10.2.1.1 The Special Report on Emission Scenarios and Constant-Concentration Commitment Scenarios

The future projections discussed in this chapter are based upon the standard A2, A1B and B2 SRES scenarios ( Nakićenović and Swart, 2000 [NPR] ). The emissions of CO₂,methane (CH₄)and SO₂,the concentrations of CO₂,CH₄ and nitrous oxide (N₂O) and the total radiative forcing for the SRES scenarios are illustrated in Figure 10.26 and summarised for the A1B scenario in Figure 10.1 .The models have been integrated to year 2100 using the projected concentrations of LLGHGs and emissions of SO₂ specified by the A1B, B1 and A2 emissions scenarios. Some of the AOGCMs do not include sulphur chemistry, and the simulations from these models are based upon concentrations of sulphate aerosols from Boucher and Pham 2002 [JoC, MoS, ARC] ; see Section 10.2.1.2 ). The simulations for the three scenarios were continued for another 100 to 200 years with all anthropogenic forcing agents held fixed at values applicable to the year 2100 . There is also a new constant-concentration commitment scenario that assumes concentrations are held fixed at year 2000 levels( Section 10.7.1 ). In this idealised scenario, models are initialised from the end of the simulations for the 20th century, the concentrations of radiatively active species are held constant at year 2000 values from these simulations, and the models are integrated to 2100 .

For comparison with this constant composition case, it is useful to note that constant emissions would lead to much larger radiative forcing. For example, constant CO₂ emissions at year 2000 values would lead to concentrations reaching about 520 ppm by 2100, close to the B1 case ( Friedlingstein and Solomon, 2005 [PoC, JoC, SRC] ; Hare and Munschausen, 2006; see also FAQ 10.3 ).

10.2.1.2 Forcing by Additional Species and Mechanisms

The forcing agents applied to each AOGCM used to make climate projections are summarised in Table 10.1 .The radiatively active species specified by the SRES scenarios are CO₂,CH₄,N₂O, chlorofluorocarbons (CFCs) and SO₂,which is listed in its aerosol form as sulphate (SO₄)in the table. The inclusion, magnitude and temporal evolution of the remaining forcing agents listed in Table 10.1 were left to the discretion of the individual modelling groups. These agents include tropospheric and stratospheric ozone, all of the non-sulphate aerosols, the indirect effects of aerosols on cloud albedo and lifetime, the effects of land use and solar variability.

The scope of the treatments of aerosol effects in AOGCMs has increased markedly since the TAR. Seven of the AOGCMs include the first indirect effects and five include the second indirect effects of aerosols on cloud properties( Section 2.4.5 ). Under the more emissions-intensive scenarios considered in this chapter, the magnitude of the first indirect (Twomey) effect can saturate. Johns et al. 2003 [JoC, MoS] ) parametrize the first indirect effect of anthropogenic sulphur (S) emissions as perturbations to the effective radii of cloud drops in simulations of the B1, B2, A2 and A1FI scenarios using UKMO-HadCM3. At 2100, the first indirect forcing ranges from –0.50 to –0.79 Wm^–2.The normalised indirect forcing (the ratio of the forcing (Wm^–2)to the mass burden of a species (mgm^–2), leaving units of W mg^–1)decreases by a factor of four, from approximately –7 W mgS^–1 in 1860 to between –1 and –2 W mgS^–1 by the year 2100 Boucher and Pham 2002 [JoC, MoS, ARC] and Pham et al. 2005 [JoC, MoS, ARC] ) find a comparable projected decrease in forcing efficiency of the indirect effect, from –9.6 W mgS^–1 in 1860 to between –2.1 and –4.4 W mgS^–1 in 2100 Johns et al. 2003 [JoC, MoS] and Pham et al. 2005 [JoC, MoS, ARC] ) attribute the projected decline to the decreased sensitivity of clouds to greater sulphate concentrations at sufficiently large aerosol burdens.

10.2.1.3 Comparison of Modelled Forcings to Estimates in Chapter2

The forcings used to generate climate projections for the standard SRES scenarios are not necessarily uniform across the multi-model ensemble. Differences among models may be caused by different projections for radiatively active species (see Section 10.2.1.2 )and by differences in the formulation of radiative transfer (see Section 10.2.1.4 ). The AOGCMs in the ensemble include many species that are not specified or constrained by the SRES scenarios, including ozone, tropospheric non-sulphate aerosols, and stratospheric volcanic aerosols. Other types of forcing that vary across the ensemble include solar variability, the indirect effects of aerosols on clouds and the effects of land use change on land surface albedo and other land surface properties( Table 10.1 ). While the time series of LLGHGs for the future scenarios are mostly identical across the ensemble, the concentrations of these gases in the 19th and early 20th centuries were left to the discretion of individual modelling groups. The differences in radiatively active species and the formulation of radiative transfer affect both the 19th- and 20th-century simulations and the scenario integrations initiated from these historical simulations. The resulting differences in the forcing complicate the separation of forcing and response across the multi-model ensemble. These differences can be quantified by comparing the range of shortwave and longwave forcings across the multi-model ensemble against standard estimates of radiative forcing over the historical record. Shortwave and longwave forcing refer to modifications of the solar and infrared atmospheric radiation fluxes, respectively, that are caused by external changes to the climate system( Section 2.2 ).

The longwave radiative forcings for the SRES A1B scenario from climate model simulations are compared against estimates using the TAR formulae (see Chapter2 )in Figure 10.2 a. The graph shows the longwave forcings from the TAR and 20 AOGCMs in the multi-model ensemble from 2000 to 2100 . The forcings from the models are diagnosed from changes in top-of-atmosphere fluxes and the forcing for doubled atmospheric CO₂ ( Forster and Taylor, 2006 [JoC, MoS, SRC] ). The TAR and median model estimates of the longwave forcing are in very good agreement over the 21st century, with differences ranging from –0.37 to +0.06 Wm^–2.For the year 2000, the global mean values from the TAR and median model differ by only –0.13 Wm^–2.However, the 5th to 95th percentile range of the models for the period 2080 to 2099 is approximately 3.1 Wm^–2,or approximately 47% of the median longwave forcing for that time period.

The corresponding time series of shortwave forcings for the SRES A1B scenario are plotted in Figure 10.2 b. It is evident that the relative differences among the models and between the models and the TAR estimates are larger for the shortwave band. The TAR value is larger than the median model forcing by 0.2 to 0.3 Wm^–2 for individual 20-year segments of the integrations. For the year 2000, the TAR estimate is larger by 0.42 Wm^–2.In addition, the range of modelled forcings is sufficiently large that it includes positive and negative values for every 20-year period. For the year 2100, the shortwave forcing from individual AOGCMs ranges from approximately –1.7 Wm^–2 to +0.4 Wm^–2 (5th to 95th percentile). The reasons for this large range include the variety of the aerosol treatments and parametrizations for the indirect effects of aerosols in the multi-model ensemble.

Figure 10.2. Radiative forcings for the period 2000 to 2100 for the SRES A1B scenario diagnosed from AOGCMs and from the TAR ( IPCC, 2001 [NPR] ) forcing formulas ( Forster and Taylor, 2006 [JoC, MoS, SRC] ). (a) Longwave forcing; (b) shortwave forcing. The AOGCM results are plotted with box-and-whisker diagrams representing percentiles of forcings computed from 20 models in the AR4 multi-model ensemble. The central line within each box represents the median value of the model ensemble. The top and bottom of each box shows the 75th and 25th percentiles, and the top and bottom of each whisker displays the 95th and 5th percentile values in the ensemble, respectively. The models included are CCSM3, CGCM3.1 (T47 and T63), CNRM-CM3, CSIRO-MK3, ECHAM5/MPI-OM, ECHO-G, FGOALS-g1.0, GFDL-CM2.0, GFDL-CM2.1, GISS-EH, GISS-ER, INM-CM3.0, IPSL-CM4, MIROC3.2 (medium and high resolution), MRI-CGCM2.3.2, PCM1, UKMO-HadCM3 and UKMO-HadGEM1 (see Table 8.1 for model details).

Since the large range in both longwave and shortwave forcings may be caused by a variety of factors, it is useful to determine the range caused just by differences in model formulation for a given (identical) change in radiatively active species. A standard metric is the global mean, annually averaged all-sky forcing at the tropopause for doubled atmospheric CO₂.Estimates of this forcing for 15 of the models in the ensemble are given in Table 10.2 .The shortwave forcing is caused by absorption in the near-infrared bands of CO₂.The range in the longwave forcing at 200 mb is 0.84 Wm^–2,and the coefficient of variation, or ratio of the standard deviation to mean forcing, is 0.09. These results suggest that up to 35% of the range in longwave forcing in the ensemble for the period 2080 to 2099 is due to the spread in forcing estimates for the specified increase in CO₂.The findings also imply that it is not appropriate to use a single best value of the forcing from doubled atmospheric CO₂ to relate forcing and response (e.g., climate sensitivity) across a multi-model ensemble. The relationships for a given model should be derived using the radiative forcing produced by the radiative parametrizations in that model. Although the shortwave forcing has a coefficient of variation close to one, the range across the ensemble explains less than 17% of the range in shortwave forcing at the end of the 21st-century simulations. This suggests that species and forcing agents other than CO₂ cause the large variation among modelled shortwave forcings.

10.2.1.4 Results from the Radiative-Transfer Model Intercomparison Project: Implications for Fidelity of Forcing Projections

Differences in radiative forcing across the multi-model ensemble illustrated in Table 10.2 have been quantified in the Radiative-Transfer Model Intercomparison Project (RTMIP, W.D. Collins et al., 2006 [Ambiguous] ). The basis of RTMIP is an evaluation of the forcings computed by 20 AOGCMs using five benchmark line-by-line (LBL) radiative transfer codes. The comparison is focused on the instantaneous clear-sky radiative forcing by the LLGHGs CO₂,CH₄,N₂O, CFC-11, CFC-12 and the increased water vapour expected in warmer climates. The results of this intercomparison are not directly comparable to the estimates of forcing at the tropopause( Chapter2 ), since the latter include the effects of stratospheric adjustment. The effects of adjustment on forcing are approximately –2% for CH₄,–4% forN₂O, +5% for CFC-11, +8% for CFC-12 and –13% for CO₂ ( IPCC, 1995 [NPR, MoS] ; Hansen et al., 1997 [JoC, MoS, SRC] ). The total (longwave plus shortwave) radiative forcings at 200 mb, a surrogate for the tropopause, are shown in Table 10.3 for climatological mid-latitude summer conditions.

Total forcings calculated from the AOGCM and LBL codes due to the increase in LLGHGs from 1860 to 2000 differ by less than 0.04, 0.49 and 0.10 Wm^–2 at the top of model, surface and pseudo-tropopause at 200mb, respectively( Table 10.3 ). Based upon the Student t-test, none of the differences in mean forcings shown in Table 10.3 is statistically significant at the 0.01 level. This indicates that the ensemble mean forcings are in reasonable agreement with the LBL codes. However, the forcings from individual models, for example from doubled atmospheric CO₂,span a range at least 10 times larger than that exhibited by the LBL models.

The forcings from doubling atmospheric CO₂ from its concentration at 1860 AD are shown in Figure 10.3 aat the top of the model (TOM), 200 hPa( Table 10.3 ), and the surface. The AOGCMs tend to underestimate the longwave forcing at these three levels. The relative differences in the mean forcings are less than 8% for the pseudo-tropopause at 200 hPa but increase to approximately 13% at the TOM and to 33% at the surface. In general, the mean shortwave forcings from the LBL and AOGCM codes are in good agreement at all three surfaces. However, the range in shortwave forcing at the surface from individual AOGCMs is quite large. The coefficient of variation (the ratio of the standard deviation to the mean) for the surface shortwave forcing from AOGCMs is 0.95. In response to a doubling in atmospheric CO₂,the specific humidity increases by approximately 20% through much of the troposphere. The changes in shortwave and longwave fluxes due to a 20% increase in water vapour are illustrated in Figure 10.3 b. The mean longwave forcing from increasing water vapour is quite well simulated with the AOGCM codes. In the shortwave, the only significant difference between the AOGCM and LBL calculations occurs at the surface, where the AOGCMs tend to underestimate the magnitude of the reduction in insolation. In general, the biases in the AOGCM forcings are largest at the surface level.

Figure 10.3. Comparison of shortwave and longwave instantaneous radiative forcings and flux changes computed from AOGCMs and line-by-line (LBL) radiative transfer codes (W.D. Collins et al., 2006 [Ambiguous] ). (a) Instantaneous forcing from doubling atmospheric CO₂ from its concentration in 1860; b) changes in radiative fluxes caused by the 20% increase in water vapour expected in the climate produced from doubling atmospheric CO₂.The forcings and flux changes are computed for clear-sky conditions in mid-latitude summer and do not include effects of stratospheric adjustment. No other well-mixed greenhouse gases are included. The minimum-to-maximum range and median are plotted for five representative LBL codes. The AOGCM results are plotted with box-and-whisker diagrams (see caption for Figure 10.2) representing percentiles of forcings from 20 models in the AR4 multi-model ensemble. The AOGCMs included are BCCRBCM2.0, CCSM3, CGCM3.1(T47 and T63), CNRMCM3, ECHAM5/MPIOM, ECHOG, FGOALSg1.0, GFDLCM2.0, GFDLCM2.1, GISSEH, GISSER, INMCM3.0, IPSLCM4, MIROC3.2 (medium and high resolution), MRICGCM2.3.2, PCM, UKMOHadCM3, and UKMOHadGEM1 (see Table 8.1 for model details). The LBL codes are the Geophysical Fluid Dynamics Laboratory (GFDL) LBL, the Goddard Institute for Space Studies (GISS) LBL3, the National Center for Atmospheric Research (NCAR)/Imperial College of Science, Technology and Medicine (ICSTM) general LBL GENLN2, the National Aeronautics and Space Administration (NASA) Langley Research Center MRTA and the University of Reading Reference Forward Model (RFM).

10.2.1.5 Implications for Range in Climate Response

The results from RTMIP imply that the spread in climate response discussed in this chapter is due in part to the diverse representations of radiative transfer among the members of the multi-model ensemble. Even if the concentrations of LLGHGs were identical across the ensemble, differences in radiative transfer parametrizations among the ensemble members would lead to different estimates of radiative forcing by these species. Many of the climate responses (e.g., global mean temperature) scale linearly with the radiative forcing to first approximation. Therefore, systematic errors in the calculations of radiative forcing should produce a corresponding range in climate responses. Assuming that the RTMIP results( Table 10.3 )are globally applicable, the range of forcings for 1860 to 2000 in the AOGCMs should introduce a ±18% relative range (the 5 to 95% confidence interval) for 2000 in the responses that scale with forcing. The corresponding relative range for doubled atmospheric CO₂,which is comparable to the change in CO₂ in the B1 scenario by 2100, is ± 25%.

Some recent studies have suggested that the global atmospheric burden of soil dust aerosols could decrease by between 20 and 60% due to reductions in desert areas associated with climate change ( Mahowald and Luo, 2003 [JoC, MoS] Tegen et al. 2004a [JoC, MoS] ,b) compared simulations by the European Centre for Medium Range Weather Forecasts/Max Planck Institute for Meteorology Atmospheric GCM (ECHAM4) and UKMO-HadCM3 that included the effects of climate-induced changes in atmospheric conditions and vegetation cover and the effects of increased CO₂ concentrations on vegetation density. These simulations are forced with identical (IS92a) time series for LLGHGs. Their findings suggest that future projections of changes in dust loading are quite model dependent, since the net changes in global atmospheric dust loading produced by the two models have opposite signs. They also conclude that dust from agriculturally disturbed soils is less than 10% of the current burden, and that climate-induced changes in dust concentrations would dominate land use changes under both minimum and maximum estimates of increased agricultural area by 2050 .

10.3.2.1 Warming

The TAR noted that much of the regional variation of the annual mean warming in the multi-model means is associated with high- to low-latitude contrast. This can be better quantified from the new multi-model mean in terms of zonal averages. A further contrast is provided by partitioning the land and ocean values based on model data interpolated to a standard grid. Figure 10.6 illustrates the late-century A2 case, with all values shown both in absolute terms and relative to the global mean warming. Warming over land is greater than the mean except in the southern mid-latitudes, where the warming over ocean is a minimum. Warming over ocean is smaller than the mean except at high latitudes, where sea ice changes have an influence. This pattern of change illustrated by the ratios is quite similar across the scenarios. The commitment case (shown), discussed in Section 10.7.1 ,has relatively smaller warming of land, except in the far south, which warms closer to the global rate. At nearly all latitudes, the A1B and B1 warming ratios lie between A2 and commitment, with A1B particularly close to the A2 results. Aside from the commitment case, the ratios for the other time periods are also quite similar to those for A2. Regional patterns and precipitation contrasts are discussed in Section 10.3.2.3 .

Figure 10.6. Zonal means over land and ocean separately, for annual mean surface warming (a, b) and precipitation (c, d), shown as ratios scaled with the global mean warming (a, c) and not scaled (b, d). Multi-model mean results are shown for two scenarios, A2 and Commitment (see Section 10.7 ), for the period 2080 to 2099 relative to the zonal means for 1980 to 1999 . Results for individual models can be seen in the Supplementary Material for this chapter.

