Comparative study of change detection methods for forest extent changes using TM and ETM+ imagery

In this study، three common change detection methods including post classification، differential NDVI and PCA were compared using 1987 TM and 2001 ETM+ imagery to determine the kind and rate of forest extent changes in Golydaghi forestry plan. After geo-referencing of images، training areas from forest and non-forest classes were selected and classification of images was done using maximum likelihood classifier. Accuracy of classified images was assessed using some samples ground truth. Results showed that overall accuracy of TM and ETM+ classified images were %92. 02 and %92. 66، respectively. The classified maps were crossed to produce the changed forest map. In differencing NDVI method، NDVI indices were created for two dates and differenced each other. Thresholds of change were determined using NDVI values of 80 control points from unchanged forested and non forested areas and the change map was created using applying two standard deviations (SD) thresholds. Results showed thatapplying two SD could mapped changes better، compared to one SD. In the one date differential PCA، first component of PCA on the TM and ETM+ images were selected and differentiated each other. In two dated PCA، third and fourth component of PCA of whole TM and ETM+ images were selected and one and two SD thresholds were applied for changes mapping. Generally، accuracy assessment of change detection methods using ground truth maps showed that post classification method could better mapped changes with overall accuracy of %85. 13 and about 0. 51 Kappa coefficient compared to other methods.