220.127.116.11 Surface Temperature and the Climate System’s Energy Budget
For models to simulate accurately the global distribution of the annual and diurnal cycles of surface temperature, they must, in the absence of compensating errors, correctly represent a variety of processes. The large-scale distribution of annual mean surface temperature is largely determined by the distribution of insolation, which is moderated by clouds, other surface heat fluxes and transport of energy by the atmosphere and to a lesser extent by the ocean. Similarly, the annual and diurnal cycles of surface temperature are governed by seasonal and diurnal changes in these factors, respectively, but they are also damped by storage of energy in the upper layers of the ocean and to a lesser degree the surface soil layers.
Figure 8.2a shows the observed time mean surface temperature as a composite of surface air temperature over regions of land and SST elsewhere. Also shown is the difference between the multi-model mean field and the observed field. With few exceptions, the absolute error (outside polar regions and other data-poor regions) is less than 2°C. Individual models typically have larger errors, but in most cases still less than 3°C, except at high latitudes (see Figure 8.2b and Supplementary Material, Figure S8.1). Some of the larger errors occur in regions of sharp elevation changes and may result simply from mismatches between the model topography (typically smoothed) and the actual topography. There is also a tendency for a slight, but general, cold bias. Outside the polar regions, relatively large errors are evident in the eastern parts of the tropical ocean basins, a likely symptom of problems in the simulation of low clouds. The extent to which these systematic model errors affect a model’s response to external perturbations is unknown, but may be significant (see Section 8.6).
Figure 8.2. (a) Observed climatological annual mean SST and, over land, surface air temperature (labelled contours) and the multi-model mean error in these temperatures, simulated minus observed (colour-shaded contours). (b) Size of the typical model error, as gauged by the root-mean-square error in this temperature, computed over all AOGCM simulations available in the MMD at PCMDI. The Hadley Centre Sea Ice and Sea Surface Temperature (HadISST; Rayner et al., 2003) climatology of SST for 1980 to 1999 and the Climatic Research Unit (CRU; Jones et al., 1999) climatology of surface air temperature over land for 1961 to 1990 are shown here. The model results are for the same period in the 20th-century simulations. In the presence of sea ice, the SST is assumed to be at the approximate freezing point of seawater (–1.8°C). Results for individual models can be seen in the Supplementary Material, Figure S8.1.
In spite of the discrepancies discussed here, the fact is that models account for a very large fraction of the global temperature pattern: the correlation coefficient between the simulated and observed spatial patterns of annual mean temperature is typically about 0.98 for individual models. This supports the view that major processes governing surface temperature climatology are represented with a reasonable degree of fidelity by the models.
An additional opportunity for evaluating models is afforded by the observed annual cycle of surface temperature. Figure 8.3 shows the standard deviation of monthly mean surface temperatures, which is dominated by contributions from the amplitudes of the annual and semi-annual components of the annual cycle. The difference between the mean of the model results and the observations is also shown. The absolute differences are in most regions less than 1°C. Even over extensive land areas of the NH where the standard deviation generally exceeds 10°C, the models agree with observations within 2°C almost everywhere. The models, as a group, clearly capture the differences between marine and continental environments and the larger magnitude of the annual cycle found at higher latitudes, but there is a general tendency to underestimate the annual temperature range over eastern Siberia. In general, the largest fractional errors are found over the oceans (e.g., over much of tropical South America and off the east coasts of North America and Asia). These exceptions to the overall good agreement illustrate a general characteristic of current climate models: the largest-scale features of climate are simulated more accurately than regional- and smaller-scale features.
Like the annual range of temperature, the diurnal range (the difference between daily maximum and minimum surface air temperature) is much smaller over oceans than over land, where it is also better observed, so the discussion here is restricted to continental regions. The diurnal temperature range, zonally and annually averaged over the continents, is generally too small in the models, in many regions by as much as 50% (see Supplementary Material, Figure S8.3). Nevertheless, the models simulate the general pattern of this field, with relatively high values over the clearer, drier regions. It is not yet known why models generally underestimate the diurnal temperature range; it is possible that in some models it is in part due to shortcomings of the boundary-layer parametrizations or in the simulation of freezing and thawing soil, and it is also known that the diurnal cycle of convective cloud, which interacts strongly with surface temperature, is rather poorly simulated.
Surface temperature is strongly coupled with the atmosphere above it. This is especially evident at mid-latitudes, where migrating cold fronts and warm fronts can cause relatively large swings in surface temperature. Given the strong interactions between the surface temperature and the temperature of the air above, it is of special interest to evaluate how well models simulate the vertical profile of atmospheric temperature. The multi-model mean absolute error in the zonal mean, annual mean air temperature is almost everywhere less than 2°C (compared with the observed range of temperatures, which spans more than 100°C when the entire troposphere is considered; see Supplementary Material, Figure S8.4). It is notable, however, that near the tropopause at high latitudes the models are generally biased cold. This bias is a problem that has persisted for many years, but in general is now less severe than in earlier models. In a few of the models, the bias has been eliminated entirely, but compensating errors may be responsible. It is known that the tropopause cold bias is sensitive to several factors, including horizontal and vertical resolution, non-conservation of moist entropy, and the treatment of sub-grid scale vertical convergence of momentum (‘gravity wave drag’). Although the impact of the tropopause temperature bias on the model’s response to radiative forcing changes has not been definitively quantified, it is almost certainly small relative to other uncertainties.
