2.2. Landsat ETM+ data
One Landsat 7 ETM+ SLC-off data of the study area on August 18,2005 was downloaded from USGS website for forest cover classification. This data was selected because ofmany land cover products were provided in 2005 and the classification result could be compared with the existed products.Furthermore, the vegetation in August was at exuberant growthstage in this study area and the ETM+ data was nearly not affectedby cloud. For the scan-line corrector of the Landsat 7 ETM+ sen-sor failed in 2003, it was resulted in about 22% of the pixels perscene not being scanned . To fill in the datagaps, another Landsat data in August 21, 2006 was downloaded from Global Land Cover Facility Website of Maryland University. The data gaps was filled using the method developed by , which was based on the assump-tion that the same class neighboring pixels around the un-scannedpixels had the similar spectral characteristics and these neigh-boring and un-scanned pixels exhibit similar patterns of spectraldifferences between dates. The projection of the data was convertedto Albers Equal Area projection with World Geodetic System 84(WGS-84) datum to keep the forest cover area in the data consistedwith the actual ground surface. Finally, a rectangle subset consistedof 5013 columns × 5542 lines × 6 visible-NIR bands were extractedas the final data for forest cover classification to reduce the influ-ence by the gaps .