The SAR identified a number of factors that limited the degree to which any human influence on climate could be quantified. It was noted that detection and attribution of anthropogenic climate change signals would be accomplished through a gradual accumulation of evidence, and that there were appreciable uncertainties in the magnitude and patterns of natural variability, and in the radiative forcing and climate response resulting from human activity.
The SAR predicted an increase in the anthropogenic contribution to global mean temperature of slightly over 0.1°C in the five years following the SAR, which is consistent with the observed change since the SAR (Chapter 2). The predicted increase in the anthropogenic signal (and the observed change) are small compared to natural variability, so it is not possible to distinguish an anthropogenic signal from natural variability on five year time-scales.
Differences in surface and free atmosphere temperature trends
There are unresolved differences between the observed and modelled temperature variations in the free atmosphere. These include apparent changes in the temperature difference between the surface and the lower atmosphere, and differences in the tropical upper troposphere. While model simulations of large-scale changes in free atmospheric and surface temperatures are generally consistent with the observed changes, simulated and observed trends in troposphere minus surface temperature differences are not consistent. It is not clear whether this is due to model or observational error, or neglected forcings in the models.
Internal climate variability
The precise magnitude of natural internal climate variability remains uncertain. The amplitude of internal variability in the models most often used in detection studies differs by up to a factor of two from that seen in the instrumental temperature record on annual to decadal time-scales, with some models showing similar or larger variability than observed (Section 12.2; Chapter 8). However, the instrumental record is only marginally useful for validating model estimates of variability on the multi-decadal time-scales that are relevant for detection. Some palaeoclimatic reconstructions of temperature suggest that multi-decadal variability in the pre-industrial era was higher than that generated internally by models (Section 12.2; Chapter 8). However, apart from the difficulties inherent in reconstructing temperature accurately from proxy data, the palaeoclimatic record also includes the climatic response to natural forcings arising, for example, from variations in solar output and volcanic activity. Including the estimated forcing due to natural factors increases the longer-term variability simulated by models, while eliminating the response to external forcing from the palaeo-record brings palaeo-variability estimates closer to model-based estimates (Crowley, 2000).
Estimates of natural forcing have now been included in simulations over the period of the instrumental temperature record. Natural climate variability (forced and/or internally generated) on its own is generally insufficient to explain the observed changes in temperature over the last few decades. However, for all but the most recent two decades, the accuracy of the estimates of forcing may be limited, being based entirely on proxy data for solar irradiance and on limited surface data for volcanoes. There are some indications that solar irradiance fluctuations have indirect effects in addition to direct radiative heating, for example due to the substantially stronger variation in the UV band and its effect on ozone, or hypothesised changes in cloud cover (see Chapter 6). These mechanisms remain particularly uncertain and currently are not incorporated in most efforts to simulate the climate effect of solar irradiance variations, as no quantitative estimates of their magnitude are currently available.
The representation of greenhouse gases and the effect of sulphate aerosols has been improved in models. However, some of the smaller forcings, including those due to biomass burning and changes in land use, have not been taken into account in formal detection studies. The major uncertainty in anthropogenic forcing arises from the indirect effects of aerosols. The global mean forcing is highly uncertain (Chapter 6, Figure 6.8). The estimated forcing patterns vary from a predominantly Northern Hemisphere forcing similar to that due to direct aerosol effects (Tett et al., 2000) to a more globally uniform distribution, similar but opposite in sign to that associated with changes in greenhouse gases (Roeckner et al., 1999). If the response to indirect forcing has a component which can be represented as a linear combination of the response to greenhouse gases and to the direct forcing by aerosols, it will influence amplitudes of the responses to these two factors estimated through optimal detection.
Estimates of response patterns
Finally, there remains considerable uncertainty in the amplitude and pattern of the climate response to changes in radiative forcing. The large uncertainty in climate sensitivity, 1.5 to 4.5°C for a doubling of atmospheric carbon dioxide, has not been reduced since the SAR, nor is it likely to be reduced in the near future by the evidence provided by the surface temperature signal alone. In contrast, the emerging signal provides a relatively strong constraint on forecast transient climate change under some emission scenarios. Some techniques can allow for errors in the magnitude of the simulated global mean response in attribution studies. As noted in Section 12.2, there is greater pattern similarity between simulations of greenhouse gases alone, and of greenhouse gases and aerosols using the same model, than between simulations of the response to the same change in greenhouse gases using different models. This leads to some inconsistency in the estimation of the separate greenhouse gas and aerosol components using different models (see Section 12.4.3).
In summary, some progress has been made in reducing uncertainty, particularly with respect to distinguishing the responses to different external influences using multi-pattern techniques and in quantifying the magnitude of the modelled and observed responses. Nevertheless, many of the sources of uncertainty identified in the SAR still remain.
Other reports in this collection