A relief-based forest cover change extraction using GF-1 images

China is one of the countries with rapidest forest increase in the world due to the large-scale afforestation and strict natural forest protection. Promptly exact extraction of forest change provides key supporting of local governors' record assessment. High-resolution image has become prior choice to detect the spatial distribution and area estimate of forest cover change. Yunyang county at the viscera of Three-Gorges Reservoir is regarded as the study area, and the new fine-resolution optical satellite of GF-1 is introduced. Using 2000 forest plots as training samples, an improved Relief is explored through space building, J-M distance and correlative degree to establish optimal features matrix covering hue, homogeneity and NDVI from original twenty features. Intelligent classifier is modeled based on BP ANN to identify suitable features composition and threshold levels, and find out new forested area and deforestation sites during 2006-2013 in a scene of GF-1 image. The study results show that: 1) the composition of hue, homogeneity and NDVI owns higher accuracy of above 87%, of which reasonable threshold point is 4.0, 0.9 and 0.4 respectively; 2) there are 1200 patches of forest gain with 2620 ha mainly formed by cropland and four-sides planting, and 300 sites of converted forest with 1120 ha for harvest and forestland requisition and occupation; 3) the average area of forest gain is 2.2 ha less than that of deforestation (3.7 ha); 4) small patches of afforestation may passively increase landscape fragmentation, and unreasonable large area of trees harvest and illegal forestland occupation aggravate fragmentation because of disconnected forests and broken ecosystems.