The very large consensus concerning the main patterns of climate change, the stability of the results through the various IPCC reports necessarily indicate that these patterns are the result of a few dominant physical processes. These processes have been identified over the years. A first issue is that of resolution, since the stability of model results, as model resolution was steadily increasing, may be seen as a surprise. It has been known since the mid-1980s that an atmospheric resolution of about T42 (for spectral models, or equivalently 300 km grid mesh, for finite difference dynamical codes) was enough to resolve the largest part of the mid-latitude energy transport by baroclinic eddies. The more recent analysis of climate models also shows that a number of atmospheric dynamical features are not strongly affected by climate change as resulting from greenhouse increase. Fig. 1 shows the geographical limits of the Hadley cells as diagnosed from the AR4 models [5,6] and the variations which they may undergo in response to a CO₂ doubling. The variations due to climate change are small compared to the differences between models. An important component of the climate response may therefore be explained simply by ignoring those changes. We may therefore distinguish: (i) a component of temperature and precipitation changes which does not depend on possible modifications of the Hadley cell extension and may offer some elements of consensus between models; and (ii) a component which does depend on these changes. This second component is less consensual, as the models tend to diverge in the assessment of these Hadley cell modifications.

In consequence, let us first consider that the limit between the traditional atmospheric circulation regimes (Hadley cells in the Tropics, mid-latitude disturbances) is little affected by climate changes. The most important changes are then driven by thermodynamic rather than dynamic features. The geographical patterns of climate change warming then primarily depend on a few amplifying or feedback factors. The larger warming over the polar regions, although not fully understood, is certainly the combined result of a decreased surface albedo associated with sea ice and snow melting, of the heat accumulation within the Arctic ocean associated with a stronger penetration of the solar flux, and a very stable atmospheric temperature lapse rate which maintains a stronger warming near the ground. The larger continental warming is generally attributed to the larger heat capacity of the oceans, although some have also pointed the different continental and oceanic share between the sensible and latent heat fluxes that participate to the surface cooling in response to radiative heating. The explanations of the global patterns of precipitation changes resulting from a global warming are probably more counterintuitive. The larger precipitation values in the equatorial regions, shown by all models, are not the result of an increasing strength of the Hadley and Walker cell circulation [7]. For example, Fig. 2 [5,6] shows that on the contrary, in most models, the intensity of the Hadley circulation tends to diminish with global warming, as will be further commented in Section 4 [19]. The increased equatorial precipitations therefore mostly result from an increased atmospheric loading in water vapor, associated with the existence of higher saturation levels in a warmer world. The same argument holds to explain larger precipitation values at mid-latitudes. The smaller precipitation values in the dry subtropical areas, a feature which is also shown very systematically by most models, reflect on the contrary, a different physical rationale, probably as robust, although more complex. In a warmer and moister world, the Hadley cells transport more heat and more moisture toward the Equator in the lower atmospheric levels. It must therefore export more geopotential in the poleward direction, to maintain a total transport of energy which is oriented from the equator toward the poles. In this process, the upper part of the Hadley cell and the detrainment of moist air from convective system have to be located at higher altitudes, where the atmosphere is also drier and the subtropical subsiding regions are also drier.

In view of the robustness of climate change patterns associated with temperature and precipitation, it may come as a surprise that the climate sensitivity displayed by the different models can, on the contrary, be very different from one model to the other. Climate sensitivity is defined as the global equilibrium warming which occurs in response to a CO₂ doubling. The Charney report, in 1979, based on the results of two models only, determined that climate sensitivity could lie within a range going from 1.5 to 4.5 °C. Fig. 3, from the IPCC Third Assessment report shows that this range has hardly changed over the years, with global equilibrium temperature changes still between 2 and 5 °C approximately. The same figure also shows a general consistency between the global temperature and precipitation changes, which constitutes a good indication that climate sensitivity is an important indicator of climate impacts.

This is why such a persisting range in climate sensitivity, also confirmed by the AR4, is very troublesome: better constraining climate sensitivity probably constitutes one of the most important open scientific questions. Over the years, it has been demonstrated that climate sensitivity is dependent on a small number of key feedback processes. The most important of them is probably water vapor feedback, which roughly doubles climate sensitivity, in most models. Following a pioneering paper par Raval and Ramanthan [18], the IPCC Second Assessment Report referred to the existence and importance of this feedback as very consensual. The complexity of the processes involved already mentioned by Lindzen [14], was strongly revisited in the TAR. The idea of the water vapor feedback is indeed based on a simple initial argument: as temperature increases, the Clausius-Clapeyron law indicates than the water vapor saturation value also increases, and that this may cause an increase in the water vapor mixing ratio itself, which in turn favors an additional greenhouse effect. But this water vapor and greenhouse increase is effective only when changes in the water vapor atmospheric content are really dictated by changes in its saturation value, which is not the case, for example, in semiarid areas, or, partly, in the upper atmosphere. In addition, any change in the atmospheric temperature (as may result from a change in the atmospheric temperature lapse rate) will also modify the infrared emission by water vapor and therefore the amplitude of water vapor feedback. The complexity of cloud feedbacks is much larger still, because clouds act both on the terrestrial part of the radiative budget (positive greenhouse feedback) and on the solar part (negative albedo feedback). While models converge in predicting a positive water vapor feedback, the net cloud effect is much less certain. Other feedbacks concern the land surface (snow and ice albedo, land cover) and also add to these uncertainties.

