Abstract: Vegetation coverage information is one of the significant indexes to represent desert eco-environment. Vegetation information of remote sensing spectral information is normally weaker due to the sparse vegetation in desert areas. There are various extraction methods of vegetation coverage using remote sensing data，which are limited in the application of arid desert. From this perspective，the comparison of different methods in desert areas will be essential way to find one feasible and suitable method of desert vegetation coverage information extraction at the larger scale within remote sensing. Taking desert areas in Xinjiang for example，six extraction methods of remote sensing （Modified maximal gradient difference model （Modified TGDVI），Pixel dichotomy model （PDM），Linear Spectral Unmixing （LSU） and vegetation index model （NDVI，EVI，MSAVI2）） were used to extract the vegetation fraction from MODIS images and the results were tested by observed data at field. It indicated that methods of vegetation index were mostly depends on the amount of observed values，so had different results. [EVI] which considered the factors of atmosphere and soil had the highest simulation accuracy while [NDVI] had the lowest accuracy of results as it only considered red and near-infrared radiation. Because it is difficult to extract pure desert vegetation pixel from MODIS images，using pixel values of crops instead of desert vegetation pixel could decreased the accuracy of PDM and LSU.  With the better accuracy（R²=0.74；RMSE=13.46），modified TGDVI could dramatically reflect the vegetation fraction differences between the northern and southern desert in Xinjiang，and was the appropriate vegetation coverage information extraction method in such large scale of desert areas.

Key words: desert of Xinjiang, remote sensing, fraction of sparse vegetation, MODIS images