The resulting simulations perform well in certain key measures relative to highly-tuned simulations without satellite data assimilation. For example, the mixing heights at sounding locations are more accurate during most of the episode with satellite data assimilation. Wind fields are also improved on critical days when the original model run had difficulty.
Nevertheless, the large subgrid scale variations in land surface characteristics may cause systematic errors in the satellite assimilation technique which would be insignificant over more homogeneous surfaces. Using idealized models, we estimate the magnitude of these errors and develop corrections for them.
The first such error is in the estimate of surface skin temperature itself. While the assimilation technique assumes that the satellite-observed skin temperature represents an average over the image pixel, the actual emissions at the relevant infrared wavelengths from surfaces within the pixel are a strongly nonlinear function of temperature. This introduces a small positive bias in the satellite temperature estimates which depends on the degree of subpixel temperature inhomogeneity.
The second error occurs is caused not by the assimilation technique but by the model itself due to the neglect of subgrid scale variability within the PBL scheme. The neglect of this process causes a small negative bias in the estimation of surface fluxes from an inhomogeneous surface. This effect is quantified with the use of a one-dimensional PBL model. This error, as well, depends on the magnitude of subgrid surface inhomogeneity, so the two errors act to oppose each other.
Supplementary URL: