 Figure 9.3: The time evolution of the globally averaged (a) temperature change relative to the control run of the CMIP2 simulations (Unit: °C). (b) ditto. for precipitation. (Unit: %). See Table 9.1 for more information on the individual models used here. |
Projections of climate change are affected by a range of uncertainties (see also Chapter 14) and there is a need to discuss and to quantify uncertainty in so far as is possible. Uncertainty in projected climate change arises from three main sources; uncertainty in forcing scenarios, uncertainty in modelled responses to given forcing scenarios, and uncertainty due to missing or misrepresented physical processes in models. These are discussed in turn below.
Forcing scenarios: The use of a range of forcing scenarios reflects uncertainties in future emissions and in the resulting greenhouse gas concentrations and aerosol loadings in the atmosphere. The complexity and cost of full AOGCM simulations has restricted these calculations to a subset of scenarios; these are listed in Table 9.1 and discussed in Section 9.3.1. Climate projections for the remaining scenarios are made with less general models and this introduces a further level of uncertainty. Section 9.3.2 discusses global mean warming for a broad range of scenarios obtained with simple models calibrated with AOGCMs. Chapter 13 discusses a number of techniques for scaling AOGCM results from a particular forcing scenario to apply to other scenarios.
Model response: The ensemble standard deviation and the range are used as available indications of uncertainty in model results for a given forcing, although they are by no means a complete characterisation of the uncertainty. There are a number of caveats associated with the ensemble approach. Common or systematic errors in the simulation of current climate (e.g., Gates et al., 1999; Lambert and Boer, 2001; Chapter 8) survive ensemble averaging and contribute error to the ensemble mean while not contributing to the standard deviation. A tendency for models to under-simulate the level of natural variability would result in an underestimate of ensemble variance. There is also the possibility of seriously flawed outliers in the ensemble corrupting the results. The ensemble approach nevertheless represents one of the few methods currently available for deriving information from the array of model results and it is used in this chapter to characterise projections of future climate.
Missing or misrepresented physics: No attempt has been made to quantify the uncertainty in model projections of climate change due to missing or misrepresented physics. Current models attempt to include the dominant physical processes that govern the behaviour and the response of the climate system to specified forcing scenarios. Studies of “missing” processes are often carried out, for instance of the effect of aerosols on cloud lifetimes, but until the results are well-founded, of appreciable magnitude, and robust in a range of models, they are considered to be studies of sensitivity rather than projections of climate change. Physical processes which are misrepresented in one or more, but not all, models will give rise to differences which will be reflected in the ensemble standard deviation.
The impact of uncertainty due to missing or misrepresented processes can, however, be limited by requiring model simulations to reproduce recent observed climate change. To the extent that errors are linear (i.e., they have proportionally the same impact on the past and future changes), it is argued in Chapter 12, Section 12.4.3.3 that the observed record provides a constraint on forecast anthropogenic warming rates over the coming decades that does not depend on any specific model’s climate sensitivity, rate of ocean heat uptake and (under some scenarios) magnitude of sulphate forcing and response.
|
Other reports in this collection |
|
