Even if quantifiable uncertainties are represented, further uncertainties in climate scenarios may still need to be documented or explicitly treated. These include the possible impact on scenarios of errors in the unforced model simulation, the possibility that current models cannot adequately simulate the enhanced greenhouse response of a climatic feature of interest, or inconsistencies between results of model simulations and emerging observed climatic trends. For these reasons climate scenarios are often annotated with a list of caveats, along with some assessment as to their importance for the scenario user.
When choosing which GCM(s) to use as the basis for climate scenario construction, one of the criteria that has often been used is the ability of the GCM to simulate present day climate. Many climate scenarios have used this criterion to assist in their choice of GCM, arguing that GCMs that simulate present climate more faithfully are likely to simulate more plausible future climates (e.g., Whetton and Pittock, 1991; Robock et al., 1993; Risbey and Stone, 1996; Gyalistras et al., 1997; Smith and Pitts, 1997; Smith and Hulme, 1998; Lal and Harasawa, 2000). A good simulation of present day climate, however, is neither a necessary nor a sufficient condition for accurate simulation of climate change (see Chapter 8). It is possible, for example, that a model with a poor simulation of present day climate could provide a more accurate simulation of climate change than one which has a good simulation of present climate, if it contains a better representation of the dominant feedback processes that will be initiated by radiative forcing. While such uncertainties are difficult to test, useful insights into the ability of models to simulate long-term climate change can also be obtained by comparing model simulations of the climate response to past changes in radiative forcing against reconstructed paleoclimates.
This approach to GCM selection, however, raises a number of questions. Over which geographic domain should the GCM be evaluated - the global domain or only over the region of study? Which climate variables should be evaluated upper air synoptic features that largely control the surface climate, or only those climate variables, mostly surface, that are used in impact studies? Recent AOGCMs simulate observed 1961 to 1990 mean climate more faithfully than earlier GCMs (Kittel et al., 1998; see also Chapter 8), but they still show large errors in simulating inter-annual climate variability in some regions (Giorgi and Francisco, 2000; Lal et al., 2000) and in replicating ENSO-like behaviour in the tropics (Knutson et al., 1997). These questions demonstrate that there is no easy formula to apply when choosing GCMs for climate scenario construction; there will always be a role for informed but, ultimately, individual judgement. This judgement, however, should be made not just on empirical grounds (for example, which model's present climate correlates best with observations) but also on the basis of understanding the reasons for good or bad model performance, particularly if those reasons are important for the particular scenario application.
Several examples of such annotations can be given. Lal and Giorgi (1997) suggested that GCMs that cannot simulate the observed interannual variability of the Indian monsoon correctly should not be used as the basis for climate scenarios. Giorgi et al. (1998) commented that model-simulated spring temperatures over the USA Central Plains were too cold in both the CSIRO GCM and in the CSIRO-driven RegCM2 control simulations and affected the credibility of the ensuing temperature climate scenarios. Finally, scenarios prepared for the Australian region have often been accompanied by the note that ENSO is an important component of Australian climate that may change in the future, but that is not yet adequately simulated in climate models (e.g., Hennessy et al., 1998). Expert judgment can also be used to place confidence estimates on scenario ranges (Morgan and Keith, 1995). For example, Jones et al. (2000) placed "high confidence" on the temperature scenarios (incorporating quantifiable uncertainty) prepared for the South Pacific, but only "moderate to low confidence" in the corresponding rainfall scenarios.
Other reports in this collection