IPCC Fourth Assessment Report: Climate Change 2007
Climate Change 2007: Working Group II: Impacts, Adaptation and Vulnerability Climate scenarios

The most recent climate projection methods and results are extensively discussed in the WG I volume (especially Christensen et al., 2007a; Meehl et al., 2007), and most of these were not available to the CCIAV studies assessed in this volume. Box 2.3 compares recent climate projections from Atmosphere-Ocean General Circulation Models (AOGCMs) with the earlier projections relied on throughout this volume. While AOGCMs are the most common source of regional climate scenarios, other methods and tools are also applied in specific CCIAV studies. Numerous regionalisation techniques[8] have been employed to obtain high-resolution, SRES-based climate scenarios, nearly always using low-resolution General Circulation Model (GCM) outputs as a starting point. Some of these methods are also used to develop scenarios of extreme weather events.

Box 2.3. SRES-based climate scenarios assumed in this report

Not all of the impact studies reported in this assessment employed SRES-based climate scenarios. Earlier scenarios are described in previous IPCC reports (IPCC, 1992, 1996; Greco et al., 1994). The remaining discussion focuses on SRES-based climate projections, which are applied in most CCIAV studies currently undertaken.

In recent years, many simulations of the global climate response to the SRES emission scenarios have been completed with AOGCMs, also providing regional detail on projected climate. Early AOGCM runs (labelled ‘pre-TAR’) were reported in the TAR (Cubasch et al., 2001) and are available from the IPCC DDC. Many have been adopted in CCIAV studies reported in this volume. A new generation of AOGCMs, some incorporating improved representations of climate system processes and land surface forcing, are now utilising the SRES scenarios in addition to other emissions scenarios of relevance for impacts and policy. The new models and their projections are evaluated in WG I (Christensen et al., 2007a; Meehl et al., 2007; Randall et al., 2007) and compared with the pre-TAR results below. Projections of global mean annual temperature change for SRES and CO2-stabilisation profiles are presented in Box 2.8.

Pre-TAR AOGCM results held at the DDC were included in a model intercomparison across the four SRES emissions scenarios (B1, B2, A2, and A1FI) of seasonal mean temperature and precipitation change for thirty-two world regions (Ruosteenoja et al., 2003).9 The inter-model range of changes by the end of the 21st century is summarised in Figure 2.6 for the A2 scenario, expressed as rates of change per century. Recent A2 projections, reported in WG I, are also shown for the same regions for comparison.

Almost all model-simulated temperature changes, but fewer precipitation changes, were statistically significant relative to 95% confidence intervals calculated from 1,000-year unforced coupled AOGCM simulations (Ruosteenoja et al., 2003; see also Figure 2.6). Modelled surface air temperature increases in all regions and seasons, with most land areas warming more rapidly than the global average (Giorgi et al., 2001; Ruosteenoja et al., 2003). Warming is especially pronounced in high northern-latitude regions in the boreal winter and in southern Europe and parts of central and northern Asia in the boreal summer. Warming is less than the global average in southern parts of Asia and South America, Southern Ocean areas (containing many small islands) and the North Atlantic (Figure 2.6a).

For precipitation, both positive and negative changes are projected, but a regional precipitation increase is more common than a decrease. All models simulate higher precipitation at high latitudes in both seasons, in northern mid-latitude regions in boreal winter, and enhanced monsoon precipitation for southern and eastern Asia in boreal summer. Models also agree on precipitation declines in Central America, southern Africa and southern Europe in certain seasons (Giorgi et al., 2001; Ruosteenoja et al., 2003; see also Figure 2.6b).

Comparing TAR projections to recent projections

The WG I report provides an extensive intercomparison of recent regional projections from AOGCMs (Christensen et al., 2007a; Meehl et al., 2007), focusing on those assuming the SRES A1B emissions scenario, for which the greatest number of simulations (21) were available. It also contains numerous maps of projected regional climate change. In summary:

  • The basic pattern of projected warming is little changed from previous assessments.
  • The projected rate of warming by 2030 is insensitive to the choice of SRES scenarios.
  • Averaged across the AOGCMs analysed, the global mean warming by 2090-2099 relative to 1980-1999 is projected to be 1.8, 2.8, and 3.4°C for the B1, A1B, and A2 scenarios, respectively. Local temperature responses in nearly all regions closely follow the ratio of global temperature response.
  • Model-average mean local precipitation responses also roughly scale with the global mean temperature response across the emissions scenarios, though not as well as for temperature.
  • The inter-model range of seasonal warming for the A2 scenario is smaller than the pre-TAR range at 2100 in most regions, despite the larger number of models (compare the red and blue bars in Figure 2.6a)
  • The direction and magnitude of seasonal precipitation changes for the A2 scenario are comparable to the pre-TAR changes in most regions, while inter-model ranges are wider in some regions/seasons and narrower in others (Figure 2.6b).
  • Confidence in regional projections is higher than in the TAR for most regions for temperature and for some regions for precipitation.

