Grassland phenology extraction and trend analysis based on MODIS data in Xinjiang

The pros and cons of vegetation phenology reconstruction and extraction methods for different regions still need to be fully elucidated. Therefore, performing a comparative analysis between different methods is necessary to achieve optimal remote sensing extraction in a specific region. In this study, the asymmetric Gaussian function fitting (Asymmetric Gaussian, AG), double logistic function fitting (Double Logistic, DL), and Savitzky-Golay filter (SG) methods were used to extract the end of season (EOS) data for grasslands in Xinjiang from 2001 to 2019. By comparing the extraction results of four methods, an optimal model suitable for extracting EOS was established and the temporal and spatial changes in the EOS data were studied in Xinjiang. The main conclusions of this study were as follows: 1) The SOS results for Xinjiang grassland extracted using the D-L method were the best. 2) The grassland area in Xinjiang estimated using the three methods showed distinct regional differences; EOS gradually advanced from north to south. 3) The grassland EOS estimated using all three methods showed an advancing trend, and the areas where EOS advanced significantly (P < 0.1) are distributed in the area south of the Altai Mountains and in the central and western marginal areas of the Junggar Basin. 4) The annual variation trend of grassland phenology extracted using A-G and D-L was consistent for all the studied grassland types.