Figure 10.7 shows the zonal mean warming for the A1B scenario at each latitude from the bottom of the ocean to the top of the atmosphere for the three 21st-century periods used in Table 10.5 .To produce this ensemble mean, the model data were first interpolated to standard ocean depths and atmospheric pressures. Consistent with the global transfer of excess heat from the atmosphere to the ocean, and the difference between warming over land and ocean, there is some discontinuity between the plotted means of the lower atmosphere and the upper ocean. The relatively uniform warming of the troposphere and cooling of the stratosphere in this multi-model mean are consistent with the changes shown in Figure 9.8 of the TAR, but now its evolution during the 21st century under this scenario can also be seen. Upper-tropospheric warming reaches a maximum in the tropics and is seen even in the early-century time period. The pattern is very similar over the three periods, consistent with the rapid adjustment of the atmosphere to the forcing. These changes are simulated with good consistency among the models. The larger values of both signs are stippled, indicating that the ensemble mean is larger in magnitude than the inter-model standard deviation. The ratio of mean to standard deviation can be related to formal tests of statistical significance and confidence intervals, if the individual model results were to be considered a sample.

Figure 10.7. Zonal means of change in atmospheric (top) and oceanic (bottom) temperatures (°C), shown as cross sections. Values are the multi-model means for the A1B scenario for three periods (a–c). Stippling denotes regions where the multi-model ensemble mean divided by the multi-model standard deviation exceeds 1.0 (in magnitude). Anomalies are relative to the average of the period 1980 to 1999 . Results for individual models can be seen in the Supplementary Material for this chapter.

The ocean warming evolves more slowly. There is initially little warming below the mixed layer, except at some high latitudes. Even as a ratio with mean surface warming, later in the century the temperature increases more rapidly in the deep ocean, consistent with results from individual models (e.g., Watterson, 2003 [JoC, MoS, SRC] ; Stouffer, 2004 [JoC, SRC] ). This rapid warming of the atmosphere and the slow penetration of the warming into the ocean has implications for the time scales of climate change commitment( Section 10.7 ). It has been noted in a five-member multi-model ensemble analysis that, associated with the changes in temperature of the upper ocean in Figure 10.7 ,the tropical Pacific Ocean heat transport remains nearly constant with increasing greenhouse gases due to the compensation of the subtropical cells and the horizontal gyre variations, even as the subtropical cells change in response to changes in the trade winds ( Hazeleger, 2005 [JoC] ). Additionally, a southward shift of the Antarctic Circumpolar Current is projected to occur in a 15-member multi-model ensemble, due to changes in surface winds in a future warmer climate ( Fyfe and Saenko, 2005 [JoC, ARC] ). This is associated with a poleward shift of the westerlies at the surface (see Section 10.3.6 )and in the upper troposphere particularly notable in the Southern Hemisphere (SH) ( Stone and Fyfe, 2005 [JoC, ARC] ), and increased relative angular momentum from stronger westerlies ( Räisänen, 2003 [JoC, SRC] ) and westerly momentum flux in the lower stratosphere particularly in the tropics and southern mid-latitudes ( Watanabe et al., 2005 [SRC] ). The surface wind changes are associated with corresponding changes in wind stress curl and horizontal mass transport in the ocean ( Saenko et al., 2005 [JoC, ARC] ).

Global-scale patterns for each of the three scenarios and time periods are given in Figure 10.8 .In each case, greater warming over most land areas is evident (e.g., Kunkel and Liang, 2005 [JoC, MoS, ARC] ). Over the ocean, warming is relatively large in the Arctic and along the equator in the eastern Pacific (see Sections 10.3.5.2 and 10.3.5.3 ), with less warming over the North Atlantic and the Southern Ocean (e.g., Xu et al., 2005 [MoS, SRC] ). Enhanced oceanic warming along the equator is also evident in the zonal means of Figure 10.6 ,and can be associated with oceanic heat flux changes ( Watterson, 2003 [JoC, MoS, SRC] ) and forced by the atmosphere ( Liu et al., 2005 [JoC] ).

Fields of temperature change have a similar structure, with the linear correlation coefficient as high as 0.994 between the late-century A2 and A1B cases. As for the zonal means, the fields normalised by the mean warming are very similar. The strict agreement between the A1B field, as a standard, and the others is quantified in Table 10.5 ,by the absolute measure M ( Watterson, 1996 [JoC, MoS, SRC] ; a transformation of a measure of ( Mielke, 1991 ) ), with unity meaning identical fields and zero meaning no similarity (the expected value under random rearrangement of the data on the grid of the measure prior to the arcsin transformation). Values of M become progressively larger later in the 21st century, with values of 0.9 or larger for the late 21st century, thus confirming the closeness of the scaled patterns in the late-century cases. The deviation from unity is approximately proportional to the mean absolute difference. The earlier warming patterns are also similar to the standard case, particularly for the same scenario A1B. Furthermore, the zonal means over land and ocean considered above are representative of much of the small differences in warming ratio. While there is some influence of differences in forcing patterns among the scenarios, and of effects of oceanic uptake and heat transport in modifying the patterns over time, there is also support for the role of atmospheric heat transport in offsetting such influences (e.g., Boer and Yu, 2003b [JoC] ; Watterson and Dix, 2005 [JoC, MoS, SRC] Dufresne et al. 2005 [JoC, SRC] ) show that aerosol contributes a modest cooling of the Northern Hemisphere (NH) up to the mid-21st century in the A2 scenario.

Such similarities in patterns of change have been described by Mitchell 2003 [JoC, MoS] and Harvey 2004 [JoC] ). They aid the efficient presentation of the broad scale multi-model results, as patterns depicted for the standard A1B 2080 to 2099 case are usually typical of other cases. This largely applies to other seasons and also other variables under consideration here. Where there is similarity of normalised changes, values for other cases can be estimated by scaling by the appropriate ratio of global means from Table 10.5 .Note that for some quantities like variability and extremes, such scaling is unlikely to work. The use of such scaled results in combination with global warmings from simple models is discussed in Section 11.10.1 .

As for the zonal means (aside from the Arctic Ocean), consistency in local warmings among the models is high (stippling is omitted in Figure 10.8 for clarity). Only in the central North Atlantic and the far south Pacific in 2011 to 2030 is the mean change less than the standard deviation, in part a result of ocean model limitations there( Section 8.3.2 ). Some regions of high-latitude surface cooling occur in individual models.

Figure 10.8. Multi-model mean of annual mean surface warming (surface air temperature change, °C) for the scenarios B1 (top), A1B (middle) and A2 (bottom), and three time periods, 2011 to 2030 (left), 2046 to 2065 (middle) and 2080 to 2099 (right). Stippling is omitted for clarity (see text). Anomalies are relative to the average of the period 1980 to 1999 . Results for individual models can be seen in the Supplementary Material for this chapter.

The surface warming fields for the extratropical winter and summer seasons, December to February (DJF) and June to August (JJA), are shown for scenario A1B in Figure 10.9 .The high-latitude warming is rather seasonal, being larger in winter as a result of sea ice and snow, as noted in Chapter 9 of the TAR. However, the relatively small warming in southern South America is more extensive in southern winter. Similar patterns of change in earlier model simulations are described by Giorgi et al. 2001 [JoC, MoS, ARC] ).

Figure 10.9. Multi-model mean changes in surface air temperature (°C, left), precipitation (mm day^–1,middle) and sea level pressure (hPa, right) for boreal winter (DJF, top) and summer (JJA, bottom). Changes are given for the SRES A1B scenario, for the period 2080 to 2099 relative to 1980 to 1999 . Stippling denotes areas where the magnitude of the multi-model ensemble mean exceeds the inter-model standard deviation. Results for individual models can be seen in the Supplementary Material for this chapter.

10.3.2.2 Cloud and Diurnal Cycle

In addition to being an important link to humidity and precipitation, cloud cover plays an important role for the sensitivity of the general circulation models (GCMs; e.g., Soden and Held, 2006 [JoC, MoS, ARC] ) and for the diurnal temperature range (DTR) over land (e.g., Dai and Trenberth, 2004 [PoC, JoC, MoS, ARC] and references therein) so this section considers the projection of these variables now made possible by multi-model ensembles. Cloud radiative feedbacks to greenhouse gas forcing are sensitive to the elevation, latitude and hence temperature of the clouds, in addition to their optical depth and their atmospheric environment (see Section 8.6.3.2 ). Current GCMs simulate clouds through various complex parametrizations (see Section 8.2.1.3 )to produce cloud cover quantified by an area fraction within each grid square and each atmospheric layer. Taking multi-model ensemble zonal means of this quantity interpolated to standard pressure levels and latitudes shows increases in cloud cover at all latitudes in the vicinity of the tropopause, and mostly decreases below, indicating an increase in the altitude of clouds overall( Figure 10.10 a). This shift occurs consistently across models. Outside the tropics the increases aloft are rather consistent, as indicated by the stippling in the figure. Near-surface amounts increase at some latitudes. The mid-level mid-latitude decreases are very consistent, amounting to as much as one-fifth of the average cloud fraction simulated for 1980 to 1999 .

Figure 10.10. Multi-model mean changes in (a) zonal mean cloud fraction (%), shown as a cross section though the atmosphere, and (b) total cloud area fraction (percent cover from all models). Changes are given as annual means for the SRES A1B scenario for the period 2080 to 2099 relative to 1980 to 1999 . Stippling denotes areas where the magnitude of the multi-model ensemble mean exceeds the inter-model standard deviation. Results for individual models can be seen in the Supplementary Material for this chapter.

The total cloud area fraction from an individual model represents the net coverage over all the layers, after allowance for the overlap of clouds, and is an output included in the data set. The change in the ensemble mean of this field is shown in Figure 10.10 b. Much of the low and middle latitudes experience a decrease in cloud cover, simulated with some consistency. There are a few low-latitude regions of increase, as well as substantial increases at high latitudes. The larger changes relate well to changes in precipitation discussed in Section 10.3.2.3 .While clouds need not be precipitating, moderate spatial correlation between cloud cover and precipitation holds for seasonal means of both the present climate and future changes.

The radiative effect of clouds is represented by the cloud radiative forcing diagnostic (see Section 8.6.3.2 ). This can be evaluated from radiative fluxes at the top of the atmosphere calculated with or without the presence of clouds that are output by the GCMs. In the multi-model mean (not shown) values vary in sign over the globe. The global and annual mean averaged over the models, for 1980 to 1999, is –22.3 Wm^–2.The change in mean cloud radiative forcing has been shown to have different signs in a limited number of previous modelling studies ( Meehl et al., 2004b [JoC, MoS, SRC] ; Tsushima et al., 2006 [JoC, MoS] ). Figure 10.11 ashows globally averaged cloud radiative forcing changes for 2080 to 2099 under the A1B scenario for individual models of the data set, which have a variety of different magnitudes and even signs. The ensemble mean change is –0.6 Wm^–2.This range indicates that cloud feedback is still an uncertain feature of the global coupled models (see Section 8.6.3.2.2 ).

The DTR has been shown to be decreasing in several land areas of the globe in 20th-century observations (see Section 3.2.2.7 ), together with increasing cloud cover (see also Section 9.4.2.3 ). In the multi-model mean of present climate, DTR over land is indeed closely spatially anti-correlated with the total cloud cover field. This is true also of the 21st-century changes in the fields under the A1B scenario, as can be seen by comparing the change in DTR shown in Figure 10.11 bwith the cloud area fraction shown in Figure 10.10 b. Changes in DTR reach a magnitude of 0.5°C in some regions, with some consistency among the models. Smaller widespread decreases are likely due to the radiative effect of the enhanced greenhouse gases including water vapour (see also Stone and Weaver, 2002 [JoC, MoS, SRC] ). Further discussion of DTR is provided in Section 10.3.6.2 .

In addition to the DTR, Kitoh and Arakawa 2005 [JoC, MoS, SRC] ) document changes in the regional patterns of diurnal precipitation over the Indonesian region, and show that over ocean, nighttime precipitation decreases and daytime precipitation increases, while over land the opposite is the case, thus producing a decrease in the diurnal precipitation amplitude over land and ocean. They attribute these changes to a larger nighttime temperature increase over land due to increased greenhouse gases.

Figure 10.11. Changes in (a) global mean cloud radiative forcing (Wm^–2)from individual models (see Table 10.4 for the list of models) and (b) multi-model mean diurnal temperature range (°C). Changes are annual means for the SRES A1B scenario for the period 2080 to 2099 relative to 1980 to 1999 . Stippling denotes areas where the magnitude of the multi-model ensemble mean exceeds the inter-model standard deviation. Results for individual models can be seen in the Supplementary Material for this chapter.

10.3.2.3 Precipitation and Surface Water

Models simulate that global mean precipitation increases with global warming. However, there are substantial spatial and seasonal variations in this field even in the multi-model means depicted in Figure 10.9 .There are fewer areas stippled for precipitation than for the warming, indicating more variation in the magnitude of change among the ensemble of models. Increases in precipitation at high latitudes in both seasons are very consistent across models. The increases in precipitation over the tropical oceans and in some of the monsoon regimes (e.g., South Asian monsoon in JJA, Australian monsoon in DJF) are notable, and while not as consistent locally, considerable agreement is found at the broader scale in the tropics ( Neelin et al., 2006 [JoC, MoS] ). There are widespread decreases in mid-latitude summer precipitation, except for increases in eastern Asia. Decreases in precipitation over many subtropical areas are evident in the multi-model ensemble mean, and consistency in the sign of change among the models is often high ( Wang, 2005 [JoC, MoS] ), particularly in some regions like the tropical Central American-Caribbean ( Neelin et al., 2006 [JoC, MoS] ). Further discussion of regional changes is presented in Chapter 11 .

The global map of the A1B 2080 to 2099 change in annual mean precipitation is shown in Figure 10.12 ,along with other hydrological quantities from the multi-model ensemble. Emori and Brown 2005 [JoC, SRC] ) show percentage changes of annual precipitation from the ensemble. Increases of over 20% occur at most high latitudes, as well as in eastern Africa, central Asia and the equatorial Pacific Ocean. The change over the ocean between 10°S and 10°N accounts for about half the increase in the global mean( Figure 10.5 ). Substantial decreases, reaching 20%, occur in the Mediterranean region ( Rowell and Jones, 2006 [JoC, MoS, ARC] ), the Caribbean region ( Neelin et al., 2006 [JoC, MoS] ) and the subtropical western coasts of each continent. Overall, precipitation over land increases by about 5%, while precipitation over ocean increases 4%, but with regional changes of both signs. The net change over land accounts for 24% of the global mean increase in precipitation, a little less than the areal proportion of land (29%). In Figure 10.12 ,stippling indicates that the sign of the local change is common to at least 80% of the models (with the alternative test shown in the Supplementary Material). This simpler test for consistency is of particular interest for quantities where the magnitudes for the base climate vary across models.

These patterns of change occur in the other scenarios, although with agreement (by the metric M) a little lower than for the warming. The predominance of increases near the equator and at high latitudes, for both land and ocean, is clear from the zonal mean changes of precipitation included in Figure 10.6 .The results for change scaled by global mean warming are rather similar across the four scenarios, an exception being a relatively large increase over the equatorial ocean for the commitment case. As with surface temperature, the A1B and B1 scaled values are always close to the A2 results. The zonal means of the percentage change map (shown in Figure 10.6 )feature substantial decreases in the subtropics and lower mid-latitudes of both hemispheres in the A2 case, even if increases occur over some regions.

Figure 10.12. Multi-model mean changes in (a) precipitation (mm day^–1), (b) soil moisture content (%), (c) runoff (mm day^–1)and (d) evaporation (mm day^–1). To indicate consistency in the sign of change, regions are stippled where at least 80% of models agree on the sign of the mean change. Changes are annual means for the SRES A1B scenario for the period 2080 to 2099 relative to 1980 to 1999 . Soil moisture and runoff changes are shown at land points with valid data from at least 10 models. Details of the method and results for individual models can be found in the Supplementary Material for this chapter.

Wetherald and Manabe 2002 [JoC, MoS] ) provide a good description of the mechanism of hydrological change simulated by GCMs. In GCMs, the global mean evaporation changes closely balance the precipitation change, but not locally because of changes in the atmospheric transport of water vapour. Annual average evaporation( Figure 10.12 )increases over much of the ocean, with spatial variations tending to relate to those in the surface warming( Figure 10.8 ). As found by Kutzbach et al. 2005 [JoC, MoS] and Bosilovich et al. 2005 [JoC] ), atmospheric moisture convergence increases over the equatorial oceans and over high latitudes. Over land, rainfall changes tend to be balanced by both evaporation and runoff. Runoff( Figure 10.12 )is notably reduced in southern Europe and increased in Southeast Asia and at high latitudes, where there is consistency among models in the sign of change (although less consistency in the magnitude of change). The larger changes reach 20% or more of the simulated 1980 to 1999 values, which range from 1 to 5 mm day^–1 in wetter regions to below 0.2 mm day^–1 in deserts. Runoff from the melting of ice sheets( Section 10.3.3 )is not included here. Nohara et al. 2006 [JoC, MoS, SRC] and Milly et al. 2005 [JoC] ) assess the impacts of these changes in terms of river flow, and find that discharges from high-latitude rivers increase, while those from major rivers in the Middle East, Europe and Central America tend to decrease.

Models simulate the moisture in the upper few metres of the land surface in varying ways, and evaluation of the soil moisture content is still difficult (See Section 8.2.3.2 ;Wang, 2005 [JoC, MoS] ; Gao and Dirmeyer, 2006 [JoC, MoS, ARC] for multi-model analyses). The average of the total soil moisture content quantity submitted to the data set is presented here to indicate typical trends. In the annual mean( Figure 10.12 ), decreases are common in the subtropics and the Mediterranean region. There are increases in east Africa, central Asia, and some other regions with increased precipitation. Decreases also occur at high latitudes, where snow cover diminishes( Section 10.3.3 ). While the magnitudes of change are quite uncertain, there is good consistency in the signs of change in many of these regions. Similar patterns of change occur in seasonal results ( Wang, 2005 [JoC, MoS] ). Regional hydrological changes are considered in Chapter 11 and in the IPCC Working Group II report.

10.3.2.4 Sea Level Pressure and Atmospheric Circulation

As a basic component of the mean atmospheric circulations and weather patterns, projections of the mean sea level pressure for the medium scenario A1B are considered. Seasonal mean changes for DJF and JJA are shown in Figure 10.9 (matching results in Wang and Swail, 2006b [JoC, MoS] ). Sea level pressure differences show decreases at high latitudes in both seasons in both hemispheres. The compensating increases are predominantly over the mid-latitude and subtropical ocean regions, extending across South America, Australia and southern Asia in JJA, and the Mediterranean in DJF. Many of these increases are consistent across the models. This pattern of change, discussed further in Section 10.3.5.3 ,has been linked to an expansion of the Hadley Circulation and a poleward shift of the mid-latitude storm tracks ( Yin, 2005 [JoC, MoS] ). This helps explain, in part, the increases in precipitation at high latitudes and decreases in the subtropics and parts of the mid-latitudes. Further analysis of the regional details of these changes is given in Chapter 11 .The pattern of pressure change implies increased westerly flows across the western parts of the continents. These contribute to increases in mean precipitation( Figure 10.9 )and increased precipitation intensity ( Meehl et al., 2005a [JoC, MoS, SRC] ).

10.3.3.1 Changes in Sea Ice Cover

Models of the 21st century project that future warming is amplified at high latitudes resulting from positive feedbacks involving snow and sea ice, and other processes( Section 8.6.3.3 ). The warming is particularly large in autumn and early winter ( Manabe and Stouffer, 1980 [JoC, MoS, SRC] ; Holland and Bitz, 2003 [JoC, MoS, SRC] ) when sea ice is thinnest and the snow depth is insufficient to blur the relationship between surface air temperature and sea ice thickness ( Maykut and Untersteiner, 1971 [JoC, MoS] ). As shown by Zhang and Walsh 2006 [JoC, MoS, ARC] ), the coupled models show a range of responses in NH sea ice areal extent ranging from very little change to a strong and accelerating reduction over the 21st century( Figure 10.13 a,b).

Figure 10.13. Multi-model simulated anomalies in sea ice extent for the 20th century (20c3m) and 21st century using the SRES A2, A1B and B1 as well as the commitment scenario for (a) Northern Hemisphere January to March (JFM), (b) Northern Hemisphere July to September (JAS). Panels (c) and (d) are as for (a) and (b) but for the Southern Hemisphere. The solid lines show the multi-model mean, shaded areas denote ±1 standard deviation. Sea ice extent is defined as the total area where sea ice concentration exceeds 15%. Anomalies are relative to the period 1980 to 2000 . The number of models is given in the legend and is different for each scenario.

An important characteristic of the projected change is for summer ice area to decline far more rapidly than winter ice area ( Gordon and O’Farrell, 1997 [JoC, MoS] ), and hence sea ice rapidly approaches a seasonal ice cover in both hemispheres (Figures 10.13 band 10.14). Seasonal ice cover is, however, rather robust and persists to some extent throughout the 21st century in most (if not all) models. Bitz and Roe 2004 [JoC, SRC] ) note that future projections show that arctic sea ice thins fastest where it is initially thickest, a characteristic that future climate projections share with sea ice thinning observed in the late 20th century ( Rothrock et al., 1999 [JoC, ARC] ). Consistent with these results, a projection by Gregory et al. 2002b [JoC, MoS, SRC] ) shows that arctic sea ice volume decreases more quickly than sea ice area (because trends in winter ice area are low) in the 21st century.

In 20th- and 21st-century simulations, antarctic sea ice cover is projected to decrease more slowly than in the Arctic (Figures 10.13 c,d and 10.14 ), particularly in the vicinity of the Ross Sea where most models predict a local minimum in surface warming. This is commensurate with the region with the greatest reduction in ocean heat loss, which results from reduced vertical mixing in the ocean ( Gregory, 2000 [JoC, SRC] ). The ocean stores much of its increased heat below 1 km depth in the Southern Ocean. In contrast, horizontal heat transport poleward of about 60°N increases in many models ( Holland and Bitz, 2003 [JoC, MoS, SRC] ), but much of this heat remains in the upper 1 km of the northern subpolar seas and Arctic Ocean ( Gregory, 2000 [JoC, SRC] ; Bitz et al., 2006 [JoC, SRC] Bitz et al. 2006 [JoC, SRC] ) argue that these differences in the depth where heat is accumulating in the high-latitude oceans have consequences for the relative rates of sea ice decay in the Arctic and Antarctic.

Figure 10.14. Multi-model mean sea ice concentration (%) for January to March (JFM) and June to September (JAS), in the Arctic (top) and Antarctic (bottom) for the periods (a) 1980 to 2000 and b) 2080 to 2100 for the SRES A1B scenario. The dashed white line indicates the present-day 15% average sea ice concentration limit. Modified from Flato et al. 2004 [JoC, MoS, ARC] ).

While most climate models share these common characteristics (peak surface warming in autumn and early winter, sea ice rapidly becomes seasonal, arctic ice decays faster than antarctic ice, and northward ocean heat transport increases into the northern high latitudes), models have poor agreement on the amount of thinning of sea ice (Flato and Participating CMIP Modeling Groups, 2004; Arzel et al., 2006 [JoC, MoS, ARC] ) and the overall climate change in the polar regions ( IPCC, 2001 [NPR] ; Holland and Bitz, 2003 [JoC, MoS, SRC] Flato 2004 [JoC, MoS, ARC] ) shows that the basic state of the sea ice and the reduction in thickness and/or extent have little to do with sea ice model physics among CMIP2 models. Holland and Bitz 2003 [JoC, MoS, SRC] and Arzel et al. 2006 [JoC, MoS, ARC] ) find serious biases in the basic state of simulated sea ice thickness and extent. Further, Rind et al. 1995 [JoC, MoS, ARC] Holland and Bitz 2003 [JoC, MoS, SRC] and Flato 2004 [JoC, MoS, ARC] ) show that the basic state of the sea ice thickness and extent have a significant influence on the projected change in sea ice thickness in the Arctic and extent in the Antarctic.

10.3.3.2 Changes in Snow Cover and Frozen Ground

Snow cover is an integrated response to both temperature and precipitation and exhibits strong negative correlation with air temperature in most areas with a seasonal snow cover (see Section 8.6.3.3 for an evaluation of model-simulated present-day snow cover). Because of this temperature association, the simulations project widespread reductions in snow cover over the 21st century (Supplementary Material, Figure S10.1). For the Arctic Climate Impact Assessment (ACIA) model mean, at the end of the 21st century the projected reduction in the annual mean NH snow cover is 13% under the B2 scenario ( ACIA, 2004 [NPR] ). The individual model projections range from reductions of 9 to 17%. The actual reductions are greatest in spring and late autumn/early winter, indicating a shortened snow cover season ( ACIA, 2004 [NPR] ). The beginning of the snow accumulation season (the end of the snowmelt season) is projected to be later (earlier), and the fractional snow coverage is projected to decrease during the snow season ( Hosaka et al., 2005 [MoS, SRC] ).

Warming at high northern latitudes in climate model simulations is also associated with large increases in simulated thaw depth over much of the permafrost regions ( Lawrence and Slater, 2005 [JoC, MoS, SRC] ; Yamaguchi et al., 2005 [MoS, SRC] ;( Kitabata et al., 2006 ) Yamaguchi et al. 2005 [MoS, SRC] ) show that initially soil moisture increases during the summer. In the late 21st century when the thaw depth has increased substantially, a reduction in summer soil moisture eventually occurs ( (Kitabata et al., 2006 ) Stendel and Christensen 2002 [JoC, ARC] ) show poleward movement of permafrost extent, and a 30 to 40% increase in active layer thickness for most of the permafrost area in the NH, with the largest relative increases concentrated in the northernmost locations.

Regionally, the changes are a response to both increased temperature and increased precipitation (changes in circulation patterns) and are complicated by the competing effects of warming and increased snowfall in those regions that remain below freezing (see Section 4.2 for a further discussion of processes that affect snow cover). In general, snow amount and snow coverage decreases in the NH (Supplementary Material, Figure S10.1). However, in a few regions (e.g., Siberia), snow amount is projected to increase. This is attributed to the increase in precipitation (snowfall) from autumn to winter ( (Meleshko et al., 2004; ) Hosaka et al., 2005 [MoS, SRC] ).

10.3.3.3 Changes in Greenland Ice Sheet Mass Balance

As noted in Section 10.6 ,modelling studies (e.g., Hanna et al., 2002 [MoS, SRC] ; Kiilsholm et al., 2003 [JoC, MoS, SRC] ; Wild et al., 2003 [JoC, SRC] ) as well as satellite observations, airborne altimeter surveys and other studies ( Abdalati et al., 2001 [JoC] ; Thomas et al., 2001 [JoC, ARC] ; Krabill et al., 2004 [JoC] ; Johannessen et al., 2005 [JoC] ;( Zwally et al., 2005; ) Rignot and Kanagaratnam, 2006 [JoC, ARC] ) suggest a slight inland thickening and strong marginal thinning resulting in an overall negative Greenland Ice Sheet mass balance which has accelerated recently (see Section 4.6.2.2 .). A consistent feature of all climate models is that projected 21st-century warming is amplified in northern latitudes. This suggests continued melting of the Greenland Ice Sheet, since increased summer melting dominates over increased winter precipitation in model projections of future climate. Ridley et al. 2005 [JoC, SRC] ) coupled UKMO-HadCM3 to an ice sheet model to explore the melting of the Greenland Ice Sheet under elevated (four times pre-industrial) levels of atmospheric CO₂ (see Section 10.7.4.3 , Figure 10.38 ). While the entire Greenland Ice Sheet eventually completely ablated (after 3 kyr), the peak rate of melting was 0.06 Sv (1 Sv = 10⁶ m³ s^–1)corresponding to about 5.5 mm yr^–1 global sea level rise (see Sections 10.3.4 and 10.6.6). Toniazzo et al. 2004 [JoC, SRC] ) further show that in UKMO-HadCM3, the complete melting of the Greenland Ice sheet is an irreversible process even if pre-industrial levels of atmospheric CO₂ are re-established after it melts.

10.3.5.1 Interannual Variability in Surface Air Temperature and Precipitation

Future changes in anthropogenic forcing will result not only in changes in the mean climate state but also in the variability of climate. Addressing the interannual variability in monthly mean surface air temperature and precipitation of 19 AOGCMs in CMIP2, Räisänen 2002 [JoC, SRC] ) finds a decrease in temperature variability during the cold season in the extratropical NH and a slight increase in temperature variability in low latitudes and in warm season northern mid-latitudes. The former is likely due to the decrease of sea ice and snow with increasing temperature. The summer decrease in soil moisture over the mid-latitude land surfaces contributes to the latter. Räisänen 2002 [JoC, SRC] ) also finds an increase in monthly mean precipitation variability in most areas, both in absolute value (standard deviation) and in relative value (coefficient of variation). However, the significance level of these variability changes is markedly lower than that for time mean climate change. Similar results were obtained from 18 AOGCM simulations under the SRES A2 scenario ( Giorgi and Bi, 2005 [JoC, MoS, ARC] ).

10.3.5.2 Monsoons

In the tropics, an increase in precipitation is projected by the end of the 21st century in the Asian monsoon and the southern part of the West African monsoon with some decreases in the Sahel in northern summer ( Cook and Vizy, 2006 [JoC, MoS] ), as well as increases in the Australian monsoon in southern summer in a warmer climate( Figure 10.9 ). The monsoonal precipitation in Mexico and Central America is projected to decrease in association with increasing precipitation over the eastern equatorial Pacific that affects Walker Circulation and local Hadley Circulation changes( Figure 10.9 ). A more detailed assessment of regional monsoon changes is provided in Chapter 11 .

As a projected global warming will be more rapid over land than over the oceans, the continental-scale land-sea thermal contrast will become larger in summer and smaller in winter. Based on this, a simple idea is that the summer monsoon will be stronger and the winter monsoon will be weaker in the future than the present. However, model results are not as straightforward as this simple consideration. Tanaka et al. 2005 [MoS] ) define the intensities of Hadley, Walker and monsoon circulations using the velocity potential fields at 200 hPa. Using 15 AOGCMs, they show a weakening of these tropical circulations by 9%, 8% and 14%, respectively, by the late 21st century compared to the late 20th century. Using eight AOGCMs, Ueda et al. 2006 [JoC, MoS] ) demonstrate that pronounced warming over the tropics results in a weakening of the Asian summer monsoon circulations in relation to a reduction in the meridional thermal gradients between the Asian continent and adjacent oceans.

Despite weakening of the dynamical monsoon circulation, atmospheric moisture buildup due to increased greenhouse gases and consequent temperature increase results in a larger moisture flux and more precipitation for the Indian monsoon ( Douville et al., 2000 [MoS] ; IPCC, 2001 [NPR] ; Ashrit et al., 2003 [MoS] ; Meehl and Arblaster, 2003 [JoC, MoS, SRC] ; May, 2004; Ashrit et al., 2005 [MoS, SRC] ). For the South Asian summer monsoon, models suggest a northward shift of lower-tropospheric monsoon wind systems with a weakening of the westerly flow over the northern Indian Ocean ( Ashrit et al., 2003 [MoS] , 2005 ). Over Africa in northern summer, multi-model analysis projects an increase in rainfall in East and Central Africa, a decrease in the Sahel, and increases along the Gulf of Guinea coast( Figure 10.9 ). However, some individual models project an increase of rainfall in more extensive areas of West Africa related to a projected northward movement of the Sahara and the Sahel ( Liu et al., 2002 [JoC, MoS, SRC] ; Haarsma et al., 2005 [JoC] ). Whether the Sahel will be more or less wet in the future is thus uncertain, although a multi-model assessment of the West African monsoon indicates that the Sahel could become marginally more dry ( Cook and Vizy, 2006 [JoC, MoS] ). This inconsistency of the rainfall projections may be related to AOGCM biases, or an unclear relationship between Gulf of Guinea and Indian Ocean warming, land use change and the West African monsoon. Nonlinear feedbacks that may exist within the West African climate system should also be considered ( Jenkins et al., 2005 [JoC, MoS, SRC] ).

Most model results project increased interannual variability in season-averaged Asian monsoon precipitation associated with an increase in its long-term mean value (e.g., Hu et al., 2000b [JoC, MoS, SRC] ; Räisänen, 2002 [JoC, SRC] ; Meehl and Arblaster, 2003 [JoC, MoS, SRC] Hu et al. 2000a [JoC] ) relate this to increased variability in the tropical Pacific SST (El Niño variability) in their model. Meehl and Arblaster 2003 [JoC, MoS, SRC] ) relate the increased monsoon precipitation variability to increased variability in evaporation and precipitation in the Pacific due to increased SSTs. Thus, the South Asian monsoon variability is affected through the Walker Circulation such that the role of the Pacific Ocean dominates and that of the Indian Ocean is secondary.

Atmospheric aerosol loading affects regional climate and its future changes (see Chapter7 ). If the direct effect of the aerosol increase is considered, surface temperatures will not get as warm because the aerosols reflect solar radiation. For this reason, land-sea temperature contrast becomes smaller than in the case without the direct aerosol effect, and the summer monsoon becomes weaker. Model simulations of the Asian monsoon project that the sulphate aerosols’ direct effect reduces the magnitude of precipitation change compared with the case of only greenhouse gas increases ( Emori et al., 1999 [MoS, SRC] ; Roeckner et al., 1999 [JoC, MoS, SRC] ;( Lal and Singh, 2001 ) ). However, the relative cooling effect of sulphate aerosols is dominated by the effects of increasing greenhouse gases by the end of the 21st century in the SRES marker scenarios( Figure 10.26 ), leading to the increased monsoon precipitation at the end of the 21st century in these scenarios (see Section 10.3.2.3 ). Furthermore, it is suggested that aerosols with high absorptivity such as black carbon absorb solar radiation in the lower atmosphere, cool the surface, stabilise the atmosphere and reduce precipitation ( Ramanathan et al., 2001 [JoC, ARC] ). The solar radiation reaching the surface decreases as much as 50% locally, which could reduce the surface warming by greenhouse gases ( Ramanathan et al., 2005 [JoC, ARC] ). These atmospheric brown clouds could cause precipitation to increase over the Indian Ocean in winter and decrease in the surrounding Indonesia region and the western Pacific Ocean ( Chung et al., 2002 [Ambiguous] ), and could reduce the summer monsoon precipitation in South and East Asia ( Menon et al., 2002 [JoC, SRC] ; Ramanathan et al., 2005 [JoC, ARC] ). However, the total influence on monsoon precipitation of temporally varying direct and indirect effects of various aerosol species is still not resolved and the subject of active research.

