Figure 8.7: Zonally averaged December-January-February total cloudiness simulated by ten AMIP1 models (a) and by revised versions of the same ten models (b). The solid black line gives observed data from the International Satellite Cloud Climatology Project (ISCCP). From Gates et al. (1999).
In this and the following two sub-sections we discuss simulations by AGCMs that are provided observed sea surface temperatures and sea-ice distributions as input boundary conditions. AOGCM control runs have not yet been thoroughly examined in studies of surface boundary fluxes or mid-tropospheric and stratospheric quantities.
Satellite observations over the past quarter of a century have provided estimates of top of atmosphere (TOA) flux that are considered reliable. Any discrepancies between models and observations are usually attributed to the inadequate modelling of clouds, since they are difficult to specify and accurately model, and account for most of the variability.
Unfortunately, there are no global estimates of surface flux that do not rely heavily on models. The best model-independent estimates come from the Global Energy Balance Archive (GEBA), a compilation of observations from more than 1,000 stations (Gilgen et al., 1998). Compared with GEBA observations, surface solar insolation is overestimated in most AGCMs (Betts et al., 1993; Garratt, 1994; Wild et al., 1997, 1998; Garratt et al., 1998). Downwelling long wave radiation, on the other hand, is underestimated (Garratt and Prata, 1996; Wild et al., 1997). The shortwave discrepancy is of more concern: it is more than a factor of two larger than the long-wave discrepancy, and could be due to missing absorption processes in the atmosphere.
The observations indicate that about 25% of the incident solar flux at the TOA is absorbed in the atmosphere, but most models underestimate this quantity by 5 to 8% of the of the incident solar flux (Arking, 1996, 1999; Li et al., 1997). The extent and the source (or sources) of this discrepancy have been intensely debated over the past five years, with investigations yielding contradictory results on whether the discrepancy is associated with clouds, aerosols, water vapour, or is an artefact of the instrumentation and/or the methods by which sensors are calibrated and deployed.
This discrepancy is important for climate modelling because it affects the partitioning of solar energy between the atmosphere and the surface. If the observations are correct, then improving the models will reduce the energy available for surface evaporation by 10 to 20% with a corresponding reduction in precipitation (Kiehl et al., 1995) and a general weakening of the hydrological cycle.
The SAR concluded that although atmospheric models adequately simulate the three-dimensional temperature distribution and wind patterns, "current models portray the large-scale latitudinal structure and seasonal change of the observed total cloud cover with only fair accuracy". Subsequent studies have confirmed both the good and bad aspects of model simulations. Throughout most of the troposphere, errors in AMIP1 ensemble simulations of temperature and zonal wind are small compared with either inter-model scatter or the observed spatial standard deviation (Gates et al., 1999). (See Section 8.8 for brief discussion of storm tracks.) On the other hand, discrepancies between models and observations that substantially exceed the observational uncertainty are evident for both clouds (Mokhov and Love, 1995; Weare et al., 1995, 1996; Weare, 2000a, 2000b) and upper tropospheric humidity (see Chapter 7).
Although solutions to these problems have proved elusive, incremental improvements have been noted since publication of the SAR. For total cloudiness, a revised subset of AMIP models exhibits noticeably less inter-model variation and significantly less average r.m.s error (Gates et al., 1999; Figure 8.7), compared with the original versions of the models. Several models adequately simulate seasonal changes in cloud radiative forcing (Cess et al., 1997). Model intercomparisons organised under the Global Energy and Water cycle Experiment (GEWEX) Cloud System Study (Stewart et al., 1998) will provide further information for improving cloud simulation. For tropospheric humidity, improved agreement with observations may result from improved numerical techniques (Section 8.9). Furthermore, even though the seasonal mean amounts of clouds and upper tropospheric water vapour are not well simulated in current climate models, variations of these quantities may be more important than absolute amounts for predicting climate changes. For example, Del Genio et al. (1994) noted that, in mid-latitudes, the seasonal cycle of upper tropospheric humidity can be simulated reasonably well by climate models. They argued that this variation provides a surrogate for decadal climate change in mid-latitudes because both are characterised by combined temperature increase and latitudinal temperature-gradient decrease, and thus both have similar effects on storms.
Examination of monsoons in climate models provides another measure of their ability to simulate hydrologic variations. Developments since publication of the SAR have been encouraging. Sperber and Palmer (1996) found that about half the original AMIP models obtained a realistic dependence of monsoon circulation on location and season. A follow-up study reveals that nearly all the revised AMIP models do so (Sperber et al., 1999; see Section 8.7.3).
Other reports in this collection