11.2.3.1 Mean Temperature

The differences in near-surface temperature between the years 2080 to 2099 and the years 1980 to 1999 in the MMD-A1B projections, averaged over the West African (WAF), East African (EAF), South African (SAF) and Saharan (SAH) sub-regions, are provided in Table 11.1 ,with the temporal evolution displayed in Figure 11.1 .The Mediterranean coast is discussed together with southern Europe in Section 11.3 .In all four regions and in all seasons, the median temperature increase lies between 3°C and 4°C, roughly 1.5 times the global mean response. Half of the models project warming within about 0.5°C of these median values. The distributions estimated by Tebaldi et al. 2004a [JoC, MoS, SRC] ,b; see also Supplementary Material Table S11.2) have a very similar half-width, but reduce the likelihood of the extreme high limit as compared to the raw quartiles in Table 11.1 .There is a strong correlation across these AOGCMs between the global mean temperature response and the response in Africa. The signal-to-noise ratio is very large for these 20-year mean temperatures and 10 years is typically adequate to obtain a clearly discernible signal, as defined in Section 11.1.2 .Regionally averaged temperatures averaged over the period 1990 to 2009 are clearly discernible from the 1980 to 1999 averages.

Figure 11.1. Temperature anomalies with respect to 1901 to 1950 for four African land regions for 1906 to 2005 (black line) and as simulated (red envelope) by MMD models incorporating known forcings; and as projected for 2001 to 2100 by MMD models for the A1B scenario (orange envelope). The bars at the end of the orange envelope represent the range of projected changes for 2091 to 2100 for the B1 scenario (blue), the A1B scenario (orange) and the A2 scenario (red). The black line is dashed where observations are present for less than 50% of the area in the decade concerned. More details on the construction of these figures are given in Box 11.1 and Section 11.1.2 .

The upper panels in Figure 11.2 show the geographical structure of the ensemble-mean projected warming for the A1B scenario in more detail. Smaller values of projected warming, near 3°C, are found in equatorial and coastal areas and larger values, above 4°C, in the western Sahara. The largest temperature responses in North Africa are projected to occur in JJA, while the largest responses in southern Africa occur in September, October and November (SON). But the seasonal structure in the temperature response over Africa is modest as compared to extratropical regions. The basic pattern of projected warming has been robust to changes in models since the TAR, as indicated by comparison with ( Hulme et al. 2001 ) ).

To date there is insufficient evidence from RCMs to modify the large-scale temperature projections from GCMs, although Tadross et al. 2005a [SRC] ) project changes in the A2 scenario for southern Africa that are lower than those in the forcing GCM and near the low end of the spread in the MMD models, likely due to a weaker drying tendency than in most of the global models.

Figure 11.2. Temperature and precipitation changes over Africa from the MMD-A1B simulations. Top row: Annual mean, DJF and JJA temperature change between 1980 to 1999 and 2080 to 2099, averaged over 21 models. Middle row: same as top, but for fractional change in precipitation. Bottom row: number of models out of 21 that project increases in precipitation.

11.2.3.2 Mean Precipitation

Figure 11.2 and Table 11.1 illustrate some of the robust aspects of the precipitation response over Africa in the MMD-A1B projections. The fractional changes in annual mean precipitation in each of the 21 models are provided in Supplementary Material Figure S11.13. With respect to the most robust features (drying in the Mediterranean and much of southern Africa, and increases in rainfall in East Africa), there is qualitative agreement with the results in ( Hulme et al. 2001 ) and Ruosteenoja et al. 2003 [NPR, MoS] ), which summarise results from the models available at the time of the TAR.

The large-scale picture is one of drying in much of the subtropics and an increase (or little change) in precipitation in the tropics, increasing the rainfall gradients. This is a plausible hydrological response to a warmer atmosphere, a consequence of the increase in water vapour and the resulting increase in vapour transport in the atmosphere from regions of moisture divergence to regions of moisture convergence (see Chapter9 and Section 11.2.1 ).

The drying along Africa’s Mediterranean coast is a component of a larger-scale drying pattern surrounding the Mediterranean and is discussed further in the section on Europe( Section 11.3 ). A 20% drying in the annual mean is typical along the African Mediterranean coast in A1B by the end of the 21st century. Drying is seen throughout the year and is generated by nearly every MMD model. The drying signal in this composite extends into the northern Sahara, and down the West Coast as far as 15°N. The processes involved include increased moisture divergence and a systematic poleward shift of the storm tracks affecting the winter rains, with positive feedback from decreasing soil moisture in summer (see Section 11.3 ).

In southern Africa, a similar set of processes produces drying that is especially robust in the extreme southwest in winter, a manifestation of a much broader-scale poleward shift in the circulation across the South Atlantic and Indian Oceans. However, the drying is subject to the caveat that strong orographic forcing may result in locally different changes (as discussed in Section 11.4.3.2 , Box 11.3 ). With the exception of the winter rainfall region in the southwest, the robust drying in winter corresponds to the dry season over most of the sub-continent and does not contribute to the bulk of the annual mean drying. More than half of the annual mean reduction occurs in the spring and is mirrored in some RCM simulations for this region (see below). To an extent, this can be thought of as a delay in the onset of the rainy season. This spring drying suppresses evaporation, contributing to the spring maximum in the temperature response.

The increase in rainfall in East Africa, extending into the Horn of Africa, is also robust across the ensemble of models, with 18 of 21 models projecting an increase in the core of this region, east of the Great Lakes. This East African increase is also evident in ( Hulme et al. 2001 ) and Ruosteenoja et al. 2003 [NPR, MoS] ). The Guinean coastal rain belts and the Sahel do not show as robust a response. A straight average across the ensemble results in modest moistening in the Sahel with little change on the Guinean coast. The composite MMD simulations have a weak drying trend in the Sahel in the 20th century that does not continue in the future projections ( Biasutti and Giannini, 2006 [JoC, MoS] ; Hoerling, et al., 2006 [JoC, ARC] ), implying that the weak 20th-century drying trend in the composite 20th-century simulations is unlikely to be forced by greenhouse gases, but is more likely forced by aerosols, as in Rotstayn and Lohmann 2002 [ARC] ), or a result of low-frequency internal variability of the climate.

Individual models generate large, but disparate, responses in the Sahel. Two outliers are GFDL/CM2.1, which projects very strong drying in the Sahel and throughout the Sahara, and MIROC3.2_midres, which shows a very strong trend towards increased rainfall in the same region (see Supplementary Figure S11.13; and see Table 8.1 for model descriptions). Cook and Vizy 2006 [MoS] ) find moderately realistic interannual variability in the Gulf of Guinea and Sahel in both models. While the drying in the GFDL model is extreme within the ensemble, it generates a plausible simulation of 20th-century Sahel rainfall trends ( Held et al., 2005 [JoC, MoS, SRC] ; Hoerling et al., 2006 [JoC, ARC] ) and an empirical downscaling from AOGCMs ( Hewitson and Crane, 2006 [JoC, MoS, SRC] ) shows a similar response (see below). More research is needed to understand the variety of modelled precipitation responses in the Sahel and elsewhere in the tropics. Progress is being made in developing new methodologies for this purpose (e.g., Chou and Neelin, 2004 [JoC] ; Lintner and Chiang, 2005 [JoC] ; Chou et al., 2007 [JoC, 2007] ), leading to better appreciation of the sources of model differences. Haarsma et al. 2005 [JoC] ) describe a plausible mechanism associated with increasing land-ocean temperature contrast and decreasing surface pressures over the Sahara, which contributes to the increase in Sahel precipitation with warming in some models.

It has been argued (e.g., Paethe and Hense, 2004 [JoC, MoS] ) that the partial amelioration of the Sahel drought since the 1990 s may be a sign of a greenhouse-gas driven increase in rainfall, providing support for those models that moisten the Sahel into the 21st century (e.g., Maynard et al., 2002 [JoC] ; Haarsma et al., 2005 [JoC] ; Kamga et al., 2005 [JoC, MoS] ). However, it is premature to take this partial amelioration as evidence of a global warming signature, given the likely influence of internal variability on the inter-hemispheric SST gradients that influence Sahel rainfall, as well as the influence of aerosol variations.

Table 11.1 provides information on the spread of model-projected precipitation change in the four African sub-regions. The regions and seasons for which the central half (25 to 75%) of the projections are uniformly of one sign are: EAF where there is an increase in DJF, March, April and May (MAM), SON and in the annual mean; SAF where there is a decrease in austral winter and spring; and SAH where there is a decrease in boreal winter and spring. The Tebaldi et al. 2004a [JoC, MoS, SRC] ,b) Bayesian estimates (Supplementary Material Table S11.2) do not change this distinction between robust and non-robust regions and seasons. The time required for emergence of a clearly discernible signal in these robust regions and seasons is typically 50 to 100 years, except in the Sahara where even longer times are required.

Land use change is a potential contributor to climate change in the 21st century (see also Section 11.7 , Box 11.4 ). C.M. ( Taylor et al. 2002 ) ) project drying over the Sahel of 4% from 1996 to 2015 due to changing land use, but suggest that the potential exists for this contribution to grow substantially further into the century. Maynard and Royer 2004a [JoC] ) suggest that estimated land use change scenarios for the mid-21st century would have only a modest compensating effect on the greenhouse-gas induced moistening in their model. Neither of these studies includes a dynamic vegetation model.

Several climate change projections based on RCM simulations are available for southern Africa but are much scarcer for other regions. Tadross et al. 2005a [SRC] ) examine two RCMs, Providing Regional Impacts for Climate Studies (PRECIS) and Mesoscale Model version 5 (MM5), nested for southern Africa in a time-slice AGCM based in turn on lower-resolution Hadley Centre Coupled Model (HadCM3) coupled simulations for the Special Report on Emission Scenarios (SRES) A2 scenario. During the early summer season, October to December, both models predict drying over the tropical western side of the continent, responding to the increase in high-pressure systems entering from the west, with MM5 indicating that the drying extends further south and PRECIS further east. The drying in the west continues into late summer, but there are increases in total rainfall towards the east in January and February, a feature barely present in the ensemble mean of the MMD models. Results obtained by downscaling one global model must be assessed in the context of the variety of responses in southern Africa among the MMD models (Supplementary Material Figure S11.13).

Hewitson and Crane 2006 [JoC, MoS, SRC] ) use empirical downscaling to provide projections for daily precipitation as a function of six GCM simulations. The degree of convergence in the downscaled results for the SRES A2 scenario near the end of the 21st century suggests more commonality in GCM-projected changes in daily circulation, on which the downscaling is based, than in the GCM precipitation responses. Figure 11.3 shows the response of mean JJA monthly total precipitation for station locations across Africa. The ensemble mean of these downscaling results shows increased precipitation in east Africa extending into southern Africa, especially in JJA, strong drying in the core Sahel in JJA with some coastal wetting, and moderate wetting in DJF. There is also drying along the Mediterranean coast, and, in most models, drying in the western portion of southern Africa. The downscaling also shows marked local-scale variation in the projected changes, for example, the contrasting changes in the west and east of Madagascar, and on the coastal and inland borders of the Sahel.

Figure 11.3. Anomaly of mean monthly precipitation (mm) using daily data empirically downscaled from six GCMs (ECHAM4.5, Hadley Centre Atmospheric Model (HadAM3), CSIRO Mk2, GFDL 2.1, MRI, MIROC; see Table 8.1 for descriptions of most of these models) to 858 station locations. The GCMs were forced by the SRES A2 scenario. Anomalies are for the future period ( 2070 to 2099 for the first three models, and 2080 to 2099 for the latter three models) minus a control 30-year period (from Hewitson and Crane, 2006 [JoC, MoS, SRC] ).

While this result is generally consistent with the underlying GCMs and the composite MMD projections, there is a tendency for greater Sahel drying than in the underlying GCMs, providing further rationale (alongside the large spread in model responses and poor coupled model performance in simulating droughts of the magnitude observed in the 20th century) for viewing with caution the projection for a modest increase in Sahel rainfall in the ensemble mean of the MMD models.

11.2.3.3 Extremes

Research on changes in extremes specific to Africa, in either models or observations, is limited. A general increase in the intensity of high-rainfall events, associated in part with the increase in atmospheric water vapour, is expected in Africa, as in other regions. Regional modelling and downscaling results ( Tadross et al., 2005a [SRC] ) both support an increase in the rainfall intensity in southern Africa. In regions of mean drying, there is generally a proportionally larger decrease in the number of rain days, indicating compensation between intensity and frequency of rain. In the downscaling results of Hewitson and Crane 2006 [JoC, MoS, SRC] and Tadross et al. 2005a [SRC] ), changes in the median precipitation event magnitude at the station scale do not always mirror the projected changes in seasonal totals.

There is little modelling guidance on possible changes in tropical cyclones affecting the southeast coast of Africa. Thermodynamic arguments for increases in precipitation rates and intensity of tropical storms (see Chapter 10 )are applicable to these Indian Ocean storms as for other regions, but changes in frequency and spatial distribution remain uncertain. In a time-slice simulation with a 20-km resolution AGCM, Oouchi et al. 2006 [MoS] ) obtain a significant reduction in the frequency of tropical storms in the Indian Ocean.

Using the definition of ‘extreme seasons’ given in Section 11.1.2 ,the probability of extremely warm, wet and dry seasons, as estimated by the MMD models, is provided in Table 11.1 .As in most tropical regions, all seasons are extremely warm by the end of the 21st century, with very high confidence under the A1B scenario. Although the mean precipitation response in West Africa is less robust than in East Africa, the increase in the number of extremely wet seasons is comparable in both, increasing to roughly 20% (i.e., 1 in 5 of the seasons are extremely wet, as compared to 1 in 20 in the control period in the late 20th century). In southern Africa, the frequency of extremely dry austral winters and springs increases to roughly 20%, while the frequency of extremely wet austral summers doubles in this ensemble of models.

11.3.3.1 Mean Temperature

The observed evolution of European temperatures in the 20th century, characterised by a warming trend modulated by multi-decadal variability, was well within the envelope of the MMD simulations( Figure 11.4 ).

Figure 11.4. Temperature anomalies with respect to 1901 to 1950 for two Europe land regions for 1906 to 2005 (black line) and as simulated (red envelope) by MMD models incorporating known forcings; and as projected for 2001 to 2100 by MMD models for the A1B scenario (orange envelope). The bars at the end of the orange envelope represent the range of projected changes for 2091 to 2100 for the B1 scenario (blue), the A1B scenario (orange) and the A2 scenario (red). More details on the construction of these figures are given in Box 11.1 and Section 11.1.2 .

In this century, the warming is projected to continue at a rate somewhat greater than its global mean, with the increase in 20-year mean temperatures (from its values in 1980 to 1999 ) becoming clearly discernible (as defined in Section 11.1.2 )within a few decades. Under the A1B scenario, the simulated area and annual mean warming from 1980 to 1999 to 2080 to 2099 varies from 2.3°C to 5.3°C in NEU and from 2.2°C to 5.1°C in SEM. The warming in northern Europe is likely to be largest in winter and that in the Mediterranean area largest in summer( Figure 11.5 ). Seasonal mean temperature changes typically vary by a factor of three among the MMD models( Table 11.1 ); however, the upper end of the range in NEU in DJF is reduced from 8.1°C to 6.7°C when one model with an extreme cold bias in present-day winter climate is excluded. Further details are given in Table 11.1 and Supplementary Material Figures S11.2 to S11.4.

Although changes in atmospheric circulation have a significant potential to affect temperature in Europe (e.g., Dorn et al., 2003 [JoC, MoS, SRC] ), they are not the main cause of the projected warming (e.g., Rauthe and Paeth, 2004 [MoS] ; Stephenson et al., 2006 [JoC, MoS] ; van Ulden et al., 2007 [NPR, 2007] ). A regression-based study using five of the MMD models ( van Ulden and van Oldenborgh, 2006 [MoS] ) indicated that in a region comprising mainly Germany, circulation changes enhanced the warming in most models in winter (due to an increase in westerly flow) and late summer (due to a decrease in westerly flow), but reduced the warming slightly in May and June. However, the circulation contribution to the simulated temperature changes (typically –1°C to 1.5°C depending on model and month) was generally much smaller than the total simulated warming in the late 21st century.

Despite a decrease in the North Atlantic MOC in most models (see Section 10.3 ), all the MMD simulations show warming in Great Britain and continental Europe, as other climatic effects of increased greenhouse gases dominate over the changes in ocean circulation. The same holds for earlier simulations with increased greenhouse gas concentrations, except for a very few ( Russell and Rind, 1999 [MoS, ARC] ; Schaeffer et al., 2004 [MoS, ARC] ) with slight cooling along the north-western coastlines of Europe but warming over the rest of the continent. The impact of MOC changes depends on the regional details of the change, being largest if ocean convection is suppressed at high latitudes where the sea ice feedback may amplify atmospheric cooling ( Schaeffer et al., 2004 [MoS, ARC] ). Sensitivity studies using AOGCMs with an artificial shutdown of the MOC and no changes in greenhouse gas concentrations typically show a 2°C to 4°C annual mean cooling in most of Europe, with larger cooling in the extreme north-western parts (e.g., Stouffer et al., 2006 [MoS, ARC] ).

Statistical downscaling (SD) studies tend to show a large-scale warming similar to that of dynamical models but with finer-scale regional details affected by factors such as distance from the coast and altitude (e.g., Benestad, 2005 [JoC, MoS, SRC] ; Hanssen-Bauer et al., 2005 [SRC] ). Comparing RCM and SD projections for Norway downscaled from the same GCM, Hanssen-Bauer et al. 2003 [SRC] ) found the largest differences between the two approaches in winter and/or spring at locations with frequent temperature inversions in the present climate. A larger warming at these locations in the SD projections was found, consistent with increased winter wind speed in the driving GCM and reduced snow cover, both of which suppress formation of ground inversions.

11.3.3.2 Mean Precipitation

Asouth-north contrast in precipitation changes across Europe is indicated by AOGCMs, with increases in the north and decreases in the south( Figure 11.5 ). The annual area-mean change from 1980 to 1999 to 2080 to 2099 in the MMD-A1B projections varies from 0 to 16% in NEU and from –4 to –27% in SEM( Table 11.1 ). The largest increases in northern and central Europe are simulated in winter. In summer, the NEU area mean changes vary in sign between models, although most models simulate increased (decreased) precipitation north (south) of about 55°N. In SEM, the most consistent and, in percentage terms, largest decreases, occur in summer, but the area mean precipitation in the other seasons also decreases in most or all models. More detailed statistics are given in Table 11.1 .Increasing evaporation makes the simulated decreases in annual precipitation minus evaporation extend a few hundred kilometres further north in central Europe than the decreases in precipitation (Supplementary Material Figure S11.1).

Figure 11.5. Temperature and precipitation changes over Europe from the MMD-A1B simulations. Top row: Annual mean, DJF and JJA temperature change between 1980 to 1999 and 2080 to 2099, averaged over 21 models. Middle row: same as top, but for fractional change in precipitation. Bottom row: number of models out of 21 that project increases in precipitation.

Both circulation changes and thermodynamic factors appear to affect the simulated seasonal cycle of precipitation changes in Europe. Applying a regression method to five of the MMD simulations, van Ulden and van Oldenborgh 2006 [MoS] ) found that in a region comprising mainly Germany, circulation changes played a major role in all seasons. In most models, increases in winter precipitation were enhanced by increased westerly winds, with decreases in summer precipitation largely due to more easterly and anticyclonic flow. However, differences in the simulated circulation changes among the individual models were accompanied by large differences in precipitation change, particularly in summer. The residual precipitation change varied less with season and among models, being generally positive as expected from the increased moisture transport capacity of a warmer atmosphere. In a more detailed study of one model, HadAM3P, Rowell and Jones 2006 [JoC, MoS, SRC] ) showed that decreases in summer precipitation in continental and southeastern Europe were mainly associated with thermodynamic factors. These included reduced relative humidity resulting from larger continental warming compared to surrounding sea areas and reduced soil moisture due mainly to spring warming causing earlier snowmelt. Given the confidence in the warming patterns driving these changes, the reliability of the simulated drying was assessed as being high.

