| 摘要: | 今年八月的一場風災重重的傷害了台灣。在短短的72 小時之間莫拉克颱風降下了2,748 毫米的驚人雨量而重創了南台灣。極端快速且大量的降雨將造成平地低窪地區大規模的淹水，大量的降水也會造成山區土壤含水量快速飽合而發生山崩、甚至引發毀滅性的地滑及土石流。極度瞬間的大量降雨是造成此次水災的主要原因，而全球暖化正是造成瞬間大量降雨的元兇。在面對常態性的極端氣候時，唯一的方法就是平時就要做好密集且準確的環境監測，再利用所得到的資訊提出因應的對策。能準確的監測山區植被的分佈自然就能大幅的降低大雨為山區所帶來的危害。面對調查偏遠、面積廣大且地形陡峭的山區，衛星遙測就是長期觀測植被變遷最好的工具。前人研究多利用遙測影像計算NDVI 植生指數來偵測土石流區域，但土石流多發生之山區會因陰影效應造成誤判。本計畫利用數值高層模型、統計參數、大氣校正及光譜比值消去陰影，以提升偵測精確度。計畫分二年進行研究。第一年針對莫拉克風災前後多光譜衛星影像加以處理，利用數值高層模型及統計參數分析，比較陰影消除程度及土石流偵測鑑別度；第二年延伸第一年的工作，將所有的處理技術從單一波段(單一維度)的處理轉換成多波段(多維度)的處理。利用相對大氣校正及光譜比值法，不但對地形陰影處理，並降低影像中之大氣效應影響，對前後期影像比對及土石流偵測之效能將有所提高。最後我們還是希望能發展出精準且可從事多維度影像的陰影去除及大氣校正演算法以供後續研究使用。 In August 2009, typhoon Morakot severely damaged Taiwan. In 72 hours (Aug 8-10), it rained 2,748 mm in Southern Taiwan which caused floods in flat areas and also landslide and mudslide in mountain areas because of the saturation of soil water content. The main reason of this disaster is the huge amount of rain fall in very short of time, and it is also a part of the extreme climate changes. To avoid disasters in this kind of repeated extreme weathers, we need to continuously and precisely monitor our environments as daily routine, and rapidly response to any anomalies. The vegetation can be an indication for landslide or mudslide. In order to monitor large area in remote and precipitous mountain regions, satellite remote sensing provide a efficient and economical approach for long-term vegetation surveillance. Previous researches proposed to detect landslide based on the monitoring of vegetation with Normalized Differential Vegetation Index (NDVI). The drawback of this index is that it will give low value in shadow area even if it is vegetated. To overcome this drawback, in this two-year project, we propose to reduce or remove the shadow effect with digital elevation model, statistical parameters, relative atmospheric calibration and/or band ratio techniques, and expect to improve the detection performance. In the first year of research, we process the satellite remote sensing images before and after typhoon Morakot with digital elevation model and statistical parameters methods and focus on the comparison of shadow removal and landslide detection performance. For the second year, we continue to expand the first year research from signal image to multispectral bands. With relative atmospheric calibration and band ratio techniques, it not only can reduce the shadow effect of terrain, but also reduce the difference of the atmospheric conditions when two images were acquired. Finally, our developed method can provide a procedure to reduce the shadow and atmospheric effect for other remote sensing applications. 研究期間：9908 ~ 10007 |
