台灣有三分之二以上土地屬於山坡地,近年來,由於平原區人口稠密,導致人為開發不斷往山坡地延伸,加上全球氣候異常,極端降雨誘發山崩災害頻傳,且伴隨著嚴重的生命及財產之大量損失,亦造成政府單位與社會相當之負擔。因此,若能有效地評估降雨誘發之山崩,勢必能減少坡地災害所帶來的衝擊與影響。 本研究以南台灣曾文溪流域部分區域為範圍,利用基因演算法自動演化類神經網路技術,針對研究區歷經2013年蘇力、潭美及康芮三場颱風降雨接續侵襲前後之衛星影像進行判釋,藉以獲取災害記錄資料及地表資訊,並運用地理資訊系統平台,結合數值高程模型及雨量資料,建置包括土地擾動程度、自然環境及降雨觸發等山崩潛勢因子資料庫。本研究並透過MATLAB平台,運用最佳數值搜尋原理,開發山崩潛勢評估模組,藉以探討山崩降雨誘發之潛勢及其影響因子,並透過地理資訊系統平台繪製研究區之山崩潛勢圖。 研究結果顯示,衛星影像之判釋結果皆達高精確程度。本研究所建置之山崩潛勢評估模式,其建構與測試平均之評估正確率達86%,有合理的評估能力;且所推估之山崩潛勢結果與現地實際崩塌情況及歷史災害點位大多吻合,可應用於相關防災單位擬定治理政策之參考,期減少傷亡人數外,也可以降低災害損失之社會成本。
Two-thirds of the land in Taiwan is hillside land. In recent years, the human development is extending towards the hillsides due to the high population of plain area. With the adverse global climate, extreme rainfall often caused landslide and accompanied by the major damage of lives and properties. It is quite a burden to the government and the society. Therefore, the strikes and effects would surely be lessened if we can estimate the probability of rainfall-induced landslide occurrence and take action in disaster prevention. We chose the Tsengwen river watershed in southern Taiwan as study region. We used Genetic Adaptive Neural Network (GANN) to classify satellite images before and after typhoon Soulik, Trami, and Kongrey that struck Taiwan and to get the information of the surface and disaster records. The geographic information system combined with digital elevation model and rainfall data was employed to establish database of potential factors of landslide which include the degree of land disturbance, nature environment and rainfall. This study developed an evaluation module for landslide potential according to optimum seeking theory through MATLAB platform. Then, the weights of potential factors of landslide were determined and the map of the potential of landslide in study area was plotted. The results of the study show that the accuracy of the satellite images classification is at high level. For training and testing, the overall accuracy of evaluation module for landslide potential is up to 86%. In addition, the results predicted by the evaluation module match the actual situation of collapsed scene. Therefore, the proposed model can be applied in practice.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。