Changes in precipitation may vary substantially on relatively small horizontal scales, particularly in areas of complex topography. Details of these variations are sensitive to changes in the atmospheric circulation, as illustrated in Figure 11.6 for two PRUDENCE simulations that only differ with respect to the driving global model. In one, an increase in westerly flow from the Atlantic Ocean (caused by a large increase in the north-south pressure gradient) is accompanied by increases of up to 70% in annual precipitation over the Scandinavian mountains. In the other, with little change in the average pressure pattern, the increase is in the range of 0 to 20%. When compared with circulation changes in the more recent MMD simulations, these two cases fall in the opposite ends of the range. Most MMD models suggest an increased north-south pressure gradient across northern Europe, but the change is generally smaller than in the top row of Figure 11.6 .

Figure 11.6. Simulated changes in annual mean sea level pressure (∆SLP), precipitation (∆Prec) and mean 10-m level wind speed (∆Wind) from the years 1961 to 1990 to the years 2071 to 2100 . The results are based on the SRES A2 scenario and were produced by the same RCM (Rossby Centre regional Atmosphere-Ocean model; RCAO) using boundary data from two global models: ECHAM4/OPYC3 (top) and HadAM3H (bottom) (redrawn from Rummukainen et al., 2004 [JoC, MoS, SRC] ).

Projections of precipitation change in Europe based on SD tend to support the large-scale picture from dynamical models (e.g., Busuioc et al., 2001 [JoC, MoS, SRC] ; Beckmann and Buishand, 2002 [JoC] ; Hanssen-Bauer et al., 2003 [SRC] , 2005; Benestad, 2005 [JoC, MoS, SRC] ; Busuioc et al., 2006 [MoS, SRC] ), although variations between SD methods and the dependence on the GCM data sets used (see Section 11.10.1.3 )make it difficult to draw quantitative conclusions. However, some SD studies have suggested a larger small-scale variability of precipitation changes than indicated by GCM and RCM results, particularly in areas of complex topography ( Hellström et al., 2001 [SRC] ).

The decrease in precipitation together with enhanced evaporation in spring and early summer is very likely to lead to reduced summer soil moisture in the Mediterranean region and parts of central Europe ( Douville et al., 2002 [JoC] ; Wang, 2005 [JoC, MoS] ). In northern Europe, where increased precipitation competes with earlier snowmelt and increased evaporation, the MMD models disagree on whether summer soil moisture will increase or decrease ( Wang, 2005 [JoC, MoS] ).

11.3.3.3 Temperature Variability and Extremes

Based on both GCM ( Giorgi and Bi, 2005 [JoC, MoS, SRC] ; Rowell, 2005 [JoC] ; Clark et al., 2006 [JoC, MoS, ARC] ) and RCM simulations ( Schär et al., 2004 [JoC, SRC] ; Vidale et al., 2007 [NPR, SRC, 2007] ), interannual temperature variability is likely to increase in summer in most areas. However, the magnitude of change is uncertain, even in central Europe where the evidence for increased variability is strongest. In some PRUDENCE simulations, interannual summer temperature variability in central Europe doubled between 1961 to 1990 and 2071 to 2100 under the A2 scenario, while other simulations showed almost no change ( Vidale et al., 2007 [NPR, SRC, 2007] ). Possible reasons for the increase in temperature variability are reduced soil moisture, which reduces the capability of evaporation to damp temperature variations, and increased land-sea contrast in average summer temperature ( Rowell, 2005 [JoC] ; Lenderink et al., 2007 [NPR, 2007] ).

Simulated increases in summer temperature variability also extend to daily time scales. Kjellström et al. 2007 [NPR, 2007] ) analyse the PRUDENCE simulations and find that the inter-model differences in the simulated temperature change increase towards the extreme ends of the distribution. However, a general increase in summer daily temperature variability is evident, especially in southern and central parts of Europe, with the highest maximum temperatures increasing more than the median daily maximum temperature (Supplementary Material Figure S11.23). Similarly, Shkolnik et al. 2006 [MoS, ARC] ) report a simulated increase in summer daily time-scale temperature variability in mid-latitude western Russia. These RCM results are supported by GCM studies of Hegerl et al. 2004 [JoC, ARC] Meehl and Tebaldi 2004 [JoC, SRC] and Clark et al. 2006 [JoC, MoS, ARC] ).

In contrast with summer, models project reduced temperature variability in most of Europe in winter, both on interannual ( Räisänen, 2001 [MoS, SRC] ; Räisänen et al., 2003 [NPR, MoS, SRC] ; Giorgi et al., 2004 [JoC, SRC] ; Giorgi and Bi, 2005 [JoC, MoS, SRC] ; Rowell, 2005 [JoC] ) and daily time scales ( Hegerl et al., 2004 [JoC, ARC] ; Kjellström et al., 2007 [NPR, 2007] ). In the PRUDENCE simulations, the lowest winter minimum temperatures increased more than the median minimum temperature especially in eastern, central and northern Europe, although the magnitude of this change was strongly model-dependent (Supplementary Material Figure S11.23). The geographical patterns of the change indicate a feedback from reduced snow cover, with a large warming of the cold extremes where snow retreats but a more moderate warming in the mostly snow-free south-western Europe ( Rowell, 2005 [JoC] ; Kjellström et al., 2007 [NPR, 2007] ).

Along with the overall warming and changes in variability, heat waves are very likely to increase in frequency, intensity and duration ( Barnett et al., 2006 [JoC, MoS] ; Clark et al., 2006 [JoC, MoS, ARC] ; Tebaldi et al., 2006 [JoC, MoS, SRC] ). Conversely, the number of frost days is very likely to decrease ( Tebaldi et al., 2006 [JoC, MoS, SRC] ).

11.3.3.4 Precipitation Variability and Extremes

In northern Europe and in central Europe in winter, where time mean precipitation is simulated to increase, high extremes of precipitation are very likely to increase in magnitude and frequency. In the Mediterranean area and in central Europe in summer, where reduced mean precipitation is projected, extreme short-term precipitation may either increase (due to the increased water vapour content of a warmer atmosphere) or decrease (due to a decreased number of precipitation days, which if acting alone would also make heavy precipitation less common). These conclusions are based on several GCM (e.g., Semenov and Bengtsson, 2002 [JoC, MoS] ; Voss et al., 2002 [JoC, MoS, ARC] ; Hegerl et al., 2004 [JoC, ARC] ; Wehner, 2004 [MoS] ; Kharin and Zwiers, 2005 [JoC, MoS, ARC] ; Tebaldi et al., 2006 [JoC, MoS, SRC] ) and RCM (e.g., Jones and Reid, 2001 [JoC, MoS, ARC] ; Räisänen and Joelsson, 2001 [JoC, MoS, SRC] ; Booij, 2002 [JoC] ; Christensen and Christensen, 2003 [JoC, ARC] , 2004; Pal et al., 2004 [JoC, MoS, SRC] ; Räisänen et al., 2004 [JoC, MoS, SRC] ; Sánchez et al., 2004 [JoC, MoS] ; Ekström et al., 2005 [MoS, ARC] ; Frei et al., 2006 [JoC, MoS, ARC] ; Gao et al., 2006a [JoC, MoS, SRC] ; Shkolnik et al., 2006 [MoS, ARC] ; Beniston et al., 2007 [NPR, SRC, 2007] ) studies. However, there is still a lot of quantitative uncertainty in the changes in both mean and extreme precipitation.

Time scale also matters. Although there are some indications of increased interannual variability, particularly in summer precipitation ( Räisänen, 2002 [SRC] ; Giorgi and Bi, 2005 [JoC, MoS, SRC] ; Rowell, 2005 [JoC] ), changes in the magnitude of long-term (monthly to annual) extremes are expected to follow the changes in mean precipitation more closely than are those in short-term extremes ( Räisänen, 2005 [JoC, SRC] ). On the other hand, changes in the frequency of extremes tend to increase with increasing time scale even when this is not the case for the changes in the magnitude of extremes ( Barnett et al., 2006 [JoC, MoS] ).

Figure 11.7 illustrates the possible characteristics of precipitation change. The eight models in this PRUDENCE study ( Frei et al., 2006 [JoC, MoS, ARC] ) projected an increase in mean precipitation in winter in both southern Scandinavia and central Europe, due to both increased wet day frequency and increased mean precipitation for the wet days. In summer, a decrease in the number of wet days led to a decrease in mean precipitation, particularly in central Europe. Changes in extreme short-term precipitation were broadly similar to the change in average wet-day precipitation in winter. In summer, extreme daily precipitation increased in most models despite the decrease in mean precipitation, although the magnitude of the change was highly model-dependent. However, this study only covered the uncertainties associated with the choice of the RCM, not those associated with the driving GCM and the emissions scenario.

Figure 11.7. Changes (ratio 2071 – 2100 / 1961 – 1990 for the A2 scenario) in domain-mean precipitation diagnostics in the PRUDENCE simulations in southern Scandinavia (5°E–20°E, 55°N–62°N) and central Europe (5°E–15°E, 48°N–54°N) in winter (top) and in summer (bottom). fre = wet-day frequency; mea = mean seasonal precipitation; int = mean wet-day precipitation; q90 = 90th percentile of wet-day precipitation; x1d.5 and x1d.50 = 5- and 50-year return values of one-day precipitation; x5d.5 and x5d.50 = 5- and 50-year return values of five-day precipitation. For each of the eight models, the vertical bar gives the 95% confidence interval associated with sampling uncertainty (redrawn from Frei et al., 2006 [JoC, MoS, ARC] ). Models are the Hadley Centre Atmospheric Model (HadAM3H), the Climate High Resolution Model (CHRM), the climate version of the ‘Lokalmodell’ (CLM), the Hadley Centre Regional Model (HadRM3H and HadRM3P), the combination of the High-Resolution Limited Area Model (HIRLAM) and the European Centre Hamburg (ECHAM4) GCM (HIRHAM), the regional climate model REMO, and the Rossby Centre regional Atmosphere-Ocean model (RCAO).

Much larger changes are expected in the recurrence frequency of precipitation extremes than in the magnitude of extremes ( Huntingford et al., 2003 [JoC, MoS] ; Barnett et al., 2006 [JoC, MoS] ; Frei et al., 2006 [JoC, MoS, ARC] ). For example, Frei et al. 2006 [JoC, MoS, ARC] ) estimate that, in Scandinavia under the A2 scenario, the highest five-day winter precipitation totals occurring once in 5 years in 2071 to 2100 would be similar to those presently occurring once in 8 to 18 years (the range reflects variation between the PRUDENCE models). In the MMD simulations, large increases occur in the frequencies of both high winter precipitation in northern Europe and low summer precipitation in southern Europe and the Mediterranean area( Table 11.1 ).

The risk of drought is likely to increase in southern and central Europe. Several model studies have indicated a decrease in the number of precipitation days (e.g., Semenov and Bengtsson, 2002 [JoC, MoS] ; Voss et al., 2002 [JoC, MoS, ARC] ; Räisänen et al., 2003 [NPR, MoS, SRC] , 2004; Frei et al., 2006 [JoC, MoS, ARC] ) and an increase in the length of the longest dry spells in this area ( Voss et al., 2002 [JoC, MoS, ARC] ; Pal et al., 2004 [JoC, MoS, SRC] ; Beniston et al., 2007 [NPR, SRC, 2007] ; Gao et al., 2006a [JoC, MoS, SRC] ; Tebaldi et al., 2006 [JoC, MoS, SRC] ). By contrast, the same studies do not suggest major changes in dry-spell length in northern Europe.

11.3.3.5 Wind Speed

Confidence in future changes in windiness in Europe remains relatively low. Several model studies (e.g., Zwiers and Kharin, 1998 [MoS, ARC] ;( Knippertz et al., 2000; ) Leckebusch and Ulbrich, 2004 [JoC] ; Pryor et al., 2005a [JoC, MoS] ; van den Hurk et al., 2006 [NPR] ) have suggested increased average and/or extreme wind speeds in northern and/or central Europe, but some studies point in the opposite direction (e.g., ( Pryor et al., 2005b ) ). The changes in both average and extreme wind speeds may be seasonally variable, but the details of this variation appear to be model-dependent (e.g., Räisänen et al., 2004 [JoC, MoS, SRC] ; Rockel and Woth, 2007 [NPR, 2007] ).

Akey factor is the change in the large-scale atmospheric circulation ( Räisänen et al., 2004 [JoC, MoS, SRC] ; Leckebusch et al., 2006 [NotFound] ). Simulations with an increased north-south pressure gradient across northern Europe (e.g., top row of Figure 11.6 )tend to indicate stronger winds in northern Europe, because of both the larger time-averaged pressure gradient and a northward shift in cyclone activity. Conversely, the northward shift in cyclone activity tends to reduce windiness in the Mediterranean area. On the other hand, simulations with little change in the pressure pattern tend to show only small changes in the mean wind speed (bottom row of Figure 11.6 ). Most of the MMD-projected pressure changes fall between the two PRUDENCE simulations shown in Figure 11.6 ,which suggests that the most likely outcome for windiness might be between these two cases.

Extreme wind speeds in Europe are mostly associated with strong winter cyclones (e.g., Leckebush and Ullbrich, 2004 ), the occurrence of which is only indirectly related to the time-mean circulation. Nevertheless, models suggest a general similarity between the changes in average and extreme wind speeds ( (Knippertz et al., 2000; ) Räisänen et al., 2004 [JoC, MoS, SRC] ). A caveat to this conclusion is that, even in most RCMs, the simulated extremes of wind speed over land tend to be too low (see Section 11.3.2 ).

11.3.3.6 Mediterranean Cyclones

Several studies have suggested a decrease in the total number of cyclones in the Mediterranean Sea ( (Lionello et al., 2002; ) Vérant, 2004 [NPR] ; Somot 2005 [NPR, MoS, SRC] ; Leckebusch et al., 2006 [NotFound] ;( Pinto et al., 2006; ) Ulbrich et al., 2006 [NPR] ), but there is no agreement on whether the number of intense cyclones will increase or decrease ( (Lionello et al., 2002; ) ( Pinto et al., 2006 ) ).

The Baltic Sea is likely to lose a large part of its seasonal ice cover during this century. Using a regional atmosphere-Baltic Sea model ( Meier et al., 2004 [JoC, MoS] ), the average winter maximum ice extent decreased by about 70% (60%) between 1961 to 1990 and 2071 to 2100 under the A2 (B2) scenario. The length of the ice season was projected to decrease by one to two months in northern parts and two to three months in the central parts. Comparable decreases in Baltic Sea ice cover were projected by earlier studies ( Haapala et al., 2001 [JoC, MoS] ; Meier, 2002 [JoC, MoS] ).

11.3.3.7 Snow and Sea Ice

The overall warming is very likely to shorten the snow season in all of Europe. Snow depth is also likely to be reduced, at least in most areas, although increases in total winter precipitation may counteract the increased melting and decreased fraction of solid precipitation associated with the warming. The changes may be large, including potentially a one-to-three month shortening of the snow season in northern Europe ( Räisänen et al., 2003 [NPR, MoS, SRC] ) and a 50 to 100% decrease in snow depth in most of Europe ( Räisänen et al., 2003 [NPR, MoS, SRC] ; Rowell, 2005 [JoC] ) by the late 21st century. However, snow conditions in the coldest parts of Europe, such as northern Scandinavia and north-western Russia ( Räisänen et al., 2003 [NPR, MoS, SRC] ; Shkolnik et al., 2006 [MoS, ARC] ) and the highest peaks of the Alps ( Beniston et al., 2003 [MoS, SRC] ) appear to be less sensitive to the temperature and precipitation changes projected for this century than those at lower latitudes and altitudes (see also Section 11.4.3.2 , Box 11.3 ).

East Asia

Simulated temperatures in most MMD models are too low in all seasons over East Asia; the mean cold bias is largest in winter and smallest in summer. Zhou and Yu 2006 [JoC, MoS, ARC] ) show that over China, the models perform reasonably in simulating the dominant variations of the mean temperature over China, but not the spatial distributions. The annual precipitation over East Asia exceeds the observed estimates in almost all models and the rain band in the mid-latitudes is shifted northward in seasons other than summer. This bias in the placement of the rains in central China also occurred in earlier models (e.g., Zhou and Li, 2002 [JoC, MoS] ; Gao et al., 2004 [MoS, SRC] ). In winter, the area-mean precipitation is overestimated by more than 50% on average due to strengthening of the rain band associated with extratropical systems over South China. The bias and inter-model differences in precipitation are smallest in summer but the northward shift of this rain band results in large discrepancies in summer rainfall distribution over Korea, Japan and adjacent seas.

Kusunoki et al. 2006 [MoS] ) find that the simulation of the Meiyu-Changma-Baiu rains in the East Asian monsoon is improved substantially with increasing horizontal resolution. Confirming the importance of resolution, RCMs simulate more realistic climatic characteristics over East Asia than AOGCMs, whether driven by re-analyses or by AOGCMs (e.g., Ding et al., 2003 [NPR, MoS, ARC] ; Oh et al., 2004 [MoS] ; Fu et al., 2005 [JoC, MoS, SRC] ; Zhang et al., 2005a [MoS, SRC] , Ding et al., 2006 [MoS, ARC] ; Sasaki et al., 2006b [MoS] ). Several studies reproduce the fine-scale climatology of small areas using a multiply nested RCM ( Im et al., 2006 [MoS, SRC] ) and a very-high resolution (5 km) RCM ( Yasunaga et al., 2006 [MoS] Gao et al. 2006b [MoS, SRC] ) report that simulated East Asia large-scale precipitation patterns are significantly affected by resolution, particularly during the mid- to late-monsoon months, when smaller-scale convective processes dominate.

Southeast Asia

The broad-scale spatial distribution of temperature and precipitation in DJF and JJA averaged across the MMD models compares well with observations. Rajendran et al. 2004 [MoS, SRC] ) examine the simulation of current climate in the MRI coupled model (see Table 8.1 for model details). Large-scale features were well simulated, but errors in the timing of peak rainfall over Indochina were considered a major shortcoming. Collier et al. 2004 [JoC, MoS] ) assess the performance of the CCSM3 model (see Table 8.1 for model details) in simulating tropical precipitation forced by observed SST. Simulation was good over the Maritime continent compared to the simulation for other tropical regions. B Wang et al. 2004 [Ambiguous] ) assess the ability of 11 AGCMs in the Asian-Australian monsoon region simulation forced with observed SST variations. They found that the models’ ability to simulate observed interannual rainfall variations was poorest in the Southeast Asian portion of the domain. Since current AOGCMs continue to have some significant shortcomings in representing ENSO variability (see Section 8.4 ), the difficulty of projecting changes in ENSO-related rainfall in this region is compounded.

Rainfall simulation across the region at finer scales has been examined in some studies. The Commonwealth Scientific and Industrial Research Organisation (CSIRO) stretched-grid Conformal-Cubic Atmospheric Model (CCAM) at 80-km resolution shows reasonable precipitation simulation in JJA, although Indochina tended to be drier than in the observations ( McGregor and Nguyen, 2003 [NPR, MoS] Aldrian et al. 2004a [JoC, MoS] ) conducted a number of simulations with the Max-Planck Institute (MPI) regional model for an Indonesian domain, forced by reanalyses and by the ECHAM4 GCM. The model was able to represent the spatial pattern of seasonal rainfall. It was found that a resolution of at least 50 km was required to simulate rainfall seasonality correctly over Sulawesi. The formulation of a coupled regional model improves regional rainfall simulation over the oceans ( Aldrian et al., 2004b [JoC, MoS] Arakawa and Kitoh 2005 [MoS, SRC] ) demonstrate an accurate simulation of the diurnal cycle of rainfall over Indonesia with an AGCM of 20-km horizontal resolution.

