جداسازي تيپ هاي پوشش گياهي با استفاده از داده هاي سنجنده LISS Ш ماهواره IRS-P6

Separating vegetation types using LISS III, IRS-P6 satellite images

امروزه استفاده از داده هاي ماهواره اي به علت قابليت هاي منحصر بفرد آن در ارزيابي وضعيت پوشش گياهي و همچنين تهيه نقشه جنگل ها و اراضي مختلف از اهميت ويژه اي برخوردار است. اين تحقيق به منظور بررسي تيپ هاي پوشش گياهي ذخيره گاه جنگلي ارس گلوچار واقع در شهرستان رابر در جنوب استان كرمان، با استفاده از داده هاي ماهواره IRS-P6، LISS III صورت گرفت. پس از تهيه داده هاي ماهواره اي، تمامي باندها از لحاظ خطاي راديومتري و هندسي كنترل شد و هيچ گونه خطاي راديومتري مشاهده نگرديد. تصحيحات هندسي با 18 نقطه كنترل زميني با استفاده از مدل ارتفاعي رقومي با دقت 35/0RMSE= پيكسل انجام شد. نقشه واقعيت زميني از طريق نمونه برداري 17 درصد از كل منطقه تهيه گرديد. از اين نقشه به منظور ارزيابي صحت نقشه طبقه بندي شده استفاده شد. طبقه بندي با استفاده از باندهاي اصلي و مصنوعي در 7 طبقه و به شيوه نظارت شده با استفاده از دو طبقه بندي كننده حداكثر شباهت و حداقل فاصله از ميانگين در جداسازي تيپ هاي پوشش گياهي انجام شد. بر اساس نتايج اين تحقيق داده هاي حاصل از هر دو طبقه بندي كننده نسبتاً مناسب است؛ بالاترين صحت كلي و ضريب كاپا به ترتيب 63/38 و 0/6174 با طبقه بندي كننده حداكثر شباهت بدست آمد. در نتيجه استفاده از داده هاي سنجنده LISS III در مطالعات ساختار جنگل مناسب مي باشد.

Background and Objectives: Due to the interaction of vegetation cover and its environment, description and analysis of vegetation has constituted an important part of the ecological studies. Today, using satellite data because of excellent abilities in the vegetation assessment and the mapping of forest and different land uses has high importance in ecological researches. Given the importance of forest ecosystems in arid and semi-arid and necessity of using modern methods, this study compared the accuracy of two classifiers, maximum likelihood and minimum distance to mean in separation of vegetation types using IRS-P6, LISS III satellite data in Galoochar juniper forest reserve located in the Rabor city in Kerman province. Materials and Methods: First, the bands of the satellite data were controlled according to radiometric and geometric errors. No radiometric distortion was found, geometric correction was performed by 18 ground control points with the digital elevation model, up to orthorectification level with precision of less than half pixel (RMSE=0.35 pixel). Ground truth map was prepared through sampling in 17% of whole area. It was used in order to evaluate the correct conclusion of classification of image. The supervised classification was performed by using basic and synthetic bands to 7 classes. Two supervised classification methods, Maximum Likelihood and Minimum Distance to Means were applied to classify the digital data in this study. Results: Classification of the vegetation types in the region was done after physiognomic-floristic survey and then seven classes of vegetation types were determined. Then this information was used for classification of satellite data. The results of user accuracy in two classifier, maximum likelihood and minimum distance to mean showed that the best results belong to maximum likelihood classifier. The most user accuracy was 84.31% and belonged to Juniperus excelsa- Amygdalus eburneam-Acer monspessulanum class. The results of producer accuracy in two classifier, maximum likelihood and minimum distance to mean showed that the best results belong to maximum likelihood classifier. The most producer accuracy was 89.36% and belonged to Acer monspessulanum- Amygdalus eburneam class. The results of Kappa coefficient and overall accuracy showed that the results of maximum likelihood classifier were better than minimum distance to mean. Conclusion: According to results of this study, accuracies of two classifiers were fairly appropriate. The most overall accuracy and Kappa coefficient were 63.38% and 61.74% respectively by using maximum likelihood classifier. As regards, maximum likelihood classifier is based on the assumption that the data belongs to normal classes and it considers more statistical parameters in classification. The results of maximum likelihood classifier in the regions with similar and dense vegetation cover were strengthened. According to the results using LISS III sensor data in studies of forest structure in arid and semiarid areas is recommended.