Wild et al. , a blue-chip study, shows that the downward longwave radiation in cold, dry climates is dramatically under-estimated in the GCMs used in IPCC TAR, as shown in the following excerpt from their Figure 4 (from one of the best GCMs). The bias is systemic.
FIG. 4. Annual cycles of model-calculated and observed DLR (W m22) at some of the most reliable high-latitude sites, midlatitude sites, and low-latitude sites (see Table 1 for more information on these sites): model calculations by ECHAM3 (dotted) and ECHAM4 (dashed), and observed (solid). MARTIN WILD, ATSUMU OHMURA, AND HANS GILGEN, JEAN-JACQUES MORCRETTE, ANTHONY SLINGO, 2001, Evaluation of Downward Longwave Radiation in General Circulation Models, Journal of Climate 14.
Abstract: The longwave radiation emitted by the atmosphere toward the surface [downward longwave radiation (DLR)] is a crucial factor in the exchange of energy between the earth surface and the atmosphere and in the context of radiation-induced climate change. Accurate modeling of this quantity is therefore a fundamental prerequisite for a reliable simulation and projection of the surface climate in coupled general circulation models (GCM). DLR climatologies calculated in a number of GCMs and in a model in assimilation mode (reanalysis) are analyzed using newly available data from 45 worldwide distributed observation sites of the Global Energy Balance Archive (GEBA) and the Baseline Surface Radiation Network (BSRN). It is shown that substantial biases are present in the GCM-calculated DLR climatologies, with the GCMs typically underestimating the DLR (estimated here to be approximately 344 W m22 globally). The biases are, however, not geographically homogeneous, but depend systematically on the prevailing atmospheric conditions. The DLR is significantly underestimated particularly at observation sites with cold and dry climates and thus little DLR emission. This underestimation gradually diminishes toward sites with more moderate climates; at sites with warm or humid atmospheric conditions and strong DLR emission, the GCM-calculated DLR is in better agreement with the observations or even overestimates them. This is equivalent to creating an excessively strong meridional gradient of DLR in the GCMs.The very same tendencies are independently found in stand-alone calculations with the GCM radiation codes in isolation, using observed atmospheric profiles of temperature and humidity for cloud-free conditions as input to the radiation schemes. A significant underestimation of DLR is calculated by the radiation schemes when driven with clear-sky atmospheric profiles of temperature and humidity representative for cold and dry climates, whereas the DLR is no longer underestimated by the radiation schemes with prescribed clear-sky profiles representative for a hot and humid atmosphere. This suggests that the biases in the GCM-calculated DLR climatologies are predominantly induced by problems in the simulated emission of the cloud-free atmosphere. The same biases are also found in the DLR fluxes calculated by the European Centre for Medium-Range Weather Forecasts (ECMWF) model in assimilation mode (reanalysis), in which the biases in the atmospheric thermal and humidity structure are minimized. This gives further support that the biases in the DLR are not primarily due to errors in the model-predicted atmospheric temperature and humidity profiles that enter the radiative transfer calculations, but rather are due to the radiation schemes themselves. A particular problem in these schemes is the accurate simulation of the thermal emission from the cold, dry, cloud-free atmosphere.