期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2004
卷号：XXXV Part B7
页码：410-412
出版社：Copernicus Publications
摘要：Landsat ETM+ data from the National Park of Khabr in Kerman province, dating May 2000, were analyzed to investigate the possibility of forest type mapping in arid and semi-arid regions.Quality evaluation of the image showed any radiometric distortion. Orthorectification was implemented to reduce relief displacement. The RMS error obtained was less than half a pixel. The gr ound truth map allocating 50% of the total area was prepared by fieldwork, using strip sampling. Differ ent forest types considering the density, were qualitatively estimated in the strips based on typology definitions,. The original and synthetic bands were obtained applying tasseled cap transformation, PCA, ratioing and bands fusion. Furthermore, the parameters of the soil line relation were applied to produce suitable vegetation indices to reduce soil reflectance. The best band set, based on the divergence between classes signatures, using sample areas were selected. Forest type classification utilizing ML, MD, PPD and SAM classifiers were performed to separate pure and dominant types of Pistacia atlantica, Acer monspessulanum, Amygdalus elaeagnifolia, A. scoparia and a mixed type. Because of spectral similarity between the pure and dominant types, these classes merged together and the classification was repeated. In this case the highest overall accuracy and kappa coefficient equal to 47% and 23% respectively, were achieved by MD classifier. Accuracy assessment and signature separability criterions showed undesirable separation between the whole for est types, except for Amygdalus scoparia. By merging all of the types but A. scoparia and perfor ming the classification again, the highest overall accuracy and kappa coefficient equal to 92% and 68% respectively were resulted utilizing MD classifier. Based on the results, in such regions, low forest canopy, increases the role of background reflection and this makes undesirable results. Therefore high resolution sensors data and improved classification methods are advised
