Forest cover classification from MODIS images in Northeastern Asia

Forest ecosystem in Eastern Siberia and Northeastern China (ESNC) has been undergoing dramatic changes during the last several decades due to forest fires and massive logging. These changes affect climate dynamics, economic activity and living heritage in local region, further, to the global carbon balance and climate changes. In this paper, a 2D feature space grid split (FSGS) algorithm was developed to identify forests cover region by combined TM/ ETM+ images and MODIS datasets, due to its dark object attributes. This no-parametric algorithm was based on statistical signatures in feature space and Bayesian rule. The producer accuracy of tree cover commission can be approximately 90%, comparing with local TM/ETM+ classification results. Then, forests cover was stratified into different biomes by a decision tree classifier. and Forests cover map was respectively compared with MODIS land cover products and Global land cover 2000(GLC2000) products derived from images observed by VEGETATION (VGT) sensor on both areal and per-pixel bases.