| description abstract | simple statistical model is used to partition uncertainty from different sources, in projections of future climate from multimodel ensembles. Three major sources of uncertainty are considered: the choice of climate model, the choice of emissions scenario, and the internal variability of the modeled climate system. The relative contributions of these sources are quantified for mid- and late-twenty-first-century climate projections, using data from 23 coupled atmosphere?ocean general circulation models obtained from phase 3 of the Coupled Model Intercomparison Project (CMIP3). Similar investigations have been carried out recently by other authors but within a statistical framework for which the unbalanced nature of the data and the small number (three) of scenarios involved are potentially problematic. Here, a Bayesian analysis is used to overcome these difficulties. Global and regional analyses of surface air temperature and precipitation are performed. It is found that the relative contributions to uncertainty depend on the climate variable considered, as well as the region and time horizon. As expected, the uncertainty due to the choice of emissions scenario becomes more important toward the end of the twenty-first century. However, for midcentury temperature, model internal variability makes a large contribution in high-latitude regions. For midcentury precipitation, model internal variability is even more important and this persists in some regions into the late century. Implications for the design of climate model experiments are discussed. | |