Central Asia and Tibet

Due to the complex topography and the associated mesoscale weather systems of the high-altitude and arid areas, GCMs typically perform poorly over the region. Importantly, the GCMs, and to a lesser extent RCMs, tend to overestimate the precipitation over arid and semi-arid areas in the north (e.g., Small et al., 1999 [JoC, MoS, SRC] ; Gao et al., 2001 [MoS, SRC] ; Elguindi and Giorgi, 2006 [JoC, MoS, SRC] ).

Over Tibet, the few available RCM simulations generally exhibit improved performance in the simulation of present-day climate compared to GCMs (e.g., Gao et al., 2003a [MoS, SRC] ,b; Zhang et al., 2005b [MoS, SRC] ). For example, the GCM simulation of Gao et al. 2003a [MoS, SRC] ) overestimated the precipitation over the north-western Tibetan Plateau by a factor of five to six, while in an RCM nested in this model, the overestimate was less than a factor of two.

South Asia

For the A1B scenario, the MMD-A1B models show a median increase of 3.3°C (see Table 11.1 )in annual mean temperature by the end of the 21st century. The median warming varies seasonally from 2.7°C in JJA to 3.6°C in DJF, and is likely to increase northward in the area, particularly in winter, and from sea to land( Figure 11.9 ). Studies based on earlier AOGCM simulations ( Douville et al., 2000 [MoS] ; Lal and Harasawa, 2001 [MoS] ; Lal et al., 2001 [MoS] ; Rupa Kumar and Ashrit, 2001 [MoS] ; Rupa Kumar et al., 2002 [NPR, MoS] , 2003; Ashrit et al., 2003 [MoS] ; May, 2004b ) support this picture. The tendency of the warming to be more pronounced in winter is also a conspicuous feature of the observed temperature trends over India ( Rupa Kumar et al., 2002 [NPR, MoS] , 2003 ).

Downscaled projections using the Hadley Centre Regional Model (HadRM2) indicate future increases in extreme daily maximum and minimum temperatures throughout South Asia due to the increase in greenhouse gas concentrations. This projected increase is of the order of 2°C to 4°C in the mid-21st century under the IPCC Scenario IS92a in both minimum and maximum temperatures ( Krishna Kumar et al., 2003 [NPR, MoS] ). Results from a more recent RCM, PRECIS, indicate that the night temperatures increase faster than the day temperatures, with the implication that cold extremes are very likely to be less severe in the future ( (Rupa Kumar et al., 2006 ) ).

Figure 11.9. Temperature and precipitation changes over Asia from the MMD-A1B simulations. Top row: Annual mean, DJF and JJA temperature change between 1980 to 1999 and 2080 to 2099, averaged over 21 models. Middle row: same as top, but for fractional change in precipitation. Bottom row: number of models out of 21 that project increases in precipitation.

11.4.3.1 Temperature

The temperature projections for the 21st century based on the MMD-A1B models( Figure 11.8 and Table 11.1 )represent a strong warming over the 21st century. Warming is similar to the global mean warming in Southeast Asia (mean warming between 1980 to 1999 and 2080 to 2099 of 2.5°C). Warming greater than the global mean is projected for South Asia (3.3°C) and East Asia (3.3°C), and much more than the global mean in the continental interior of Asia (3.7°C in central Asia, 3.8°C in Tibet and 4.3°C in northern Asia). In four out of the six regions, the largest warming occurs in DJF, but in central Asia, the maximum occurs in JJA. In Southeast Asia, the warming is nearly the same throughout the year. Model-to-model variation in projected warming is typically about three-quarters of the mean warming (e.g., 2.0°C to 4.7°C for annual mean warming in South Asia). The 5 to 95% ranges based on Tebaldi et al. 2004a [JoC, MoS, SRC] ) suggest a slightly smaller uncertainty than the full range of the model results (Supplementary Material Table S11.2). Because the projected warming is large compared to the interannual temperature variability, a large majority of individual years and seasons in the late 21st century are likely to be extremely warm by present standards( Table 11.1 ). The projections of changes in mean temperature and, where available, temperature extremes, are discussed below in more detail for individual Asian regions.

Figure 11.8. Temperature anomalies with respect to 1901 to 1950 for six Asian land regions for 1906 to 2005 (black line) and as simulated (red envelope) by MMD models incorporating known forcings; and as projected for 2001 to 2100 by MMD models for the A1B scenario (orange envelope). The bars at the end of the orange envelope represent the range of projected changes for 2091 to 2100 for the B1 scenario (blue), the A1B scenario (orange) and the A2 scenario (red). The black line is dashed where observations are present for less than 50% of the area in the decade concerned. More details on the construction of these figures are given in Box 11.1 and Section 11.1.2 .

East Asia

The MMD-A1B models project a median warming of 3.3°C( Table 11.1 )by the end of the 21st century, which varies seasonally from 3.0°C in JJA to 3.6°C in DJF. The warming tends to be largest in winter, especially in the northern inland area( Figure 11.9 ), but the area-mean difference from the other seasons is not large. There is no obvious relationship between model bias and the magnitude of the warming. The spatial pattern of larger warming over northwest EAS( Figure 11.9 )is very similar to the ensemble mean of pre-MMD models. Regional Climate Model simulations show mean temperature increases similar to those simulated by AOGCMs ( Gao et al., 2001 [MoS, SRC] , 2002; Kwon et al., 2003 [NPR, SRC] ; Jiang, 2005 [NPR, MoS] ; Kurihara et al., 2005 [MoS] ; Y.L. ( Xu et al., 2005 ) ).

Daily maximum and minimum temperatures are very likely to increase in East Asia, resulting in more severe warm but less severe cold extremes ( Gao et al., 2002 [MoS, SRC] ; Mizuta et al., 2005 [MoS] ; Y.L. ( Xu et al., 2005; ) Boo et al., 2006 [JoC, MoS, SRC] Mizuta et al. 2005 [MoS] ) analyse temperature-based extreme indices over Japan with a 20-km mesh AGCM and find the changes in the indices to be basically those expected from the mean temperature increase, with changes in the distribution around the mean not playing a large role. Boo et al. 2005 [SRC] ) report similar results for Korea. Gao et al. 2002 [MoS, SRC] ) and Y.L. ( Xu et al. 2005 ) ) find a reduced diurnal temperature range in China and larger increases in daily minimum than maximum temperatures.

The projected decrease in mean precipitation in central Asia is accompanied by an increase in the frequency of very dry spring, summer and autumn seasons; conversely, in winter, where models project increases in the mean precipitation, very high precipitation becomes more common( Table 11.1 ). The projections of changes in mean precipitation and, where available, precipitation extremes, are discussed in more detail below for individual Asian regions. Where appropriate, the connection to changes in circulation systems that bring precipitation is also discussed. Smaller (slightly larger) changes are generally projected for the B1 (A2) scenario, but the inter-scenario differences are small compared with the inter-model differences.

11.4.3.2 Precipitation and Associated Circulation Systems

The MMD models indicate an increase in annual precipitation in most of Asia during this century, the percentage increase being largest and most consistent among models in North and East Asia( Figure 11.9 and Table 11.1 ). The main exception is central Asia, particularly its western parts, where most models simulate reduced precipitation in the summer. Based on these simulations, sub-continental boreal winter precipitation is very likely to increase in northern Asia and the Tibetan Plateau, and likely to increase in eastern Asia. It is also likely to increase in the southern parts of Southeast Asia. Summer precipitation is likely to increase in North, South, Southeast and East Asia, but decrease in central Asia. Probability estimates from Tebaldi et al. 2004a [JoC, MoS, SRC] ; see Supplementary Material Table S11.2) support these judgments.

Southeast Asia

In the MMD-A1B simulations, the median warming for the region is 2.5°C by the end of the 21st century, with little seasonal variation( Table 11.1 ). Simulations by the CSIRO Division of Atmospheric Research Limited Area Model (DARLAM; McGregor et al., 1998 [NPR, MoS] ) and more recently by the CSIRO stretched-grid model ( McGregor and Dix, 2001 [NPR, MoS, ARC] ) centred on the Indochina Peninsula ( AIACC, 2004 [NPR] ) at a resolution of 14 km have demonstrated the potential for significant local variation in warming, particularly the tendency for warming to be significantly stronger over the interior of the landmasses than over the surrounding coastal regions. A tendency for the warming to be stronger over Indochina and the larger landmasses of the archipelago is also visible in the MMD models (Figures 10.8 and 11.9 ). As in other regions, the magnitude of the warming depends on the forcing scenario.

Central Asia and Tibet

In the MMD-A1B simulations, central Asia warms by a median of 3.7°C, and Tibet by 3.8°C( Table 11.1 )by the end of the 21st century. The seasonal variation in the simulated warming is modest. Findings from earlier multi-model studies ( Zhao et al., 2002 [NPR, MoS, SRC] ; Xu et al., 2003a [MoS, ARC] ,b; ( Meleshko et al., 2004; ) Y. ( Xu et al., 2005 ) ) are consistent with the MMD models’ results.

An RCM study by Gao et al. 2003b [SRC] ) indicates greater warming over the Plateau compared to surrounding areas, with the largest warming at highest altitudes, for example, over the Himalayas (see also Box 11.3 ). The higher temperature increase over high-altitude areas can be explained by the decrease in surface albedo associated with the melting of snow and ice ( Giorgi et al., 1997 [JoC, MoS, SRC] ). This phenomenon is found to different extents in some, although not all, of the MMD models, and it is visible in the multi-model mean changes, particularly in the winter( Figure 11.9 ).

South Asia

Most of the MMD-A1B models project a decrease in precipitation in DJF (the dry season), and an increase during the rest of the year. The median change is 11% by the end of the 21st century, and seasonally is –5% in DJF and 11% in JJA, with a large inter-model spread( Table 11.1 ). The probabilistic method of Tebaldi et al. 2004a [JoC, MoS, SRC] ) similarly shows a large spread, although only 3 of the 21 models project a decrease in annual precipitation. This qualitative agreement on increasing precipitation for most of the year is also supported by earlier AOGCM simulations ( Lal and Harasawa, 2001 [MoS] ; Lal et al., 2001 [MoS] ; Rupa Kumar and Ashrit, 2001 [MoS] ; Rupa Kumar et al., 2002 [NPR, MoS] , 2003; Ashrit et al., 2003 [MoS] ; May, 2004b ).

In a study with four GCMs, Douville et al. 2000 [MoS] ) find a significant spread in the summer monsoon precipitation anomalies despite a general weakening of the monsoon circulation (see also May, 2004b ). They conclude that the changes in atmospheric water content, precipitation and land surface hydrology under greenhouse forcing could be more important than the increase in the land-sea thermal gradient for the future evolution of monsoon precipitation. ( Stephenson et al. 2001 ) ) propose that the consequences of climate change could manifest in different ways in the physical and dynamical components of monsoon circulation. Douville et al. 2000 [MoS] ) also argue that the weakening of the ENSO-monsoon correlation could be explained by a possible increase in precipitable water as a result of global warming, rather than by an increased land-sea thermal gradient. However, model diagnostics using ECHAM4 to investigate this aspect indicate that both the above mechanisms can play a role in monsoon changes in a greenhouse-gas warming scenario. Ashrit et al. 2001 [JoC] ) show that the monsoon deficiency due to El Niño might not be as severe, while the favourable impact of la Niña seems to remain unchanged. In a later study using the Centre National de Recherches Météorologiques (CNRM) GCM, Ashrit et al. 2003 [MoS] ) find that the simulated ENSO-monsoon teleconnection shows a strong modulation on multi-decadal time scales, but no systematic change with increasing amounts of greenhouse gases.

Time-slice experiments with ECHAM4 indicate a general increase in the intensity of heavy rainfall events in the future, with large increases over the Arabian Sea and the tropical Indian Ocean, in northern Pakistan and northwest India, as well as in northeast India, Bangladesh and Myanmar (May, 2004a ). The HadRM2 RCM shows an overall decrease by up to 15 days in the annual number of rainy days over a large part of South Asia, under the IS92a scenario in the 2050 s, but with an increase in the precipitation intensity as well as extreme precipitation ( Krishna Kumar et al., 2003 [NPR, MoS] ). Simulations with the PRECIS RCM also project substantial increases in extreme precipitation over a large area, particularly over the west coast of India and west central India ( (Rupa Kumar et al., 2006 ) Dairaku and Emori 2006 [JoC, SRC] ) show from a high-resolution AGCM simulation (about 1.5 degrees) that the increased extreme precipitation over land in South Asia would arise mainly from dynamic effects, that is, enhanced upward motion due to the northward shift of monsoon circulation.

Based on regional HadRM2 simulations, Unnikrishnan et al. 2006 [MoS, ARC] ) report increases in the frequency as well as intensities of tropical cyclones in the 2050 s under the IS92a scenario in the Bay of Bengal, which will cause more heavy precipitation in the surrounding coastal regions of South Asia, during both southwest and northeast monsoon seasons.

East Asia

The MMD-A1B models project an increase in precipitation in East Asia in all seasons. The median change at the end of the 21st century is +9% in the annual mean with little seasonal difference, and a large model spread in DJF( Table 11.1 ). In winter, this increase contrasts with a decrease in precipitation over the ocean to the southeast, where reduced precipitation corresponds well with increased mean sea level pressure. While the projections have good qualitative agreement, there remain large quantitative differences among the models, which is consistent with previous studies (e.g., Giorgi et al., 2001a [NPR, MoS, SRC] ; Hu et al., 2003 [JoC] ; Min et al., 2004 [MoS] ).

Based on the MMD models, Kimoto 2005 [JoC, MoS, ARC] ) projects increased Meiyu-Changma-Baiu activity associated with the strengthening of anticyclonic cells to its south and north, and Kwon et al. 2005 [NPR, SRC] ) show increased East Asia summer precipitation due to an enhanced monsoon circulation in the decaying phase of El Niño. A 20-km mesh AGCM simulation shows that Meiyu-Changma-Baiu rainfall increases over the Yangtze River valley, the East China Sea and western Japan, while rainfall decreases to the north of these areas mostly due to the lengthening of the Meiyu-Changma-Baiu ( Kusunoki et al., 2006 [MoS] ). Simulations by RCMs support the results from AOGCMs. For example, Kurihara et al. 2005 [MoS] ) show an increase in precipitation over western Japan in summer.

Kitoh and Uchiyama 2006 [MoS, SRC] ) investigated the onset and withdrawal times of the Asian summer rainfall season in 15 MMD simulations( Figure 11.10 ). They find a delay in early summer rain withdrawal over the region extending from Taiwan to the Ryukyu Islands to the south of Japan, but an earlier withdrawal over the Yangtze Basin, although the latter is not significant due to large inter-model variation. Changes in onset dates are smaller.

Figure 11.10. (a) The ensemble mean change in withdrawal date of the summer rainy season between the MMD-A1B projections in 2081 to 2100 as compared with the 1981 to 2000 period in the 20C3M simulations. A positive value indicates a later withdrawal date in the A1B scenario. Units are 5 days. (b) Fraction of the models projecting a positive difference in withdrawal date. ( Kitoh and Uchiyama, 2006 [MoS, SRC] ).

Yasunaga et al. 2006 [MoS] ) used a 5-km mesh cloud-resolving RCM to investigate summer rainfall in Japan. They find no changes in June rainfall but increased July rainfall in a warmer climate. The increase in July can be attributed to the more frequent large-precipitation systems.

Intense precipitation events are very likely to increase in East Asia, consistent with the historical trend in this region ( (Fujibé et al., 2005; ) Zhai et al., 2005 [ARC] Kanada et al. 2005 [MoS] ) show, using a 5-km resolution RCM, that the confluence of disturbances from the Chinese continent and from the East China Sea would often cause extremely heavy precipitation over Japan’s Kyushu Island in July in a warmer climate. An increase in the frequency and intensity of heavy precipitation events also occurs in Korea in the long RCM simulation of Boo et al. 2006 [JoC, MoS, SRC] ). Similarly based on RCM simulations, Y.L. ( Xu et al. 2005 ) ) report more extreme precipitation events over China. Gao et al. 2002 [MoS, SRC] ) find a simulated increase in the number of rainy days in northwest China, and a decrease in rain days but an increase in days with heavy rain over South China. Kitoh et al. 2005 [MoS, SRC] ) report similar results in South China from an AOGCM simulation.

Kimoto et al. 2005 [JoC, MoS, ARC] ) suggest that the frequencies of non-precipitating and heavy (≥30 mm day–1) rainfall days would increase significantly at the expense of relatively weak (1–20 mm day–1) rainfall days in Japan. Mizuta et al. 2005 [MoS] ) find significantly more days with heavy precipitation and stronger average precipitation intensity in western Japan and Hokkaido Island. Hasegawa and Emori 2005 [MoS, SRC] ) show that daily precipitation associated with tropical cyclones over the western North Pacific would increase.

The previously noted weakening of the East Asian winter monsoon (e.g., Hu et al., 2000 [JoC] ) is further confirmed by recent studies (e.g., Kimoto, 2005 [JoC, MoS, ARC] ; Hori and Ueda, 2005 [NotFound] ).

Southeast Asia

Area-mean precipitation over Southeast Asia increases in most MMD model simulations, with a median change of about 7% in all seasons( Table 11.1 ), but the projected seasonal changes vary strongly within the region. The seasonal confidence intervals based on the methods of Tebaldi et al. 2004a [JoC, MoS, SRC] ,b) are similar for DJF and JJA (roughly –4% to 17%). The strongest and most consistent increases broadly follow the ITCZ, lying over northern Indonesia and Indochina in JJA, and over southern Indonesia and Papua New Guinea in DJF( Figure 11.9 ). Away from the ITCZ, precipitation decreases are often simulated. The pattern is broadly one of wet season rainfall increase and dry season decrease.

Earlier studies of precipitation change in the area in some cases have suggested a worse inter-model agreement than found for the MMD models. Both Giorgi et al. 2001a [NPR, MoS, SRC] and Ruosteenoja et al. 2003 [NPR, MoS] ) find inconsistency in the simulated direction of precipitation change in the region, but a relatively narrow range of possible changes; similar results were found over an Indonesian domain by Boer and Faqih 2004 [NPR, MoS] ). Compositing the projections from a range of earlier simulations forced by the IS92a scenario, Hulme and Sheard 1999a [NPR] ,b) find a pattern of rainfall increase across northern Indonesia and the Philippines, and decrease over the southern Indonesian archipelago. More recently, Boer and Faqih 2004 [NPR, MoS] ) compared patterns of change across Indonesia from five AOGCMS and obtained highly contrasting results. They conclude that ‘no generalisation could be made on the impact of global warming on rainfall’ in the region.

The regional high-resolution simulations of McGregor et al. 1998 [NPR, MoS] McGregor and Dix 2001 [NPR, MoS, ARC] and AIACC 2004 [NPR] ) have demonstrated the potential for significant local variation in projected precipitation change. The simulations showed considerable regional detail in the simulated patterns of change, but little consistency across the three simulations. The authors related this result to significant deficiencies in the current-climate simulations of the models for this region.

Rainfall variability will be affected by changes in ENSO and its effect on monsoon variability, but this is not well understood (see Section 10.3 ). However, as Boer and Faqih 2004 [NPR, MoS] ) note, those parts of Indonesia that experience a mean rainfall decrease are likely to also experience increases in drought risk. The region is also likely to share the general tendency for daily extreme precipitation to become more intense under enhanced greenhouse conditions, particularly where the mean precipitation is projected to increase. This has been demonstrated in a range of global and regional studies (see Section 10.3 ), but needs explicit study for the Southeast Asian region.

The northern part of the Southeast Asian region will be affected by any change in tropical cyclone characteristics. As noted in Section 10.3 ,there is evidence in general of likely increases in tropical cyclone intensity, but less consistency about how occurrence will change (see also ( Walsh, 2004 ) ). The likely increase in intensity (precipitation and winds) is supported for the northwest Pacific (and other regions) by the recent modelling study of Knutson and Tuleya 2004 [JoC, MoS, ARC] ). The high-resolution time-slice modelling experiment of Hasegawa and Emori 2005 [MoS, SRC] ) also demonstrates an increase in tropical cyclone precipitation in the western North Pacific, but not an increase in tropical cyclone intensity. Wu and Wang 2004 [ARC] ) examined possible changes in tracks in the northwest Pacific due to changes in steering flow in two Geophysical Fluid Dynamics Laboratory (GFDL) enhanced greenhouse gas experiments. Tracks moved more north-easterly, possibly reducing tropical cyclone frequency in the Southeast Asian region. Since most of the tropical cyclones form along the monsoon trough and are also influenced by ENSO, changes in the occurrence, intensity and characteristics of tropical cyclones and their interannual variability will be affected by changes in ENSO (see Section 10.3 ).

Central Asia and Tibet

Precipitation over central Asia increases in most MMD-A1B projections for DJF but decreases in the other seasons. The median change by the end of the 21st century is –3% in the annual mean, with +4% in DJF and –13% in JJA (the dry season)( Table 11.1 ). This seasonal variation in the changes is broadly consistent with the earlier multi-model study of ( Meleshko et al. 2004 ) ), although they find an increase in summer precipitation in the northern part of the area.

Over the Tibetan Plateau, all MMD-A1B models project increased precipitation in DJF (median 19%). Most but not all models also simulate increased precipitation in the other seasons( Table 11.1 ). Earlier studies both by AOGCMs and RCMs are consistent with these findings ( Gao et al., 2003b [SRC] ; Y. Xu et al., 2003a [MoS, ARC] ,b, 2005 ).

11.5.3.1 Surface Air Temperature

The ensemble mean of the MMD models projects a generalised warming for the entire continent with the magnitude projected to increase almost linearly with time( Figure 11.11 ). On an annual-mean basis, projected surface air temperature warming varies from 2°C to 3°C along the western, southern and eastern continental edges (where at least 16 out of the 21 models project a warming in excess of 2°C) up to more than 5°C in the northern region (where 16 out of the 21 AOGCMs project a warming in excess of 4°C). This warming exceeds the spread among models by a factor of three to four over most of the continent. The warming in the USA is projected to exceed 2°C by nearly all the models, and to exceed 4°C by more than 5 AOGCMs out of 21. More regional and seasonal detail on ranges of projected warming is provided in Table 11.1 and Supplementary Table S11.2.

Figure 11.11. Temperature anomalies with respect to 1901 to 1950 for five North American land regions for 1906 to 2005 (black line) and as simulated (red envelope) by MMD models incorporating known forcings; and as projected for 2001 to 2100 by MMD models for the A1B scenario (orange envelope). The bars at the end of the orange envelope represent the range of projected changes for 2091 to 2100 for the B1 scenario (blue), the A1B scenario (orange) and the A2 scenario (red). The black line is dashed where observations are present for less than 50% of the area in the decade concerned. More details on the construction of these figures are given in Box 11.1 and Section 11.1.2 .

The largest warming is projected to occur in winter over northern parts of Alaska and Canada, reaching 10°C in the northernmost parts, due to the positive feedback from a reduced period of snow cover. The ensemble-mean northern warming varies from more than 7°C in winter (nearly all AOGCMs project a warming exceeding 4°C) to as little as 2°C in summer. In summer, ensemble-mean projected warming ranges between 3°C and 5°C over most of the continent, with smaller values near the coasts. In western, central and eastern regions, the projected warming has less seasonal variation and is more modest, especially near the coast, consistent with less warming over the oceans. The warming could be larger in winter over elevated areas as a result of snow-albedo feedback, an effect that is poorly modelled by AOGCMs due to insufficient horizontal resolution (see also Box 11.3 ). In winter, the northern part of the eastern region is projected to warm most while coastal areas are projected to warm by only 2°C to 3°C.

The climate change response of RCMs is sometimes different from that of the driving AOGCM. This appears to be the result of a combination of factors, including the use of different parametrizations (convection and land surface processes are particularly important over North America in summer) and resolution (different resolution may lead to differing behaviour of the same parametrization). For example, Chen et al. 2003 [Ambiguous] ) find that two RCMs project larger temperature changes in summer than their driving AOGCM. In contrast, the projected warming of an RCM compared to its driving AOGCM was found to be 1.5°C less in the central USA ( (Pan et al., 2004; ) Liang et al., 2006 [JoC, MoS] ), a region where observations have shown a cooling trend in recent decades. This resulted in an area of little warming that may have been due to a changing pattern of the LLJ frequency and associated moisture convergence. It is argued that the improved simulation of the LLJ in the RCM is made possible owing to its increased horizontal and vertical resolution. However, other RCMs with similar resolution do not produce the same response.

11.5.3.2 Precipitation

As a consequence of the temperature dependence of the saturation vapour pressure in the atmosphere, the projected warming is expected to be accompanied by an increase in atmospheric moisture flux and its convergence/divergence intensity. This results in a general increase in precipitation over most of the continent except the most south-westerly part( Figure 11.12 ). The ensemble mean of MMD models projects an increase in annual mean precipitation in the north reaching +20%, which is twice the inter-model spread, so likely significant; the projected increase reaches as much as +30% in winter. Because the increased saturation vapour pressure can also yield greater evaporation, projected increases in annual precipitation are partially offset by increases in evaporation; regions in central North America may experience net surface drying as a result (see Supplementary Material Figure S11.1). See Table 11.1 and Supplementary Table S11.2 for more regional and seasonal details, noting that regional averaging hides important north-south differences.

Figure 11.12. Temperature and precipitation changes over North America from the MMD-A1B simulations. Top row: Annual mean, DJF and JJA temperature change between 1980 to 1999 and 2080 to 2099, averaged over 21 models. Middle row: same as top, but for fractional change in precipitation. Bottom row: number of models out of 21 that project increases in precipitation.

In keeping with the projected northward displacement of the westerlies and the intensification of the Aleutian Low( Section 11.5.3.3 ), northern region precipitation is projected to increase, by the largest amount in autumn and by the largest fraction in winter. Due to the increased precipitable water, the increase in precipitation amount is likely to be larger on the windward slopes of the mountains in the west with orographic precipitation. In western regions, modest changes in annual mean precipitation are projected, but the majority of AOGCMs indicate an increase in winter and a decrease in summer. Models show greater consensus on winter increases (ensemble mean maximum of 15%) to the north and on summer decreases (ensemble mean maximum of –20%) to the south. These decreases are consistent with enhanced subsidence and flow of drier air masses in the southwest USA and northern Mexico resulting from an amplification of the subtropical anticyclone off the West Coast due to the land-sea contrast in warming (e.g., Mote and Mantua, 2002 [JoC, MoS, ARC] ). However, this reduction is close to the inter-model spread so it contains large uncertainty, an assessment that is reinforced by the fact that some AOGCMs project an increase in precipitation.

In central and eastern regions, projections from the MMD models show the same characteristics as in the west, with greater consensus for winter increases to the north and summer decreases to the south. The line of zero change is oriented more or less west-to-east and moves north from winter to summer. The line of zero change is also projected to lie further to the north under SRES scenarios with larger greenhouse gas amounts. However, uncertainty around the projected changes is large and the changes do not scale well across different SRES scenarios.

Govindasamy 2003 [JoC, MoS] ) finds that, averaged over the USA, the few existing time-slice simulations with high-resolution AGCM results do not significantly differ from those obtained with AOGCMs. Available RCM simulations provide little extra information on average changes. Some RCMs project precipitation changes of different sign, either locally ( Chen et al., 2003 [Ambiguous] ) or over the entire continental USA ( Han and Roads, 2004 [JoC, MoS] , where in summer the AOGCM generally produced a small increase and the RCM a substantial decrease). In contrast, Plummer et al. 2006 [MoS, SRC] ) find only small differences in precipitation responses using two sets of physical parametrizations in their RCM, despite the fact that one corrected significant summer precipitation excess present in the other.

11.5.3.3 Temperature and Precipitation Extremes

Several RCM studies focused particularly on changes in extreme temperature events. Bell et al. 2004 [JoC, MoS] ) examine changes in temperature extremes in their simulations centred on California. They find increases in extreme temperature events, both as distribution percentiles and threshold events, prolonged hot spells and increased diurnal temperature range. Leung et al. 2004 [JoC] ) examine changes in extremes in their RCM simulations of the western USA; in general, they find increases in diurnal temperature range in six sub-regions of their domain in summer. Diffenbaugh et al. 2005 [JoC, SRC] ) find that the frequency and magnitude of extreme temperature events changes dramatically under SRES A2, with increases in extreme hot events and a decrease in extreme cold events.

In a study of precipitation extremes over California, Bell et al. 2004 [JoC, MoS] ) find that changes in precipitation exceeding the 95th percentile followed changes in mean precipitation, with decreases in heavy precipitation in most areas. Leung et al. 2004 [JoC] ) find that extremes in precipitation during the cold season increase in the northern Rockies, the Cascades, the Sierra Nevada and British Columbia by up to 10% for 2040 to 2060, although mean precipitation was mostly reduced, in accord with earlier studies ( Giorgi et al., 2001a [NPR, MoS, SRC] ). In a large river basin in the Pacific Northwest, increases in rainfall over snowfall and rain-on-snow events increased extreme runoff by 11%, which would contribute to more severe flooding. In their 25-km RCM simulations covering the entire USA, Diffenbaugh et al. 2005 [JoC, SRC] ) find widespread increases in extreme precipitation events under SRES A2, which they determine to be significant.

11.5.3.4 Atmospheric Circulation

In general, the projected climate changes over North America follow the overall features of those over the NH( Chapter 10 ). The MMD models project a northward displacement and strengthening of the mid-latitude westerly flow, most pronounced in autumn and winter. Surface pressure is projected to decrease in the north, with a northward displacement of the Aleutian low-pressure centre and a north-westward displacement of the Labrador Sea trough, and to decrease slightly in the south. The reductions in surface pressure in the north are projected to be strongest in winter, reaching –1.5 to –3 hPa, in part as a result of the warming of the continental arctic air mass. On an annual basis, the pressure decrease in the north exceeds the spread among models by a factor of 3 on an annual-mean basis and a factor of 1.5 in summer, so it is significant. The East Pacific subtropical anticyclone is projected to intensify in summer, particularly off the coast of California and Baja California, resulting in an increased air mass subsidence and drier airflow over south-western North America. The pressure increase (less than 0.5 hPa) is small compared to the spread among models, so this projection is rather uncertain.

11.5.3.5 Snowpack, Snowmelt and River Flow

The ensemble mean of the MMD models projects a general decrease in snow depth( Chapter 10 )as a result of delayed autumn snowfall and earlier spring snowmelt. In some regions where winter precipitation is projected to increase, the increased snowfall can more than make up for the shorter snow season and yield increased snow accumulation. Snow depth increases are projected by some GCMs over some land around the Arctic Ocean (Figure S10.1) and by some RCMs in the northernmost part of the Northwest Territories( Figure 11.13 ). In principle a similar situation could arise at lower latitudes at high elevations in the Rocky Mountains, although most models project a widespread decrease of snow depth there ( Kim et al., 2002 [JoC] ; Snyder et al., 2003 [JoC, MoS] ; Leung et al., 2004 [JoC] ; see also Box 11.3 ).

Much SD research activity has focused on resolving future water resources in the complex terrain of the western USA. Studies typically point to a decline in winter snowpack and hastening of the onset of snowmelt caused by regional warming ( Hayhoe et al., 2004 [JoC] ; Salathé, 2005 [JoC, MoS] ). Comparable trends towards increased annual mean river flows and earlier spring peak flows have also been projected by two SD techniques for the Saguenay watershed in northern Québec, Canada ( Dibike and Coulibaly, 2005 [MoS] ). Such changes in the flow regime also favour increased risk of winter flooding and lower summer soil moisture and river flows. However, differences in snowpack behaviour derived from AOGCMs depend critically on the realism of downscaled winter temperature variability and its interplay with precipitation and snowpack accumulation and melt ( Salathé, 2005 [JoC, MoS] Hayhoe et al. 2004 [JoC] ) produced a standard set of statistically downscaled temperature and precipitation scenarios for California; under both the A1F1 and B1 scenarios, they find overall declines in snowpack.

Figure 11.13. Percent snow depth changes in March (only calculated where climatological snow amounts exceed 5 mm of water equivalent), as projected by the Canadian Regional Climate Model (CRCM; Plummer et al., 2006 [MoS, SRC] ), driven by the Canadian General Circulation Model (CGCM), for 2041 to 2070 under SRES A2 compared to 1961 to 1990 .

11.6.3.1 Temperature

The warming as simulated by the MMD-A1B projections increases approximately linearly with time during this century, but the magnitude of the change and the inter-model range are greater over CAM and AMZ than over SSA( Figure 11.14 ). The annual mean warming under the A1B scenario between 1980 to 1999 and 2080 to 2099 varies in the CAM region from 1.8°C to 5.0°C, with half of the models within 2.6°C to 3.6°C and a median of 3.2°C. The corresponding numbers for AMZ are 1.8°C to 5.1°C, 2.6°C to 3.7°C and 3.3°C, and those for SSA 1.7°C to 3.9°C, 2.3°C to 3.1°C and 2.5°C( Table 11.1 ). The median warming is close to the global ensemble mean in SSA but about 30% above the global mean in the other two regions. As in the rest of the tropics, the signal-to-noise ratio is large for temperature, and it requires only 10 years for a 20-year mean temperature, growing at the rate of the median A1B response, to be clearly discernible above the models’ internal variability.

The simulated warming is generally largest in the most continental regions, such as inner Amazonia and northern Mexico( Figure 11.15 ). Seasonal variation in the regional area mean warming is relatively modest, except in CAM where there is a difference of 1°C in median values between DJF and MAM( Table 11.1 ). The warming in central Amazonia tends to be larger in JJA than in DJF, while the reverse is true over the Altiplano where, in other words, the seasonal cycle of temperature is projected to increase( Figure 11.15 ). Similar results were found by Boulanger et al. 2006 [JoC, MoS] ), who studied the regional thermal response over South America by applying a statistical method based on neural networks and Bayesian statistics to find optimal weights for a linear combination of MMD models.

Figure 11.14. Temperature anomalies with respect to 1901 to 1950 for three Central and South American land regions for 1906 to 2005 (black line) and as simulated (red envelope) by MMD models incorporating known forcings; and as projected for 2001 to 2100 by MMD models for the A1B scenario (orange envelope). The bars at the end of the orange envelope represent the range of projected changes for 2091 to 2100 for the B1 scenario (blue), the A1B scenario (orange) and the A2 scenario (red). The black line is dashed where observations are present for less than 50% of the area in the decade concerned. More details on the construction of these figures are given in Box 11.1 and Section 11.1.2 .

Figure 11.15. Temperature and precipitation changes over Central and South America from the MMD-A1B simulations. Top row: Annual mean, DJF and JJA temperature change between 1980 to 1999 and 2080 to 2099, averaged over 21 models. Middle row: same as top, but for fractional change in precipitation. Bottom row: number of models out of 21 that project increases in precipitation.

For the variation of seasonal warming between the individual models, see Table 11.1 .As an alternative approach to estimating uncertainty in the magnitude of the warming, the 5th and 95th percentiles for temperature change at the end of the 21st century, assessed using the method of Tebaldi et al. 2004a [JoC, MoS, SRC] ), are typically within ±1°C of the median value in all three of these regions (Supplementary Material Table S11.2).

11.6.3.2 Precipitation

The MMD models suggest a general decrease in precipitation over most of Central America, consistent with Neelin et al. 2006 [JoC, MoS] ), where the median annual change by the end of the 21st century is –9% under the A1B scenario, and half of the models project area mean changes from –16 to –5%, although the full range of the projections extends from –48 to 9%. Median changes in area mean precipitation in Amazonia and southern South America are small and the variation between the models is also more modest than in Central America, but the area means hide marked regional differences( Table 11.1 , Figure 11.15 ).

Area mean precipitation in Central America decreases in most models in all seasons. It is only in some parts of north-eastern Mexico and over the eastern Pacific, where the ITCZ forms during JJA, that increases in summer precipitation are projected( Figure 11.15 ). However, since tropical storms can contribute a significant fraction of the rainfall in the hurricane season in this region, these conclusions might be modified by the possibility of increased rainfall in storms not well captured by these global models. In particular, if the number of storms does not change, Knutson and Tuleya 2004 [JoC, MoS, ARC] ) estimate nearly a 20% increase in average precipitation rate within 100 km of the storm centre at the time of atmospheric carbon dioxide (CO2) doubling.

For South America, the multi-model mean precipitation response( Figure 11.15 )indicates marked regional variations. The annual mean precipitation is projected to decrease over northern South America near the Caribbean coasts, as well as over large parts of northern Brazil, Chile and Patagonia, while it is projected to increase in Colombia, Ecuador and Peru, around the equator and in south-eastern South America. The seasonal cycle modulates this mean change, especially over the Amazon Basin where monsoon precipitation increases in DJF and decreases in JJA. In other regions (e.g., Pacific coasts of northern South America, a region centered over Uruguay, Patagonia) the sign of the response is preserved throughout the seasonal cycle.

As seen in the bottom panels of Figure 11.15 ,most models project a wetter climate near the Rio de la Plata and drier conditions along much of the southern Andes, especially in DJF. However, when estimating the likelihood of this response, the qualitative consensus within this set of models should be weighed against the fact that most models show considerable biases in regional precipitation patterns in their control simulations.

The poleward shift of the South Pacific and South Atlantic subtropical anticyclones is a robust response across the models. Parts of Chile and Patagonia are influenced by the polar boundary of the subtropical anticyclone in the South Pacific and experience particularly strong drying because of the combination of the poleward shift in circulation and increase in moisture divergence. The strength and position of the subtropical anticyclone in the South Atlantic is known to influence the climate of south-eastern South America and the South Atlantic Convergence Zone ( Robertson et al., 2003 [MoS] ; Liebmann et al., 2004 [JoC] ). The increase in rainfall in south-eastern South America is related to a corresponding poleward shift in the Atlantic storm track ( Yin, 2005 [JoC, MoS] ).

Some projected changes in precipitation (such as the drying over east-central Amazonia and northeast Brazil and the wetter conditions over south-eastern South America) could be a partial consequence of the El-Niño like response projected by the models( Section 10.3 ). The accompanying shift and alterations in the Walker Circulation would directly affect tropical South America ( Cazes Boezio et al., 2003 [JoC] ) and affect southern South America through extratropical teleconnections ( Mo and Nogués-Paegle, 2001 [JoC] ).

Although feedbacks from carbon cycle and dynamic vegetation are not included in MMD models, a number of coupled carbon cycle-climate projections have been performed since the TAR (see Sections 7.2 and 10.4.1 ). The initial carbon-climate simulations suggest that drying of the Amazon potentially contributes to acceleration of the rate of anthropogenic global warming by increasing atmospheric CO2 ( Cox et al., 2000 [JoC, MoS, ARC] ; Friedlingstein et al., 2001 [JoC, MoS, ARC] ; Dufresne et al., 2002 [JoC, MoS] ; Jones et al., 2003 [JoC, MoS, SRC] ). These models display large uncertainty in climate projections and differ in the timing and sharpness of the changes ( Friedlingstein et al., 2003 [JoC, ARC] ). Changes in CO2 are related to precipitation changes in regions such as the northern Amazon ( Zeng et al., 2004 [JoC] ). In a version of the HadCM3 model with dynamic vegetation and an interactive global carbon cycle ( Betts et al., 2004 [MoS, ARC] ), a tendency to a more El Niño-like state contributes to reduced rainfall and vegetation dieback in the Amazon ( Cox et al., 2004 [MoS, ARC] ). But the version of HadCM3 participating in the MMD projects by far the largest reduction in annual rainfall over AMZ (–21% for the A1B scenario). This stresses the necessity of being very cautious in interpreting carbon cycle impacts on the regional climate and ecosystem change until there is more convergence among models on rainfall projections for the Amazon with fixed vegetation. Box 11.4 summarises some of the major issues related to regional land use/land changes in the context of climate change.

11.7.3.1 Mean Temperature

In both the southern and northern Australia regions, the projected MMD-A1B warming in the 21st century represents a significant acceleration of warming over that observed in the 20th century( Figure 11.16 ). The warming is larger than over the surrounding oceans, but only comparable to, or slightly larger than the global mean warming. Averaging over the region south of 30°S (SAU), the median 2100 warming among all of the models is 2.6°C (with an inter-quartile range of 2.4°C to 2.9°C) whereas the median warming averaged over the region north of 30°S (NAU) is 3.0°C (range of 2.8°C to 3.5°C). The seasonal cycle in the warming is weak, but with larger values (and larger spread among model projections) in summer (DJF). Across the MMD models, the warming is well correlated with the global mean warming, with a correlation coefficient of 0.79, so that more than half of the variance among models is controlled by global rather than local factors, as in many other regions. The range of responses is comparable but slightly smaller than the range in global mean temperature responses, and warming over equivalent time periods under the B1, A1B, and A2 scenarios is close to the ratios of the global mean responses The warming varies sub-regionally, with less warming in coastal regions, Tasmania and the South Island of New Zealand, and greater warming in central and northwest Australia (see Figure 10.8 ).

Figure 11.16. Temperature anomalies with respect to 1901 to 1950 for two Australian land regions for 1906 to 2005 (black line) and as simulated (red envelope) by MMD models incorporating known forcings; and as projected for 2001 to 2100 by MMD models for the A1B scenario (orange envelope). The bars at the end of the orange envelope represent the range of projected changes for 2091 to 2100 for the B1 scenario (blue), the A1B scenario (orange) and the A2 scenario (red). More details on the construction of these figures are given in Box 11.1 and Section 11.1.2 .

These results are broadly (and in many details) similar to those described in earlier studies, so other aspects of these earlier studies can be assumed to remain relevant. For the CSIRO 2001 [NPR, MoS] ) projections, pattern-scaling methods were used to provide patterns of change rescaled by the range of global warming given by IPCC 2001 [NPR] ) for 2030 and 2070 based on the SRES scenarios. By 2030, the warming is 0.4°C to 2°C over most of Australia, with slightly less warming in some coastal areas and Tasmania, and slightly more warming in the northwest. By 2070, annual average temperatures increase by 1°C to 6°C over most of Australia with spatial variations similar to those for 2030 . Dynamically downscaled mean temperature change typically does not differ very significantly from the picture based on AOGCMs (e.g., see Whetton et al., 2002 [NPR, SRC] ). Projected warming over New Zealand (allowing for the IPCC 2001 [NPR] ) range of global warming and differences in the regional results of six GCMs used for downscaling) is 0.2°C to 1.3°C by the 2030 s and 0.5°C to 3.5°C by the 2080 s (Ministry for the Environment, 2004 ).

11.7.3.2 Mean Precipitation

Asummary of projected precipitation changes from the MMD models is presented in Figure 11.17 and Table 11.1 .The most robust feature is the reduction in rainfall along the south coast in JJA (not including Tasmania) and in the annual mean, and a decrease is also strongly evident in SON. The percentage JJA change in 2100 under the A1B scenario for southern Australia has an inter-quartile range of –26 to –7% and by comparison the same range using the probabilistic method of Tebaldi et al. 2004b [MoS, SRC] ) is –13 to –6% (Supplemental Material Table S11.2). There are large reductions to the south of the continent in all seasons, due to the poleward movement of the westerlies and embedded depressions ( Cai et al., 2003a [JoC, SRC] ; Yin, 2005 [JoC, MoS] ; Chapter 10 ), but this reduction extends over land during winter when the storm track is placed furthest equatorward. Due to poleward drift of the storm track as it crosses Australian longitudes, the strongest effect is in the southwest, where the ensemble mean drying is in the 15 to 20% range. Hope 2006a [JoC] ,b) shows a southward or longitudinal shift in storms away from south-western Australia in the MMD simulations. To the east of Australia and over New Zealand, the primary storm track is more equatorward, and the north/south drying/moistening pattern associated with the poleward displacement is shifted equatorward as well. The result is a robust projection of increased rainfall on the South Island (especially its southern half), possibly accompanied by a decrease in the north part of the North Island. The South Island increase is likely to be modulated by the strong topography (see Box 11.3 )and to appear mainly upwind of the main mountain range.

Figure 11.17. Temperature and precipitation changes over Australia and New Zealand from the MMD-A1B simulations. Top row: Annual mean, DJF and JJA temperature change between 1980 to 1999 and 2080 to 2099, averaged over 21 models. Middle row: same as top, but for fractional change in precipitation. Bottom row: number of models out of 21 that project increases in precipitation.

Other aspects of simulated precipitation change appear less robust. On the east coast of Australia, there is a tendency in the models for an increase in rain in the summer and a decrease in winter, with a slight annual decrease. However, consistency among the models on these features is weak.

These results are broadly consistent with results based on earlier GCM simulations. In the CSIRO 2001 [NPR, MoS] ) projections based on a range of nine simulations, projected ranges of annual average rainfall change tend towards a decrease in the southwest and south but show more mixed results elsewhere ( Whetton et al., 2005 [NPR, MoS, SRC] ). Seasonal results showed that rainfall tended to decrease in southern and eastern Australia in winter and spring, increase inland in autumn and increase along the east coast in summer. Moise et al. 2005 [MoS, ARC] ) also find a tendency for winter rainfall decreases across southern Australia and a slight tendency for rainfall increases in eastern Australia in 18 CMIP2 simulations under a 1% yr–1 atmospheric CO2 increase.

Whetton et al. 2001 [MoS, SRC] ) demonstrate that inclusion of high-resolution topography could reverse the simulated direction of rainfall change in parts of Victoria (see Box 11.3 ). In a region of strong rainfall decrease as simulated directly by the GCMs, two different downscaling methods ( Charles et al., 2004 [MoS] ; Timbal, 2004 [MoS] ) have been applied to obtain the characteristics of rainfall change at stations ( IOCI, 2002 [NPR] , 2005; Timbal, 2004 [MoS] ). The downscaled results continued to show the simulated decrease, although the magnitude of the changes was moderated relative to the GCM in the Timbal 2004 [MoS] ) study. Downscaled rainfall projections for New Zealand (incorporating differing results from some six GCMs) showed a strong variation across the islands (Ministry for the Environment, 2004 ). The picture that emerges is that the pattern of precipitation changes described above in the global simulations is still present, but with the precipitation changes focused on the upwind sides of the islands, with the increase in rainfall in the south concentrated in the west, and the decrease in the north concentrated in the east.

11.7.3.3 Snow Cover

The likelihood that precipitation will fall as snow will decrease as temperature rises. Hennessy et al. 2003 [NPR] ) modelled snowfall and snow cover in the Australian Alps under the CSIRO 2001 [NPR, MoS] ) projected temperature and precipitation changes and obtained very marked reductions in snow. The total alpine area with at least 30 days of snow cover decreases 14 to 54% by 2020, and 30 to 93% by 2050 . Because of projected increased winter precipitation over the southern Alps, it is less clear that mountain snow will be reduced in New Zealand (Ministry for the Environment, 2004; see also Box 11.3 ). However, marked decreases in average snow water over New Zealand (60% by 2040 under the A1B scenario) have been simulated by Ghan and Shippert 2006 [JoC, MoS] ) using a high-resolution sub-grid scale orography in a global model that simulates little change in precipitation.

Using the method of Walsh et al. 1999 [NPR] ), changes in potential evaporation in the Australian region have been calculated for a range of enhanced greenhouse climate model simulations ( Whetton et al., 2002 [NPR, SRC] ; Cai et al., 2003b [NPR, ARC] ; McInnes et al., 2003 [NPR, SRC] , 2004; Hennessy et al., 2004a [NPR] ,b). In all cases, increases in potential evaporation were simulated, and in almost all cases, the moisture balance deficit became larger. This has provided a strong indication of the Australian environment becoming drier under enhanced greenhouse conditions.

11.7.3.4 Potential Evaporation

11.7.3.5 Temperature and Precipitation Extremes

Where the analysis has been done for Australia (e.g., Whetton et al., 2002 [NPR, SRC] ), the effect on changes in extreme temperature due to simulated changes in variability is small relative to the effect of the change in the mean. Therefore, most regional assessments of changes in extreme temperatures have been based on adding a projected mean temperature change to each day of a station-observed data set. Based on the CSIRO 2001 [NPR, MoS] ) projected mean temperature change scenarios, the average number of days over 35°C each summer in Melbourne would increase from 8 at present to 9 to 12 by 2030 and 10 to 20 by 2070 ( CSIRO, 2001 [NPR, MoS] ). In Perth, such hot days would rise from 15 at present to 16 to 22 by 2030 and 18 to 39 by 2070 ( CSIRO, 2001 [NPR, MoS] ). On the other hand, cold days become much less frequent. For example, Canberra’s current 44 winter days of minimum temperature below 0°C is projected to be 30 to 42 by 2030 and 6 to 38 by 2070 ( CSIRO, 2001 [NPR, MoS] ).

Changes in extremes in New Zealand have been assessed using a similar methodology and simulations ( Mullan et al., 2001b [NPR, MoS, SRC] ). Decreases in the annual frequency of days below 0°C of 5 to 30 days by 2100 are projected for New Zealand, particularly for the lower North Island and the South Island. Increases in the annual number of days above 25°C of 10 to 50 days by 2100 are projected.

Arange of GCM and regional modelling studies in recent years have identified a tendency for daily rainfall extremes to increase under enhanced greenhouse conditions in the Australian region (e.g., Hennessy et al., 1997 [JoC, ARC] ; Whetton et al., 2002 [NPR, SRC] ; McInnes et al., 2003 [NPR, SRC] ; Watterson and Dix, 2003 [JoC, MoS, ARC] ; Hennessy et al., 2004b [NPR] ; Suppiah et al., 2004 [NPR, SRC] ; Kharin and Zwiers, 2005 [JoC, MoS, ARC] ). Commonly, return periods of extreme rainfall events halve in late 21st-century simulations. This tendency can apply even when average rainfall is simulated to decrease, but not necessarily when this decrease is marked (see Timbal, 2004 [MoS] ). Recently, Abbs 2004 [NPR, MoS] ) dynamically downscaled to a resolution of 7 km current and enhanced greenhouse cases of extreme daily rainfall occurrence in northern New South Wales and southern Queensland as simulated by the CSIRO GCM. The downscaled extreme events for a range of return periods compared well with observations and the enhanced greenhouse simulations for 2040 showed increases of around 30% in magnitude, with the 1-in-40 year event becoming the 1-in-15 year event. Less work has been done on projected changes in rainfall extremes in New Zealand, although the recent analysis of Ministry for the Environment ( 2004 ) based on Semenov and Bengtsson 2002 [JoC, MoS] ) indicates the potential for extreme winter rainfall (95th percentile) to change by between –6% and +40%.

Where GCMs simulate a decrease in average rainfall, it may be expected that there would be an increase in the frequency of dry extremes (droughts). Whetton and Suppiah 2003 [NPR, MoS, SRC] ) examine simulated monthly frequencies of serious rainfall deficiency for Victoria, which show strong average rainfall decreases in most simulations considered. There is a marked increase in the frequency of rainfall deficiencies in most simulations, with doubling in some cases by 2050 . Using a slightly different approach, likely increases in the frequency of drought have also been established for the states of South Australia, New South Wales and Queensland ( Walsh et al., 2002 [MoS, ARC] ; McInnes et al., 2003 [NPR, SRC] ; Hennessy et al., 2004c [NPR] Mullan et al. 2005 [NPR, ARC] ) show that by the 2080 s in New Zealand there may be significant increases in drought frequency in the east of both islands.

11.7.3.6 Tropical Cyclones

Anumber of recent regional model-based studies of changes in tropical cyclone behaviour in the Australian region have examined aspects of number, tracks and intensities under enhanced greenhouse conditions (e.g., Walsh and Katzfey, 2000 [MoS] ;( Walsh and Ryan, 2000; ) ( Walsh et al., 2004 ) ). There is no clear picture with respect to regional changes in frequency and movement, but increases in intensity are indicated. For example, ( Walsh et al. 2004 ) ) obtained, under tripled CO2 conditions, a 56% increase in storms with a maximum wind speed greater than 30 m s–1. It should also be noted that ENSO fluctuations have a strong impact on patterns of tropical cyclone occurrence in the region, and therefore uncertainty with respect to future ENSO behaviour (see Section 10.3 )contributes to uncertainty with respect to tropical cyclone behaviour ( (Walsh, 2004 ) ). See Section 10.3.6.3 for a global assessment of changes in tropical cyclone characteristics.

The MMD ensemble mean projected change in winter sea level pressure is shown in Figure 10.9 .Much of Australia lies to the north of the centre of the high-pressure anomaly. With the mean latitude of maximum pressure near 30°S at this season, this corresponds to a modest strengthening of the mean wind over inland and northern areas and a slight weakening of the mean westerlies on the southern coast, consistent with Hennessy et al. 2004b [NPR] ). Studies of daily extreme winds in the region using high-resolution model output ( McInnes et al., 2003 [NPR, SRC] ) indicate increases of up to 10% across much of the northern half of Australia and the adjacent oceans during summer by 2030 . In winter, the pressure gradient is projected to increase over the South Island of New Zealand (see Figure 10.9 ), implying increased windiness. This increase is present in all of the MMD-A1B projections.

11.7.3.7 Winds

Temperature

The simulated spatial patterns of the MMD ensemble mean temperatures agree closely with those of the observations throughout the annual cycle. Generally, the simulations are 1°C to 2°C colder than the European Centre for Medium-Range Weather Forecasts 40-year (ERA40) reanalyses with the exception of a cold bias maximum of 6°C to 8°C in the Barents Sea (particularly in winter/spring) caused by overestimated sea ice in this region ( Chapman and Walsh, 2007 [JoC, MoS, ARC, 2007] ; see also Section 8.3 ). Compared with earlier model versions, the annual temperature simulations improved in the Barents and Norwegian Seas and Sea of Okhotsk, but some deterioration is noted in the central Arctic Ocean and the high terrain areas of Alaska and northwest Canada ( Chapman and Walsh, 2007 [JoC, MoS, ARC, 2007] ). The mean model ensemble bias is relatively small compared to the across-model scatter of temperatures. The annual mean root-mean-squared error in the individual MMD models ranges from 2°C to 7°C ( Chapman and Walsh, 2007 [JoC, MoS, ARC, 2007] ). Compared with previous models, the MMD-simulated temperatures are more consistent across the models in winter, but somewhat less so in summer. There is considerable agreement between the modelled and observed interannual variability both in magnitude and spatial pattern.

11.8.1.1 Key Processes

Arctic climate is characterised by a distinctive complexity due to numerous nonlinear interactions between and within the atmosphere, cryosphere, ocean, land and ecosystems. Sea ice plays a crucial role in the arctic climate, particularly through its albedo. Reduction of ice extent leads to warming due to increased absorption of solar radiation at the surface. Substantial low-frequency variability is evident in various atmosphere and ice parameters ( (Polyakov et al., 2003a, ) b), complicating the detection and attribution of arctic changes. Natural multi-decadal variability has been suggested as partly responsible for the large warming in the 1920 s to 1940 s ( Bengtsson et al., 2004 [JoC] ; Johannessen et al., 2004 [JoC, MoS] ) followed by cooling until the 1960 s. In both models and observations, the interannual variability of monthly temperatures is at a maximum at high latitudes ( Räisänen, 2002 [SRC] ). Natural atmospheric patterns of variability on annual and decadal time scales play an important role in the arctic climate. Such patterns include the NAM, the NAO, the Pacific-North American (PNA) pattern and the Pacific Decadal Oscillation (PDO), which are associated with prominent arctic regional precipitation and temperature anomalies (see Box 3.4 and Section 3.6 ). For instance, the positive NAM/NAO phase is associated with warmer, wetter winters in Siberia and colder, drier winters in western Greenland and north-eastern Canada. The NAM/NAO showed a trend towards its positive phase over the last three to four decades, although it returned to near its long-term mean state in the last five years (see Section 3.6 ). In the future, global models project a positive trend in the NAO/NAM during the 21st century (see Section 10.3 ). There was substantial decadal-to-inter-decadal atmospheric variability in the North Pacific over the 20th century, associated with fluctuations in the strength of the winter Aleutian Low that co-vary with North Pacific SST in the PDO (see Section 3.6 ). A deeper and eastward-shifted Aleutian Low advects warmer and moister air into Alaska. While some studies have suggested that the Brooks Range effectively isolates arctic Alaska from much of the variability associated with North Pacific teleconnection patterns (e.g., L’Heureux et al., 2004 [JoC] ), other studies find relationships between the Alaskan and Beaufort-Chukchi region’s climate and North Pacific variability ( Stone, 1997 [JoC] ; Curtis et al., 1998 [JoC] ; Lynch et al., 2004 [JoC, SRC] ). Patterns of variability in the North Pacific, and their implications for climate change, are especially difficult to sort out due to the presence of several patterns (NAM, PDO, PNA) with potentially different underlying mechanisms (see Chapter3 ).

11.8.1.2 Skill of Models in Simulating Present Climate

Many processes are still poorly understood and thus continue to pose a challenge for climate models ( ACIA, 2005 [NPR] ). In addition, evaluating simulations of the Arctic is difficult because of the uncertainty in the observations. The few available observations are sparsely distributed in space and time and different data sets often differ considerably ( (Serreze and Hurst, 2000; ) ACIA, 2005 [NPR] ; Liu et al., 2005 [JoC] ; Wyser and Jones, 2005 [JoC, MoS] ). This holds especially for precipitation measurements, which are problematic in cold environments ( Goodison et al., 1998 [NPR] ; Bogdanova et al., 2002 [JoC, MoS] ).

Few pan-arctic atmospheric RCMs are in use. When driven by analysed lateral and sea ice boundary conditions, RCMs tend to show smaller temperature and precipitation biases in the Arctic compared to GCMs, indicating that sea ice simulation biases and biases originating from lower latitudes contribute substantially to the contamination of GCM results in the Arctic (e.g., Dethloff et al., 2001 [JoC, MoS, SRC] ; Wei et al., 2002 [MoS, SRC] ; Lynch et al., 2003 [JoC, SRC] ; Semmler et al., 2005 [ARC] ). However, even under a very constrained experimental design, there can be considerable across-model scatter in RCM simulations ( Tjernström et al., 2005 [MoS] ; Rinke et al., 2006 [JoC, MoS, SRC] ). The construction of coupled atmosphere-ice-ocean RCMs for the Arctic is a recent development ( Maslanik et al., 2000 [JoC, MoS, SRC] ; Debernard et al., 2003 [NPR, MoS] ; Rinke et al., 2003 [JoC, MoS, SRC] ; Mikolajewicz et al., 2005 [MoS] ).

Precipitation

The AOGCM-simulated monthly precipitation varies substantially among the models throughout the year but the MMD ensemble mean monthly means are within the range of different observational data sets. This is an improvement compared to earlier simulations ( Walsh et al., 2002 [MoS, ARC] ; ACIA, 2005 [NPR] ), particularly from autumn to spring ( Kattsov et al., 2007 [JoC, MoS, ARC, 2007] ). The ensemble mean bias varies with season and remains greatest in spring and smallest in summer. The annual bias pattern (positive over most parts of the Arctic) can be partly attributed to coarse orography and to biased atmospheric storm tracks and sea ice cover (see Chapter8 ). The MMD models capture the observed increase in the annual precipitation through the 20th century (see Section 3.3 ).

Sea Ice and Ocean

Arctic sea ice biases in present-day MMD simulations are discussed in Section 8.3 .Arctic ocean-sea ice RCMs under realistic atmospheric forcing are increasingly capable of reproducing the known features of the Arctic Ocean circulation and observed sea ice drift patterns. The inflow of the two branches of Atlantic origin via the Fram Strait and the Barents Sea and their subsequent passage at mid-depths in several cyclonic circulation cells are present in most recent simulations ( Karcher et al., 2003 [JoC, SRC] ; Maslowski et al., 2004 [JoC, MoS] ; Steiner et al., 2004 [JoC, MoS] ). Most of the models are biased towards overly salty values in the Beaufort Gyre and thus too little freshwater storage in the arctic halocline. Several potential causes have been identified, among them a biased simulation of arctic shelf processes and wind forcing. Most hindcast simulations with these RCMs show a reduction in the arctic ice volume over recent decades ( Holloway and Sou, 2002 [JoC] ).

11.8.1.3 Climate Projections

Temperature

Anorthern high-latitude maximum in the warming (‘polar amplification’) is consistently found in all AOGCM simulations (see Section 10.3 ). The simulated annual mean arctic warming exceeds the global mean warming by roughly a factor of two in the MMD models, while the winter warming in the central arctic is a factor of four larger than the global annual mean when averaged over the models. These magnitudes are comparable to those obtained in previous studies ( Holland and Bitz, 2003 [JoC, MoS, ARC] ; ACIA, 2005 [NPR] ). The consistency between observations and the ensemble mean 20th-century simulations( Figure 11.18 ), combined with the fact that the near-future projections ( 2010 – 2029 ) continue the late 20th-century trends in temperature, ice extent and thickness with little modification ( Serreze and Francis, 2006 [JoC] ), increases confidence in this basic polar-amplified warming pattern, despite the inter-model differences in the amount of polar amplification.

Figure 11.18. Top panels: Temperature anomalies with respect to 1901 to 1950 for the whole Arctic for 1906 to 2005 (black line) as simulated (red envelope) by MMD models incorporating known forcings; and as projected for 2001 to 2100 by MMD models for the A1B scenario (orange envelope). The bars at the end of the orange envelope represent the range of projected changes for 2091 to 2100 for the B1 scenario (blue), the A1B scenario (orange) and the A2 scenario (red). The black line is dashed where observations are present for less than 50% of the area in the decade concerned. Bottom panels: The same for Antarctic land, but with observations for 1936 to 2005 and anomalies calculated with respect to 1951 to 2000 . More details on the construction of these figures are given in Box 11.1 and Section 11.1.2 .

At the end of the 21st century, the projected annual warming in the Arctic is 5°C, estimated by the MMD-A1B ensemble mean projection( Section 11.8.2.3 , Figure 11.21 ). There is a considerable across-model range of 2.8°C to 7.8°C( Table 11.1 ). Larger (smaller) mean warming is found for the A2 (B1) scenario of 5.9°C (3.4°C), with a proportional across-model range. The across-model and across-scenario variability in the projected temperatures are both considerable and of comparable amplitude ( Chapman and Walsh, 2007 [JoC, MoS, ARC, 2007] ).

Over both ocean and land, the largest (smallest) warming is projected in winter (summer)( Table 11.1 , Figure 11.19 ). But the seasonal amplitude of the temperature change is much larger over ocean than over land due the presence of melting sea ice in summer keeping the temperatures close to the freezing point. The surface air temperature over the Arctic Ocean region is generally warmed more than over arctic land areas (except in summer). The range between the individual simulated changes remains large( Figure 11.19 , Table 11.1 ). By the end of the century, the mean warming ranges from 4.3°C to 11.4°C in winter, and from 1.2°C to 5.3°C in summer under the A1B scenario. The corresponding 5 to 95% confidence intervals are given in Supplementary Material Table S11.2. In addition to the overall differences in global warming, difficulties in simulating sea ice, partly related to biases in the surface wind fields, as well as deficiencies in cloud schemes, are likely responsible for much of the inter-model scatter. Internal variability plays a secondary role when examining these late-21st century responses.

Figure 11.19. Annual cycle of arctic area mean temperature and percentage precipitation changes (averaged over the area north of 60°N) for 2080 to 2099 minus 1980 to 1999, under the A1B scenario. Thick lines represent the ensemble median of the 21 MMD models. The dark grey area represents the 25 and 75% quartile values among the 21 models, while the light grey area shows the total range of the models.

The annual mean temperature response pattern at the end of the 21st century under the A1B scenario (Supplementary Material Figures S11.27 and S11.11) is characterised by a robust and large warming over the central Arctic Ocean (5°C to 7°C), dominated by the warming in winter/autumn associated with the reduced sea ice. The maximum warming is projected over the Barents Sea, although this could result from an overestimated albedo feedback caused by removal of the present-day simulations’ excessive sea ice cover. A region of reduced warming (<2°C, even slight cooling in several models) is projected over the northern North Atlantic, which is consistent among the models. This is due to weakening of the MOC (see Section 10.3 ).

While the natural variability in arctic temperatures is large compared to other regions, the signals are still large enough to emerge quickly from the noise( Table 11.1 ). Looking more locally, as described by Chapman and Walsh 2007 [JoC, MoS, ARC, 2007] ), Alaska is perhaps the land region with the smallest signal-to-noise ratio, and is the only arctic region in which the 20-year mean 2010 to 2029 temperature is not clearly discernible from the 1981 to 2000 mean in the MMD models. But even here the signal is clear by mid-century in all three scenarios.

The regional temperature responses are modified by changes in circulation patterns( Chapter 10 ). In winter, shifts in NAO phase can induce inter-decadal temperature variations of up to 5°C in the eastern Arctic ( Dorn et al., 2003 [JoC, MoS, SRC] ). The MMD models project winter circulation changes consistent with an increasingly positive NAM/NAO (see Section 10.3 ), which acts to enhance the warming in Eurasia and western North America. In summer, circulation changes are projected to favour warm anomalies north of Scandinavia and extending into the eastern Arctic, with cold anomalies over much of Alaska ( Cassano et al., 2006 [JoC, SRC] ). However, deficiencies in the arctic summer synoptic activity in these models reduce confidence in the detailed spatial structure. In addition, these circulation-induced temperature changes are not large enough to change the relatively uniform summer warming seen in the MMD models

The patterns of temperature changes simulated by RCMs are quite similar to those simulated by GCMs. However, they show an increased warming along the sea ice margin possibly due to a better description of the mesoscale weather systems and air-sea fluxes associated with the ice edge ( ACIA, 2005 [NPR] ). The warming over most of the central Arctic and Siberia, particularly in summer, tends to be lower in RCM simulations (by up to 2°C) probably due to more realistic present-day snowpack simulations ( ACIA, 2005 [NPR] ). The warming is modulated by the topographical height, snow cover and associated albedo feedback as shown for the region of northern Canada and Alaska (see Section 11.5.3 ).

Precipitation

The MMD simulations show a general increase in precipitation over the Arctic at the end of the 21st century( Table 11.1 ;Supplementary Material Figure S11.28). The precipitation increase is robust among the models( Table 11.1 ;Supplementary Material Figure S11.19) and qualitatively well understood, attributed to the projected warming and related increased moisture convergence( Section 10.3 ). The very strong correlation between the temperature and precipitation changes (approximately a 5% precipitation increase per degree celsius warming) across the model ensemble is worth noting( Figure 11.20 ). Thus, both the sign and the magnitude (per degree warming) of the percentage precipitation change are robust among the models.

Figure 11.20. Relationship between the change in annual precipitation (%) and temperature (°C) ( 2080 – 2099 minus 1980 – 1999 ) in the Arctic (averaged over the area north of 60°N) in the MMD-A1B projections. Each point represents one model. The model ensemble mean response is indicated by the circle.

The spatial pattern of the projected change (Supplementary Material Figure S11.28) shows the greatest percentage increase over the Arctic Ocean (30 to 40%) and smallest (and even slight decrease) over the northern North Atlantic (<5%). By the end of the 21st century, the projected change in the annual mean arctic precipitation varies from 10 to 28%, with an MMD-A1B ensemble median of 18%( Table 11.1 ). Larger (smaller) mean precipitation increases are found for the A2 (B1) scenario with 22% (13%). The percentage precipitation increase is largest in winter and smallest in summer, consistent with the projected warming( Figure 11.19 ; Table 11.1 ). The across-model scatter of the precipitation projections is substantial( Figure 11.19 ; Table 11.1 ). The Tebaldi et al. 2004a [JoC, MoS, SRC] ) 5th to 95th percentile confidence interval of percentage precipitation change in winter is 13 to 36% and in summer 5 to 19% (Supplementary Material Table S11.2).

Differences between the projections for different scenarios are small in the first half of the 21st century but increase later. Differences among the models increase rapidly as the spatial domain becomes smaller ( ACIA, 2005 [NPR] ). The geographical variation of precipitation changes is determined largely by changes in the synoptic circulation patterns. During winter, the MMD models project a decreased (increased) frequency of strong Arctic high (Icelandic low) pressure patterns that favour precipitation increases along the Canadian west coast, southeast Alaska and North Atlantic extending into Scandinavia ( Cassano et al., 2006 [JoC, SRC] ). Projections with RCMs support the broad-scale messages while adding expected local and regional detail ( ACIA, 2005 [NPR] ).

By the end of the 21st century, the MMD-A1B ensemble-projected precipitation increase is significant( Table 11.1 ), particularly the annual and cold season (winter/autumn) precipitation. However, local precipitation changes in some regions and seasons (particularly in the Atlantic sector and generally in summer) remain difficult to discern from natural variability ( ACIA, 2005 [NPR] ).

Extremes of Temperature and Precipitation

Very little work has been done in analysing future changes in extreme events in the Arctic. However, the MMD simulations indicate that the increase in mean temperature and precipitation will be combined with an increase in the frequency of very warm and wet winters and summers. Using the definition of extreme season in Section 11.1.2 ,every DJF and JJA season, in all model projections, is ‘extremely’ warm in the period 2080 to 2099 ( Table 11.1 ). The corresponding numbers for extremely wet seasons are 90 and 85% for DJF and JJA. For the other scenarios, the frequency of extremes is very similar, except that for the wet seasons under B1, which is smaller (~63%).

Northern Hemisphere sea ice, snow and permafrost projections are discussed in Section 10.3 ;projected changes in the surface mass balance of arctic glaciers and of the Greenland Ice Sheet are discussed in Sections 10.3 and 10.7 .

Cryosphere

Arctic Ocean

Asystematic analysis of future projections for the Arctic Ocean circulation is still lacking. Coarse resolution in global models prevents the proper representation of local processes that are of global importance (such as the convection in the Greenland Sea that affects the deep waters in the Arctic Ocean and the intermediate waters that form overflow waters). The MMD models project a reduction in the MOC in the Atlantic Ocean (see Section 10.3 ). Correspondingly, the northward oceanic heat transport decreases south of 60°N in the Atlantic. However, the CMIP2 model assessment showed a projected increase in the oceanic heat transport at higher latitudes, associated with a stronger subarctic gyre circulation in the models ( Holland and Bitz, 2003 [JoC, MoS, ARC] ). The Atlantic Ocean north of 60°N freshens during the 21st century, in pronounced contrast to the observed development in the late 20th century ( Wu et al., 2003 [NPR, JoC, ARC] ).

Compared to the SAM, the Southern Oscillation (SO) shows weaker association with surface temperature over Antarctica but the correlation with SST and sea ice variability in the Pacific sector of the Southern Ocean is significant (e.g., Kwok and Comiso, 2002b [JoC, SRC] ;( Renwick, 2002; ) Bertler et al., 2004 [JoC, ARC] ;( Yuan, 2004 ) ). Correlation between the SO index and antarctic precipitation and accumulation has also been studied but the persistence of the signal is not clear ( Bromwich et al., 2000 [JoC, SRC] , 2004a; Genthon and Cosme, 2003 [JoC] ; Guo et al., 2004 [JoC, MoS, SRC] ; Genthon et al., 2005 [JoC] ). Recent work suggests that this intermittence is due to nonlinear interactions between ENSO and SAM that vary on decadal time scales ( Fogt and Bromwich, 2006 [JoC, SRC] ; L’Heureux and Thompson, 2006 [JoC] ). The SO index has a negative trend over recent decades (corresponding to a tendency towards more El Niño-like conditions in the equatorial Pacific; see Section 3.6 )associated with sea ice cover anomalies in the Pacific sector, namely negative (positive) anomalies in the Ross and Amundsen Seas (Bellingshausen and Weddell Seas) ( Kwok and Comiso, 2002a [JoC, SRC] ). However, a definitive assessment of ENSO amplitude and frequency changes in the 21st century cannot be made (see Chapter 10 ).

Temperature

Compared to National Centers for Environmental Prediction (NCEP) reanalyses, the MMD ensemble annual surface temperatures are in general slightly warmer in the Southern Ocean to the north of the sea ice region. The mean bias is predominantly less than 2°C ( Carril et al., 2005 [JoC, MoS] ), which may indicate a slight improvement compared to previous models due to better simulation of the position and depth of the Antarctic trough ( Carril et al., 2005 [JoC, MoS] ; Raphael and Holland, 2006 [JoC, MoS, ARC] ). The temperature bias over sea ice is larger. Biases over the continent are several degrees where the model topography is erroneous ( Turner et al., 2006 [JoC, MoS, ARC] ). However, as emphasized above, the biases have to be viewed in the context of the uncertainty in the observations. Changes in cloud and radiation parametrizations have been shown to change the temperature simulation significantly ( Hines et al., 2004 [JoC, MoS, SRC] ). A lateral nudging of a stretched-grid GCM (imposing the correct synoptic cyclones from 60°S and lower latitudes) brings the model in better agreement with observations but significant biases remain ( Genthon et al., 2002 [JoC, MoS] ).

The spread in the individual MMD-simulated patterns of surface temperature trends over the past 50 years is very large, but in contrast to previous models, the multi-model composite of the MMD models qualitatively captures the observed enhanced warming trend over the Antarctic Peninsula ( Chapman and Walsh, 2006 [JoC, ARC] ). The general improvements in resolution, sea ice models and cloud-radiation packages have evidently contributed to improved simulations. The ensemble-mean temperature trends show similarity to the observed spatial pattern of the warming, for both annual and seasonal trends. For the annual trend, this includes the warming of the peninsula and near-coastal Antarctica and neutral or slight cooling over the sea-ice covered regions of the Southern Ocean. While the large spread among the models is not encouraging, this level of agreement suggests that some confidence in the ensemble mean 21st-century projection is appropriate.

11.8.2.1 Key Processes

Over Antarctica, there is special interest in changes in snow accumulation expected to accompany global climate change as well as the pattern of temperature change, particularly any differences in warming over the peninsula and the interior of the ice sheet. As in the Arctic, warming of the troposphere is expected to increase precipitation. However, circulation changes in both ocean and atmosphere can alter the pattern of air masses, which would modify both precipitation and temperature patterns substantially over the region.

The dominant patterns controlling the atmospheric seasonal-to-interannual variability of the SH extratropics are the Southern Annular Mode (SAM) and ENSO (see Section 3.6 ). Signatures of these patterns in the Antarctic have been revealed in many studies (reviews by Carleton, 2003 [JoC] and Turner, 2004 [JoC] ). The positive phase of the SAM is associated with cold anomalies over most of Antarctica and warm anomalies over the Antarctic Peninsula ( Kwok and Comiso, 2002a [JoC, SRC] ). Over recent decades, a drift towards the positive phase in the SAM is evident (see Section 3.6 ). Observational studies have presented evidence of pronounced warming over the Antarctic Peninsula, but little change over the rest of the continent during the last half of the 20th century (see Sections 3.6 and 4.6 ). The response of the SAM in transient warming simulations is a robust positive trend, but the response to the ozone hole in the late 20th century, which is also a positive perturbation to the SAM, makes any simple extrapolation of current trends into the future uncertain (see Section 10.3 ).

11.8.2.2 Skill of Models in Simulating Present Climate

Evaluating temperature and precipitation simulations over Antarctica is difficult due to sparse observations and often relies on numerical weather prediction (re)analyses. However, significant differences between those have been found, and comparisons with station observations show that the surface temperature can be subject to considerable biases ( Connolley and Harangozo, 2001 [JoC, MoS] ; Bromwich and Fogt, 2004 [JoC, SRC] ). Marked improvement in the bias is seen after the satellite era (~ 1978 ( Simmons et al., 2004 [JoC, ARC] ), and parts of the bias are explained by the reanalyses’ smoothing of the sharp changes in the terrain near coastal stations. Satellite-derived monthly surface temperatures agree with antarctic station data with an accuracy of 3°C ( Comiso, 2000 [JoC, SRC] ). Precipitation evaluation is even more challenging and the different (re)analyses differ significantly ( Connolley and Harangozo, 2001 [JoC, MoS] ; Zou et al., 2004 [JoC, MoS] ). Very few direct precipitation gauge and detailed snow accumulation data are available, and these are uncertain to varying degrees (see Section 4.6 ).

Major challenges face the simulation of the atmospheric conditions and precipitation patterns of the polar desert in the high interior of East Antarctica ( Guo et al., 2003 [JoC, MoS, SRC] ; Bromwich et al., 2004a [JoC, MoS, SRC] ; Pavolonis et al., 2004 [JoC, MoS] , van de Berg et al., 2005 [MoS] ). Driven by analysed boundary conditions, RCMs tend to show smaller temperature and precipitation biases in the Antarctic compared to the GCMs ( Bailey and Lynch, 2000 [JoC, MoS, SRC] ; van Lipzig et al., 2002a [JoC, MoS, ARC] ,b; van den Broeke and van Lipzig, 2003 [JoC] ; Bromwich et al., 2004b [JoC, MoS, SRC] ; Monaghan et al., 2006 [JoC, MoS, SRC] Krinner et al. 1997 [JoC, MoS] ) show the value of a stretched model grid with higher horizontal resolution over the Antarctic as compared to standard GCM formulations. Despite these promising developments, since the TAR there has been no coordinated comparison of the performance of GCMs, RCMs and other alternatives to global GCMs over Antarctica.

Precipitation

The MMD models simulate the position of the storm tracks reasonably well but nearly all show some deficiency in the distribution and level of cyclone activity compared to reanalyses (see Section 8.3 ). Regional Climate Models generally capture the cyclonic events affecting the coast and the associated synoptic variability of precipitation with more fidelity ( Adams, 2004 [MoS] ; Bromwich et al., 2004a [JoC, MoS, SRC] ). Over the 20th century, the MMD models simulate changes in storm track position that are generally consistent with observed changes (i.e., poleward displacement of the storm tracks; see Sections 9.5 and 10.3 ).

The precipitation simulations by both GCMs and RCMs contain uncertainty, on all time scales ( Covey et al., 2003 [JoC, MoS, ARC] ; Bromwich et al., 2004a [JoC, MoS, SRC] ,b; van de Berg et al., 2005 [MoS] ), as a result of model physics limitations. All atmospheric models, including the models underlying the reanalyses, have incomplete parametrizations of polar cloud microphysics and ice crystal precipitation. The simulated precipitation depends, among other things, on the simulated sea ice concentrations, and is strongly affected by biases in the sea ice simulations ( Weatherly, 2004 [MoS] ). Recent RCM simulations driven by observed sea ice conditions demonstrate good precipitation skill ( Van de Berg et al., 2005 [MoS] ; Monaghan et al., 2006 [JoC, MoS, SRC] ). However, as emphasized above, the observational uncertainty contributes to uncertainty in the differences between observations and simulations.

Sea Ice

The performance biases of SH sea ice conditions in present-day MMD simulations are discussed in Section 8.3 .

Very little effort has been spent to model the future climate of Antarctica at a spatial scale finer than that of GCMs.

11.8.2.3 Climate Projections

Temperature

At the end of the 21st century, the annual warming over the Antarctic continent is moderate but significant( Figure 11.21 ; Table 11.1 ;Chapman and Walsh, 2006 [JoC, ARC] ). It is estimated to be 2.6°C by the median of the MMD-A1B models with a range from 1.4°C to 5.0°C across the models( Table 11.1 ). Larger (smaller) warming is found for the A2 (B1) scenario with mean value of 3.1°C (1.8°C). These warming magnitudes are similar to previous estimates ( Covey et al., 2003 [JoC, MoS, ARC] ). The annual mean MMD model projections show a relatively uniform warming over the entire continent (with a maximum in the Weddell Sea)( Figure 11.21 ;Carril et al., 2005 [JoC, MoS] ; Chapman and Walsh, 2006 [JoC, ARC] ). They do not show a local maximum warming over the Antarctic Peninsula. This is a robust feature among the individual models (Supplementary Material Figure S11.12). Thus, the pattern of observed temperature trends in the last half of the 20th century (warming over the Antarctic Peninsula, little change over the rest of the continent) is not projected to continue throughout the 21st century, despite a projected positive SAM trend (see Section 10.3 ). It has been argued that two distinct factors have contributed to the observed SAM trend: greenhouse gas forcing and the ozone hole formation ( Stone et al., 2001 [MoS, SRC] ; Shindell and Schmidt, 2004 [PoC, JoC, ARC] ). Their relative importance for the peninsular warming is not readily understood (see Chapter 10 ).

Figure 11.21. Annual surface temperature change between 1980 to 1999 and 2080 to 2099 in the Arctic and Antarctic from the MMD-A1B projections.

The mean antarctic temperature change does not show a strong seasonal dependency; the MMD-A1B ensemble mean winter (summer) warming is 2.8°C (2.6°C)( Table 11.1 ;Supplementary Material Figure S11.29; Chapman and Walsh, 2006 [JoC, ARC] ). This is also illustrated by how close the Tebaldi et al. 2004a [JoC, MoS, SRC] ) 5 to 95% confidence interval for the two seasons is: 0.1°C to 5.7°C in summer and 1.0°C to 4.8°C in winter (Supplementary Material Table S11.2). However, over the Southern Oceans, the temperature change is larger in winter/autumn than in summer/spring, which can primarily be attributed to the sea ice retreat (see Section 10.3 ).

Precipitation

Almost all MMD models simulate a robust precipitation increase in the 21st century (Supplementary Material Figures S11.29 and S11.30; Table 11.1 ). However, the scatter among the individual models is considerable. By the end of the 21st century, the projected change in the annual precipitation over the Antarctic continent varies from –2% to 35%, with a MMD-A1B ensemble median of 14%( Table 11.1 ). Similar (smaller) mean precipitation increase is found for the A2 (B1) scenario, with values of 15% (10%). The spatial pattern of the annual change is rather uniform (Supplementary Material Figure S11.30). The projected relative precipitation change shows a seasonal dependency, and is larger in winter than in summer (Supplementary Material Figure S11.29). The Tebaldi et al. 2004a [JoC, MoS, SRC] ) 5 to 95% confidence interval for winter is –1 to 34% and in summer –6 to 22% (Supplementary Material Table S11.2). The projected increase in precipitation over Antarctica and thus greater accumulation of snow, without substantial surface melting, will contribute negatively to sea level rise relative to the present day (see Section 10.6 ). It is notable that the most recent model studies of antarctic precipitation show no significant contemporary trends ( Van de Berg et al., 2005 [MoS] ; Monaghan et al., 2006 [JoC, MoS, SRC] ; van den Broeke et al., 2006 [JoC] ; see Section 4.6 ).

The moisture transport to the continent by synoptic activity represents a large fraction of net precipitation ( Noone and Simmonds, 2002 [JoC, ARC] ; Massom et al., 2004 [JoC] ). During summer and winter, a systematic shift towards strong cyclonic events is projected by the MMD models (see Section 10.3 ). In particular, the frequency of occurrence of deep cyclones in the Ross Sea to Bellingshausen Sea sector is projected to increase by 20 to 40% (63%) in summer (winter) by the middle of the 21st century ( Lynch et al., 2006 [JoC, SRC] ). Related to this, the precipitation over the sub-antarctic seas and Antarctic Peninsula is projected to increase.

Extremes of Temperature and Precipitation

Very little work has been done in analysing future changes in extreme events in the Antarctic. However, the MMD simulations indicate that the increase in mean temperature and precipitation will be combined with an increase in the frequency of very warm and wet winters and summers. Using the definition of ‘extreme’ seasons provided in Section 11.1.2 ,the MMD models predict extremely warm seasons in about 85% of all DJF and 83% of all JJA seasons in the period 2080 to 2099, as averaged over all models( Table 11.1 ). The corresponding numbers for extremely wet seasons are 34% and 59%. For the B1 scenario, the frequency of extremes is smaller, with little difference between A1B and A2.

Sea Ice and Antarctic Ice Sheet

Southern Hemisphere sea ice projections are discussed in Section 10.3 .The projections of the Antarctic Ice Sheet surface mass balance are discussed in Section 10.6 .

11.9.1.1 Caribbean

The Caribbean region spans roughly the area between 10°N to 25°N and 85°W to 60°W. Its climate can be broadly characterised as dry winter/wet summer with orography and elevation being significant modifiers at the sub-regional scale ( Taylor and Alfero, 2005 [NPR] ). The dominant synoptic influence is the North Atlantic subtropical high (NAH). During the winter, the NAH is southernmost and the region is generally at its driest. With the onset of the spring, the NAH moves northward, the trade wind intensity decreases and the equatorial flank of the NAH becomes convergent. Concurrently, easterly waves traverse the Atlantic from the coast of Africa into the Caribbean. These waves frequently mature into storms and hurricanes under warm SSTs and low vertical wind shear, generally within a 10°N to 20°N latitudinal band. They represent the primary rainfall source and their onset in June and demise in November roughly coincide with the mean Caribbean rainy season. In the coastal zones of Venezuela and Columbia, the wet season occurs later, from October to January ( Martis et al., 2002 [JoC, MoS] ). Interannual variability of the rainfall is influenced mainly by ENSO events through their effect on SSTs in the Atlantic and Caribbean Basins. The late rainfall season tends to be drier in El Niño years and wetter in la Niña years ( Giannini et al., 2000 [JoC] , Martis et al., 2002 [JoC, MoS] , M. ( Taylor et al., 2002 ) ) and tropical cyclone activity diminishes over the Caribbean during El Niño summers ( Gray, 1984 [JoC] ). However, the early rainfall season in the central and southern Caribbean tends to be wetter in the year after an El Niño and drier in a la Niña year ( Chen and Taylor, 2002 [JoC, SRC] ). The phase of the NAO modulates the behaviour of warm ENSO events ( Giannini et al., 2001 [JoC] ). A positive NAO phase implies a stronger than normal NAH and amplifies the drying during a warm ENSO. On the other hand, a negative NAO phase amplifies the precipitation in the early rainfall season in the year after an El Niño.

11.9.1.2 Indian Ocean

The Indian Ocean region refers to the area between 35°S to 17.5°N and 50°E to 100°E. The climate of the region is influenced primarily by the Asian monsoons (see Section 11.4.1 for processes influencing monsoons). During January, the ITCZ is located primarily in the SH. The region north of the ITCZ then experiences north-easterly trade winds (northeast monsoons) and the region to the south, the south-easterly trades. During northern summer, the ITCZ is located in the north and virtually covers the entire Bay of Bengal, the surrounding lands and the eastern Arabian Sea. The winds in the north turn into strong south-westerlies (southwest monsoons), while the south-easterlies persist in the south. Precipitation and wind stress bring about a response that is distinctly different in the northern and southern parts of the Indian Ocean (International CLIVAR Project Office, 2006 ). The wet (dry) season in the Maldives occurs during the southwest (northeast) monsoons. From May to October, the southeast trades dominate in the Seychelles and the climate is relatively cool and dry, and December to March is the principal wet season with winds mainly from west to northwest.

While the monsoons recur each year, their irregularity at a range of time scales from weeks to years depends on feedback from the ocean in ways that are not fully understood. Intra-seasonal variability is associated with the Monsoon Intra-Seasonal Oscillation (MISO) and the Madden-Julian Oscillation (MJO), which are long-lasting weather patterns that evolve in a systematic way for periods of four to eight weeks. On an interannual and decadal scale, statistical methods have shown that while there are periods of high correlation between ENSO and monsoon variation, there are decades where there appears to be little or no association (International CLIVAR Project Office, 2006; see also Section 10.3.5.4 ). A modulating factor is the Indian Ocean Dipole or Indian Ocean Zonal Mode (IOZM), a large interannual variation in zonal SST gradient (see Section 3.6 ). The magnitude of the secondary rainfall maximum from October to December in East Africa is strongly correlated with IOZM events, and the positive phase of IOZM, with higher SSTs in the west, counters the drying effect that ENSO has on monsoon rainfall ( Ashok et al., 2001 [JoC] ).

11.9.1.3 Pacific

The Pacific region refers to equatorial, tropical and subtropical region of the Pacific in which there is a high density of inhabited small islands. Broadly, it is the region between 20°N and 30°S and 120°E to 120°W. The major climatic processes that play a key role in the climate of this region are the easterly trade winds (both north and south of the equator), the SH high pressure belt, the ITCZ and the South Pacific Convergence Zone (SPCZ; see Vincent, 1994 [JoC] ), which extends from the ICTZ near the equator due north of New Zealand south-eastward to at least 21°S, 130°W. The region has a warm, highly maritime climate and rainfall is abundant. The highest rainfall follows the seasonal migration of the ITCZ and SPCZ. Year-to-year climatic variability in the region is very strongly affected by ENSO events. During El Niño conditions, rainfall increases in the zone northeast of the SPCZ ( Vincent, 1994 [JoC] ). Tropical cyclones are also a feature of the climate of the region, except within 10 degrees of the equator, and are associated with extreme rainfall, strong winds and storm surge. Many islands in the region are very low lying, but there are also many with strong topographical variations. In the case of the latter, orographic effects on rainfall amount and seasonal distribution can be strong.

The ability of the MMD models to simulate present climate in the Caribbean, Indian Ocean and North and South Pacific Ocean is summarised in Supplementary Material Table S11.1. In general, the biases in about half of the temperature simulations are less than 1°C in all seasons, so that the model performances are, on the whole, satisfactory. There are, however, large spreads in precipitation simulations. During the last decade, steady progress has been made in simulating and predicting ENSO using coupled GCMs. However, serious systematic errors in both the simulated mean climate and the natural variability persist (see Section 8.4.7 )

11.9.2.1 Caribbean

Simulations of the annual Caribbean temperature in the 20th century ( 1980 – 1999 ) by the MMD models give an average that agrees closely with climatology, differing by less than 0.1°C. The inter-quartile range difference between individual models and climatology ranged from –0.3°C to +0.3°C. Thus, the models have good skill in simulating annual temperature. The average of the MMD simulations of precipitation, however, underestimates the observed precipitation by approximately 30%. The deviations in individual models range from –64 to +20%, a much greater range than the deviations in temperature simulations. Recently the Parallel Climate Model (at T42 resolution – about 3.75 degrees), a fully coupled global climate model, was found to be capable of simulating the main climate features over the Caribbean region ( Angeles et al., 2007 [JoC, MoS, 2007] ), but it also underestimated the area average precipitation across the Caribbean. Martinez-Castro et al. 2006 [MoS] ), in a sensitivity experiment, conclude that the Regional Climate Model (RegCM3), using the Anthes-Kuo cumulus parametrization scheme, can be used for long-term area-averaged climatology.

For annual temperature in the Indian Ocean in the 20th century ( 1980 – 1999 ), the mean value of the MMD outputs overestimated the climatology by 0.6°C, with 50% of deviations ranging from 0.2°C to 1.0°C. For rainfall, the multi-model ensemble average was only slightly below the mean precipitation by 3%, and the model deviations ranged from –22 to +20%. There are, however, problems with the simulation of year-to-year variation. Many of the important climatic effects of the MJO, including its impacts on rainfall variability in the monsoons, are still poorly simulated by contemporary climate models (see Section 8.4 ).

11.9.2.2 Indian Ocean

11.9.2.3 Pacific

Climate model simulations of current-climate means of temperature and precipitation were investigated by Jones et al. 2000 [Ambiguous] , 2002 and Lal et al. 2002 [MoS] ) for the South Pacific. The AOGCMs available at the time of these studies simulated well the broad-scale patterns of temperature and precipitation across the region, with the precipitation patterns more variable than for temperature in the models considered, and with some significantly underestimating or overestimating of the intensity of rainfall in the high-rainfall zones. All models simulated a broad rainfall maximum stretching across the SPCZ and ITCZ, but not all models resolved a rainfall minimum between these two regions. A problem of simulating the spatial structure of the MJO resulting in tendencies for the convective anomaly to split into double ITCZs in the Pacific is also discussed in Section 8.4.8 .

Analysis of the MMD simulations shows that the average model value overestimated the annual mean temperature from 1980 to 1999 by 0.9°C over a southern Pacific region, with 50% of the deviations varying from 0.6°C to 1.2°C. Over the North Pacific, the simulated ensemble average temperature for the same period was only 0.7°C above the climatology, with half of the model deviations from the climatology ranging from 0.2°C to 1.0°C. Average precipitation was overestimated by 10%, but individual model values varied from –7 to 31% in the southern Pacific region, whereas in the northern Pacific the mean model output for precipitation almost agreed with the climatology, while the individual models deviated from –13 to 13%. Thus, the models are better at simulating rainfall in the northern Pacific than in the southern Pacific and the quality of the simulations, both north and south, were not much different from those for the Indian Ocean.

11.9.3.1 Caribbean

The MMD-simulated annual temperature increases at the end of the 21st century range from 1.4°C to 3.2°C with a median of 2.0°C, somewhat below the global average. Fifty percent of the models give values differing from the median by only ±0.4°C. Statistical downscaling of HadCM3 results using the A2 and B2 emission scenarios gives around a 2°C rise in temperature by the 2080 s, approximately the same as the HadCM3 model. The agreement between the AOGCMs and the downscaling analysis gives a high level of confidence in the temperature simulations. The downscaling was performed with the use of the Statistical DownScaling Model (SDSM) developed by Wilby et al. 2002 [SRC] ) as part of an Assessments of Impacts and Adaptations to Climate Change in Multiple Regions and Sectors (AIACC) Small Island States SIS06 project (http://www.aiaccproject.org). Angeles et al. 2007 [JoC, MoS, 2007] ) also simulate an approximately 1°C rise in SST up to the 2050 s using the IS92a scenario. There were no noticeable differences in monthly changes (see Supplementary Material Figure S11.31). Observations suggest that warming is ongoing ( Peterson et al., 2002 [JoC, ARC] ).

According to Table 11.1 ,most models project decreases in annual precipitation and a few increases, varying from –39 to +11%, with a median of –12%. Figure 11.23 shows that the annual mean decrease is spread across the entire region. In DJF, some areas of increases are noted and in JJA, the region-wide decrease is enhanced, especially in the region of the Greater Antilles, where the model consensus is also strong. Monthly changes in the Caribbean are shown in Supplementary Material Figure S11.32, which also shows that the simulations for the Caribbean have a greater spread compared to the other oceanic regions (IND, NPA and SPA in S11.32). Results from HadCM3 downscaled for the A2 and B2 emission scenarios using the SDSM also show a near-linear decrease in summer precipitation to the 2080 s for a station in Jamaica. Downscaled results from the SDSM for stations in Barbados and Trinidad, however, show increases rather than decreases. Thus, there is consensus between the MMD results and the downscaled results for the Greater Antilles in JJA but not for the other islands, and also not on an annual basis. Angeles et al. 2007 [JoC, MoS, 2007] ) also simulate decreases up to the middle of the century in the vicinity of the Greater Antilles but not in the other islands in the late rainfall season. Table 11.1 shows that the decrease in JJA has the largest signal-to-noise ratio. The decrease is in agreement with the expected drying in the subtropics discussed in Sections 9.5 and 11.1 .In the multi-model analysis, most models show shift to a more positive phase of the NAO (see Section 10.3 ), and consensus on temperature changes in the Pacific indicates an El Niño-like pattern with higher temperatures in the eastern Pacific (see Section 10.3 ). These conditions are associated with drying in the Caribbean. Observed trends in precipitation are unclear. While Peterson et al. 2002 [JoC, ARC] ) find no statistically significant trends in mean precipitation amounts from the 1950 s to 2000, Neelin et al. 2006 [JoC, MoS] ) note a modest but statistically significant summer drying trend over recent decades in the Caribbean in several observational data sets.

Figure 11.23. Precipitation changes over the Caribbean from the MMD-A1B simulations. Top row: Annual mean, DJF and JJA fractional precipitation change between 1980 to 1999 and 2080 to 2099, averaged over 21 models. Bottom row: number of models out of 21 that project increases in precipitation.

11.9.3.2 Indian Ocean

Based on the MMD ensemble mean the annual temperature is projected to increase by about 2.1°C, somewhat below the global average, with individual models ranging from 1.4°C to 3.7°C and at least half of the models giving values quite close to the mean. All models show temperature increases in all months with no significant seasonal variation (Supplementary Material Figure S11.31). Evidence of temperature increases from 1961 to 1990 in the Seychelles is provided by Easterling et al. 2003 [JoC, ARC] ), who find that the percentage of time when the minimum temperature is below the 10th percentile is decreasing, and the percentage of time where the minimum temperature exceeds the 90th percentile is increasing. Similar results were obtained for the maximum temperatures. This is consistent with general patterns of warming elsewhere (see Chapter3 ).

The annual precipitation changes projected by individual MMD models varied from –2 to 20% with a median of 4% and 50% of the models projecting changes between 3 and 5%. Thus, there is some level of confidence in the precipitation results although not as high as for temperature. Figure 11.24 shows that the annual increase is restricted mainly to the north Indian Ocean, where the model consensus is greatest, especially in the vicinity of the Maldives. In DJF, some increases are noted in the south. Model agreement on increases is greatest for the Seychelles in DJF and for the Maldives in JJA. There is also strong agreement on decreases in the vicinity of Mauritius in JJA. Sections 10.3.5 and 11.4 discuss changes in monsoon behaviour in a warmer climate. There is an emerging consensus that the effect of enhanced moisture convergence in a warmer atmosphere will dominate over possible weaker monsoonal flows and tropical large-scale circulation in global warming simulations, resulting in increased monsoonal precipitation. Easterling et al. 2003 [JoC, ARC] ) find evidence that extreme rainfall tended to increase from 1961 to 1990 (see also Section 11.4.3 ,South Asia projections).

Figure 11.24. As for Figure 11.23 but for the Indian Ocean.

11.9.3.3 Pacific

Projected regional temperature changes in the South Pacific from a range of AOGCMs have been prepared by Lal et al. 2002 [MoS] ( Lal 2004 ) and Ruosteenoja et al. 2003 [NPR, MoS] Jones et al. 2000 [Ambiguous] , 2002 and Whetton and Suppiah 2003 [NPR, MoS, SRC] ) also consider patterns of change. Broadly, simulated warming in the South Pacific closely follows the global average warming rate. However, there is a tendency in many models for the warming to be a little stronger in the central equatorial Pacific (North Polynesia) and a little weaker to the south (South Polynesia).

The MMD-A1B projections for the period 2080 to 2099 show an increase in annual temperature of 1.8°C, somewhat below the global average over the South Pacific( Table 11.1 ). The individual model values vary from 1.4°C to 3.1°C and at least half of the models project values very close to the mean. All models show increases, slightly less in the second half of the year compared to the first (Supplementary Material Figure S11.31). Over the North Pacific, the models simulate an increase in temperature of 2.3°C, slightly below the global average, with values ranging from 1.5°C to 3.7°C and 50% of the models within ±0.4°C of the mean. All models show increases, more in the second half of the year compared to the first (Supplementary Material Figure S11.31).

For the same period, 2080 to 2099, annual precipitation increases over the southern Pacific when averaged over all MMD models are close to 3%, with individual models projecting values from –4 to +11% and 50% of the models showing increases between 3 and 6%. The time to reach a discernible signal is relatively short.( Table 11.1 ). Most of these increases were in the first half of the year (Supplementary Material Figure S11.32). For precipitation in the northern Pacific, an increase of 5% is found, with individual models projecting values from 0 to 19% and at least half of the models within –2 to +5% of the median. The time to reach a discernible signal is relatively long. Most of these increases were in the latter half of the year (Supplementary Material Figure S11.32). Figure 11.25 illustrates the spatial distribution of annual, DJF and JJA rainfall changes and inter-model consistency. The figure shows that the tendency for precipitation increase in the Pacific is strongest in the region of the ITCZ due to increased moisture transport described in Section 11.1.3.1. Griffiths et al. 2003 [JoC] ) find an increasing trend from 1961 to 2000 in mean rainfall in and northeast of the SPCZ in the southern Pacific. As for the Indian Ocean, there is some level of confidence in the precipitation results for the Pacific, but not as high as for the temperature results.

Figure 11.25. As for Figure 11.23 but for the northern and southern Pacific Ocean.

Changes in rainfall variability in the South Pacific were analysed by Jones et al. 2000 [Ambiguous] ) using IPCC 1996 [NPR] ) scenarios, but more recent simulations have not been examined. These changes will be strongly driven by changes in ENSO, and this is not well understood (see Section 10.3 ).

11.10.1.1 High-Resolution Atmosphere-Only GCMs

Atmosphere-only GCMs (AGCMs) include interactive land surface schemes as in an AOGCM but require information on SST and sea ice as a lower boundary condition. Given the short time scales associated with the atmosphere and land surface components compared to those in the ocean, relatively short time slices (a few decades) can be run at high resolution. The SST and sea ice information required can be derived from observations or AOGCMs. The use of observations can improve simulations of current climate but combining these with AOGCM-derived changes for the future climate (e.g., Rowell, 2005 [JoC] ) increases the risk of inconsistency in the projected climate. The absence of two-way feedback between the atmosphere and ocean in AGCMs can cause a significant distortion of the climatic variability ( Bretherton and Battisti, 2000 [JoC, MoS, ARC] ), as documented over regions such as the Indian Ocean and the South Asian monsoon ( Douville, 2005 [JoC, MoS] ; Inatsu and Kimoto, 2005 [ARC] ). The large-scale climate responses of AGCMs and AOGCMs appear to be similar in many regions; when and where they differ, the consistency of the oceanic surface boundary condition may be questioned (May and Roeckner, 2001 [JoC, ARC] ; Govindasamy et al., 2003 [JoC, MoS] ). Further research is required to determine if the similarity is sufficient for the time-slice approach with AGCMs to be considered a robust downscaling technique.

Model grids of 100 km and finer have become feasible and 50 km will likely be the norm in the near future ( Bengtsson, 1996 [NPR] ; May and Roeckner, 2001 [JoC, ARC] ; Déqué and Gibelin, 2002 [NPR, MoS] ; Govindaswamy, 2003 ). High-performance computer systems now allow global computations at 20 km (e.g., May, 2004a; Mizuta et al., 2006 [MoS] ), although for short time slices only. Evaluated on the scale typical of current AOGCMs, nearly all quantities simulated by high-resolution AGCMs agree better with observations, but the improvements vary significantly for different regions ( Duffy et al., 2003 [JoC, MoS] ) and specific variables, and extensive recalibration of parametrizations is often required. Notable improvements occur in orographic precipitation and dynamics of mid-latitude weather systems (see Chapter 10 ). The highest resolution offers the prospect of credible simulations of the climatology of tropical cyclones (e.g., May, 2004a; Mizuta et al., 2006 [MoS] ). Coordinated multi-model experiments are needed, however, to optimise the value of these high-resolution studies for general assessment.

An alternative to uniform high-resolution AGCMs is Variable-Resolution AGCMs (VRGCMs; e.g., Déqué and Piedelievre, 1995 [JoC, MoS] ; Krinner et al., 1997 [JoC, MoS] ; Fox-Rabinovitz et al., 2001 [JoC, MoS] ; McGregor et al., 2002 [NPR, MoS] ; Gibelin and Déqué, 2003 [JoC, MoS] ). The VRGCM approach is attractive as it permits, within a unified modelling framework, a regional increase in resolution while retaining the interaction of all regions of the atmosphere. Numerical artefacts due to stretching have been shown to be small when using modest stretching factors (e.g., Lorant and Royer, 2001 [JoC, MoS] ). The results from VRGCMs capture, over the high-resolution region, finer-scale details than uniform-resolution models while retaining global skill similar to uniform-resolution simulations with the same number of grid points.

11.10.1.2 Nested Regional Climate Models

The principle behind nested modelling is that, consistent with the large-scale atmospheric circulation, realistic regional climate information can be generated by integrating an RCM if the following premises are satisfied: time-varying large-scale atmospheric fields (winds, temperature and moisture) are supplied as lateral boundary conditions (LBCs) and SST and sea ice as lower boundary conditions; the control from the LBCs keeps the interior solution of the RCM consistent with the driving atmospheric circulation; and sub-grid scale physical processes are suitably parametrized, including fine-scale surface forcing such as orography, land-sea contrast and land use.

Atypical RCM grid for climate change projections is around 50 km, although some climate simulations have been performed using grids of 15 or 20 km (e.g., Leung et al., 2003 [JoC, SRC] , 2004; Christensen and Christensen, 2004 [JoC, ARC] ; Kleinn et al., 2005 [JoC, MoS] ). Recently, projections of climate changes for East Asia were completed with a 5-km non-hydrostatic RCM ( Kanada et al., 2005 [MoS] ; Yoshizaki et al., 2005 [MoS] ; Yasunaga et al., 2006 [MoS] ), but only for short simulations. Following the trend in global modelling, RCMs are increasingly coupled interactively with other components of the climate system, such as regional ocean and sea ice (e.g., Bailey and Lynch 2000 [JoC, MoS, SRC] ; Döscher et al., 2002 [MoS] ; Rinke et al., 2003 [JoC, MoS, SRC] ; Bailey et al., 2004 [NPR, JoC, MoS, SRC] ; Meier et al., 2004 [JoC, MoS] ; Sasaki et al., 2006a [MoS] ), hydrology, and with interactive vegetation ( Gao and Yu, 1998 [JoC, MoS] ; Xue et al., 2000 [MoS] ).

Multi-decadal RCM experiments are becoming standard (e.g., Whetton et al., 2000 [NPR, MoS, SRC] ; Kwon et al., 2003 [NPR, SRC] ; Leung et al., 2004 [JoC] ; Kjellström et al., 2007 [NPR, 2007] ; Plummer et al., 2006 [MoS, SRC] ), including the use of ensembles ( Christensen et al., 2002 [JoC, SRC] ), enabling a more thorough validation and exploration of projected changes. In multi-year ensemble simulations driven by reanalyses of atmospheric observations, Vidale et al. 2003 [JoC, MoS] ) show that RCMs have skill in reproducing interannual variability in precipitation and surface air temperature. The use of ensemble simulations has enabled quantitative estimates regarding the sources of uncertainty in projections of regional climate changes ( Rowell, 2005 [JoC] ; Déqué et al., 2005 [JoC, MoS] , 2007; Beniston et al., 2007 [NPR, SRC, 2007] ; Frei et al., 2006 [JoC, MoS, ARC] ; Graham et al., 2007 [NPR, SRC, 2007] ). Combining information from four RCM simulations, Christensen et al. 2001 [JoC, MoS, ARC] and Rummukainen et al. 2003 [SRC] ) demonstrate that it is feasible to explore not only uncertainties related to projections in the mean climate state, but also for higher-order statistics.

The difficulties associated with the implementation of LBCs in nested models are well documented (e.g., Davies, 1976 [JoC, MoS] ; Warner et al., 1997 [JoC, MoS] ). As time progresses in a climate simulation, the RCM solution gradually turns from an initial-value problem more into a boundary value problem. The mathematical interpretation is that nested models represent a fundamentally ill-posed boundary value problem ( Staniforth, 1997 [MoS, ARC] ; Laprise, 2003 [JoC, MoS, SRC] ). The control exerted by LBCs on the internal solution generated by RCMs appears to vary with the size of the computational domain (e.g., Rinke and Dethloff, 2000 [MoS, SRC] ), as well as location and season (e.g., Caya and Biner, 2004 [JoC, MoS, SRC] ). In some applications, the flow developing within the RCM domain may become inconsistent with the driving LBC. This may ( Jones et al., 1997 [JoC, MoS, SRC] ) or may not ( Caya and Biner, 2004 [JoC, MoS, SRC] ) affect climate statistics. Normally, RCMs are only driven by LBCs with high time resolution to capture the temporal variations of large-scale flow. Some RCMs also use nudging or relaxation of large scales in the interior of the domain (e.g., Kida et al., 1991 [MoS] ; Biner et al., 2000 [NPR, MoS, SRC] ; von Storch et al., 2000 [JoC] ). This has proved useful to minimise the distortion of the large scales in RCMs ( von Storch et al., 2000 [JoC] ; Mabuchi et al., 2002 [MoS] ; Miguez-Macho et al., 2004 [JoC, MoS, ARC] ), although it can also hide model biases. One-way RCM-GCM coupling is mostly used, although recently a two-way nested RCM has been developed ( Lorenz and Jacob, 2005 [JoC, MoS, ARC] ) thus achieving interaction with the global atmosphere as with variable-resolution AGCMs.

The ability of RCMs to simulate the regional climate depends strongly on the realism of the large-scale circulation that is provided by the LBCs (e.g., Pan et al., 2001 [JoC, MoS] Latif et al. 2001 [JoC, MoS, ARC] and Davey et al. 2002 [JoC] ) show that strong biases in the tropical climatology of AOGCMs can negatively affect downscaling studies for several regions of the world. Nonetheless, the reliability of nested models, that is, their ability to generate meaningful fine-scale structures that are absent in the LBCs, is clear. A number of studies have shown that the climate statistics of atmospheric small scales can be re-created with the right amplitude and spatial distribution, even if these small scales are absent in the LBCs ( Denis et al., 2002 [JoC, MoS, SRC] , 2003; Antic et al., 2005 [JoC, MoS, SRC] ; Dimitrijevic and Laprise, 2005 [JoC, MoS, SRC] ). This implies that RCMs can add value at small scales to climate statistics when driven by AOGCMs with accurate large scales. Overall, the skill at simulating current climate has improved with the MMD AOGCMs( Chapter8 ), which will lead to higher quality LBCs for RCMs.

11.10.1.3 Empirical and Statistical Downscaling Methods

Acomplementary technique to RCMs is to use derived relationships linking large-scale atmospheric variables (predictors) and local/regional climate variables (predictands). The local/regional-scale climate change information is then obtained by applying the relationships to equivalent predictors from AOGCM simulations. The guidance document ( Wilby et al., 2004 [NPR, SRC] ) from the IPCC Task Group on Data and Scenario Support for Impact and Climate Analysis (TGICA) provides a comprehensive background on this approach and covers important issues in using SD applications. Statistical downscaling methods cover regression-type models including both linear and nonlinear relationships, unconditional or conditional weather generators for generating synthetic sequences of local variables, techniques based on weather classification that draw on the more skilful attributes of models to simulate circulation patterns, and analogue methods that seek equivalent weather states from the historical record; a combination of these techniques possibly being most appropriate. An extension to SD is the statistical-dynamical downscaling technique (e.g., ( Fuentes and Heimann, 2000 ) ), which combines weather classification with RCM simulations. A further development is the application of SD to high-resolution climate model output ( (Lionello et al., 2003; ) Imbert and Benestad, 2005 [MoS, SRC] ).

Research on SD has shown an extensive growth in application, and includes an increased availability of generic tools for the impact community (e.g., SDSM, Wilby et al., 2002 [SRC] ; the clim.pact package, Benestad, 2004b [JoC, MoS, SRC] ; the pyclimate package, ( Fernández and Sáenz, 2003 ) ); applications in new regions (e.g., Asia, Chen and Chen, 2003 [JoC] ); the use of techniques to address exotic variables such as phenological series ( (Matulla et al., 2003 ) ), extreme heat-related mortality ( Hayhoe et al., 2004 [JoC] ), ski season ( (Scott et al., 2003 ) ), land use ( Solecki and Oliveri, 2004 [MoS] ), streamflow or aquatic ecosystems ( Cannon and Whitfield, 2002 [MoS] ;( Blenckner and Chen, 2003 ) ); the treatment of climate extremes (e.g., Katz et al., 2002 [SRC] ; Seem, 2004 [MoS] ; X.L. Wang et al., 2004 [Ambiguous] ; Caires et al., 2006 [JoC, MoS] ); intercomparison studies evaluating methods (e.g., STAtistical and Regional dynamical Downscaling of EXtremes for European regions (STARDEX), Haylock et al., 2006 [JoC, MoS] ; Schmidli et al., 2006 [JoC, ARC] ); application to multi-model and multi-ensemble simulations in order to express model uncertainty alongside other key uncertainties (e.g., Benestad, 2002a [MoS, SRC] ,b; Hewitson and Crane, 2006 [JoC, MoS, SRC] ; Wang and Swail, 2006b [NPR, MoS] ); assessing non-stationarity in climate relationships ( Hewitson and Crane, 2006 [JoC, MoS, SRC] ); and spatial interpolation using geographical dependencies ( Benestad, 2005 [JoC, MoS, SRC] ). In some cases SD methods have been used to project statistical attributes instead of raw values of the predictand, for example, the probability of rainfall occurrence, precipitation, wind or wave height distribution parameters and extreme event frequency (e.g., Beckmann and Buishand, 2002 [JoC] ; Buishand et al., 2004 [JoC] ; Busuioc and von Storch, 2003 [MoS, SRC] ; Abaurrea and Asin, 2005 [MoS] ; Diaz-Nieto and Wilby, 2005 [JoC, SRC] ; Pryor et al., 2005a [JoC, MoS] ,b; Wang and Swail, 2006a [JoC, MoS] ,b).

Evaluation of SD is done most commonly through cross-validation with observational data for a period that represents an independent or different ‘climate regime’ (e.g., Busuioc et al. 2001 [JoC, MoS, SRC] ;( Trigo and Palutikof, 2001; ) Bartman et al., 2003 [JoC, MoS] ; Hanssen-Bauer et al., 2003 [SRC] ). Stationarity, that is, whether the statistical relationships are valid under future climate regimes, remains a concern with SD methods. This is only weakly assessed through cross-validation tests because future changes in climate are likely to be substantially larger than observed historical changes. This issue was assessed in Hewitson and Crane 2006 [JoC, MoS, SRC] ) where, within the SD method used, the non-stationarity was shown to result in an underestimation of the magnitude of the change. In general, the most effective SD methods are those that combine elements of deterministic transfer functions and stochastic components (e.g., ( Hansen and Mavromatis, 2001; ) ( Palutikof et al., 2002; ) Beersma and Buishand, 2003 [MoS] ; Busuioc and von Storch, 2003 [MoS, SRC] ; Katz et al., 2003 [JoC, MoS, SRC] ;( Lionello et al., 2003; ) Wilby et al., 2003 [MoS, SRC] ; X.L. Wang et al., 2004 [Ambiguous] ; Hewitson and Crane, 2006 [JoC, MoS, SRC] ). Regarding the predictors, the best choice appears to combine dynamical and moisture variables, especially in cases where precipitation is the predictand (e.g., Wilby et al., 2003 [MoS, SRC] ).

Pattern scaling is a simple statistical method for projecting regional climate change, which involves normalising AOGCM response patterns according to the global mean temperature. These normalised patterns are then rescaled using global mean temperature responses estimated for different emissions scenarios from a simple climate model (see Chapter 10 ). Some developments were made using various versions of scaling techniques (e.g., Christensen et al., 2001 [JoC, MoS, ARC] ; Mitchell, 2003 [JoC, MoS] ; Salathé, 2005 [JoC, MoS] ; Ruosteenoja et al., 2007 [NPR, 2007] ). For example, Ruosteenoja et al. 2007 [NPR, 2007] ) developed a pattern-scaling method using linear regression to represent the relationship between the local AOGCM-simulated temperature and precipitation response and the global mean temperature change. Another simple statistical technique is to use the GCM output for the variable of interest (i.e., the predictand) as the predictor and then apply a simple local change factor/scaling procedure (e.g., Chapter 13 of IPCC, 2001 [NPR] ; Hanssen-Bauer et al., 2003 [SRC] ; Widmann et al., 2003 [MoS, SRC] ; Diaz-Nieto and Wilby, 2005 [JoC, SRC] ).

Many studies have been performed since the TAR comparing various SD methods. In general, conclusions about one method compared to another are dependent on region and the criteria used for comparison, and on the inherent attributes of each method. For example, Diaz-Nieto and Wilby 2005 [JoC, SRC] ) downscale river flow and find that while two methods give comparable results, they differ in responses as a function of how the methods treat multi-decadal variability.

When comparing the merits of SD methods based on daily and monthly downscaling models, in terms of their ability to predict monthly means, daily models are better (e.g., Buishand et al., 2004 [JoC] ). In terms of nonlinearity in downscaling relationships, ( Trigo and Palutikof 2001 ) ) note that complex nonlinear models may not be better than simpler linear/slightly nonlinear approaches for some applications. However, Haylock et al. 2006 [JoC, MoS] ) find that SD methods based on nonlinear artificial neural networks are best at modelling the interannual variability of heavy precipitation but they underestimate extremes. Much downscaling work remains unreported, as SD activities are often implemented pragmatically for serving specific project needs, rather than for use by a broader scientific community; this is especially the case in developing nations. In some cases, this work is only found within the “grey” literature, for example, the AIACC project (http://www.aiaccproject.org/), which supports impact studies in developing nations.

11.10.1.4 Intercomparison of Downscaling Methods

At the time of the TAR, SD methods were viewed as a complementary technique to RCMs for downscaling regional climate, each approach having distinctive strengths and weaknesses. The conclusion of the TAR that SD methods and RCMs are comparable for simulating current climate still holds.

Since the TAR, a few additional studies have systematically compared the SD and RCM approaches (e.g., Huth et al., 2001 [JoC, MoS] ; Hanssen-Bauer et al., 2003 [SRC] , 2005; Wood et al., 2004 [JoC, MoS] , Busuioc et al., 2006 [MoS, SRC] ; Haylock et al., 2006 [JoC, MoS] ; Schmidli et al., 2006 [JoC, ARC] ). These related mainly to the similarity of the climate change signal (e.g., Hanssen-Bauer et al., 2003 [SRC] ). A more complex study considered additional information about the RCM skill in simulating the current regional climate features and reproducing the connection between large- and regional-scale patterns used for fitting the SD method ( Busuioc et al., 2006 [MoS, SRC] ). Other studies following the STARDEX project (e.g., Haylock et al., 2006 [JoC, MoS] ; Schmidli et al., 2006 [JoC, ARC] ) compared the two approaches in terms of their skill in reproducing current climate features, as well as in terms of the climate change signal derived from their outputs, focusing on climate extremes and complex topography processes over Europe.

11.10.2.1 Sources of Regional Uncertainty

Most sources of uncertainty at regional scales are similar to those at the global scale( Section 10.5 and Box 10.2 ), but there are both changes in emphasis and new issues that arise in the regional context. Spatial inhomogeneity of both land use and land cover change ( (De Fries et al., 2002; ) Chapter2 , Section 7.2 and Box 11.4 )and aerosol forcing adds to regional uncertainty. When analysing studies involving models to add local detail, the full cascade of uncertainty through the chain of models has to be considered. The degree to which these uncertainties influence the regional projections of different climate variables is not uniform. An indication of this is, for example, that models agree more readily on the sign and magnitude of temperature changes than of precipitation changes.

The regional impact of these uncertainties in climate projections has been illustrated by several authors. For example, incorporating a model of the carbon cycle into a coupled AOGCM gave a dramatically enhanced response to climate change over the Amazon Basin ( Cox et al., 2000 [JoC, MoS, ARC] ; Jones et al., 2003 [JoC, MoS, SRC] ) and Borneo ( Kumagi et al., 2004 [MoS] ). Further, the scale of the resolved processes in a climate model can significantly affect its simulation of climate over large regional scales ( Pope and Stratton, 2002 [JoC, MoS] ; Lorenz and Jacob, 2005 [JoC, MoS, ARC] Frei et al. 2003 [JoC, MoS, ARC] ) show that models with the same representation of resolved processes but different representations of sub-grid scale processes can represent the climate differently. The regional impact of changes in the representation of the land surface feedback is demonstrated by, for example, Oleson et al. 2004 [JoC, MoS, ARC] and Feddema et al. 2005 [JoC, MoS, SRC] ) (see also Box 11.4 ).