10.3.5.3 Mean Tropical Pacific Climate Change

This subsection assesses changes in mean tropical Pacific climate. Enhanced greenhouse gas concentrations result in a general increase in SST, which will not be spatially uniform in association with a general reduction in tropical circulations in a warmer climate (see Section 10.3.5.2 ). Figures 10.8 and 10.9 indicate that SST increases more over the eastern tropical Pacific than over the western tropical Pacific, together with a decrease in the sea level pressure (SLP) gradient along the equator and an eastward shift of the tropical Pacific rainfall distribution. These background tropical Pacific changes can be called an El Niño-like mean state change (upon which individual El Niño-Southern Oscillation (ENSO) events occur). Although individual models show a large scatter of ‘ENSO-ness’ (Collins and The CMIP Modelling Groups, 2005; Yamaguchi and Noda, 2006 [SRC] ), an ENSO-like global warming pattern with positive polarity (i.e., El Niño-like mean state change) is simulated based on the spatial anomaly patterns of SST, SLP and precipitation( Figure 10.16 ;Yamaguchi and Noda, 2006 [SRC] ). The El Niño-like change may be attributable to the general reduction in tropical circulations resulting from the increased dry static stability in the tropics in a warmer climate ( Knutson and Manabe, 1995 [JoC, MoS, SRC] ; Sugi et al., 2002 [MoS, SRC] ; Figure 10.7 ). An eastward displacement of precipitation in the tropical Pacific accompanies an intensified and south-westward displaced subtropical anticyclone in the western Pacific, which can be effective in transporting moisture from the low latitudes to the Meiyu/Baiu region, thus generating more precipitation in the East Asian summer monsoon ( Kitoh and Uchiyama, 2006 [MoS, SRC] ).

In summary, the multi-model mean projects a weak shift towards conditions which may be described as ‘El Niño-like’, with SSTs in the central and eastern equatorial Pacific warming more than those in the west, and with an eastward shift in mean precipitation, associated with weaker tropical circulations.

Figure 10.16. Base state change in average tropical Pacific SSTs and change in El Niño variability simulated by AOGCMs (see Table 8.1 for model details). The base state change (horizontal axis) is denoted by the spatial anomaly pattern correlation coefficient between the linear trend of SST in the 1% yr^–1 CO₂ increase climate change experiment and the first Empirical Orthogonal Function (EOF) of SST in the control experiment over the area 10°S to 10°N, 120°E to 80°W (reproduced from Yamaguchi and Noda, 2006 [SRC] ). Positive correlation values indicate that the mean climate change has an El Niño-like pattern, and negative values are la Niña-like. The change in El Niño variability (vertical axis) is denoted by the ratio of the standard deviation of the first EOF of sea level pressure (SLP) between the current climate and the last 50 years of the SRES A2 experiments ( 2051 – 2100 ), except for FGOALS-g1.0 and MIROC3.2(hires), for which the SRES A1B was used, and UKMO-HadGEM1 for which the 1% yr^–1 CO₂ increase climate change experiment was used, in the region 30°S to 30°N, 30°E to 60°W with a five-month running mean (reproduced from van Oldenborgh et al., 2005 [Ambiguous] ). Error bars indicate the 95% confidence interval. Note that tropical Pacific base state climate changes with either El Niño-like or La Niña-like patterns are not permanent El Niño or La Niña events, and all still have ENSO inter- annual variability superimposed on that new average climate state in a future warmer climate.

10.3.5.4 El Niño

This subsection addresses the projected change in the amplitude, frequency and spatial pattern of El Niño. Guilyardi 2006 [JoC, MoS] ) assessed mean state, coupling strength and modes (SST mode resulting from local SST-wind interaction or thermocline mode resulting from remote wind-thermocline feedbacks), using the pre-industrial control and stabilised 2 × and 4 × atmospheric CO₂ simulations in a multi-model ensemble. The models that exhibit the largest El Niño amplitude change in scenario experiments are those that shift towards a thermocline mode. The observed 1976 climate shift in the tropical Pacific actually involved such a mode shift ( Fedorov and Philander, 2001 [JoC] ). The mean state change, through change in the sensitivity of SST variability to surface wind stress, plays a key role in determining the ENSO variance characteristics (Z. Hu et al., 2004 [Ambiguous] ; Zelle et al., 2005 [JoC, MoS] ). For example, a more stable ENSO system is less sensitive to changes in the background state than one that is closer to instability ( Zelle et al., 2005 [JoC, MoS] ). Thus, GCMs with an improper simulation of present-day climate mean state and air-sea coupling strength are not suitable for ENSO amplitude projections. van Oldenborgh et al. 2005 [Ambiguous] ) calculate the change in ENSO variability by the ratio of the standard deviation of the first Empirical Orthogonal Function (EOF) of SLP between the current climate and in the future( Figure 10.16 ), which shows that changes in ENSO interannual variability differ from model to model. They categorised 19 models based on their skill in the present-day ENSO simulations. Using the most realistic 6 out of 19 models, they find no statistically significant changes in the amplitude of ENSO variability in the future. Large uncertainty in the skewness of the variability limits the assessment of the future relative strength of El Niño and la Niña events. Merryfield 2006 [JoC, MoS] ) also analysed a multi-model ensemble and finds a wide range of behaviour for future El Niño amplitude, ranging from little change to larger El Niño events to smaller El Niño events, although several models that simulated some observed aspects of present-day El Niño events showed future increases in El Niño amplitude. However, significant multi-decadal fluctuations in El Niño amplitude in observations and in long coupled model control runs add another complicating factor to attempting to discern whether any future changes in El Niño amplitude are due to external forcing or are simply a manifestation of internal multi-decadal variability ( Meehl et al., 2006a [JoC, MoS, SRC] ). Even with the larger warming scenario under 4 × atmospheric CO₂ climate, Yeh and Kirtman 2005 [JoC, ARC] ) find that despite the large changes in the tropical Pacific mean state, the changes in ENSO amplitude are highly model dependent. Therefore, there are no clear indications at this time regarding future changes in El Niño amplitude in a warmer climate. However, as first noted in the TAR, ENSO teleconnections over North America appear to weaken due at least in part to the mean change of base state mid-latitude atmospheric circulation ( Meehl et al., 2006a [JoC, MoS, SRC] ).

In summary, all models show continued ENSO interannual variability in the future no matter what the change in average background conditions, but changes in ENSO interannual variability differ from model to model. Based on various assessments of the current multi-model archive, in which present-day El Niño events are now much better simulated than in the TAR, there is no consistent indication at this time of discernible future changes in ENSO amplitude or frequency.

10.3.5.5 ENSO-Monsoon Relationship

The El Niño-Southern Oscillation affects interannual variability throughout the tropics through changes in the Walker Circulation. Analysis of observational data finds a significant correlation between ENSO and tropical circulation and precipitation such that there is a tendency for less Indian summer monsoon rainfall in El Niño years and above normal rainfall in La Niña years. Recent analyses have revealed that the correlation between ENSO and the Indian summer monsoon has decreased recently, and many hypotheses have been put forward (see Chapter3 ). With respect to global warming, one hypothesis is that the Walker Circulation (accompanying ENSO) shifted south-eastward, reducing downward motion in the Indian monsoon region, which originally suppressed precipitation in that region at the time of El Niño, but now produces normal precipitation as a result ( Krishna Kumar et al., 1999 [JoC] ). Another explanation is that as the ground temperature of the Eurasian continent has risen in the winter-spring season, the temperature difference between the continent and the ocean has increased, thereby causing more precipitation, and the Indian monsoon is normal in spite of the occurrence of El Niño ( Ashrit et al., 2001 [JoC] ).

An earlier version of an AOGCM developed at the Max Planck Institute (MPI) ( Ashrit et al., 2001 [JoC] ) and the Action de Recherche Petite Echelle Grande Echelle/Océan Parallélisé (ARPEGE/OPA) model ( Ashrit et al., 2003 [MoS] ) simulated no global-warming related change in the ENSO-monsoon relationship, although a decadal-scale fluctuation is seen, suggesting that a weakening of the relationship might be part of the natural variability. However, Ashrit et al. 2001 [JoC] ) show that while the impact of La Niña does not change, the influence of El Niño on the monsoon becomes small, suggesting the possibility of asymmetric behaviour of the changes in the ENSO-monsoon relationship. On the other hand, the MRI-CGCM2 (see Table 8.1 for model details) indicates a weakening of the correlation into the 21st century, particularly after 2050 ( Ashrit et al., 2005 [MoS, SRC] ). The MRI-CGCM2 model results support the above hypothesis that the Walker Circulation shifts eastward and no longer influences India at the time of El Niño in a warmer climate. Camberlin et al. 2004 [JoC, MoS] and van Oldenborgh and Burgers 2005 [JoC] ) find decadal fluctuations in the effect of ENSO on regional precipitation. In most cases, these fluctuations may reflect natural variability in the ENSO teleconnection, and long-term correlation trends may be comparatively weaker.

The Tropospheric Biennial Oscillation (TBO) has been suggested as a fundamental set of coupled interactions in the Indo-Pacific region that encompasses ENSO and the Asian-Australian monsoon, and the TBO has been shown to be simulated by current AOGCMs (see Chapter8 Nanjundiah et al. 2005 [MoS, ARC] ) analyse a multi-model data set to show that, for models that successfully simulate the TBO for present-day climate, the TBO becomes more prominent in a future warmer climate due to changes in the base state climate, although, as with ENSO, there is considerable inherent decadal variability in the relative dominance of TBO and ENSO.

In summary, the ENSO-monsoon relationship can vary due to natural variability. Model projections suggest that a future weakening of the ENSO-monsoon relationship could occur in a future warmer climate.

10.3.5.6 Annular Modes and Mid-Latitude Circulation Changes

Many simulations project some decrease in the arctic surface pressure in the 21st century, as seen in the multi-model average (see Figure 10.9 ). This contributes to an increase in indices of the Northern Annular Mode (NAM) or the Arctic Oscillation (AO), as well as the NAO, which is closely related to the NAM in the Atlantic sector (see Chapter8 ). In the recent multi-model analyses, more than half of the models exhibit a positive trend in the NAM ( Rauthe et al., 2004 [JoC, MoS] ; Miller et al., 2006 [PoC, JoC, MoS, SRC] ) and/or NAO ( Osborn, 2004 [PoC, JoC, MoS, ARC] ; Kuzmina et al., 2005 [JoC, MoS] ). Although the magnitude of the trends shows a large variation among different models, Miller et al. 2006 [PoC, JoC, MoS, SRC] ) find that none of the 14 models exhibits a trend towards a lower NAM index and higher arctic SLP. In another multi-model analysis, Stephenson et al. 2006 [JoC, MoS] ) show that of the 15 models able to simulate the NAO pressure dipole, 13 predict a positive increase in the NAO index with increasing CO₂ concentrations, although the magnitude of the response is generally small and model dependent. However, the multi-model average from the larger number (21) of models shown in Figure 10.9 indicates that it is likely that the NAM index would not notably decrease in a future warmer climate. The average of IPCC-AR4 simulations from 13 models suggests the increase of the NAM index becomes statistically significant early in the 21st century( Figure 10.17 a, Miller et al., 2006 [PoC, JoC, MoS, SRC] ).

Figure 10.17. (a) Multi-model mean of the regression of the leading EOF of ensemble mean Northern Hemisphere sea level pressure (NH SLP, thin red line). The time series of regression coefficients has zero mean between year 1900 and 1970 . The thick red line is a 10-year low-pass filtered version of the mean. The grey shading represents the inter-model spread at the 95% confidence level and is filtered. A filtered version of the observed SLP from the Hadley Centre (HadSLP1) is shown in black. The regression coefficient for the winter following a major tropical eruption is marked by red, blue and black triangles for the multi-model mean, the individual model mean and observations, respectively. (b) As in (a) for Southern Hemisphere SLP for models with (red) and without (blue) ozone forcing. Adapted from Miller et al. 2006 [PoC, JoC, MoS, SRC] ).

The spatial patterns of the simulated SLP trends vary among different models, in spite of close correlations of the models’ leading patterns of interannual (or internal) variability with the observations ( Osborn, 2004 [PoC, JoC, MoS, ARC] ; Miller et al., 2006 [PoC, JoC, MoS, SRC] ). However, at the hemispheric scale of SLP change, the reduction in the Arctic is seen in the multi-model mean( Figure 10.9 ), although the change is smaller than the inter-model standard deviation. Besides the decrease in the arctic region, increases over the North Pacific and the Mediterranean Sea exceed the inter-model standard deviation; the latter suggests an association with a north-eastward shift of the NAO’s centre of action ( Hu and Wu, 2004 [JoC, MoS] ). The diversity of the patterns seems to reflect different responses in the Aleutian Low ( Rauthe et al., 2004 [JoC, MoS] ) in the North Pacific. Yamaguchi and Noda 2006 [SRC] ) discuss the modelled response of ENSO versus AO, and find that many models project a positive AO-like change. In the North Pacific at high latitudes, however, the SLP anomalies are incompatible between the El Niño-like change and the positive AO-like change, because models that project an El Niño-like change over the Pacific simulate a non-AO-like pattern in the polar region. As a result, the present models cannot fully determine the relative importance of the mechanisms inducing the positive AO-like change and those inducing the ENSO-like change, leading to scatter in global warming patterns at regional scales over the North Pacific. Rauthe et al. 2004 [JoC, MoS] ) suggest that the effects of sulphate aerosols contribute to a deepening of the Aleutian Low resulting in a slower or smaller increase in the AO index.

Analyses of results from various models indicate that the NAM can respond to increasing greenhouse gas concentrations through tropospheric processes ( Fyfe et al., 1999 [JoC, ARC] ; Gillett et al., 2003 [Ambiguous] ; Miller et al., 2006 [PoC, JoC, MoS, SRC] ). Greenhouse gases can also drive a positive NAM trend through changes in the stratospheric circulation, similar to the mechanism by which volcanic aerosols in the stratosphere force positive annular changes ( Shindell et al., 2001 [PoC, JoC, MoS, SRC] ). Models with their upper boundaries extending farther into the stratosphere exhibit, on average, a relatively larger increase in the NAM index and respond consistently to the observed volcanic forcing( Figure 10.17 a, Miller et al., 2006 [PoC, JoC, MoS, SRC] ), implying the importance of the connection between the troposphere and the stratosphere.

Aplausible explanation for the cause of the upward NAM trend simulated by the models is an intensification of the polar vortex resulting from both tropospheric warming and stratospheric cooling mainly due to the increase in greenhouse gases ( Shindell et al., 2001 [PoC, JoC, MoS, SRC] ; Sigmond et al., 2004 [JoC, MoS, ARC] ; Rind et al., 2005a [JoC, ARC] ). The response may not be linear with the magnitude of radiative forcing ( Gillett et al., 2002 [JoC, ARC] ) since the polar vortex response is attributable to an equatorward refraction of planetary waves ( Eichelberger and Holton, 2002 [JoC, MoS] ) rather than radiative forcing itself. Since the long-term variation in the NAO is closely related to SST variations ( Rodwell et al., 1999 [PoC, JoC, MoS, ARC] ), it is considered essential that the projection of the changes in the tropical SST ( Hoerling et al., 2004 [JoC] ; Hurrell et al., 2004 [JoC, ARC] ) and/or meridional gradient of the SST change ( Rind et al., 2005b [JoC, ARC] ) is reliable.

