parameters in models of GHG cycles, radiative forcing, and the climate system. The models then sample repeatedly from the uncertainty distributions for inputs and model parameters, in order to produce a pdf of outcomes, e.g., global temperature and precipitation change. Either simple climate models (e.g., Wigley and Raper, 2001) or climate models of intermediate complexity (Forest et al., 2002) have been applied.
Alternative methods of developing pdfs for emissions are described in Nakicenovic et al. (2007), but they all require subjective judgement in the weighting of different future outcomes, which is a matter of considerable debate (Parson et al., 2006). Some argue that this should be done by experts, otherwise decision-makers will inevitably assign probabilities themselves without the benefit of established techniques to control well-known biases in subjective judgements (Schneider, 2001, 2002; Webster et al., 2002, 2003). Others argue that the climate change issue is characterised by 'deep uncertainty' - i.e., system models, parameter values, and interactions are unknown or contested - and therefore the elicited probabilities may not accurately represent the nature of the uncertainties faced (Grübler and Nakicenovic, 2001; Lempert et al., 2004).
The most important uncertainties to be represented in pdfs of regional climate change, the scale of greatest relevance for impact assessments, are GHG emissions, climate sensitivity, and inter-model differences in climatic variables at the regional scale. Other important factors include downscaling techniques, and regional forcings such as aerosols and land-cover change (e.g., Dessai, 2005). A rapidly growing literature reporting pdfs of climate sensitivity is providing a significant methodological advance over the long-held IPCC estimate of 1.5°C to 4.5°C for the (non-probabilistic) range of global mean annual temperature change for a doubling of atmospheric CO2 (see Meehl et al., 2007, for a detailed discussion). For regional change, recent methods of applying different weighting schemes to multi-model ensemble projections of climate are described in Christensen et al. (2007a). Other work has examined the full chain of uncertainties from emissions to regional climate. For example, Dessai et al. (2005b) tested the sensitivity of probabilistic regional climate changes to a range of uncertainty sources including climate sensitivity, GCM simulations, and emissions scenarios. The ENSEMBLES research project is modelling various sources of uncertainty to produce regional probabilities of climate change and its impacts for Europe (Hewitt and Griggs, 2004).
Methods to translate probabilistic climate changes for use in impact assessment (e.g., New and Hulme, 2000; Wilby and Harris, 2006; Fowler et al., 2007) include those assessing probabilities of impact threshold exceedance (e.g., Jones, 2000, 2004; Jones et al., 2007). Wilby and Harris (2006) combined information from various sources of uncertainty (emissions scenarios, GCMs, statistical downscaling, and hydrological model parameters) to estimate probabilities of low flows in the River Thames basin, finding the most important uncertainty to be the differences between the GCMs, a conclusion supported in water resources assessments in Australia (Jones and Page, 2001; Jones et al., 2005). Scholze et al. (2006) quantified risks of changes in key ecosystem processes on a global scale, by grouping scenarios according to ranges of global mean temperature change rather than considering probabilities of individual emissions scenarios. Probabilistic impact studies sampling across emissions, climate sensitivity, and regional climate change uncertainties have been conducted for wheat yield (Howden and Jones, 2004; Luo et al., 2005), coral bleaching (Jones, 2004; Wooldridge et al., 2005), water resources (Jones and Page, 2001; Jones et al., 2005), and freshwater ecology (Preston, 2006).
Was this article helpful?