產損失，面對上述課題，以往皆以災害潛勢分析，災害風險評估作為擬定與推動防救災計畫之依據，以期能降低相關災害之威脅，但近幾年來由於全球氣候變遷所導致的多起重大災害，揭露了傳統的災害潛勢及災害風險分析其在評估與考量上的不足。災害風險分析包含危害(Hazard)分析與脆弱度(Vulnerability)分析之考量，以往之思維主要著重於危害(Hazard)分析之探討，而對於脆弱度(Vulnerability)之著墨較少，有鑑於此，本研究將針對脆弱度層面進行深入探討，結合社會性，經濟性，物理性與環境性等不同觀點，進行地區災害脆弱度之評估，並進行災害脆弱區域之指認，以了解地區脆弱度之分布概況。
本研究透過回顧國內外都市災害相關文獻、災害脆弱度相關文獻，地理資訊系統(GIS)相關文獻以及類神經網路於環境資訊之鑑識、推估及預測相關領域之研究成果等內容為基礎，作為建立脆弱度評估指標架構及建構地區環境脆弱度評估模式之參考依據，並選定台北市為模擬實證地區，運用地理資訊系統(GIS)進行分析運算以取得相關脆弱度數值，並運用類神經網路進行數值推估模式之建構，藉由上述技術與方法進行都市脆弱區域之指認與模擬分析。
而本研究之研究成果主要包含脆弱度評估成果以及類神經網路數值推估模式建置成果兩部分，依序分述如下：
脆弱度評估成果部分，災害暴露性脆弱度最高者為北投區秀山里，其次為內湖區大湖里；地理自然環境脆弱度最高者為信義區中坡里，其次為內湖區金瑞里；土地使用與建築環境脆弱度最高者為文山區景行里，其次為南港區東明里；社會經濟脆弱度最高者為士林區平等里，其次為士林區溪山里；而最後加總之總和脆弱度最高者則為北投區秀山里，其次為士林區溪山里。
類神經網路數值推估模式建置成果部分，整體訓練樣本之誤差均方根(RMSE)為0.08307、誤判率(Error rate)為0.46250，而測試樣本之誤差均方根(RMSE)為0.08992、誤判率(Error rate)為0.46281，顯示整體網路配適度及準確性尚可，即此網路模式於脆弱度指標數值推估模擬上具有一定之可行性。
Taiwan is the special geographical environment, small and densely populated, about seventy percent of the population living in the area of less than two percent of urban areas, but also natural disasters, China is a high potential of the country, together with a high degree of population caused by over-exploitation of the environment urban living environment less secure, but also have many hazards, making the city vulnerable to natural and man-made disasters, greatly increased the probability of abuse, and every time disaster comes, often resulting in very severe loss of life, the face of these issues, both past Analysis of the potential disaster, disaster preparation and risk assessment as a basis for promoting disaster prevention plan in order to reduce the threat of disaster-related, but in recent years due to global climate change caused by more than a major disaster, the traditional disclose potential hazards Potential disaster risk analysis and assessment and examination of its volume is not on foot. Disaster risk analysis includes hazard (Hazard) Analysis and vulnerable level (Vulnerability) Analysis of the consideration, in the past mainly focused on the hazards of thinking (Hazard) Analysis of, and for the degree of vulnerability (Vulnerability) of Zhumo smaller view of this, Research will focus on vulnerability levels depth, combined with social, economic, physical nature and the environment of the different point of view, the regional disaster vulnerability assessment and disaster vulnerable regions of identify, to understand the regional vulnerability of the distribution profile
This study reviewed the literature at home and abroad urban disaster, disaster vulnerability literature, geographic information system (GIS) neural network literature and the forensic information on the environment, estimation and forecast results, and other related fields of study based on the content, Vulnerability assessment indicators as a framework for the establishment and construction of regional environmental vulnerability assessment model of reference, and selected areas in Taipei for the simulation evidence, the use of geographic information system (GIS) analysis of expression values to obtain the relevant vulnerability and the use of artificial neural network Road construction of the numerical model to estimate, by the above techniques and methods to identify the urban vulnerable areas of analysis and simulation.
The results of this study include vulnerability assessment of key results and the estimation model of neural network to build numerical results of two parts, the order are as follows:
Vulnerability assessment results section, the disaster exposed the vulnerability of the highest in the Beitou District Xiushan, followed by a large lake within the lake; geographical vulnerability of the natural environment, the highest slope in the Xinyi District, followed in Neihu Rui; land vulnerability of the built environment using the highest mountain Kageyuki in an article, followed by the South East Ming Lane; the highest degree of socio-economic vulnerability as an equal in the Shihlin district, followed in the Shihlin District Mountains; and finally add up the total vulnerability Xiushan was highest in Beitou District, followed in the Shihlin District Mountains.
Neural network results of numerical estimation model to build part of the overall root mean square error of training samples (RMSE) is 0.08307, false positive rate (Error rate) to 0.46250, root mean square error of the test samples (RMSE) is 0.08992, false positive rate (Error rate) was 0.46281, indicating the overall accuracy of the network can still be fit and that this network model to estimate values of the indicators in the vulnerability of certain of the feasibility of the simulation. |