Figure 2.6Figure 2.6Figure 2.6

Figure 2.6. AOGCM projections of seasonal changes in (a) mean temperature (previous page) and (b) precipitation up to the end of the 21st century for 32 world regions. For each region two ranges between minimum and maximum are shown. Red bar: range from 15 recent AOGCM simulations for the A2 emissions scenario (data analysed for Christensen et al., 2007a). Blue bar: range from 7 pre-TAR AOGCMs for the A2 emissions scenario (Ruosteenoja et al., 2003). Seasons: DJF (December–February); MAM (March–May); JJA (June–August); SON (September–November). Regional definitions, plotted on the ECHAM4 model grid (resolution 2.8 * 2.8°), are shown on the inset map (Ruosteenoja et al., 2003). Pre-TAR changes were originally computed for 1961-1990 to 2070-2099 and recent changes for 1979-1998 to 2079-2098, and are converted here to rates per century for comparison; 95% confidence limits on modelled 30-year natural variability are also shown based on millennial AOGCM control simulations with HadCM3 (mauve) and CGCM2 (green) for constant forcing (Ruosteenoja et al., 2003). Numbers on precipitation plots show the number of recent A2 runs giving negative/positive precipitation change. Percentage changes for the SAH region (Sahara) exceed 100% in JJA and SON due to low present-day precipitation. Key for (a) and (b):

Figure 2.6

Scenarios from high-resolution models

The development and application of scenarios from high-resolution regional climate models and global atmospheric models (time-slices) since the TAR confirms that improved resolution allows a more realistic representation of the response of climate to fine-scale topographic features (e.g., lakes, mountains, coastlines). Impact models will often produce different results utilising high-resolution scenarios compared with direct GCM outputs (e.g., Arnell et al., 2003; Mearns et al., 2003; Stone et al., 2003; Leung et al., 2004; Wood et al., 2004). However, most regional model experiments still rely on only one driving AOGCM and scenarios are usually available from only one or two regional climate models (RCMs).

More elaborate and extensive modelling designs have facilitated the exploration of multiple uncertainties (across different RCMs, AOGCMs, and emissions scenarios) and how those uncertainties affect impacts. The PRUDENCE project in Europe produced multiple RCM simulations based on the ECHAM/OPYC AOGCM and HadAM3H AGCM simulations for two different emissions scenarios (Christensen et al., 2007b). Uncertainties due to the spatial scale of the scenarios and stemming from the application of different RCMs versus different GCMs (including models not used for regionalisation) were elaborated on in a range of impact studies (e.g., Ekstrom et al., 2007; Fronzek and Carter, 2007; Hingray et al., 2007; Graham et al., 2007; Olesen et al., 2007). For example, Olesen et al. (2007) found that the variation in simulated agricultural impacts was smaller across scenarios from RCMs nested in a single GCM than it was across different GCMs or across the different emissions scenarios.

The construction of higher-resolution scenarios (now often finer than 50 km), has encouraged new types of impact studies. For example, studies examining the combined impacts of increased heat stress and air pollution are now more feasible because the resolution of regional climate models is converging with that of air-quality models (e.g., Hogrefe et al., 2004). Furthermore, scenarios developed from RCMs (e.g., UKMO, 2001) are now being used in many more regions of the world, particularly the developing world (e.g., Arnell et al., 2003; Gao et al., 2003; Anyah and Semazzi, 2004; Government of India, 2004; Rupa Kumar et al., 2006). Results of these regional modelling experiments are reported in Christensen et al. (2007a).

Statistical downscaling (SD)

Much additional work has been produced since the TAR using methods of statistical downscaling (SD) for climate scenario generation (Wilby et al., 2004b; also see Christensen et al., 2007a). Various SD techniques have been used in downscaling directly to (physically-based) impacts and to a greater variety of climate variables than previously (e.g., wind speed), including extremes of variables. For example, Wang et al. (2004) and Caires and Sterl (2005) have developed extreme value models for projecting changes in wave height.

