變遷分析是了解環境與生態變化的重要參考依據，近40年來遙測技術快速發展，提供了大尺度變遷研究的良好途徑，但是影像拍攝的狀況不一致往往使變遷偵測的正確性受到考驗，而且在土地利用複雜的地區，使用空間解析度較低的衛星影像，容易產生混合像元的資訊，造成分析上的誤差，因此分析前的輻射校正與混合像元處理為變遷分析相關研究的重要課題。本研究利用多期影像，計算植生指標並進行變遷比較，探討不同輻射校正對變遷分析的影響，研究亦檢驗混合像元對變遷分析的干擾並透過統計極端值的概念去除混合像元，以提高變遷分析的可信度。分析結果顯示，在以森林為主的植生變遷分析中，相對輻射校正法中的暗元去除法有高估變遷的狀況，而直方圖匹配法不但容易實施，且能有效改善變遷的誤判。而極端值去除法，不僅可以分離衛星影像中的混合像元，亦可應用於偵測衛星影像中未被去除乾淨的薄雲、霧及遮蔽處，有助提高變遷分析的正確度，可為影像提供有效的前置處理。

Accurate change detection is an important approach to understand the environmental and ecological changes. The rapid development of remote sensing technology over the last four decades has yielded effective tools for large-scale change research. However, change detection accuracy is affected by atmospheric conditions and by mixed pixels resulting from the complex land use or lower spatial resolution of satellite images. Therefore, radiometric correction and mixed pixel processing methods are needed for accurate change detection. The relative radiometric correction processes used in this study included Dark-object subtraction (DOS), Histogram Match (HM), and DOS+HM. After performing the relative radiometric correction processes, this study calculated the normalized difference vegetation index (NDVI) and analyzed the changes ratio to find a better approach. Extreme values were also calculated as mixed pixels and removed to improve the reliability of change detection analysis. The experimental results showed that relative radiometric correction by DOS method tended to overestimate the changes whereas HM method is easily implemented and substantially improved accuracy in change detection. Removing extreme values eliminates not only mixed pixels, but also thin clouds and fog in satellite images. Both methods are effective for image pre-processing and can improve change detection accuracy.