Evaluating Landsat ETM+ Data Capability to Produce Forest Cover Type Maps in the Timberline of Northern Forests of Iran

評估Landsat ETM+衛星資料使用於伊朗北部森林林木界線地區森林被覆型之製圖功效

Hooman Latifi ； Djafar Oladi

《台灣林業科學》 21卷3期 (2006/09) Pp. 363-375

https://doi.org/10.7075/TJFS.200609.0363

摘要

提供森林被覆型圖為經營森林地區的必要條件。為評估ETM+遙感探測資料用於繪製伊朗北部接近林木界線附近森林被覆型圖，本研究所蒐集的遙測資料涵蓋了位於馬贊達蘭省(Mazandaran)之內卡紮摩墩(Neka-Zalemroud)區域之2,700ha土地範圍。本研究藉由正射化處理，糾正影像的幾何誤差；使用全色增進技術融合多光譜和全色光譜段影像。採取1ha的地面樣區進行隨機系統取樣，以獲取地面真實資料。採用自動或使用者選擇方式選取適合的光譜波段組合。從研究區全面選取訓練樣區，並藉由樣區分離度予以評定。採用最大概似分類式進行影像分類，於原始及融合的資料分出7個森林被覆型（例如：山毛櫸混合林、櫟樹混合林等）。分類結果以kappa係數和使用者及生產者量測方式，評定整體分類的準確度；就原始資料分類，整體的準確度最佳達36%、kappa係數為0.22。經最後決定可分類型數後，以最大概似分類法歸類成4個主要森林被覆型。就原始資料分類，結果顯示整體準確度最佳達56%、kappa係數為0.26。基於上述結果結論，ETM+資料無法充分反映出研究區域的森林被覆型之光譜差異；然而，使用較高解析力的光譜及空間特徵之遙測資料，應用在此區域或許能獲致較佳的分類成果。

關鍵字

ETM+ ； 森林被覆型 ； 最大概似分類式 ； 內卡紮里摩墩地區名稱

並列摘要

Producing forest cover type maps is considered a prerequisite for managing forest areas. In order to evaluate the capability of ETM+ remotely sensed data to produce cover type maps in areas near the timberline of northern forests of Iran, data were collected in an area of 2,700 ha located in the Neka-Zalemroud region of Mazandaran. An ortho-rectification process was carried out to correct the geometric errors in the images. Multispectral and panchromatic bands were fused using the Pansharp method. The ground-truthed data were collected using 1-ha field plots in a systematic random-sampling grid. Automatic and user-based band selections were used to choose the appropriate channel set. Training areas were chosen throughout the study area and were evaluated in terms of the separability. Classification was performed using a maximum likelihood (ML) classifier on both the original and fused data sets over 7 recognized types (i.e. mixed Fagus, mixed Quercus, etc.). The results were evaluated using the overall accuracy, kappa coefficient, and user and producer accuracy measures, with the best results of an overall accuracy of 36% and a kappa coefficient of 0.22 for the original dataset. After final class reconsideration, classification was performed using the ML classifier on 4 major cover types. The results revealed an overall accuracy of 56% and a kappa coefficient of 0.26 for the original dataset. Based upon these results, it could be concluded that ETM+ data did not sufficiently fulfill the spectral variations of cover types over the study area. However, using data sets with higher spectral and spatial features may result in a better classification of the cover types of such areas.

並列關鍵字

ETM+ ； forest cover types ； timberline ； maximum likelihood classifier ； Neka-Zalemroud