Considerable advances have been made in the understanding of processes and feedbacks in the climate system. This has led to a better representation of processes and feedbacks in numerical climate models, which have become much more comprehensive. Because of the presence of non-linear processes in the climate system, deterministic projections of changes are potentially subject to uncertainties arising from sensitivity to initial conditions or to parameter settings. Such uncertainties can be partially quantified from ensembles of climate change integrations, made using different models starting from different initial conditions. They necessarily give rise to probabilistic estimates of climate change. This results in more quantitative estimates of uncertainties and more reliable projections of anthropogenic climate change. While improved parametrizations have built confidence in some areas, recognition of the complexity in other areas has not indicated an overall reduction or shift in the current range of uncertainty of model response to changes in atmospheric composition.

Atmospheric feedbacks largely control climate sensitivity. Important progress has been made in the understanding of those processes, partly by utilising new data against which models can be compared. Since the Second Assessment Report (IPCC, 1996) (hereafter SAR), there has been a better appreciation of the complexity of the mechanisms controlling water vapour distribution. Within the boundary layer, water vapour increases with increasing temperatures. In the free troposphere above the boundary layer, where the greenhouse effect of water vapour is most important, the situation is less amenable to straightforward thermodynamic arguments. In models, increases in water vapour in this region are the most important reason for large responses to increased greenhouse gases.

Water vapour feedback, as derived from current models, approximately doubles the warming from what it would be for fixed water vapour. Since the SAR, major improvements have occurred in the treatment of water vapour in models, although detrainment of moisture from clouds remains quite uncertain and discrepancies exist between model water vapour distributions and those observed. It is likely that some of the apparent discrepancy is due to observational error and shortcomings in intercomparison methodology. Models are capable of simulating the moist and very dry regions observed in the tropics and sub-tropics and how they evolve with the seasons and from year to year, indicating that the models have successfully incorporated the basic processes governing water vapour distribution. While reassuring, this does not provide a definitive check of the feedbacks, though the balance of evidence favours a positive clear-sky water vapour feedback of a magnitude comparable to that found in simulations.

Probably the greatest uncertainty in future projections of climate arises from clouds and their interactions with radiation. Cloud feedbacks depend upon changes in cloud height, amount, and radiative properties, including short-wave absorption. The radiative properties depend upon cloud thickness, particle size, shape, and distribution and on aerosol effects. The evolution of clouds depends upon a host of processes, mainly those governing the distribution of water vapour. The physical basis of the cloud parametrizations included into the models has also been greatly improved. However, this increased physical veracity has not reduced the uncertainty attached to cloud feedbacks: even the sign of this feedback remains unknown. A key issue, which also has large implications for changes in precipitation, is the sensitivity of sub-grid scale dynamical processes, turbulent and convective, to climate change. It depends on sub-grid features of surface conditions such as orography. Equally important are microphysical processes, which have only recently been introduced explicitly in the models, and carry major uncertainties. The possibility that models underestimate solar absorption in clouds remains controversial, as does the effect of such an underestimate on climate sensitivity. The importance of the structure of the stratosphere and both radiative and dynamical processes have been recognised, and limitations in representing stratospheric processes add some uncertainty to model results.

Considerable improvements have taken place in modelling ocean processes. In conjunction with an increase in resolution, these improvements have, in some models, allowed a more realistic simulation of the transports and air-sea fluxes of heat and fresh water, thereby reducing the need for flux adjustments in coupled models. These improvements have also contributed to better simulations of natural large-scale circulation patterns such as El Niño-Southern Oscillation (ENSO) and the oceanic response to atmospheric variability associated with the North Atlantic Oscillation (NAO). However, significant deficiencies in ocean models remain. Boundary currents in climate simulations are much weaker and wider than in nature, though the consequences of this fact for the global climate sensitivity are not clear. Improved parametrizations of important sub-grid scale processes, such as mesoscale eddies, have increased the realism of simulations but important details are still under debate. Major uncertainties still exist with the representation of small-scale processes, such as overflows and flow through narrow channels (e.g., between Greenland and Iceland), western boundary currents (i.e., large-scale narrow currents along coastlines), convection, and mixing.

Continues on next page

IPCC Homepage