The Sensitivity of a Land Surface Parameterization Scheme to the Choice of Remotely Sensed Land-Cover Datasets

The characteristics of satellite-derived land-cover data for climate models vary depending on sensor properties and processing options. To better understand the first-order effects of differences in land-cover data on a land surface parameterization scheme (VBATS), stand-alone model runs were performed for two adjacent 2.8° X 2.8° GCM grid cells in Wyoming using land cover from two satellite-derived maps (AVHRR, TM) and a global land-cover dataset commonly used in GCMs. The dominant cover type by area differed among the datasets for both grid cells. In the western grid cell, these differences resulted in substantially different surface fluxes simulated by VBATS. At spatial resolutions of 0.2° and 0.4°, the two satellite-derived datasets agreed on only 54%-62% of the land-cover types in both grid cells. Despite this disagreement, the VBATS simulated surface fluxes averaged over the grid cell were similar in the eastern grid cell for the two satellite-derived datasets. In the western grid cell, the partitioning of net radiation into sensible and latent heat fluxes was influenced by the dataset prescriptions of land-cover heterogeneity. In particular, the relative proportions of wet cover types (i.e., inland water and irrigated crop) had an effect on this partitioning, emphasizing the importance of accounting for the presence of wet cover types within a GCM grid cell in arid regions. Spatial aggregation of the satellite-derived datasets reduced the number of land-cover types prescribed for each GCM grid cell. In the western grid cell, the reduction in the number of cover types from 11 to 2 resulted in differences in annual averages of sensible and latent heat fluxes of about 10%. Other simulations involving these datasets suggest that these differences could be reduced if one accounted for the wet cover types. In this respect, fine spatial resolution is required for some cover types, whereas coarser resolution may be adequate for other types. Land-cover classifications for land surface modeling need to be based more on model sensitivities than on traditional vegetation-type schemes.