We show that GHG signal uncertainties are associated with errors in simulating the current climate in uncoupled and coupled climate models, the possible omission of relevant feedbacks, the non-uniqueness of the signal (due to the twin problems of the model's internally-generated natural variability and its sensitivity to initial conditions), uncertainties regarding the future GHG forcing and atmospheric GHG concentrations, and the so-called "cold start" error. Results from recent time-dependent greenhouse warming experiments are used to illustrate some of these points. We then discuss how energy-balance models, stochastically-forced ocean GCMs, and fully-coupled atmosphere-ocean GCMs have been used to derive estimates of decadal- to century-time scale natural variability, and consider some of the uncertainties associated with these estimates.
This review illustrates that it will be necessary to reduce both model signal and model natural variability uncertainties in order to detect a climate change signal and attribute this convincingly to change in carbon dioxide and other greenhouse gases. (pdf file)