The regional response of climate models to small perturbations is shown to be highly dependent on the unperturbed simulation. an experiment in which CO2 concentrations are doubled and sea surface temperatures are enhanced by 2 K has been carried out with two general circulation models which differ considerably in their control climates. the resulting changes in tropical precipitation in each model simulation are related to the increase in atmospheric water vapour which leads to enhanced precipitation in the main regions of low-level atmospheric convergence. Since these regions of convergence occur in slightly different locations in the unperturbed simulations, the distribution of changes is also different.

Differences in control simulations must be taken into account when comparing results from different models (for example, on doubling atmospheric CO2); otherwise unduly pessimistic conclusions may be reached concerning the consistency of model results. One may be able to make subjective allowance for the effect of known deficiencies in the unperturbed simulation on the model's response before using the simulated changes in, for example, impact studies.

A detailed examination of one of the experiments reveals that the change in precipitation is limited by the heat balance of the atmosphere, and indicates the importance of treating accurately the radiative perturbation due to changes in water vapour. the magnitude of the model's response is shown to be consistent with that found in three-dimensional climate models which include a simple representation of the ocean.