The future trend in the Southern Annular Mode (SAM) or the Antarctic Oscillation (AAO) has been projected in a number of model simulations ( Gillett and Thompson, 2003 [JoC, MoS, ARC] ; Shindell and Schmidt, 2004 [PoC, JoC, SRC] ; Arblaster and Meehl, 2006 [JoC, MoS, SRC] ; Miller et al., 2006 [PoC, JoC, MoS, SRC] ). According to the latest multi-model analysis ( Miller et al., 2006 [PoC, JoC, MoS, SRC] ), most models indicate a positive trend in the SAM index, and a declining trend in the antarctic SLP (as seen in Figure 10.9 ), with a higher likelihood than for the future NAM trend. On average, a larger positive trend is projected during the late 20th century by models that include stratospheric ozone changes than those that do not( Figure 10.17 b), although during the 21st century, when ozone changes are smaller, the SAM trends of models with and without ozone are similar. The cause of the positive SAM trend in the second half of the 20th century is mainly attributed to stratospheric ozone depletion, evidenced by the fact that the signal is largest in the lower stratosphere in austral spring through summer ( Thompson and Solomon, 2002 [PoC, JoC, ARC] ; Arblaster and Meehl, 2006 [JoC, MoS, SRC] ). However, increases in greenhouse gases are also important factors ( Shindell and Schmidt, 2004 [PoC, JoC, SRC] ; Arblaster and Meehl, 2006 [JoC, MoS, SRC] ) for the year-round positive SAM trend induced by meridional temperature gradient changes ( Brandefelt and Källén, 2004 [JoC, MoS] ). During the 21st century, although the ozone amount is expected to stabilise or recover, the polar vortex intensification is likely to continue due to the increases in greenhouse gases ( Arblaster and Meehl, 2006 [JoC, MoS, SRC] ).

It is implied that the future change in the annular modes leads to modifications of the future change in various fields such as surface temperatures, precipitation and sea ice with regional features similar to those for the modes of natural variability (e.g., Hurrell et al., 2003 [NPR, ARC] ). For instance, the surface warming in winter would be intensified in northern Eurasia and most of North America while weakened in the western North Atlantic, and winter precipitation would increase in northern Europe while decreasing in southern Europe. The atmospheric circulation change would also affect the ocean circulations. Sakamoto et al. 2005 [JoC, MoS] ) simulate an intensification of the Kuroshio Current but no shift in the Kuroshio Extension in response to an AO-like circulation change for the 21st century. However, Sato et al. 2006 [MoS] ) simulate a northward shift of the Kuroshio Extension, which leads to a strong warming off the eastern coast of Japan.

In summary, the future changes in the extratropical circulation variability are likely to be characterised by increases in positive phases of both the NAM and the SAM. The response in the NAM to anthropogenic forcing might not be distinct from the larger multi-decadal internal variability in the first half of the 21st century. The change in the SAM would appear earlier than in the NAM since stratospheric ozone depletion acts as an additional forcing. The positive trends in annular modes would influence the regional changes in temperature, precipitation and other fields, similar to those that accompany the NAM and the SAM in the present climate, but would be superimposed on the global-scale changes in a future warmer climate.

10.3.6.1 Precipitation Extremes

Along-standing result from global coupled models noted in the TAR is a projected increase in the chance of summer drying in the mid-latitudes in a future warmer climate with associated increased risk of drought. This is shown in Figure 10.12 ,and has been documented in the more recent generation of models ( Burke et al., 2006 [JoC, MoS] ; Meehl et al., 2006b [JoC, MoS, SRC] ; Rowell and Jones, 2006 [JoC, MoS, ARC] ). For example, Wang 2005 [JoC, MoS] ) analyse 15 recent AOGCMs and show that in a future warmer climate, the models simulate summer dryness in most parts of the northern subtropics and mid-latitudes, but with a large range in the amplitude of summer dryness across models. Droughts associated with this summer drying could result in regional vegetation die-offs ( Breshears et al., 2005 [JoC] ) and contribute to an increase in the percentage of land area experiencing drought at any one time, for example, extreme drought increasing from 1% of present-day land area to 30% by the end of the century in the A2 scenario ( Burke et al., 2006 [JoC, MoS] ). Drier soil conditions can also contribute to more severe heat waves as discussed in Section 10.3.6.2 ( Brabson et al., 2005 [JoC, MoS, ARC] ).

Associated with the risk of drying is a projected increase in the chance of intense precipitation and flooding. Although somewhat counter-intuitive, this is because precipitation is projected to be concentrated into more intense events, with longer periods of little precipitation in between. Therefore, intense and heavy episodic rainfall events with high runoff amounts are interspersed with longer relatively dry periods with increased evapotranspiration, particularly in the subtropics as discussed in Section 10.3.6.2 in relation to Figure 10.19 ( Frei et al., 1998 [JoC, ARC] ; Allen and Ingram, 2002 [JoC, MoS, SRC] ; Palmer and Räisänen, 2002 [JoC, SRC] ; Christensen and Christensen, 2003 [JoC, ARC] ; Beniston, 2004 [JoC, MoS, ARC] ; Christensen and Christensen, 2004 [JoC, ARC] ; Pal et al., 2004 [JoC, MoS, ARC] ; Meehl et al., 2005a [JoC, MoS, SRC] ). However, increases in the frequency of dry days do not necessarily mean a decrease in the frequency of extreme high rainfall events depending on the threshold used to define such events ( Barnett et al., 2006 [JoC, MoS] ). Another aspect of these changes has been related to the mean changes in precipitation, with wet extremes becoming more severe in many areas where mean precipitation increases, and dry extremes where the mean precipitation decreases ( Kharin and Zwiers, 2005 [JoC, MoS, ARC] ; Meehl et al., 2005a [JoC, MoS, SRC] ; Räisänen, 2005a [JoC, SRC] ; Barnett et al., 2006 [JoC, MoS] ). However, analysis of the 53-member perturbed physics ensemble indicates that the change in the frequency of extreme precipitation at an individual location can be difficult to estimate definitively due to model parametrization uncertainty ( Barnett et al., 2006 [JoC, MoS] ). Some specific regional aspects of these changes in precipitation extremes are discussed further in Chapter 11 .

Climate models continue to confirm the earlier results that in a future climate warmed by increasing greenhouse gases, precipitation intensity (e.g., proportionately more precipitation per precipitation event) is projected to increase over most regions ( Wilby and Wigley, 2002 [PoC, JoC, MoS, SRC] ; Kharin and Zwiers, 2005 [JoC, MoS, ARC] ; Meehl et al., 2005a [JoC, MoS, SRC] ; Barnett et al., 2006 [JoC, MoS] ), and the increase in precipitation extremes is greater than changes in mean precipitation ( Kharin and Zwiers, 2005 [JoC, MoS, ARC] ). As discussed in Chapter9 ,this is related to the fact that the energy budget of the atmosphere constrains increases in large-scale mean precipitation, but extreme precipitation relates to increases in moisture content and thus the nonlinearities involved with the Clausius-Clapeyron relationship such that, for a given increase in temperature, increases in extreme precipitation can be more than the mean precipitation increase (e.g., Allen and Ingram, 2002 [JoC, MoS, SRC] ). Additionally, time scale can play a role whereby increases in the frequency of seasonal mean rainfall extremes can be greater than the increases in the frequency of daily extremes ( Barnett et al., 2006 [JoC, MoS] ). The increase in mean and extreme precipitation in various regions has been attributed to contributions from both dynamic and thermodynamic processes associated with global warming ( Emori and Brown, 2005 [JoC, SRC] ). The greater increase in extreme precipitation compared to the mean is attributed to the greater thermodynamic effect on the extremes due to increases in water vapour, mainly over subtropical areas. The thermodynamic effect is important nearly everywhere, but changes in circulation also contribute to the pattern of precipitation intensity changes at middle and high latitudes ( Meehl et al., 2005a [JoC, MoS, SRC] Kharin and Zwiers 2005 [JoC, MoS, ARC] ) show that changes in both the location and scale of the extreme value distribution produce increases in precipitation extremes substantially greater than increases in annual mean precipitation. An increase in the scale parameter from the gamma distribution represents an increase in precipitation intensity, and various regions such as the NH land areas in winter showed particularly high values of increased scale parameter ( Semenov and Bengtsson, 2002 [JoC, MoS] ; Watterson and Dix, 2003 [JoC, MoS, SRC] ). Time-slice simulations with a higher-resolution model (~1°) show similar results using changes in the gamma distribution, namely increased extremes in the hydrological cycle ( Voss et al., 2002 [JoC, MoS, SRC] ). However, some regional decreases are also projected such as over the subtropical oceans ( Semenov and Bengtsson, 2002 [JoC, MoS] ).

Anumber of studies have noted the connection between increased rainfall intensity and an implied increase in flooding. McCabe et al. 2001 [JoC] and Watterson 2005 [MoS, SRC] ) show a projected increase in extreme rainfall intensity with the extra-tropical surface lows, particularly over NH land, with an implied increase in flooding. In a multi-model analysis of the CMIP models, Palmer and Räisänen 2002 [JoC, SRC] ) show an increased likelihood of very wet winters over much of central and northern Europe due to an increase in intense precipitation associated with mid-latitude storms, suggesting more floods across Europe (see also Chapter 11 ). They found similar results for summer precipitation with implications for greater flooding in the Asian monsoon region in a future warmer climate. Similarly, Milly et al. 2002 [JoC, ARC] Arora and Boer 2001 [JoC, MoS, ARC] and Voss et al. 2002 [JoC, MoS, SRC] ) relate the increased risk of floods in a number of major river basins in a future warmer climate to an increase in spring river discharge related to increased winter snow depth in some regions. Christensen and Christensen 2003 [JoC, ARC] ) conclude that there could be an increased risk of summer flooding in Europe.

Globally averaged time series of the ( Frich et al. 2002 ) ) indices in the multi-model analysis of Tebaldi et al. 2006 [JoC, MoS, SRC] ) show simulated increases in precipitation intensity during the 20th century continuing through the 21st century( Figure 10.18 a,b), along with a somewhat weaker and less consistent trend of increasing dry periods between rainfall events for all scenarios( Figure 10.18 c,d). Part of the reason for these results is shown in the geographic maps for these quantities, where precipitation intensity increases almost everywhere, but particularly at middle and high latitudes where mean precipitation also increases ( Meehl et al., 2005a [JoC, MoS, SRC] ; compare Figure 10.18 bto Figure 10.9 ). However, in Figure 10.18 d, there are regions of increased runs of dry days between precipitation events in the subtropics and lower mid-latitudes, but decreased runs of dry days at higher mid-latitudes and high latitudes where mean precipitation increases (compare Figure 10.9 with Figure 10.18 d). Since there are areas of both increases and decreases in consecutive dry days between precipitation events in the multi-model average( Figure 10.9 ), the global mean trends are smaller and less consistent across models as shown in Figure 10.18 .Consistency of response in a perturbed physics ensemble with one model shows only limited areas of increased frequency of wet days in July, and a larger range of changes in precipitation extremes relative to the control ensemble mean in contrast to the more consistent response of temperature extremes( Section 10.6.3.2 ), indicating a less consistent response for precipitation extremes in general compared to temperature extremes ( Barnett et al., 2006 [JoC, MoS] ). Analysis of the ( Frich et al. 2002 ) ) precipitation indices in a 20-km resolution global model shows similar results to those in Figure 10.18 ,with particularly large increases in precipitation intensity in South Asia and West Africa ( Kamiguchi et al., 2005 [MoS] ).

Figure 10.18. Changes in extremes based on multi-model simulations from nine global coupled climate models, adapted from Tebaldi et al. 2006 [JoC, MoS, SRC] ). (a) Globally averaged changes in precipitation intensity (defined as the annual total precipitation divided by the number of wet days) for a low (SRES B1), middle (SRES A1B) and high (SRES A2) scenario. (b) Changes in spatial patterns of simulated precipitation intensity between two 20-year means ( 2080 – 2099 minus 1980 – 1999 ) for the A1B scenario. (c) Globally averaged changes in dry days (defined as the annual maximum number of consecutive dry days). (d) Changes in spatial patterns of simulated dry days between two 20-year means ( 2080 – 2099 minus 1980 – 1999 ) for the A1B scenario. Solid lines in (a) and (c) are the 10-year smoothed multi-model ensemble means; the envelope indicates the ensemble mean standard deviation. Stippling in (b) and (d) denotes areas where at least five of the nine models concur in determining that the change is statistically significant. Extreme indices are calculated only over land following ( Frich et al. 2002 ) ). Each model’s time series was centred on its 1980 to 1999 average and normalised (rescaled) by its standard deviation computed (after de-trending) over the period 1960 to 2099 . The models were then aggregated into an ensemble average, both at the global and at the grid-box level. Thus, changes are given in units of standard deviations.

10.3.6.2 Temperature Extremes

The TAR concluded that there was a very likely risk of increased high temperature extremes (and reduced risk of low temperature extremes) with more extreme heat episodes in a future climate. The latter result has been confirmed in subsequent studies ( Yonetani and Gordon, 2001 [JoC, MoS] Kharin and Zwiers 2005 [JoC, MoS, ARC] ) show in a single model that future increases in temperature extremes follow increases in mean temperature over most of the world except where surface properties change (melting snow, drying soil). Furthermore, they show that in most instances warm extremes correspond to increases in daily maximum temperature, but cold extremes warm up faster than daily minimum temperatures, although this result is less consistent when model parameters are varied in a perturbed physics ensemble where there are increased daily temperature maxima for nearly the entire land surface. However, the range in magnitude of increases was substantial indicating a sensitivity to model formulations ( Clark et al., 2006 [JoC, MoS, SRC] ).

Weisheimer and Palmer 2005 [JoC, SRC] ) examine changes in extreme seasonal (DJF and JJA) temperatures in 14 models for three scenarios. They show that by the end of 21st century, the probability of such extreme warm seasons is projected to rise in many areas. This result is consistent with the perturbed physics ensemble where, for nearly all land areas, extreme JJA temperatures were at least 20 times and in some areas 100 times more frequent compared to the control ensemble mean, making these changes greater than the ensemble spread.

Since the TAR, possible future cold air outbreaks have been studied. Vavrus et al. 2006 [JoC, ARC] ) analyse seven AOGCMs run with the A1B scenario, and define a cold air outbreak as two or more consecutive days when the daily temperatures are at least two standard deviations below the present-day winter mean. For a future warmer climate, they document a 50 to 100% decline in the frequency of cold air outbreaks in NH winter in most areas compared to the present, with the smallest reductions occurring in western North America, the North Atlantic and southern Europe and Asia due to atmospheric circulation changes associated with the increase in greenhouse gases.

No studies at the time of the TAR specifically documented changes in heat waves (very high temperatures over a sustained period of days, see Chapter3 ). Several recent studies address possible future changes in heat waves explicitly, and find an increased risk of more intense, longer-lasting and more frequent heat waves in a future climate ( Meehl and Tebaldi, 2004 [JoC, SRC] ; Schär et al., 2004 [JoC, ARC] ; Clark et al., 2006 [JoC, MoS, SRC] Meehl and Tebaldi 2004 [JoC, SRC] ) show that the pattern of future changes in heat waves, with greatest intensity increases over western Europe, the Mediterranean and the southeast and western USA, is related in part to base state circulation changes due to the increase in greenhouse gases. An additional factor leading to extreme heat is drier soils in a future warmer climate ( Brabson et al., 2005 [JoC, MoS, ARC] ; Clark et al., 2006 [JoC, MoS, SRC] Schär et al. 2004 [JoC, ARC] Stott et al. 2004 [JoC, SRC] and Beniston 2004 [JoC, MoS, ARC] ) use the European 2003 heat wave as an example of the types of heat waves that are likely to become more common in a future warmer climate. Schär et al. 2004 [JoC, ARC] ) note that the increase in the frequency of extreme warm conditions is also associated with a change in interannual variability, such that the statistical distribution of mean summer temperatures is not merely shifted towards warmer conditions but also becomes wider. A multi-model ensemble shows that heat waves are simulated to have been increasing over the latter part of the 20th century, and are projected to increase globally and over most regions( Figure 10.19 ;Tebaldi et al., 2006 [JoC, MoS, SRC] ), although different model parameters can contribute to the range in the magnitude of this response ( Clark et al., 2006 [JoC, MoS, SRC] ).

Adecrease in DTR in most regions in a future warmer climate was reported in the TAR, and is substantiated by more recent studies (e.g., Stone and Weaver, 2002 [JoC, MoS, SRC] ; also discussed in relation to Figure 10.11 band in Chapter 11 ). For a quantity related to the DTR, the TAR concluded that it would be likely that a future warmer climate would also be characterised by a decrease in the number of frost days, although there were no studies at that time from global coupled climate models that addressed this issue explicitly. It has since been shown that there would indeed be decreases in frost days in a future warmer climate in the extratropics ( Meehl et al., 2004a [JoC, MoS, SRC] ), with the pattern of the decreases dictated by the changes in atmospheric circulation due to the increase in greenhouse gases ( Meehl et al., 2004a [JoC, MoS, SRC] ). Results from a nine-member multi-model ensemble show simulated decreases in frost days for the 20th century continuing into the 21st century globally and in most regions( Figure 10.19 ). A quantity related to frost days in many mid- and high-latitude areas, particularly in the NH, is growing season length as defined by ( Frich et al. 2002 ) ), and this has been projected to increase in future climate ( Tebaldi et al., 2006 [JoC, MoS, SRC] ). This result is also shown in a nine-member multi-model ensemble where the simulated increase in growing season length in the 20th century continues into the 21st century globally and in most regions( Figure 10.19 ). The globally averaged extremes indices in Figures 10.18 and 10.19 have non-uniform changes across the scenarios compared to the more consistent relative increases in Figure 10.5 for globally averaged temperature. This indicates that patterns that scale well by radiative forcing for temperature (e.g., Figure 10.8 )would not scale for extremes.

Figure 10.19. Changes in extremes based on multi-model simulations from nine global coupled climate models, adapted from Tebaldi et al. 2006 [JoC, MoS, SRC] ). (a) Globally averaged changes in the frost day index (defined as the total number of days in a year with absolute minimum temperature below 0°C) for a low (SRES B1), middle (SRES A1B) and high (SRES A2) scenario. (b) Changes in spatial patterns of simulated frost days between two 20-year means ( 2080 – 2099 minus 1980 – 1999 ) for the A1B scenario. (c) Globally averaged changes in heat waves (defined as the longest period in the year of at least five consecutive days with maximum temperature at least 5°C higher than the climatology of the same calendar day). (d) Changes in spatial patterns of simulated heat waves between two 20-year means ( 2080 – 2099 minus 1980 – 1999 ) for the A1B scenario. (e) Globally averaged changes in growing season length (defined as the length of the period between the first spell of five consecutive days with mean temperature above 5°C and the last such spell of the year). (f) Changes in spatial patterns of simulated growing season length between two 20-year means ( 2080 – 2099 minus 1980 – 1999 ) for the A1B scenario. Solid lines in (a), (c) and (e) show the 10-year smoothed multi-model ensemble means; the envelope indicates the ensemble mean standard deviation. Stippling in (b), (d) and (f) denotes areas where at least five of the nine models concur in determining that the change is statistically significant. Extreme indices are calculated only over land. Frost days and growing season are only calculated in the extratropics. Extremes indices are calculated following ( Frich et al. 2002 ) ). Each model’s time series was centred around its 1980 to 1999 average and normalised (rescaled) by its standard deviation computed (after de-trending) over the period 1960 to 2099 . The models were then aggregated into an ensemble average, both at the global and at the grid-box level. Thus, changes are given in units of standard deviations.

10.3.6.3 Tropical Cyclones (Hurricanes)

Earlier studies assessed in the TAR showed that future tropical cyclones would likely become more severe with greater wind speeds and more intense precipitation. More recent modelling experiments have addressed possible changes in tropical cyclones in a warmer climate and generally confirmed those earlier results. These studies fall into two categories: those with model grid resolutions that only roughly represent some aspects of individual tropical cyclones, and those with model grids of sufficient resolution to reasonably simulate individual tropical cyclones.

In the first category, a number of climate change experiments with global models have started to simulate some characteristics of individual tropical cyclones, although classes of models with 50 to 100 km resolution or lower cannot accurately simulate observed tropical cyclone intensities due to the limitations of the relatively coarse grid spacing (e.g., Yoshimura et al., 2006 [SRC] ). A study with roughly 100-km grid spacing shows a decrease in tropical cyclone frequency globally and in the North Pacific but a regional increase over the North Atlantic and no significant changes in maximum intensity ( Sugi et al., 2002 [MoS, SRC] Yoshimura et al. 2006 [SRC] ) use the same model but different SST patterns and two different convection schemes, and show a decrease in the global frequency of relatively weak tropical cyclones but no significant change in the frequency of intense storms. They also show that the regional changes are dependent on the SST pattern, and precipitation near the storm centres could increase in the future. Another study using a 50 km resolution model confirms this dependence on SST pattern, and also shows a consistent increase in precipitation intensity in future tropical cyclones ( Chauvin et al., 2006 [JoC, MoS] ). Another global modelling study with roughly a 100-km grid spacing finds a 6% decrease in tropical storms globally and a slight increase in intensity, with both increases and decreases regionally related to the El Niño-like base state response in the tropical Pacific to increased greenhouse gases ( McDonald et al., 2005 [JoC, MoS] ). Another study with the same resolution model indicates decreases in tropical cyclone frequency and intensity but more mean and extreme precipitation from the tropical cyclones simulated in the future in the western north Pacific ( Hasegawa and Emori, 2005 [MoS, SRC] ). An AOGCM analysis with a coarser-resolution atmospheric model (T63, or about 200-km grid spacing) shows little change in overall numbers of tropical storms in that model, but a slight decrease in medium-intensity storms in a warmer climate ( Bengtsson et al., 2006 [JoC, SRC] ). In a global warming simulation with a coarse-resolution atmospheric model (T42, or about 300-km grid spacing), the frequency of global tropical cyclone occurrence did not change significantly, but the mean intensity of the global tropical cyclones increased significantly ( (Tsutsui, 2002 ) ). Thus, from this category of coarser-grid models that can only represent rudimentary aspects of tropical cyclones, there is no consistent evidence for large changes in either frequency or intensity of these models’ representation of tropical cyclones, but there is a consistent response of more intense precipitation from future storms in a warmer climate. Also note that the decreasing tropical precipitation in future climate in Yoshimura et al. 2006 [SRC] ) is for SSTs held fixed as atmospheric CO₂ is increased, a situation that does not occur in any global coupled model.

In the second category, studies have been performed with models that have been able to credibly simulate many aspects of tropical cyclones. For example, Knutson and Tuleya 2004 [JoC, MoS, SRC] ) use a high-resolution (down to 9 km) mesoscale hurricane model to simulate hurricanes with intensities reaching about 60 to 70 ms^–1,depending on the treatment of moist convection in the model. They use mean tropical conditions from nine global climate models with increased CO₂ to simulate tropical cyclones with 14% more intense central pressure falls, 6% higher maximum surface wind speeds and about 20% greater near-storm rainfall after an idealised 80-year buildup of CO₂ at 1% yr^–1 compounded (warming given by TCR shown for models in Chapter8 ). Using a multiple nesting technique, an AOGCM was used to force a regional model over Australasia and the western Pacific with 125-km grid resolution, with an embedded 30-km resolution model over the south-western Pacific ( (Walsh et al., 2004 ) ). At that 30-km resolution, the model is able to closely simulate the climatology of the observed tropical cyclone lower wind speed threshold of 17 ms^–1.Tropical cyclone occurrence (in terms of days of tropical cyclone activity) is slightly greater than observed, and the somewhat weaker than observed pressure gradients near the storm centres are associated with lower than observed maximum wind speeds, likely due to the 30-km grid spacing that is too coarse to capture extreme pressure gradients and winds. For 3 × atmospheric CO₂ in that model configuration, the simulated tropical cyclones experienced a 56% increase in the number of storms with maximum wind speed greater than 30 ms^–1 and a 26% increase in the number of storms with central pressures less than 970 hPa, with no large changes in frequency and movement of tropical cyclones for that southwest Pacific region. It should also be noted that ENSO fluctuations have a strong impact on patterns of tropical cyclone occurrence in the southern Pacific ( Nguyen and Walsh, 2001 [JoC, MoS] ), and that uncertainty with respect future ENSO behaviour( Section 10.3.5.1 )contributes to uncertainty with respect to tropical cyclones ( (Walsh, 2004 ) ).

In another experiment with a high resolution global model that is able to generate tropical cyclones that begin to approximate real storms, a global 20-km grid atmospheric model was run in time slice experiments for a present-day 10-year period and a 10-year period at the end of the 21st century for the A1B scenario to examine changes in tropical cyclones. Observed climatological SSTs were used to force the atmospheric model for the 10-year period at the end of the 20th century, time-mean SST anomalies from an AOGCM simulation for the future climate were added to the observed SSTs and atmospheric composition was changed in the model to be consistent with the A1B scenario. At that resolution, tropical cyclone characteristics, numbers and tracks were relatively well simulated for present-day climate, although simulated wind speed intensities were somewhat weaker than observed intensities ( Oouchi et al., 2006 [MoS] ). In that study, tropical cyclone frequency decreased 30% globally (but increased about 34% in the North Atlantic). The strongest tropical cyclones with extreme surface winds increased in number while weaker storms decreased. The tracks were not appreciably altered, and maximum peak wind speeds in future simulated tropical cyclones increased by about 14% in that model, although statistically significant increases were not found in all basins. As noted above, the competing effects of greater stabilisation of the tropical troposphere (less storms) and greater SSTs (the storms that form are more intense) likely contribute to these changes except for the tropical North Atlantic where there are greater SST increases than in the other basins in that model. Therefore, the SST warming has a greater effect than the vertical stabilisation in the Atlantic and produces not only more storms but also more intense storms there. However, these regional changes are largely dependent on the spatial pattern of future simulated SST changes ( Yoshimura et al., 2006 [SRC] ).

Sugi et al. 2002 [MoS, SRC] ) show that the global-scale reduction in tropical cyclone frequency is closely related to weakening of tropospheric circulation in the tropics in terms of vertical mass flux. They note that a significant increase in dry static stability in the tropical troposphere and little increase in tropical precipitation (or convective heating) are the main factors contributing to the weakening of the tropospheric circulation. Sugi and Yoshimura 2004 [JoC, SRC] ) investigate a mechanism of this tropical precipitation change. They show that the effect of CO₂ enhancement (without changing SST conditions, which is not realistic as noted above) is a decrease in mean precipitation ( Sugi and Yoshimura, 2004 [JoC, SRC] ) and a decrease in the number of tropical cyclones as simulated in an atmospheric model with about 100 km resolution ( Yoshimura and Sugi, 2005 [SRC] ). Future changes in the large-scale steering flow as a mechanism to deduce possible changes in tropical cyclone tracks in the western North Pacific ( Wu and Wang, 2004 [JoC, SRC] ) were analysed to show different shifts at different times in future climate change experiments along with a dependence of such shifts on the degree of El Niño-like mean climate change in the Pacific (see Section 10.3.5 ).

Asynthesis of the model results to date indicates that, for a future warmer climate, coarse-resolution models show few consistent changes in tropical cyclones, with results dependent on the model, although those models do show a consistent increase in precipitation intensity in future storms. Higher-resolution models that more credibly simulate tropical cyclones project some consistent increase in peak wind intensities, but a more consistent projected increase in mean and peak precipitation intensities in future tropical cyclones. There is also a less certain possibility of a decrease in the number of relatively weak tropical cyclones, increased numbers of intense tropical cyclones and a global decrease in total numbers of tropical cyclones.

10.3.6.4 Extratropical Storms and Ocean Wave Height

The TAR noted that there could be a future tendency for more intense extratropical storms, although the number of storms could be less. A more consistent result that has emerged more recently, in agreement with earlier results (e.g., Schubert et al., 1998 [JoC, MoS] ), is a tendency for a poleward shift of several degrees latitude in mid-latitude storm tracks in both hemispheres ( Geng and Sugi, 2003 [JoC] ; Fischer-Bruns et al., 2005 [JoC, MoS, ARC] ; Yin, 2005 [JoC, MoS] ; Bengtsson et al., 2006 [JoC, SRC] ). Consistent with these shifts in storm track activity, Cassano et al. 2006 [JoC, ARC] ), using a 10-member multi-model ensemble, show a future change to a more cyclonically dominated circulation pattern in winter and summer over the Arctic, and increasing cyclonicity and stronger westerlies in the same multi-model ensemble for the Antarctic ( Lynch et al., 2006 [JoC, ARC] ).

Some studies have shown little change in extratropical cyclone characteristics ( Kharin and Zwiers, 2005 [JoC, MoS, ARC] ; Watterson, 2005 [MoS, SRC] ). But a regional study showed a tendency towards more intense systems, particularly in the A2 scenario in another global coupled climate model analysis ( Leckebusch and Ulbrich, 2004 [JoC] ), with more extreme wind events in association with those deepened cyclones for several regions of Western Europe, with similar changes in the B2 simulation although less pronounced in amplitude. Geng and Sugi 2003 [JoC] ) use a higher-resolution (about 100 km resolution) atmospheric GCM (AGCM) with time-slice experiments and find a decrease in cyclone density (number of cyclones in a 4.5° by 4.5° area per season) in the mid-latitudes of both hemispheres in a warmer climate in both the DJF and JJA seasons, associated with the changes in the baroclinicity in the lower troposphere, in general agreement with earlier results and coarser GCM results (e.g., Dai et al., 2001b [JoC, MoS, ARC] ). They also find that the density of strong cyclones increases while the density of weak and medium-strength cyclones decreases. Several studies have shown a possible reduction in mid-latitude storms in the NH but a decrease in central pressures in these storms ( Lambert and Fyfe, 2006 [JoC, MoS, ARC] , for a 15-member multi-model ensemble) and in the SH ( Fyfe, 2003 [JoC, ARC] , with a possible 30% reduction in sub-antarctic cyclones). The latter two studies did not definitively identify a poleward shift of storm tracks, but their methodologies used a relatively coarse grid that may not have been able to detect shifts of several degrees latitude and they used only identification of central pressures which could imply an identification of semi-permanent features like the sub-antarctic trough. More regional aspects of these changes were addressed for the NH in a single model study by Inatsu and Kimoto 2005 [SRC] ), who show a more active storm track in the western Pacific in the future but weaker elsewhere. Fischer-Bruns et al. 2005 [JoC, MoS, ARC] ) document storm activity increasing over the North Atlantic and Southern Ocean and decreasing over the Pacific Ocean.

By analysing stratosphere-troposphere exchanges using time-slice experiments with the middle atmosphere version of ECHAM4, Land and Feichter 2003 [JoC, MoS, ARC] ) suggest that cyclonic and blocking activity becomes weaker poleward of 30°N in a warmer climate at least in part due to decreased baroclinicity below 400 hPa, while cyclonic activity becomes stronger in the SH associated with increased baroclinicity above 400 hPa. The atmospheric circulation variability on inter-decadal time scales may also change due to increasing greenhouse gases and aerosols. One model result ( Hu et al., 2001 [JoC] ) showed that inter-decadal variability of the SLP and 500 hPa height fields increased over the tropics and decreased at high latitudes due to global warming.

In summary, the most consistent results from the majority of the current generation of models show, for a future warmer climate, a poleward shift of storm tracks in both hemispheres that is particularly evident in the SH, with greater storm activity at higher latitudes.

Anew feature that has been studied related to extreme conditions over the oceans is wave height. Studies by Wang et al. 2004 [JoC, ARC] Wang and Swail 2006a [NPR, MoS] ,b) and Caires et al. 2006 [JoC, MoS] ) have shown that for many regions of the mid-latitude oceans, an increase in extreme wave height is likely to occur in a future warmer climate. This is related to increased wind speed associated with mid-latitude storms, resulting in higher waves produced by these storms, and is consistent with the studies noted above that showed decreased numbers of mid-latitude storms but more intense storms.

10.5.2.1 Comprehensive AOGCMs

The way a climate model responds to changes in external forcing, such as an increase in anthropogenic greenhouse gases, is characterised by two standard measures: (1) ‘equilibrium climate sensitivity’ (the equilibrium change in global surface temperature following a doubling of the atmospheric equivalent CO₂ concentration; see Glossary), and (2) ‘transient climate response’ (the change in global surface temperature in a global coupled climate model in a 1% yr^–1 CO₂ increase experiment at the time of atmospheric CO₂ doubling; see Glossary). The first measure provides an indication of feedbacks mainly residing in the atmospheric model but also in the land surface and sea ice components, and the latter quantifies the response of the fully coupled climate system including aspects of transient ocean heat uptake (e.g., Sokolov et al., 2003 [JoC, SRC] ). These two measures have become standard for quantifying how an AOGCM will react to more complicated forcings in scenario simulations.

Historically, the equilibrium climate sensitivity has been given in the range from 1.5°C to 4.5°C. This range was reported in the TAR with no indication of a probability distribution within this range. However, considerable recent work has addressed the range of equilibrium climate sensitivity, and attempted to assign probabilities to climate sensitivity.

Equilibrium climate sensitivity and TCR are not independent( Figure 10.25 a). For a given AOGCM, the TCR is smaller than the equilibrium climate sensitivity because ocean heat uptake delays the atmospheric warming. A large ensemble of the BERN2.5D EMIC has been used to explore the relationship of TCR and equilibrium sensitivity over a wide range of ocean heat uptake parametrizations ( Knutti et al., 2005 [JoC, MoS, SRC] ). Good agreement with the available results from AOGCMs is found, and the BERN2.5D EMIC covers almost the entire range of structurally different models. The percent change in precipitation is closely related to the equilibrium climate sensitivity for the current generation of AOGCMs( Figure 10.25 b), with values from the current models falling within the range of the models from the TAR. Figure 10.25 cshows the percent change in globally averaged precipitation as a function of TCR at the time of atmospheric CO₂ doubling, as simulated by 1% yr^–1 transient CO₂ increase experiments with AOGCMs. The figure suggests a broadly positive correlation between these two quantities similar to that for equilibrium climate sensitivity, with these values from the new models also falling within the range of the previous generation of AOGCMs assessed in the TAR. Note that the apparent relationships may not hold for other forcings or at smaller scales. Values for an ensemble with perturbations made to parameters in the atmospheric component of UKMO-HadCM3 (M. Collins et al., 2006 [Ambiguous] ) cover similar ranges and are shown in Figure 10.25 for comparison.

