tions). A reduction of uncertainties in these reconstructions, along with a resolution of differences among competing estimates, is essential to improve knowledge of the precise history of large-scale mean temperature changes in past centuries, and hence of radiative forcing effects. Such a resolution is likely to come from the availability of increased high-quality proxy reconstructions in key regions, particularly in the data-sparse regions of the tropical oceans and Southern Hemisphere. Improved specification of physical differences and limitations of various temperature proxies (tree rings versus boreholes versus corals) is also needed.
There is a broadly consistent view between different climate models and empirical proxy-based reconstructions of hemispheric mean surface temperature changes in past centuries. The models indicate that greenhouse gases explain the observed 0.6°C global surface warming in the past three decades and that some combination of solar and volcanic forcings is likely responsible for temperature fluctuations of a few tenths of a degree Celsius in the preindustrial period (IPCC, 2001). Model and observational studies suggest that land-cover change may account for some of the surface temperature variation over land (e.g., Kalnay and Cai, 2003; Marshall et al., 2003).
However, there are also significant differences among the model simulations. These differences arise from a number of sources (see Jones and Mann, 2004), including (1) differences in the sensitivities of the models to radiative forcing, which vary by as much as a factor of two; (2) differences in the reconstructed radiative forcings used to drive the model simulations; and (3) differences in the way that radiative forcing estimates are represented in the model. For example, in the case of volcanic aerosols, some models impose a fixed annual mean TOA radiative forcing simply by changing the solar constant (Gonzalez-Rouco et al., 2003), while others (e.g., Shindell et al., 2003, 2004) specify the forcing on a seasonally, latitudinally, and vertically resolved basis. It is clear that improved estimates of past radiative forcing changes and a more organized community-wide effort to perform a controlled set of simulations using common forcing estimates could help to resolve these differences.
Spatial patterns of climate change are difficult to compare between models and observations. The dearth of proxy data over large parts of the oceans in past centuries restricts the spatial detail available in current proxy-based reconstructions (Jones and Mann, 2004). Moreover, at regional spatial scales, the role of internal, unforced variability in the climate (which is intrinsically irreproducible by a forced simulation) is likely to be greater, and observed variations may be dominated by influences from large-scale modes of atmospheric circulation such as the North Atlantic Oscillation (NAO) and ENSO. Although there has been some success in reproducing past reconstructed changes in model simulations, including an NAO-like