Evaluation of the potential of Landsat ETM+ for forest density mapping in zagros forests of Iran

Abstract

Forest in semi-arid regions as susceptible ecosystem needs to be managed based on accurate information. One of the most important information is forest density. In this study the potential of Landsat ETM+ data in forest density (canopy area percentage) mapping was investigated. ETM image from a rugged forested area (50000 ha) in west of Iran, dating may 2000, was analyzed. The quality of the image was first evaluated. No noticeable radiometric and geometric distortion was detected. Image orthorectification was performed using the satellite ephemeris data, an accurate digital elevation model and 14 ground control points. The RMS error was less than half of a pixel. A precise ground truth map with four forest density classes was prepared for 20% of the study area. 26 aerial photographs (1:40000) dated June 1997 were interpreted visually based on fieldwork. The resulted forest density polygons were digitized to generate the ground truth map. Various synthetic bands such as bands resulted from image fusion, principal component analysis, tasseled cap transformation and band rationing were used. The best spectral band-sets for classification were selected using Bhattacharrya criterion based on training areas. The supervised classification of data was performed using spectral angle mapper (SAM), maximum likelihood, minimum distance to mean and parallelepiped classifiers. Along with four density classes (very thin, thin, semi-dense and dense) MD classifier showed the highest overall accuracy and kappa coefficient equal to 53% and 0.39 respectively. Signature separability and classification accuracies showed that the second and the third classes had the most spectral reflection similarity. After merging these two classes the classification was repeated. In this case the ML classifier showed the highest overall accuracy and kappa coefficient equal to 66% and 0.50 respectively. Based on these results, in such regions, low forest canopy increases the role of background reflection. High spatial resolution image and advanced classification methods, such as object-base classification should be considered to improve the potential of this application.