Figure 8.3. Observed standard deviation (labelled contours) of SST and, over land, surface air temperature, computed over the climatological monthly mean annual cycle, and the multi-model mean error in the standard deviations, simulated minus observed (colour-shaded contours). In most regions, the standard deviation provides a measure of the amplitude of the seasonal range of temperature. The observational data sets, the model results and the climatological periods are as described in Figure 8.2. Results for individual models can be seen in the Supplementary Material, Figure S8.2.
18.104.22.168.2 The balance of radiation at the top of the atmosphere
The primary driver of latitudinal and seasonal variations in temperature is the seasonally varying pattern of incident sunlight, and the fundamental driver of the circulation of the atmosphere and ocean is the local imbalance between the shortwave (SW) and longwave (LW) radiation at the top of the atmosphere. The impact on temperature of the distribution of insolation can be strongly modified by the distribution of clouds and surface characteristics.
Considering first the annual mean SW flux at the ‘top’ of the atmosphere (TOA), the insolation is determined by well-known orbital parameters that ensure good agreement between models and observations. The annual mean insolation is strongest in the tropics, decreasing to about half as much at the poles. This largely drives the strong equator-to-pole temperature gradient. As for outgoing SW radiation, the Earth, on average, reflects about the same amount of sunlight (~100 W m–2 in the annual mean) at all latitudes. At most latitudes, the difference between the multi-model mean zonally averaged outgoing SW radiation and observations is in the annual mean less than 6 W m–2 (i.e., an error of about 6%; see Supplementary Material, Figure S8.5). Given that clouds are responsible for about half the outgoing SW radiation, these errors are not surprising, for it is known that cloud processes are among the most difficult to simulate with models (see Section 22.214.171.124.3).
There are additional errors in outgoing SW radiation due to variations with longitude and season, and these can be quantified by means of the root-mean-square (RMS) error, calculated for each latitude over all longitudes and months and plotted in Figure 8.4a (see also Supplementary Material, Figure S8.6). Errors in the complete two-dimensional fields (see Supplementary Material, Figure S8.6) tend to be substantially larger than the zonal mean errors of about 6 W m–2, an example of the common result that model errors tend to increase as smaller spatial scales and shorter time scales are considered. Figure 8.4a also illustrates a common result that the errors in the multi-model average of monthly mean fields are often smaller than the errors in the individual model fields. In the case of outgoing SW radiation, this is true at nearly all latitudes. Calculation of the global mean RMS error, based on the monthly mean fields and area-weighted over all grid cells, indicates that the individual model errors are in the range 15 to 22 W m–2, whereas the error in the multi-model mean climatology is only 13.1 W m–2. Why the multi-model mean field turns out to be closer to the observed than the fields in any of the individual models is the subject of ongoing research; a superficial explanation is that at each location and for each month, the model estimates tend to scatter around the correct value (more or less symmetrically), with no single model consistently closest to the observations. This, however, does not explain why the results should scatter in this way.
Figure 8.4. Root-mean-square (RMS) model error, as a function of latitude, in simulation of (a) outgoing SW radiation reflected to space and (b) outgoing LW radiation. The RMS error is calculated over all longitudes and over all 12 months of a climatology formed from several years of data. The RMS statistic labelled ‘Mean Model’ is computed by first calculating the multi-model monthly mean fields, and then calculating the RMS error (i.e., it is not the mean of the individual model RMS errors). The Earth Radiation Budget Experiment (ERBE; Barkstrom et al., 1989) observational estimates used here are for the period 1985 to 1989 from satellite-based radiometers, and the model results are for the same period in the 20th-century simulations in the MMD at PCMDI. See Table 8.1 for model descriptions. Results for individual models can be seen in the Supplementary Material, Figures S8.5 to S8.8.
At the TOA, the net SW radiation is everywhere partially compensated by outgoing LW radiation (i.e., infrared emissions) emanating from the surface and the atmosphere. Globally and annually averaged, this compensation is nearly exact. The pattern of LW radiation emitted by earth to space depends most critically on atmospheric temperature, humidity, clouds and surface temperature. With a few exceptions, the models can simulate the observed zonal mean of the annual mean outgoing LW within 10 W m–2 (an error of around 5%; see Supplementary Material, Figure S8.7). The models reproduce the relative minimum in this field near the equator where the relatively high humidity and extensive cloud cover in the tropics raises the effective height (and lowers the effective temperature) at which LW radiation emanates to space.
The seasonal cycle of the outgoing LW radiation pattern is also reasonably well simulated by models (see Figure 8.4b). The RMS error for most individual models varies from about 3% of the outgoing LW radiation (OLR) near the poles to somewhat less than 10% in the tropics. The errors for the multi-model mean simulation, ranging from about 2 to 6% across all latitudes, are again generally smaller than those in the individual models.
For a climate in equilibrium, any local annual mean imbalance in the net TOA radiative flux (SW plus LW) must be balanced by a vertically integrated net horizontal divergence of energy carried by the ocean and atmosphere. The fact that the TOA SW and LW fluxes are well simulated implies that the models must also be properly accounting for poleward transport of total energy by the atmosphere and ocean. This proves to be the case, with most models correctly simulating poleward energy transport within about 10%. Although superficially this would seem to provide an important check on models, it is likely that in current models compensating errors improve the agreement of the simulations with observations. There are theoretical and model studies that suggest that if the atmosphere fails to transport the observed portion of energy, the ocean will tend to largely compensate (e.g., Shaffrey and Sutton, 2004).