Large-scale land cover datasets comprise an important foundation of the research on land surface processes

ecosystem assessment

and environmental modeling. The evaluation of existing land cover datasets provides a guide to dataset use and new dataset production. Five kinds of global land cover datasets（ IGBP DISCover

UMD

GLC2000

MOD12 Q1

and GlobCover2005） over China and adjacent regions are evaluated in this paper. First

the categories of five land cover datasets are translated into the International Geosphere-Biosphere Programme-IGBP scheme based on the correlation coefficients of the corresponding classes

which is computed according to the class definition in each land cover dataset. Second

the spatial agreements of the five land c over datasets are analyzed using visual comparison and per-pixel comparison. Finally

the classification accuracy of the five land cover datasets is evaluated based on validation samples collected through Google Earth high-resolution satellite images. The results show large areas of disagreement among the five land cover datasets

and the overall consistency among them is low. GLC2000 has the highest overall accuracy and Kappa coefficient

whereas GlobCover 2005 has the lowest overall accuracy and Kappa coefficient.