Figure 10.25. (a) TCR versus equilibrium climate sensitivity for all AOGCMs (red), EMICs (blue), a perturbed physics ensemble of the UKMO-HadCM3 AOGCM (green; an updated ensemble based on M. Collins et al., 2006 [Ambiguous] ) and from a large ensemble of the Bern2.5D EMIC ( Knutti et al., 2005 [JoC, MoS, SRC] ) using different ocean vertical diffusivities and mixing parametrizations (grey lines). (b) Global mean precipitation change (%) as a function of global mean temperature change at equilibrium for doubled CO₂ in atmospheric GCMs coupled to a non-dynamic slab ocean (red all AOGCMS, green from a perturbed physics ensemble of the atmosphere-slab ocean version of UKMO-HadCM3 ( Webb et al., 2006 [JoC, ARC] )). (c) Global mean precipitation change (%) as a function of global mean temperature change (TCR) at the time of CO₂ doubling in a transient 1% yr^–1 CO₂ increase scenario, simulated by coupled AOGCMs (red) and the UKMO-HadCM3 perturbed physics ensemble (green). Black crosses in (b) and (c) mark ranges covered by the TAR AOGCMs ( IPCC, 2001 [NPR] ) for each quantity.

Fitting normal distributions to the results, the 5 to 95% uncertainty range for equilibrium climate sensitivity from the AOGCMs is approximately 2.1°C to 4.4°C and that for TCR is 1.2°C to 2.4°C (using the method of Räisänen, 2005b [MoS, SRC] ). The mean for climate sensitivity is 3.26°C and that for TCR is 1.76°C. These numbers are practically the same for both the normal and the lognormal distribution (see Box 10.2 ). The assumption of a (log) normal fit is not well supported by the limited sample of AOGCM data. In addition, the AOGCMs represent an ‘ensemble of opportunity’ and are by design not sampled in a random way. However, most studies aiming to constrain climate sensitivity with observations do indeed indicate a similar to lognormal probability distribution of climate sensitivity and an approximately normal distribution of the uncertainty in future warming and thus TCR (see Box 10.2 ). Those studies also suggest that the current AOGCMs may not cover the full range of uncertainty for climate sensitivity. An assessment of all the evidence on equilibrium climate sensitivity is provided in Box 10.2 .The spread of the AOGCM climate sensitivities is discussed in Section 8.6 and the AOGCM values for climate sensitivity and TCR are listed in Table 8.2 .

The nonlinear relationship between TCR and equilibrium climate sensitivity shown in Figure 10.25 aalso indicates that on time scales well short of equilibrium, the model’s TCR is not particularly sensitive to the model’s climate sensitivity. The implication is that transient climate change is better constrained than the equilibrium climate sensitivity, that is, models with different sensitivity might still show good agreement for projections on decadal time scales. Therefore, in the absence of unusual solar or volcanic activity, climate change is well constrained for the coming few decades, because differences in some feedbacks will only become important on long time scales (see also Section 10.5.4.5 )and because over the next few decades, about half of the projected warming would occur as a result of radiative forcing being held constant at year 2000 levels (constant composition commitment, see Section 10.7 ).

Comparing observed thermal expansion with those AR4 20th-century simulations that have natural forcings indicates that ocean heat uptake in the models may be 25% larger than observed, although both could be consistent within their uncertainties. This difference is possibly due to a combination of overestimated ocean heat uptake in the models, observational uncertainties and limited data coverage in the deep ocean (see Sections 9.5.1.1 , 9.5.2 ,and 9.6.2.1 ). Assigning this difference solely to overestimated ocean heat uptake, the TCR estimates could increase by 0.6°C at most. This is in line with evidence for a relatively weak dependence of TCR on ocean mixing based on SCMs and EMICS ( Allen et al., 2000 [JoC, MoS, SRC] ; Knutti et al., 2005 [JoC, MoS, SRC] ). The range of TCR covered by an ensemble with perturbations made to parameters in the atmospheric component of UKMO-HadCM3 is 1.5 to 2.6°C (M. Collins et al., 2006 [Ambiguous] ), similar to the AR4 AOGCM range. Therefore, based on the range covered by AOGCMs, and taking into account structural uncertainties and possible biases in transient heat uptake, TCR is assessed as very likely larger than 1°C and very unlikely greater than 3°C (i.e., 1.0°C to 3.0°C is a 10 to 90% range). Because the dependence of TCR on sensitivity becomes small as sensitivity increases, uncertainties in the upper bound on sensitivity only weakly affect the range of TCR (see Figure 10.25 ; Chapter9 ;Knutti et al., 2005 [JoC, MoS, SRC] ; Allen et al., 2006b [NPR, SRC] ). Observational constraints based on detection and attribution studies provide further support for this TCR range (see Section 9.6.2.3 ).

10.5.2.2 Earth System Models of Intermediate Complexity

Over the last few years, a range of climate models has been developed that are dynamically simpler and of lower resolution than comprehensive AOGCMs, although they might well be more ‘complete’ in terms of climate system components that are included. The class of such models, usually referred to as EMICs ( Claussen et al., 2002 [JoC, MoS, ARC] ), is very heterogeneous, ranging from zonally averaged ocean models coupled to energy balance models ( Stocker et al., 1992a [JoC, MoS, SRC] ) or to statistical-dynamical models of the atmosphere ( Petoukhov et al., 2000 [JoC, MoS, ARC] ), to low resolution three-dimensional ocean models, coupled to energy balance or simple dynamical models of the atmosphere ( Opsteegh et al., 1998 [JoC, MoS] ; Edwards and Marsh, 2005 [JoC, MoS, SRC] ; Müller et al., 2006 [PoC, JoC, MoS, SRC] ). Some EMICs have a radiation code and prescribe greenhouse gases, while others use simplified equations to project radiative forcing from projected concentrations and abundances ( Joos et al., 2001 [PoC, JoC, SRC] ; see Chapter2 and the TAR, Appendix II, Table II.3.11). Compared to comprehensive models, EMICs have hardly any computational constraints, and therefore many simulations can be performed. This allows for the creation of large ensembles, or the systematic exploration of long-term changes many centuries hence. However, because of the reduced resolution, only results at the largest scales (continental to global) are to be interpreted ( Stocker and Knutti, 2003 [MoS, SRC] ). Table 8.3 lists all EMICs used in this section, including their components and resolution.

Aset of simulations is used to compare EMICs with AOGCMs for the SRES A1B scenario with stable atmospheric concentrations after year 2100 (see Section 10.7.2 ). For global mean temperature and sea level, the EMICs generally reproduce the AOGCM behaviour quite well. Two of the EMICs have values for climate sensitivity and transient response below the AOGCM range. However, climate sensitivity is a tuneable parameter in some EMICs, and no attempt was made here to match the range of response of the AOGCMs. The transient reduction of the MOC in most EMICs is also similar to the AOGCMs (see also Sections 10.3.4 and 10.7.2 and Figure 10.34 ), providing support that this class of models can be used for both long-term commitment projections (see Section 10.7 )and probabilistic projections involving hundreds to thousands of simulations (see Section 10.5.4.5 ). If the forcing is strong enough, and lasts long enough (e.g., 4 × CO₂), a complete and irreversible collapse of the MOC can be induced in a few models. This is in line with earlier results using EMICs ( Stocker and Schmittner, 1997 [JoC, SRC] ; Rahmstorf and Ganopolski, 1999 [PoC, JoC, MoS, ARC] ) or a coupled model ( Stouffer and Manabe, 1999 [JoC, MoS, SRC] ).

10.5.4.1 The Multi-Model Ensemble Approach

The use of ensembles of AOGCMs developed at different modelling centres has become established in climate prediction/ projection on both seasonal-to-interannual and centennial time scales. To the extent that simulation errors in different AOGCMs are independent, the mean of the ensemble can be expected to outperform individual ensemble members, thus providing an improved ‘best estimate’ forecast. Results show this to be the case, both in verification of seasonal forecasts ( Palmer et al., 2004 [JoC, MoS, SRC] ; Hagedorn et al., 2005 [JoC, MoS, SRC] ) and of the present-day climate from long term simulations ( Lambert and Boer, 2001 [JoC, MoS] ). By sampling modelling uncertainties, ensembles of AOGCMs should provide an improved basis for probabilistic projections compared with ensembles of a single model sampling only uncertainty in the initial state ( Palmer et al., 2005 [Ambiguous] ). However, members of a multi-model ensemble share common systematic errors ( Lambert and Boer, 2001 [JoC, MoS] ), and cannot span the full range of possible model configurations due to resource constraints. Verification of future climate change projections is not possible, however, Räisänen and Palmer 2001 [JoC, MoS, SRC] ) used a ‘perfect model approach’ (treating one member of an ensemble as truth and predicting its response using the other members) to show that the hypothetical economic costs associated with climate events can be reduced by calculating the probability of the event across the ensemble, rather than using a deterministic prediction from an individual ensemble member.

An additional strength of multi-model ensembles is that each member is subjected to careful testing in order to obtain a plausible and stable control simulation, although the process of tuning model parameters to achieve this( Section 8.1.3.1 )involves subjective judgement, and is not guaranteed to identify the optimum location in the model parameter space.

10.5.4.2 Perturbed Physics Ensembles

The AOGCMs featured in Section 10.5.2 are built by selecting components from a pool of alternative parametrizations, each based on a given set of physical assumptions and including a number of uncertain parameters. In principle, the range of predictions consistent with these components could be quantified by constructing very large ensembles with systematic sampling of multiple options for parametrization schemes and parameter values, while avoiding combinations likely to double-count the effect of perturbing a given physical process. Such an approach has been taken using simple climate models and EMICs ( Wigley and Raper, 2001 [PoC, JoC, MoS, SRC] ; Knutti et al., 2002 [Ambiguous] and Murphy et al. 2004 [JoC, MoS, SRC] and Stainforth et al. 2005 [JoC, MoS, SRC] ) describe the first steps in this direction using AOGCMs, constructing large ensembles by perturbing poorly constrained parameters in the atmospheric component of UKMO-HadCM3 coupled to a mixed layer ocean. These experiments quantify the range of equilibrium responses to doubled atmospheric CO₂ consistent with uncertain parameters in a single GCM. Murphy et al. 2004 [JoC, MoS, SRC] ) perturbed 29 parameters one at a time, assuming that effects of individual parameters were additive but making a simple allowance for additional uncertainty introduced by nonlinear interactions. They find a probability distribution for climate sensitivity with a 5 to 95% range of 2.4°C to 5.4°C when weighting the models with a broadly based metric of the agreement between simulated and observed climatology, compared to 1.9°C to 5.3°C when all model versions are assumed equally reliable( Box 10.2 ,Figure 1c).

Stainforth et al. 2005 [JoC, MoS, SRC] ) deployed a distributed computing approach ( Allen, 1999 [JoC, MoS, SRC] ) to run a very large ensemble of 2,578 simulations sampling combinations of high, intermediate and low values of six parameters known to affect climate sensitivity. They find climate sensitivities ranging from 2°C to 11°C, with 4.2% of model versions exceeding 8°C, and show that the high-sensitivity models cannot be ruled out, based on a comparison with surface annual mean climatology. By utilising multivariate linear relationships between climate sensitivity and spatial fields of several present-day observables, the 5 to 95% range of climate sensitivity is estimated at 2.2°C to 6.8°C from the same data set ( Piani et al., 2005 [JoC, MoS, SRC] ; Box 10.2 Figure 1c). In this ensemble, Knutti et al. 2006 [JoC, SRC] ) find a strong relationship between climate sensitivity and the amplitude of the seasonal cycle in surface temperature in the present-day simulations. Most of the simulations with high sensitivities overestimate the observed amplitude. Based on this relationship, the 5 to 95% range of climate sensitivity is estimated at 1.5°C to 6.4°C( Box 10.2 ,Figure 1c). The differences between the PDFs in Box 10.2 ,Figure 1c, which are all based on the same climate model, reflect uncertainties in methodology arising from choices of uncertain parameters, their expert-specified prior distributions and alternative applications of observational constraints. They do not account for uncertainties associated with changes in ocean circulation, and do not account for structural model errors ( Smith, 2002 [JoC, MoS] ; Goldstein and Rougier, 2004 [MoS] )

Annan et al. 2005a [JoC, MoS, SRC] ) use an ensemble Kalman Filter technique to obtain uncertainty ranges for model parameters in an EMIC subject to the constraint of minimising simulation errors with respect to a set of climatological observations. Using this method, Hargreaves and Annan 2006 [JoC, MoS, SRC] ) find that the risk of a collapse in the Atlantic MOC (in response to increasing CO₂)depends on the set of observations to which the EMIC parameters are tuned. Section 9.6.3 assesses perturbed physics studies of the link between climate sensitivity and cooling during the Last Glacial Maximum ( Annan et al., 2005b [MoS, SRC] ; Schneider von Deimling et al., 2006 [PoC, JoC, MoS, ARC] ).

10.5.4.3 Diagnosing Drivers of Uncertainty from Ensemble Results

Figure 10.27 ashows the agreement between annual changes simulated by members of the AR4 multi-model ensemble for 2080 to 2099 relative to 1980 to 1999 for the A1B scenario, calculated as in Räisänen 2001 [JoC, SRC] ). For precipitation, the agreement increases with spatial scale. For surface temperature, the agreement is high even at local scales, indicating the robustness of the simulated warming (see also Figure 10.8 ,discussed in Section 10.3.2.1 ). Differences in model formulation are the dominant contributor to ensemble spread, though the role of internal variability increases at smaller scales( Figure 10.27 b). The agreement between AR4 ensemble members is slightly higher compared with the earlier CMIP2 ensemble of Räisänen 2001 [JoC, SRC] ) (also reported in the TAR), and internal variability explains a smaller fraction of the ensemble spread. This is expected, given the larger forcing and responses in the A1B scenario for 2080 to 2099 compared to the transient response to doubled CO₂ considered by Räisänen 2001 [JoC, SRC] ), although the use of an updated set of models may also contribute. For seasonal changes, internal variability is found to be comparable with model differences as a source of uncertainty in local precipitation and SLP changes (although not for surface temperature) in both multi-model and perturbed physics ensembles ( Räisänen, 2001 [JoC, SRC] ; Murphy et al., 2004 [JoC, MoS, SRC] ). Consequently the local seasonal changes for precipitation and SLP are not consistent in the AR4 ensemble over large areas of the globe (i.e., the multi-model mean change does not exceed the ensemble standard deviation; see Figure 10.9 ), whereas the surface temperature changes are consistent almost everywhere, as discussed in Section 10.3.2.1 .

Figure 10.27. Statistics of annual mean responses to the SRES A1B scenario, for 2080 to 2099 relative to 1980 to 1999, calculated from the 21-member AR4 multi-model ensemble using the methodology of Räisänen 2001 [JoC, SRC] ). Results are expressed as a function of horizontal scale on the x axis (‘Loc’: grid box scale; ‘Hem’: hemispheric scale; ‘Glob’: global mean) plotted against the y axis showing (a) the relative agreement between ensemble members, a dimensionless quantity defined as the square of the ensemble-mean response (corrected to avoid sampling bias) divided by the mean squared response of individual ensemble members, and (b) the dimensionless fraction of internal variability relative to the ensemble variance of responses. Values are shown for surface air temperature, precipitation and sea level pressure. The low agreement of SLP changes at hemispheric and global scales reflects problems with the conservation of total atmospheric mass in some of the models, however, this has no practical significance because SLP changes at these scales are extremely small.

Wang and Swail 2006b [JoC, MoS] ) examine the relative importance of internal variability, differences in radiative forcing and model differences in explaining the transient response of ocean wave height using three AOGCMs each run for three plausible forcing scenarios, and find model differences to be the largest source of uncertainty in the simulated changes.

Selten et al. 2004 [JoC, MoS] ) report a 62-member initial condition ensemble of simulations of 1940 to 2080 including natural and anthropogenic forcings. They find an individual member that reproduces the observed trend in the NAO over the past few decades, but no trend in the ensemble mean, and suggest that the observed change can be explained through internal variability associated with a mode driven by increases in precipitation over the tropical Indian Ocean. Terray et al. 2004 [JoC, MoS, SRC] ) find that the ARPEGE coupled ocean-atmosphere model shows small increases in the residence frequency of the positive phase of the NAO in response to SRES A2 and B2 forcing, whereas larger increases are found when SST changes prescribed from the coupled experiments are used to drive a version of the atmosphere model with enhanced resolution over the North Atlantic and Europe ( Gibelin and Déqué, 2003 [JoC, MoS] ).

Figure 10.25 compares global mean transient and equilibrium changes simulated by the AR4 multi-model ensembles against perturbed physics ensembles (M. Collins et al., 2006 [Ambiguous] ; Webb et al., 2006 [JoC, ARC] ) designed to produce credible present-day simulations while sampling a wide range of multiple parameter perturbations and climate sensitivities. The AR4 ensembles partially sample structural variations in model components, whereas the perturbed physics ensembles sample atmospheric parameter uncertainties for a fixed choice of model structure. The results show similar relationships between TCR, climate sensitivity and precipitation change in both types of ensemble. The perturbed physics ensembles contain several members with sensitivities higher than the multi-model range, while some of the multi-model transient simulations give TCR values slightly below the range found in the perturbed physics ensemble( Figure 10.25 a,b).

