Sources of discrepancies between satellite-derived and land surface model estimates of latent heat fluxes

Monthly-average estimates of latent heat flux have been derived from a combination of satellite-derived microwave emissivities, day-night differences in land surface temperature (from microwave AMSR-E), downward solar and infrared fluxes from ISCCP cloud analysis, and MODIS visible and near-infrared surface reflectances. The estimates, produced with a neural network, were compared with data from the Noah land surface model, as produced for GLDAS-2, and with two alternative estimates derived from different datasets and methods. Areas with extensive, persistent, substantial discrepancies between the satellite and land surface model fluxes have been analyzed with the aid of data from flux towers. The sources of discrepancies were found to include problems with the model surface roughness length and turbulent exchange coefficients for midlatitude cropland areas in summer, inaccuracies in the precipitation data that were used as forcing for the land surface model, and model underestimation of transpiration in some forests during dry periods. At the tower sites analyzed, agreement with tower data was generally closer for our satellite-derived fluxes than for the land surface model fluxes, in terms of monthly averages.