Origins of Model–Data Discrepancies in Optimal Fingerprinting

Abstract

Two approaches to distinguishing anthropogenic greenhouse gas and sulfate aerosol signals in the observed surface temperature record are compared. Both rely on a variant of general regression called ?optimal fingerprinting.? One approach is equivalent to a stepwise regression procedure estimating, first, a greenhouse gas signal and, in a second step, the sulfate aerosol signal. This is different from multiple regression, under which both signals are estimated simultaneously and treated symmetrically. The stepwise regression approach is a more powerful means of detecting greenhouse gas influence in the presence of a small and possibly poorly simulated sulfate aerosol signal. However, when both signals are of comparable size, multiple regression provides estimates of the amplitude of the greenhouse and sulfate responses that are, in principle, independent of each other, making it generally simpler to interpret. It is shown that there is a simple linear transform relating the stepwise and multiple regression approaches. Application of this transform to previous results of stepwise regression illustrates that estimated responses to anthropogenic greenhouse gas forcing are very similar between different climate models and are generally consistent with the signal estimated from the observations. The sulfate component of the anthropogenic signal appears to be responsible for the most prominent discrepancies between observations and some of the model simulations considered. The estimated contribution of anthropogenic greenhouse gases to the observed warming over the period of 1949?98 lies in the range of 0.39?1.29 K (50 yr)?1 or 0.28?1.16 K (50 yr)?1 (5%?95% range), depending on the model used to estimate the signal. These ranges depend only on the accuracy of the spatial pattern and the sign of the modeled sulfate forcing and response, not on its amplitude.