Soden and Held 2006 [JoC, MoS, ARC] ) find that differences in cloud feedback are the dominant source of uncertainty in the transient response of surface temperature in the AR4 ensemble (see also Section 8.6.3.2 ), as in previous IPCC assessments. Webb et al. 2006 [JoC, ARC] ) compare equilibrium radiative feedbacks in a 9-member multi-model ensemble against those simulated in a 128-member perturbed physics ensemble with multiple parameter perturbations. They find that the ranges of climate sensitivity in both ensembles are explained mainly by differences in the response of shortwave cloud forcing in areas where changes in low-level clouds predominate. Bony and Dufresne 2005 [JoC, MoS, ARC] ) find that marine boundary layer clouds in areas of large-scale subsidence provide the largest source of spread in tropical cloud feedbacks in the AR4 ensemble. Narrowing the uncertainty in cloud feedback may require both improved parametrizations of cloud microphysical properties (e.g., Tsushima et al., 2006 [JoC, MoS] ) and improved representations of cloud macrophysical properties, through improved parametrizations of other physical processes (e.g., Williams et al., 2001 [JoC, MoS, ARC] ) and/or increases in resolution ( Palmer, 2005 [Ambiguous] ).

10.5.4.4 Observational Constraints

Arange of observables has been used since the TAR to explore methods for constraining uncertainties in future climate change in studies using simple climate models, EMICs and AOGCMs. Probabilistic estimates of global climate sensitivity have been obtained from the historical transient evolution of surface temperature, upper-air temperature, ocean temperature, estimates of the radiative forcing, satellite data, proxy data over the last millennium, or a subset thereof ( Wigley et al., 1997a [PoC, JoC, SRC] ; Tol and de Vos, 1998 [JoC] ; Andronova and Schlesinger, 2001 [JoC, MoS, ARC] ; Forest et al., 2002 [JoC, ARC] ; Gregory et al., 2002a [JoC, MoS, SRC] ; Knutti et al., 2002 [Ambiguous] , 2003; Frame et al., 2005 [JoC, MoS] ; Forest et al., 2006 [JoC, MoS, SRC] ; Forster and Gregory, 2006 [JoC, SRC] ; Hegerl et al., 2006 [PoC, JoC, SRC] ; see Section 9.6 ). Some of these studies also constrain the transient response to projected future emissions (see section 10.5.4.5 ). For climate sensitivity, further probabilistic estimates have been obtained using statistical measures of the correspondence between simulated and observed fields of present-day climate ( Murphy et al., 2004 [JoC, MoS, SRC] ; Piani et al., 2005 [JoC, MoS, SRC] ), the climatological seasonal cycle of surface temperature ( Knutti et al., 2006 [JoC, SRC] ) and the response to palaeoclimatic forcings ( Annan et al., 2005b [MoS, SRC] ; Schneider von Deimling et al., 2006 [PoC, JoC, MoS, ARC] ). For the purpose of constraining regional climate projections, spatial averages or fields of time-averaged regional climate have been used ( Giorgi and Mearns, 2003 [JoC, ARC] ; Tebaldi et al., 2004 [JoC, MoS, SRC] , 2005; Laurent and Cai, 2007 [JoC, 2007] ), as have past regional- or continental-scale trends in surface temperature ( Greene et al., 2006 [JoC, MoS] ; Stott et al., 2006a [JoC, MoS, SRC] ).

Further observables have been suggested as potential constraints on future changes, but are not yet used in formal probabilistic estimates. These include measures of climate variability related to cloud feedbacks ( Bony et al., 2004 [JoC, ARC] ; Bony and Dufresne, 2005 [JoC, MoS, ARC] ; Williams et al., 2005 [JoC, MoS] ), radiative damping of the seasonal cycle ( Tsushima et al., 2005 [JoC, MoS] ), the relative entropy of simulated and observed surface temperature variations ( Shukla et al., 2006 [JoC, MoS, ARC] ), major volcanic eruptions ( Wigley et al., 2005 [PoC, JoC, SRC] ; Yokohata et al., 2005 [JoC, MoS] ; see Section 9.6 )and trends in multiple variables derived from reanalysis data sets ( Lucarini and Russell, 2002 [JoC, MoS, SRC] ).

Additional constraints could also be found, for example, from evaluation of ensemble climate prediction systems on shorter time scales for which verification data exist. These could include assessment of the reliability of seasonal to interannual probabilistic forecasts ( Palmer et al., 2004 [JoC, MoS, SRC] ; Hagedorn et al., 2005 [JoC, MoS, SRC] ) and the evaluation of model parametrizations in short-range weather predictions ( Phillips et al., 2004 [JoC, MoS, ARC] ; Palmer, 2005 [Ambiguous] Annan and Hargreaves 2006 [JoC, MoS, SRC] ) point out the potential for narrowing uncertainty by combining multiple lines of evidence. This will require objective quantification of the impact of different constraints and their degree of independence, estimation of the effects of structural modelling errors and the development of comprehensive probabilistic frameworks in which to combine these elements (e.g., Rougier, 2007 [JoC, MoS, 2007] ).

10.5.4.5 Probabilistic Projections - Global Mean

Anumber of methods for providing probabilistic climate change projections, both for global means (discussed in this section) and geographical depictions (discussed in the following section) have emerged since the TAR.

Methods of constraining climate sensitivity using observations of present-day climate are discussed in Section 10.5.4.2 .Results from both the AR4 multi-model ensemble and from perturbed physics ensembles suggest a very low probability for a climate sensitivity below 2°C, despite exploring the effects of a wide range of alternative modelling assumptions on the global radiative feedbacks arising from lapse rate, water vapour, surface albedo and cloud ( Bony et al., 2006 [JoC, ARC] ; Soden and Held, 2006 [JoC, MoS, ARC] ; Webb et al., 2006 [JoC, ARC] ; Box 10.2 ). However, exclusive reliance on AOGCM ensembles can be questioned on the basis that models share components, and therefore errors, and may not sample the full range of possible outcomes (e.g., Allen and Ingram, 2002 [JoC, MoS, SRC] ).

Observationally constrained probability distributions for climate sensitivity have also been derived from physical relationships based on energy balance considerations, and from instrumental observations of historical changes during the past 50 to 150 years or proxy reconstructions of surface temperature during the past millennium( Section 9.6 ). The results vary according to the choice of verifying observations, the forcings considered and their specified uncertainties, however, all these studies report a high upper limit for climate sensitivity, with the 95th percentile of the distributions invariably exceeding 6°C( Box 10.2 Frame et al. 2005 [JoC, MoS] ) demonstrate that uncertainty ranges for sensitivity are dependent on the choices made about prior distributions of uncertain quantities before the observations are applied. Frame et al. 2005 [JoC, MoS] and Piani et al. 2005 [JoC, MoS, SRC] ) show that many observable variables are likely to scale inversely with climate sensitivity, implying that projections of quantities that are inversely related to sensitivity will be more strongly constrained by observations than climate sensitivity itself, particularly with respect to the estimated upper limit ( Allen et al., 2006b [NPR, SRC] ).

In the case of transient climate change, optimal detection techniques have been used to determine factors by which hindcasts of global surface temperature from AOGCMs can be scaled up or down while remaining consistent with past changes, accounting for uncertainty due to internal variability( Section 9.4.1.6 ). Uncertainty is propagated forward in time by assuming that the fractional error found in model hindcasts of global mean temperature change will remain constant in projections of future changes. Using this approach, Stott and Kettleborough 2002 [JoC, MoS, SRC] ) find that probabilistic projections of global mean temperature derived from UKMO-HadCM3 simulations were insensitive to differences between four representative SRES emissions scenarios over the first few decades of the 21st century, but that much larger differences emerged between the response to different SRES scenarios by the end of the 21st century (see also Section 10.5.3 and Figure 10.28 Stott et al. 2006b [JoC, MoS, SRC] ) show that scaling the responses of three models with different sensitivities brings their projections into better agreement. Stott et al. 2006a [JoC, MoS, SRC] ) extend their approach to obtain probabilistic projections of future warming averaged over continental-scale regions under the SRES A2 scenario. Fractional errors in the past continental warming simulated by UKMO-HadCM3 are used to scale future changes, yielding wide uncertainty ranges, notably for North America and Europe where the 5 to 95% ranges for warming during the 21st century are 2°C to 12°C and 2°C to 11°C respectively. These estimates do not account for potential constraints arising from regionally differentiated warming rates. Tighter ranges of 4°C to 8°C for North America and 4°C to 7°C for Europe are obtained if fractional errors in past global mean temperature are used to scale the future continental changes, although this neglects uncertainty in the relationship between global and regional temperature changes.

Allen and Ingram 2002 [JoC, MoS, SRC] ) suggest that probabilistic projections for some variables may be made by searching for ‘emergent constraints’. These are relationships between variables that can be directly constrained by observations, such as global surface temperature, and variables that may be indirectly constrained by establishing a consistent, physically based relationship which holds across a wide range of models. They present an example in which future changes in global mean precipitation are constrained using a probability distribution for global temperature obtained from a large EMIC ensemble ( Forest et al., 2002 [JoC, ARC] ) and a relationship between precipitation and temperature obtained from multi-model ensembles of the response to doubled atmospheric CO₂.These methods are designed to produce distributions constrained by observations, and are relatively model independent ( Allen and Stainforth, 2002 [JoC, MoS, SRC] ; Allen et al., 2006a [NPR, MoS, SRC] ). This can be achieved provided the inter-variable relationships are robust to alternative modelling assumptions Piani et al. 2005 [JoC, MoS, SRC] and Knutti et al. 2006 [JoC, SRC] ) (described in Section 10.5.4.2 )follow this approach, noting that in these cases the inter-variable relationships are derived from perturbed versions of a single model, and need to be confirmed using other models.

Asynthesis of published probabilistic global mean projections for the SRES scenarios B1, A1B and A2 is given in Figure 10.28 .Probability density functions are given for short-term projections ( 2020 – 2030 ) and the end of the century ( 2090 – 2100 ). For comparison, normal distributions fitted to results from AOGCMs in the multi-model archive (see Section 10.3.1 )are also given, although these curve fits should not be regarded as PDFs. The five methods of producing PDFs are all based on different models and/or techniques, described in Section 10.5 .In short, Wigley and Raper 2001 [PoC, JoC, MoS, SRC] ) use a large ensemble of a simple model with expert prior distributions for climate sensitivity, ocean heat uptake, sulphate forcing and the carbon cycle, without applying constraints. Knutti et al. 2002 [Ambiguous] , 2003 ) use a large ensemble of EMIC simulations with non-informative prior distributions, consider uncertainties in climate sensitivity, ocean heat uptake, radiative forcing and the carbon cycle, and apply observational constraints. Neither method considers natural variability explicitly. Stott et al. 2006b [JoC, MoS, SRC] ) apply the fingerprint scaling method to AOGCM simulations to obtain PDFs which implicitly account for uncertainties in forcing, climate sensitivity and internal unforced as well as forced natural variability. For the A2 scenario, results obtained from three different AOGCMs are shown, illustrating the extent to which the Stott et al. PDFs depend on the model used. Harris et al. 2006 [JoC, MoS, SRC] ) obtain PDFs by boosting a 17-member perturbed physics ensemble of the UKMO-HadCM3 model using scaled equilibrium responses from a larger ensemble of simulations. Furrer et al. 2007 [NPR, MoS, SRC, 2007] ) use a Bayesian method described in Section 10.5.4.7 to calculate PDFs from the AR4 multi-model ensemble. The Stott et al. 2006b [JoC, MoS, SRC] Harris et al. 2006 [JoC, MoS, SRC] and Furrer et al. 2007 [NPR, MoS, SRC, 2007] ) methods neglect carbon cycle uncertainties.

Figure 10.28. Probability density functions from different studies for global mean temperature change for the SRES scenarios B1, A1B and A2 and for the decades 2020 to 2029 and 2090 to 2099 relative to the 1980 to 1999 average ( Wigley and Raper, 2001 [PoC, JoC, MoS, SRC] ; Knutti et al., 2002 [Ambiguous] ; Furrer et al., 2007 [NPR, MoS, SRC, 2007] ; Harris et al., 2006 [JoC, MoS, SRC] ; Stott et al., 2006b [JoC, MoS, SRC] ). A normal distribution fitted to the multi-model ensemble is shown for comparison.

Two key points emerge from Figure 10.28 .For the projected short-term warming (i) there is more agreement among models and methods (narrow width of the PDFs) compared to later in the century (wider PDFs), and (ii) the warming is similar across different scenarios, compared to later in the century where the choice of scenario significantly affects the projections. These conclusions are consistent with the results obtained with SCMs( Section 10.5.3 ).

Additionally, projection uncertainties increase close to linearly with temperature in most studies. The different methods show relatively good agreement in the shape and width of the PDFs, but with some offsets due to different methodological choices. Only Stott et al. 2006b [JoC, MoS, SRC] ) account for variations in future natural forcing, and hence project a small probability of cooling over the next few decades not seen in the other PDFs. The results of Knutti et al. 2003 [PoC, JoC, MoS, SRC] ) show wider PDFs for the end of the century because they sample uniformly in climate sensitivity (see Section 9.6.2 and Box 10.2 ). Resampling uniformly in observables ( Frame et al., 2005 [JoC, MoS] ) would bring their PDFs closer to the others. In sum, probabilistic estimates of uncertainties for the next few decades seem robust across a variety of models and methods, while results for the end of the century depend on the assumptions made.

10.5.4.6 Synthesis of Projected Global Temperature at Year 2100

All available estimates for projected warming by the end of the 21st century are summarised in Figure 10.29 for the six SRES non-intervention marker scenarios. Among the various techniques, the AR4 AOGCM ensemble provides the most sophisticated set of models in terms of the range of processes included and consequent realism of the simulations compared to observations (see Chapters 8 and 9 ). On average, this ensemble projects an increase in global mean surface air temperature of 1.8°C, 2.8°C and 3.4°C in the B1, A1B and A2 scenarios, respectively, by 2090 to 2099 relative to 1980 to 1999 (note that in Table 10.5 ,the years 2080 to 2099 were used for those globally averaged values to be consistent with the comparable averaging period for the geographic plots in Section 10.3 ;this longer averaging period smoothes spatial noise in the geographic plots). A scaling method is used to estimate AOGCM mean results for the three missing scenarios B2, A1T and A1FI. The ratio of the AOGCM mean values for B1 relative to A1B and A2 relative to A1B are almost identical to the ratios obtained with the MAGICC SCM, although the absolute values for the SCM are higher. Thus, the AOGCM mean response for the scenarios B2, A1T and A1FI can be estimated as 2.4°C, 2.4°C and 4.0°C by multiplying the AOGCM A1B mean by the SCM-derived ratios B2/A1B, A1T/A1B and A1FI/A1B, respectively (for details see Appendix 10 .A.1).

The AOGCMs cannot sample the full range of possible warming, in particular because they do not include uncertainties in the carbon cycle. In addition to the range derived directly from the AR4 multi-model ensemble, Figure 10.29 depicts additional uncertainty estimates obtained from published probabilistic methods using different types of models and observational constraints: the MAGICC SCM and the BERN2.5CC coupled climate-carbon cycle EMIC tuned to different climate sensitivities and carbon cycle settings, and theC⁴MIP coupled climate-carbon cycle models. Based on these results, the future increase in global mean temperature is likely to fall within –40 to +60% of the multi-model AOGCM mean warming simulated for each scenario. This range results from an expert judgement of the multiple lines of evidence presented in Figure 10.29 ,and assumes that the models approximately capture the range of uncertainties in the carbon cycle. The range is well constrained at the lower bound since climate sensitivity is better constrained at the low end (see Box 10.2 ), and carbon cycle uncertainty only weakly affects the lower bound. The upper bound is less certain as there is more variation across the different models and methods, partly because carbon cycle feedback uncertainties are greater with larger warming. The uncertainty ranges derived from the above percentages for the warming by 2090 to 2099 relative to 1980 to 1999 are 1.1°C to 2.9°C, 1.4°C to 3.8°C, 1.7°C to 4.4°C, 1.4°C to 3.8°C, 2.0°C to 5.4°C and 2.4°C to 6.4°C for the scenarios B1, B2, A1B, A1T, A2 and A1FI, respectively. It is not appropriate to compare the lowest and highest values across these ranges against the single range given in the TAR, because the TAR range resulted only from projections using an SCM and covered all SRES scenarios, whereas here a number of different and independent modelling approaches are combined to estimate ranges for the six illustrative scenarios separately. Additionally, in contrast to the TAR, carbon cycle uncertainties are now included in these ranges. These uncertainty ranges include only anthropogenically forced changes.

10.5.4.7 Probabilistic Projections - Geographical Depictions

Tebaldi et al. 2005 [JoC, MoS, SRC] ) present a Bayesian approach to regional climate prediction, developed from the ideas of Giorgi and Mearns 2002 [JoC, MoS, ARC] , 2003 ). Non-informative prior distributions for regional temperature and precipitation are updated using observations and results from AOGCM ensembles to produce probability distributions of future changes. Key assumptions are that each model and the observations differ randomly and independently from the true climate, and that the weight given to a model prediction should depend on the bias in its present-day simulation and its degree of convergence with the weighted ensemble mean of the predicted future change. