SPOT衛星影像應用於桃園海岸林植生監測

Abstract

桃園海岸長期處於人為開發狀態，如建造海水浴場、漁港、工業區等，加上東北季風與颱風的侵襲，使得海岸林常遭受到嚴重的破壞。人為與自然的推力造成土地覆蓋變遷迅速，影響到海岸林木健全生長。因此，本研究地區僅限桃園沿海四鄉(蘆竹鄉、大園鄉、觀音鄉和新屋鄉)之西濱快速道路以西的海岸林帶，面積約1963公頃，應用衛星影像監測不同植生光譜變遷。首先由航照數化建立土地覆蓋基本資料，得知混合林佔大多數14.58%，黃槿純林12.16%，而木麻黃純林只有6％。由不同季節衛星影像分類結果，夏季分類之準確度最高(85%)，最適於海岸林影像分類時間選用。由2000年至2005年土地覆蓋變遷偵測得知，木麻黃明顯減少(19公頃)，而混合林增加(36公頃)。進一步比對航照數化與衛星影像分類變遷分析，其仍存在相當差異性。海岸土地覆蓋變化直接或間接影響到海岸林的生長，因此為了解不同植生的變化，研究抽取木麻黃、黃槿與混合林區之紅光與近紅外光光譜值，以雙光譜圖觀測植生光譜變化，結果顯示原始DN值光譜雜亂，無法突顯植物物候的差異；而經絕對增益值校正之反射輻射值，可以明顯區分出不同季節的植生光譜狀況；直方圖匹配之相對輻射糾正值效果最佳，但若植物遭受環境壓力，直方圖匹配將無法偵測到植生的異樣。多變量變異數之Duncan檢定的結果發現紅光皆無法區分出三種植生，而以近紅外光與NDVI可有效區分出三種植生。雙光譜圖分析的結果可看出混合林介於木麻黃跟黃槿中間，因此檢視由木麻黃與黃槿所組成的混合林，發現混合林中黃槿的比例逐漸增加。最後，光譜監測結果發現冬季影像海岸林地區NDVI異常小於零，由氣象資料分析結果推測應是降雨導致土壤濕度過高以及雨水無法快速排出。研究結果發現應用衛星影像可快速、有效的監測大面積植生的動態變化，未來可以擴大應用於整個西部海岸林，建立海岸生態監測系統，以達經營管理之效。

Taoyuan coast has long been suffering from the damage of seasonal monsoon and typhoons, along with the urbanization. The environmental force and the human development resulting in landcover change have affected the growth of the coastal forests. In order to assess the damages, satellite imagery and ortho-image are integrated to monitor the changes. Our research area ranges throughout four counties along Taoyuan coast which covers 1963 ha. First of all, according to the digitized landcover, Hibiscus tiliaceus L. was the main species. Next, we applied supervised classification to compare the effect of seasonal images. Summer image came out to be higher overall accuracy (85%) in classification. With the established landcover data, change analysis was processed. The result demonstrated that Casuarina equisetifolia L. had decreased 19 ha whereas mixed stand increased 36 ha inversely from 2000 to 2005. However, there was a discrepancy between satellite image and aerial photo in post classification comparison. Furthermore, Bi-spectral plot was adopted to monitor Casuarina equisetifolia L., Hibiscus tiliaceus L. and mixed stand to assess the physiology change. The results showed the original digital number was jumbled, radiometric value could clearly distinguish seasonal vegetation spectral, and the histrogram matching yielded the best result. Nevertheless, if the vegetation was under environmental stress, histogram matching was unable to detect the extraordinary phenomena. Besides, multivariate analysis of variance and Duncan test explained that Red band could not discriminate among seasons and vegetations, whereas NIR was efficient for separating seasons and vegetations. The use of spectral distance further explained the increasing proportion of Hibiscus tiliaceus L. in mixed stand. Finally, the NDVI in coastal forest was less than zero so that we suspected it was due to abnormal precipitation. Accordingly, the study concluded that satellite imagery could monitor dynamic vegetation change in a large scale to reduce the expenditure for achieving efficient sustainable management.