IPCC_Managing Risks of Extreme Events.pdf - Climate Access

Changes in Climate Extremes and their Impacts on the Natural Physical EnvironmentChapter 3Box 3-2 | Variations in Confidence in Projections of Climate Change:Mean versus Extremes, Variables, ScaleComparisons of observed and simulated climate demonstrate good agreement for some climate variables such as mean temperature,especially at large horizontal scales (e.g., Räisänen, 2007). For instance, Figure 9.12 of the AR4 (Hegerl et al., 2007) compares the abilityof 14 climate models to simulate the temporal variations of mean temperature through the 20th century. When the models includedboth natural and anthropogenic forcings, they consistently reproduced the decadal variations in global mean temperature. Without theanthropogenic influences the models consistently failed to reproduce the multi-decadal temperature variations. However, when the samemodels’ abilities to simulate the temperature variations for smaller domains were assessed, although the mean temperature produced bythe ensemble generally tracked the observed temperature changes, the consistency among the models was poorer than was the case forthe global mean (Figure 9.12; Hegerl et al., 2007), partly because averaging over global scales smoothes internal variability or ‘noise’more than averaging over smaller domains (see also Section 3.2.2.1). We can conclude that the smaller the spatial domain for whichsimulations or projections are being prepared, the less confidence we should have in these projections (although in some limited casesregional-scale projections can have higher reliability than larger-scale projections; see Section 3.1.6).This increased uncertainty at smaller scales results from larger internal variability at smaller scales or ‘noise’ (i.e., natural variabilityunrelated to external forcings) and increased model uncertainty, both of which lead to lower model consistency at these scales (Hawkinsand Sutton, 2009). The latter factor is largely due to the role of unresolved processes (representations of clouds, convection, land surfaceprocesses; see also Section 3.2.3). Hawkins and Sutton (2009) also point out regional variations in these aspects: in the tropics thetemperature signal expected from anthropogenic factors is large relative to the model uncertainty and the natural variability, comparedwith higher latitudes. Figure 9.12 from AR4 (Hegerl et al., 2007) also shows that the models are more consistent in reproducing decadaltemperature variations in the tropics than at higher latitudes, even though the magnitudes of the temperature trends are larger at higherlatitudes.Uncertainty in projections also depends on the variables, phenomena, or impacts considered (Sections 3.3. to 3.5.). There is more modeluncertainty for variables other than temperature, for instance precipitation (Räisänen, 2007; Hawkins and Sutton, 2011; see alsoSection 3.2.3). And the situation is more difficult again for extremes. For instance, climate models simulate observed changes in extremetemperatures relatively well, but the frequency, distribution, and intensity of heavy precipitation is more poorly simulated (Randall et al.,2007) as are observed changes in heavy precipitation (e.g., Alexander and Arblaster, 2009). Also, projections of changes in temperatureextremes tend to be more consistent across climate models (in terms of sign) than for (wet and dry) precipitation extremes (Tebaldi etal., 2006; Orlowsky and Seneviratne, 2011; see also Figures 3-3 through 3-7 and 3-10) and significant inconsistencies are also found forprojections of agricultural (soil moisture) droughts (Wang, 2005; see also Box 3-3; Figure 3-10). For some other extremes, such as tropicalcyclones, differences in the regional-scale climate change projections between models can lead to marked differences in projected tropicalcyclone activity associated with anthropogenic climate change (Knutson et al., 2010), and thus decrease confidence in projections ofchanges in that extreme.The relative importance of various causes of uncertainties in projections is somewhat different for earlier compared with later futureperiods. For some variables (mean temperature, temperature extremes), the choice of emission scenario becomes more critical thanmodel uncertainty for the second part of the 21st century (Tebaldi et al., 2006; Hawkins and Sutton, 2009, 2011) though this does notapply for mean precipitation and some precipitation-related extremes (Tebaldi et al., 2006; Hawkins and Sutton, 2009, 2011), and has inparticular not been evaluated in detail for a wide range of extremes. Users need to be aware of such issues in deciding the range ofuncertainties that is appropriate to consider for their particular risk or impacts assessmentIn summary, confidence in climate change projections depends on the (temporal and spatial) scale and variable being considered andwhether one considers extremes or mean quantities. Confidence is highest for temperature, especially at the global scale, and decreaseswhen other variables are considered, and when we focus on smaller spatial domains (Tables 3-1 and 3-3). Confidence in projections forextremes is generally weaker than for projections of long-term averages.Collins et al., 2006; Murphy et al., 2007) and used to examine projectedchanges in extremes and their uncertainties (Barnett et al., 2006; Clarket al., 2006, 2010; Burke and Brown, 2008). Advances have also beenmade in developing probabilistic information at regional scales fromthe GCM simulations, but there has been rather less developmentextending this to probabilistic downscaled regional information and to132