火災災後復育工作，可依火燒嚴重度（Fire severity)的評估與圖層繪製訂 定。火燒嚴重度一般用以描述火勢所引起的環境變化程度，而火燒嚴重度圖之繪 製，則可作為未來復育計畫和監測工作的基礎數據。本研究以2009年大埔事業 區所發生之火災為研究對象，採用火災發生前後期之航照數位多光譜影像為材 料，除分析災區各植生類別光譜特徵值之變化外，同時透過災害前後影像常態化 差異植生指標之差異（dNDVI)與主軸轉換法（principal component transform, PCT)第二主軸影像之差異（dPC2)分別萃取火災分布區域。火災分布區萃取後 依前人研究與現地狀況進行分級，同時透過航照判釋選取訓練樣本，並以最大概 似法（maximum likelihood method)進行火災嚴重度分類。研究結果顯示，不同 植生類別其火災前後的光譜值變異極大，近紅外光像元值皆呈大幅下降，而紅光 像元值則大幅提升，至於NDVI則呈現下降變化；dNDVI與dPC2皆能有效萃取 火災分布區域，惟以dNDVI進行火災嚴重度分類，其所得精度較佳。

Post-fire restoration can be performed by mapping the burned area and fire severity. A fire severity can illustrate the degree of environmental changes caused by plans. This investigation considers the restoration of the Dapu area, in which a forest fire occurred in 2009, using pre- and post- fire airborne multispectral images as the materials. The difference in spectral characteristics of various vegetation types between pre- and post- fire in the burned areas are analyzed, while the burned areas are extracted using differenced normalized difference vegetation index (dNDVI) and Differenced Principal Component Transform 2-axis (dPC2). Previous research and the field situation, the fire severity can be divided into 3 degrees. The training samples are selected by analyzing the aerial photographs of the burned area. The fire severity map is then generated using maximum likelihood method image classification. Analytical results indicate major variation between the pre- and post- fire vegetation types. The near-infrared pixel values show a significant fall, but the red pixel value is significantly improved. The NDVI analysis shows a downward change in pixel values. Both dNDVI and dPC2 can obtain the required accuracy for extraction of burned areas; however, dNDVI in fire severity classification can obtain better overall accuracy than dPC2.