SPOT資料在地面上的空間解析力較大地衛星的MSS及TM為優，且 因本省山區地形多陡峭，樹種繁多又複雜，利用空間解析力粗放的遙測資料作森 林調查頗感困難，因此SPOT資料對我們林業人員具有無限的鼓舞，故本研究乃 針對SPOT多譜資料作一林地被覆型分類之探討，一則探討其在林地被覆型分類 之潛力，二則探討不同分類方法對SPOT資料分類之適用性。據本研究結果發 現： 1.SPOT資料由於空間解析力佳，利用肉眼觀察甚為方便，因此地面控制點選取 容易，影像幾何糾正效果良好； 2.利用監察分類法監選訓練樣區時，亦因空間解析力好，均質性的樣區圈選容 易，頗有利於工作人員操作； 3.空間解析力雖然提高，但光譜解析力卻反而不如大地衛星資料，尤其是紅光段 及綠光段的灰度值極接近，且在各樹種間的變化微小，分類時易產生混淆現象， 故SPOT資料對林地被覆型分類而言，並非理想的資料； 4.監察分類法較無監察分類法的分類結果為優，且三個原始波段的分類結果比再 多加兩個比例影像的分類結果為優; 5.利用事後分類法，即將容易發生混淆的地區、配合已知地理資訊重加分類後， 再以其結果取代其分類不良之像元，確可提高分類準確度； 6.陰影或白雲覆蓋區域之資料難以重建(retrieval)，分類後大都隨其濃度變化而成 環狀現象，不具代表地被型之意義，故應予棄卻。
Photogrammetric engineering and remote sensing have been widely used in forest resource inventory. However, current application of remote sensing on forestry is still only in the research stage because of poor image resolution, difficult photo interpretation, climatic factors etc. Spatial resolution is particularly a problem in Taiwan due to the complexity of forest species and terrain. Recently, SPOT image has received considerable attention because of its fine resolution. Also, there is few research which uses SPOT image for forest applications. Therefore, the reason for this study was to consider this point. The objective of this study was to investigate wether SPOT image is really helpful in forest cover type classification or not. This study consists of three different kinds of approaches. They are supervised classification, unsupervised classification, and post classification. The classification results and the comparisons between them indicates as follows. 1.Post classification which incorporates with geographic information system performs well in separating the overlapping classes and indirectly improves the classification results. Thus, this approach is recommended in the applications of remote sensing. 2.The comparison between the original three bands and the five bands (three original bands and two ratio bands) shows that the three bands is better than the five bands in the classification results. This result indicates that the addition of ratio bands into the original bands has no effect on the forest cover type classification. 3.Classification results show that there are great misclassifications between conifer plantation and natural hardwood in one small area. The problem posed by different scanning seasons or SPOT image itself is still unknown. Therefore, further investigations are needed when using SPOT image for forest applications. 4.The use of ratio bands are not helpful in removing the shadow and the cloudy problems in this study. However, overcoming these two problems is important in Taiwan and is worthy for further studies.
