The utility of atmospheric general circulation model (GCM) integrations with prescribed sea surface temperatures (SSTs) is increasingly being questioned in the contexts of climate diagnosis, climate model error diagnosis, and short-term climate predictions. The basic issue is to what extent the errors in surface heat fluxes caused by decoupling air-sea interactions in this manner affect climate variability and the mean climate. This issue is addressed here by generating and comparing multi-century coupled GCM simulations with corresponding atmospheric GCM simulations with prescribed SSTs obtained from the coupled simulations. When the SST time series is prescribed at the full (half-hourly) temporal resolution of the coupled model output, the uncoupled simulations have a negligibly small mean climate bias, small variance errors on subseasonal scales, and slightly larger variance errors on interannual and decadal scales. Even on decadal scales, the errors are notably smaller than anticipated from reduced local thermal damping considerations alone.

A simple linear inverse modeling analysis of observed daily SSTs and air temperatures shows that the popular notion of "reduced local thermal damping" in coupled versus prescribed SST integrations may be valid in a rather limited parameter space. In reality, the oceanic stochastic forcing neglected in previous studies can counterbalance the impacts of atmospheric stochastic forcing. In such situations, opposite to the conventional view, atmospheric low frequency variability in coupled integrations can be substantially lower than in prescribed SST integrations.

Overall, these results show that the errors introduced by prescribing SSTs, though not negligible, are generally much smaller than the atmospheric response to the SSTs themselves. To that extent, they justify performing and using such uncoupled integrations for diagnostic and prediction purposes.