The Potential Effect of GCM Uncertainties and Internal Atmospheric Variability on Anthropogenic Signal Detection

Abstract

When long integrations of climate models forced by observed boundary conditions are compared against observations, differences appear that have spatial and temporal coherence. These differences are due to several causes, the largest of which are fundamental model errors and the internal variability inherent in a GCM integration. Uncertainties in the observations themselves are small in comparison. The present paper constitutes a first attempt to compare the time dependence of these spatial difference patterns with the time dependence of simulated spatial patterns of climate change associated with anthropogenic sources. The analysis procedure was to project the model minus observed near-surface temperature difference fields onto estimates of the anthropogenic ?signal? (in this case the response to greenhouse-gas and sulfate-aerosol forcing). The temporal behavior of this projection was then compared with the estimated temporal evolution of the anthropogenic signal. Such comparisons were performed on timescales of 10, 20, and 30 yr. For trends of only 10 yr in length, the model minus observed spatial difference patterns are of the same magnitude and have the same time rate of change as the expected anthropogenic signal. In the case of 20- and 30-yr trends, the prospects are favorable for discriminating between temperature changes due to anthropogenic signal changes and changes associated with model minus observed difference structures. This suggests that attempts to quantitatively detect anthropogenic climate change should be based on temporal samples of at least several decades in length. This study also shows the importance of distinguishing between purely statistical detection and what the authors term practical prediction. It is found that the results of the detection analysis are sensitive to the spatial resolution at which it is performed: for the specific case of near-surface temperature, higher spatial resolution improves ability to discriminate between an anthropogenic signal and the type of model error/internal variability ?noise? considered here.