Climate Change 2001:
Working Group I: The Scientific Basis
Other reports in this collection Optimal detection studies that use multiple fixed signal patterns

Surface temperature patterns
Hegerl et al. (1997) applied a two-fingerprint approach, using a greenhouse gas fingerprint and an additional sulphate aerosol fingerprint that is made spatially independent (orthogonalised) of the greenhouse fingerprint. They analysed 50-year trends in observed northern summer temperatures. The influence of greenhouse gas and sulphate aerosol signals were both detected simultaneously in the observed pattern of 50-year temperature trends, and the amplitudes of both signals were found to be consistent between model and observations. Simulations forced with greenhouse gases alone and solar irradiance changes alone were not consistent with observations.

Hegerl et al. (2000) repeated this analysis using parallel simulations from a different climate model. The combined effect of greenhouse gases and aerosols was still detectable and consistent with observations, but the separate influence of sulphate aerosol forcing, as simulated by this second model, was not detectable. This was because the sulphate response was weaker in the second model, and closely resembled one of the main modes of natural variability. Hence, the detection of the net anthropogenic signal is robust, but the detection of the sulphate aerosol component is very sensitive to differences in model-simulated responses.

As in the single-pattern case, this study has been extended to include seven model GS simulations and to take into account observational sampling error (Figure 12.10b,c, see also Barnett et al., 1999; Hegerl et al. 2001). A simple linear transformation allows results to be displayed in terms of individual greenhouse and sulphate signal amplitudes, which assists comparison with other results (see Figure 12.10; Hegerl and Allen, 2000). The amplitudes of the greenhouse gas and sulphate components are simultaneously consistent with the observed amplitudes in 10 of the fourteen GS cases (seven models for two sets of fingerprints) displayed. This contrasts with eleven out of fourteen in the combined amplitude test described in Section If the trends to 1995 are used (Figure 12.10c), the results are similar, though in this case, the ellipse just includes the origin and six out of the fourteen GS cases are consistent with observations. The inconsistency can be seen to be mainly due to large variations in the amplitudes of the model-simulated responses to sulphate aerosols (indicated by the vertical spread of results). Model-simulated responses to greenhouse gases are generally more consistent both with each other and with observations. Two of the cases of disagreement are based on a single simulation rather than an ensemble mean and should therefore be viewed with caution (see Barnett et al., 2000). Barnett et al. (1999) found that the degree of agreement between the five models and observations they considered was similar, whether or not the global mean response was removed from the patterns. Signal amplitudes from simulations with greenhouse gas forcing only are generally inconsistent with those estimated from observations (Figure 12.10b,c).

In most of the cases presented here, the response to natural forcings was neglected. In a similar analysis to that just described, Hegerl et al. (2000); see also Barnett et al., 1999) also assessed simulations of the response to volcanic and solar forcing. They find, in agreement with Tett et al. (1999), that there is better agreement between observations and simulations when these natural forcings are included, particularly in the early 20th century, but that natural forcings alone cannot account for the late-century warming.

In summary, the estimation of the contribution of individual factors to recent climate change is highly model dependent, primarily due to uncertainties in the forcing and response due to sulphate aerosols. However, although the estimated amplitude varies from study to study, all studies indicate a substantial contribution from anthropogenic greenhouse gases to the changes observed over the latter half of the 20th century.

Vertical patterns of temperature
Allen and Tett (1999) also used spatial fingerprints in the vertical derived from simulations with greenhouse gas forcing alone and simulations with greenhouse gas, sulphate aerosol and strato-spheric ozone forcing. These authors show that, even if both greenhouse and other anthropogenic signals are estimated simultaneously in the observed record, a significant response to greenhouse gases remains detectable. Hill et al. (2001) extended this analysis to include model-simulated responses to both solar and volcanic forcing, and again found that the response to greenhouse gases remains detectable. Results with non-optimised fingerprints are consistent with the optimised case, but the uncertainty range is larger.

In summary, the fixed pattern studies indicate that the recent warming is unlikely (bordering on very unlikely) to be due to internal climate variability. A substantial response to anthropogenic greenhouse gases appears to be necessary to account for recent temperature trends but the majority of studies indicate that greenhouse gases alone do not appear to be able to provide a full explanation. Inclusion of the response to the direct effect of sulphate aerosols usually leads to a more satisfactory explanation of the observed changes, although the amplitude of the sulphate signal depends on the model used. These studies also provide some evidence that solar variations may have contributed to the early century warming.

Other reports in this collection