Delineation of Optimal Subdivisions for Watershed Sediment Disaster Management Using Environmental Variables

台灣位處板塊交界帶，地質型態較為破碎、脆弱，因屬海島國家雨量豐沛，集水區泥砂產量大，加上近年極端氣候影響，極端降雨事件增加，每有颱風過境，即易造成大規模土砂災害。陳有蘭溪為台灣最長河川¬-濁水溪的最長支流，其河床與斷層帶分布近乎吻合，顯示沿河床地質較為破碎，也因陳有蘭溪集水區坡級緩，河流沿岸農業開發行為及人為活動頻繁，若遇極端降雨事件產生土砂災害，極易造成集水區內保護對象生命財產損害。 為減免災害造成相同損害重複發生，集水區應做有效經營管理，並未雨綢繆。於尚未發生災害時，可依集水區地形、地貌或特性，分析預測致災率，並劃定管理分區。後續加入極端降雨指標，可預測未來若再遭受極端事件時，集水區內各分區相應之致災率等級。以此作為管理依據，可針對致災率高者加強管理或增加預防措施，更能將資源集中於確實需管理分區。 陳有蘭溪集水區土砂災害以1996年賀伯颱風事件最為著名，本研究採集水區掃描劃定重點集水區，並使用集水區資訊系統(WinGrid)自動集水區劃分功能，劃分集水區分區。以精確度-分區個數曲線，取得最適管理分區個數。結果顯示陳有蘭溪集水區劃分為23個分區個數最適，且以此模式分析之精確度高達87.0%，可供相關管理單位參考。Taiwan is an island located at the junction of the tectonic plate. There is a large amount of sediment yield in the watersheds due to fragile geological property, and the watershed also suffers extreme climate impacts in recent year which results in increases of extreme rainfall events. Typhoons usually cause massive sediment disasters in Taiwan. Chenyoulan Creek is the longest tributary of Taiwan's longest river. The distribution of its riverbed and fault zone is close to coincidence, and it shows that the geology along the riverbed is fragmented. Because of the gentle gradient of the riparian areas in Chenyoulan watershed, the agricultural development behavior and human activities along the river bank are frequent. The sediment disasters caused by extreme rainfall event are threating the protected object in the watershed. To avoid repeated occurrence of the disasters, the watershed should be managed effectively. Prediction of a watershed disaster rate should be made according to the topography, landform or characteristics of the watershed prior to debris disaster occurrence and delineate the management subdivisions in advance. In addition, added the extreme rainfall indicator can predict the corresponding disaster rating of each subdivision within the watershed if encountering the extreme event in future. As a management basis, it is possible to strengthen management or increase preventive measures for those with high disaster rates, and to concentrate resources on subdivisions that do need to be managed. The most notorious sediment disaster in the Chenyoulan watershed is Typhoon Herb event in 1996. In this study, a hotspot scanning system was applied to select the interested watershed, and an automatic watershed division system (WinGrid) was used to delineate the subdivisions of the watershed. The optimal subdivisions were determined from the curve of the accuracy-subdivision number. The results show that the optimal number of subdivisions of Chenyoulan watershed is 23, and the accuracy could up to 87.0%. The methods developed in this study can be as references of the related authorities.摘要 i ABSTRACT ii 目錄 iii 圖目錄 v 表目錄 vi 第一章 前言 1 第二章 文獻回顧 2 第一節 遙測影像相關研究 2 2.1.1衛星影像 2 2.1.2數值高程模型 4 2.1.3土砂災害區位判釋 4 2.1.4集水區自動劃分 5 第二節 環境因子 6 第三章 研究樣區及方法 10 第一節 樣區概述 10 3.1.1坡級 12 3.1.2地質 14 第二節 研究流程 17 第三節 研究材料與方法 19 3.3.1極端降雨指標因子 19 3.3.2道路因子 19 3.3.3植生因子 20 3.3.4順向坡 22 3.3.5影像相減法 23 3.3.6土砂災害區位萃取 23 3.3.7多變量分析 25 3.3.8主成分分析 25 3.3.9 K-means群集分類法 26 3.3.10判別分析 26 3.3.11精確度-分區個數曲線 27 第四章 結果與討論 28 第一節 土砂災害區位萃取 28 4.1.1 以NDVI影像相減萃取土砂災害區位 28 4.1.2重點集水區選定 29 4.1.3土砂災害致災率分級 33 4.1.4地文因子之萃取 33 4.1.5判別分析 33 4.1.6最適管理分區個數選定 33 4.1.7以地文因子預測致災率結果分析 36 4.1.8以環境因子預測致災率結果分析 38 4.1.9各環境因子敏感度分析 41 第五章 結論與建議 45 第一節 結論 45 第二節 建議 46 參考文獻 47 附錄 5