Estimation And Dynamic Change Analysis Of Vegetation Fractional Coverage Based On Mixed-pixel Unmixing Model In The Loess Plateau

The reconstruction of vegetation is an important task of the ecological environmental improvement; it is also a fundamental strategy to control soil erosion in the Loess Plateau. As an indicator of surface vegetation coverage, Extracting and monitoring of vegetation fractional coverage is significant in ecological environmental construction and soil and water conservation.In this paper, taking the Loess Plateau as studying area and 500 m resolution MODIS MOD09A1 products of 2006 as data sources, linear spectral mixture model was used to unmix in the area. Then, vegetation fractional coverage was extracted and accuracy assessment was made, and danymic changes of vegetation were analyzed. Following are main results:1. The mixed-pixels are unmixed in the Loess Plateau by using linear spectral mixture model. The linear spectral mixture model was selected among a variety of mixed-pixel unmixing models. According to MNF transformation and PPI calculation, soil,vegetation,water were identified as the ultimate endmembers, and unmixing accuracy was evaluated based on TM classification. The pixel accuracy of soil,vegetation and water are respectively 92.31%,91.54%,89.21%.2. Taking the average of the vegetation abundance map based on unmixing results as the vegetation fractional coverage, extracting accuracy of vegetation fractional coverage by MOD09A1 product were evaluated. The values of vegetation fractional coverage based on the mixed-pixel unmixing model are slightly higher than results based on the dimidiate pixel model. On the other hand, the vegetation fractional coverage based on the mixed-pixel unmixing model are 8.46% lower compared to the results derived from TM, the accuracy is 86.56%.3. The dynamic changes of vegetation fractional coverage were analyzed in the Loess Plateau. The vegetation fractional coverage of the four reasons are extracted and compared by unmixing and the dimidiate pixel model. In addition, the vegetation fractional coverage per month of 2006 are extracted by the dimidiate pixel model, and the time-series of vegetation fractional coverage was drew to analysis vegetation phenology in the area. The time-series curve in 2006 show that the vegetation fractional coverage is the lowest in April, and come into growing and get to the peak by late July to the early August. At last, senescent period begins in October. The classification show that the vegetation fractional coverage is mainly distributed in south-east part of the Loess Plateau, and has shown the trend of gradually decreasing from the southeast to northwest, there are basically desert grassland landscape in the northwest. The seasonal and annual changes mainly distributed in the northern Shaanxi and majority of northern Shanxi, Ningxia - hetao plain, the Yellow River and Huangshui Valley in the eastern Qinghai.