Frey, KE, Smith, LC (2007). How well do we know northern land cover? Comparison of four global vegetation and wetland products with a new ground-truth database for West Siberia. GLOBAL BIOGEOCHEMICAL CYCLES, 21(1), GB1016.
An unprecedented collection of 2161 geolocated, irregularly spaced field observations of land cover spanning similar to 10(6) km(2) throughout West Siberia suggests that currently available land cover classification products are remarkably poor indicators of vegetation type and water body extent in this northern wetland environment. The ground-truth data are compared with (1) the Global Land Cover Characteristics database derived from Advanced Very High Resolution Radiometer data (GLCC.AVHRR), (2) the Global Land Cover Classification derived from Moderate Resolution Imaging Spectroradiometer data (GLCC.MODIS), (3) the Global Lakes and Wetlands Database (GLWD), and (4) the West Siberian Lowland Peatland Database (WSLPD) using: (1) all land cover categories and (2) permanent wetland and water body categories only. Overall agreement with ground observations of land cover is only 21% for the GLCC.AVHRR database and 11% for the GLCC.MODIS database. However, at a much broader scale (one degree of latitude) there is improved qualitative agreement between vegetation classes, with some notable exceptions: (1) at low latitudes (similar to 54 degrees N-58 degrees N), both the GLCC.AVHRR and GLCC.MODIS databases underestimate woody savannas in favor of croplands; (2) at midlatitudes (similar to 58 degrees N-64 degrees N), the GLCC.AVHRR database underestimates evergreen needleleaf forest in favor of mixed forest; and (3) at high latitudes (similar to 64 degrees N-68 degrees N), both the GLCC.AVHRR and GLCC.MODIS databases underestimate deciduous needleleaf forest in favor of woody savannas or open shrublands, respectively. It is clear that all four data databases significantly underestimate permanent wetlands and water bodies, although those specifically developed for this purpose (GLWD and WSPLD) perform better than the more widely used, multiclass land cover products. For permanent wetlands, agreement with our ground data is only 2% for GLCC.MODIS, 23% for GLCC.AVHRR, 45% for GLWD and 56% for WSLPD. Agreement with open water bodies is even poorer (0-5%). These results raise into question the efficacy of incorporating currently available land cover products into terrestrial ecosystem models in northern wetland environments.