The concept of uncertainty is implicit in the philosophy of climate scenario development, and characterization and quantification of uncertainty has become one of the most vigorous and dynamic branches of climate scenario research. Some important sources of uncertainty are detailed in TAR WGI Chapter 13, of which three major sources are:

1. Uncertainties in future GHG and aerosol emissions. The IS92 and SRES emissions scenarios described in Section 3.8 exemplify these uncertainties; each scenario implies different atmospheric compositions and hence different radiative forcing.
2. Uncertainties in global climate sensitivity⁴, mainly as a result of differences in the way physical processes and feedbacks are simulated in different models. This means that some GCMs simulate greater mean global warming per unit of radiative forcing than others.
3. Uncertainties in regional climate changes, which are apparent from differences in regional estimates of climate change by different GCMs for the same mean global warming.

Many early impact studies employed a climate scenario derived from a single GCM. However, it was recognized early on that different GCMs yield different regional climate responses, even when they are perturbed with identical forcing (e.g., Smith and Tirpak, 1989). Therefore, various approaches have been used to capture this range of responses in impact studies. These approaches include using all available GCM results (e.g., Santer, 1985; Yohe et al., 1999); using a selected subset of GCM experiments, in some cases based on the performance of the GCMs at simulating the current climate (e.g., Robock et al., 1993; Risbey and Stone, 1996; Smith et al., 1996); using results from different GCMs that have been "pattern-scaled" in conjunction with simple climate models to represent different types of uncertainty (e.g., Barrow et al., 2000; see also Section 3.8.3); or using the mean or median GCM response (e.g., Rotmans et al., 1994). The effect is to generate a range of future impacts. Much of the quantitative, scenario-based, impacts literature assessed in IPCC (1990) and IPCC (1996b) reported these kinds of analyses. More recently, impact studies have begun to consider the impacts of anthropogenic climate change alongside the effects of natural multi-decadal climate variability (Hulme et al., 1999a). This creates a distribution of impact indicator values for the present day to compare with the range of future impacts under alternative climate scenarios.

There have been a few preliminary attempts to derive frequency distributions of future climate by using expert judgment (Morgan and Keith, 1995; Hulme and Carter, 1999) or by projecting the statistical fit of modeled versus observed 20th-century climate onto modeled future changes (Allen et al., 2000). This information may be useful for impact assessment because it offers an opportunity to express impacts in terms of risk—for example, the risk of exceeding a given threshold impact (Jones, 2000; Pittock, 1999).

IPCC Homepage