Relating those modeled feedbacks to actual observations is a very difficult task: future climates are of course not documented at all and past climate observations, although very precious, do not cover many atmospheric parameters and processes. The only time span for which we possess a sufficient knowledge of water vapor and cloud modifications is the period in which satellite data have been available, therefore the last 25 years approximately. During this period the earth has experienced a number of climate fluctuations. The most important one is of course the seasonal cycle, but the satellite era is also marked by interannual variability, including a number of ENSO events, volcanic eruptions such as the Pinatubo, etc. Using these changes to assess water vapor and cloud feedbacks is now a classical approach, already developed in the mid-1990s [3], for water vapor feedbacks; [4], for methodological aspects concerning cloud feedbacks. Such a method has been used recently by Bony and Dufresne [2] to show the specific contribution of low clouds to climate sensitivity. This approach is based on the idea that, in the Tropics at least, there is a certain robustness of the thermodynamic feedbacks, which can be characterized according to the atmospheric circulation regimes in which they tend to develop. This approach is applicable to determine future climate changes if we can assume that the atmospheric circulation will not change too much in the future or that we will be able to predict those changes if they occur.

We have shown in the preceding sections that a large part of the current consensus concerning model results, e.g. the stability of the global patterns of precipitation and temperature changes or the best hopes to better constrain climate sensitivity are all based on the idea that the atmospheric circulation will not be deeply changed in response to climate change.

Most models confirm the idea that, indeed, atmospheric circulation changes may be held as a second order effect if one considers the global climate response. If this was not the case, the spread between model estimates of climate change would be much larger, in particular at regional scale. In Fig. 2, for example, we have shown the diagnostic of the winter Hadley cell intensity, by a number of models which participated to the last AR4 simulations and to the World Climate Research Programme (WCRP) Coupled Model Intercomparison Programme (CMIP) [5,6]. The range of the results is very large and, knowing the huge importance of the Hadley cell in shaping up the climate distribution over the Earth, this very large spread obtained using recent models only, is really troublesome. In addition, when compared with such variations, the modifications brought up by a CO₂ doubling appear to be small and are themselves not characterized by a real consensus between models: although most models indicate a weakening of the Hadley circulations in a warmer world, with a poleward extension of both the Hadley cells and mid-latitude circulations [20], some models show an opposite response.

Such changes in the atmospheric circulation may have an impact larger than anticipated. Pierrehumbert [17] has shown that climate sensitivity results in a large part from the competition between the continuous destabilization of equatorial regions, submitted to a run-away greenhouse effect, and the stabilizing effect of the subtropics. Dynamical processes are necessarily involved in this competition and changes, for example, of the area covered by ascending motions are likely to modify climate sensitivity [1].

In addition, changes in the Hadley cell intensity are generally accompanied by (equally small) changes in the Hadley cell limits, which tend to extend poleward [15]. Gastineau [5] or Gastineau et al. [6] have reviewed some of the mechanisms which may explain the lack of consensus of current models in simulating both the Hadley limits and their changes and how this may in turn induce uncertainties concerning, for example, local changes in precipitations. These uncertainty factors are numerous: the Hadley cell intensity and extension are affected by the atmospheric vertical stratification, or the latitudinal temperature gradients at mid-latitude, which also govern in part the dynamics of mid-latitude perturbations.

These uncertainties are even larger and more systematic if one considers not only the mean meridional circulation but also the more complex features which correspond to longitudinal asymmetries: monsoons, ENSO ... The role of dynamical changes in determining the local aspects of climate modifications can be very important and their study is only beginning.

The assessment of the simulated climate changes from various climate models has been carried so far with the objective of alerting government and decision makers on the reality of climatic risk. This alert has been greatly favored because a number of key features in the climate response are organized following very robust patterns, which are confirmed by recent models, with a higher resolution and a more comprehensive physics. This robustness is partly due to the fact that many global aspects of climate changes are not strongly affected by changes in the atmospheric dynamics.

Assessing local climate changes is a much more difficult challenge because it does imply a good understanding of how atmospheric dynamics may be affected by a climate warming. This is a task which is largely in front of us and, at the local or regional levels, model results are still affected by large uncertainties.

A key question for the future is to know what we can expect from the continuing increase in model resolution. A number of parameterization problems affecting climate sensitivity (for example: low cloud formation and dissipation) will not be solved unless model resolution becomes so high that it may capture small turbulent and convective cells in the atmospheric boundary layer. Likewise the nonlinear interactions which, especially at mid-latitudes, condition the predictability, or lack or predictability, of regional climate change patterns will not necessarily be better represented in models with an increased resolution. Increasing model resolution is an important and necessary challenge, but it may not be sufficient. It is one of the necessary advancements if we wish to better explore and predict the regional impacts of climate changes. But it is not the only way of improving models: better formulations and tests of the model physical parameterizations are also necessary. Even with such efforts, the global amplitude of future climate changes, the regional distribution of these changes may very well remain uncertain: success in substantially improving climate projections cannot be warranted at this stage.