While statistical downscaling has mostly been applied for single locations, Hewitson (2003) developed empirical downscaling for point-scale precipitation at numerous sites and on a 0.1°-resolution grid over Africa. Finally, the wider availability of statistical downscaling tools is being reflected in wider application; for example, the Statistical Downscaling Model (SDSM) tool of Wilby et al. (2002), which has been used to produce scenarios for the River Thames basin (Wilby and Harris, 2006). Statistical downscaling does have some limitations; for example, it cannot take account of small-scale processes with strong time-scale dependencies (e.g., land-cover change). See Christensen et al. (2007a) for a complete discussion of the strengths and weaknesses of both statistical and dynamical downscaling.

Scenarios of extreme weather events

The improved availability of high-resolution scenarios has facilitated new studies of event-driven impacts (e.g., fire risk – Moriondo et al., 2006; low-temperature impacts on boreal forests – Jönsson et al., 2004). Projected changes in extreme weather events have been related to projected changes in local mean climate, in the hope that robust relationships could allow the prediction of extremes on the basis of changes in mean climate alone. PRUDENCE RCM outputs showed non-linear relationships between mean maximum temperature and indices of drought and heatwave (Good et al., 2006), while changes in maximum 1-day and 5-day precipitation amounts were systematically enhanced relative to changes in seasonal mean precipitation across many regions of Europe (Beniston et al., 2007). In a comprehensive review (citing over 200 papers) of the options available for developing scenarios of weather extremes for use in Integrated Assessment Models (IAMs), Goodess et al. (2003) list the advantages and disadvantages of applying direct GCM outputs, direct RCM outputs, and SD techniques. Streams of daily data are the outputs most commonly used from these sources, and these may pose computational difficulties for assessing impacts in IAMs (which commonly consider only large-scale, period-averaged climate), requiring scenario analysis to be carried out offline. Interpretation of impacts then becomes problematic, requiring a method of relating the large-scale climate change represented in the IAM to the impacts of associated changes in weather extremes modelled offline. Goodess et al. suggest that a more direct, but untested, approach could be to construct conditional damage functions (cdfs), by identifying the statistical relationships between the extreme events themselves (causing damage) and large-scale predictor variables. Box 2.4 offers a global overview of observed and projected changes in extreme weather events.

Box 2.4. SRES-based projections of climate variability and extremes

Possible changes in variability and the frequency/severity of extreme events are critical to undertaking realistic CCIAV assessments. Past trends in extreme weather and climate events, their attribution to human influence, and projected (SRES-forced) changes have been summarised globally by WG I (IPCC, 2007) and are reproduced in Table 2.2.

Table 2.2. Recent trends, assessment of human influence on the trend, and projections for extreme weather events for which there is an observed late 20th century trend. Source: IPCC, 2007, Table SPM-2.

Phenomenon and direction of trend Likelihooda that trend occurred in late 20th century (typically post-1960) Likelihooda of a human contribution to observed trend Likelihooda of future trends based on projections for 21st century using SRES scenarios 
Warmer and fewer cold days and nights over most land areas Very likelyb Likelyc Virtually certainc 
Warmer and more frequent hot days and nights over most land areas Very likelyd Likely (nights)c Virtually certainc 
Warm spells/heatwaves. Frequency increases over most land areas Likely More likely than note Very likely 
Heavy precipitation events. Frequency (or proportion of total rainfall from heavy falls) increases over most areas Likely More likely than note Very likely 
Area affected by droughts increases Likely in many regions since 1970s More likely than not Likely 
Intense tropical cyclone activity increases Likely in some regions since 1970 More likely than note Likely 
Increased incidence of extreme high sea level (excludes tsunamis)f Likely More likely than note,g Likelyh 


a The assessed likelihood, using expert judgement, of an outcome or a result: Virtually certain >99% probability of occurrence, Extremely likely >95%, Very likely >90%, Likely >66%, More likely than not >50%.

b Decreased frequency of cold days and nights (coldest 10%).

c Warming of the most extreme days and nights each year.

d Increased frequency of hot days and nights (hottest 10%).

e Magnitude of anthropogenic contributions not assessed. Attribution for these phenomena based on expert judgement rather than formal attribution studies.

f Extreme high sea level depends on average sea level and on regional weather systems. It is defined here as the highest 1% of hourly values of observed sea level at a station for a given reference period.

g Changes in observed extreme high sea level closely follow the changes in average sea level. It is very likely that anthropogenic activity contributed to a rise in average sea level.

h In all scenarios, the projected global average sea level at 2100 is higher than in the reference period. The effect of changes in regional weather systems on sea-level extremes has not been assessed.

  1. ^  Defined in the TAR as “techniques developed with the goal of enhancing the regional information provided by coupled AOGCMs and providing fine-scale climate information” (Giorgi et al., 2001).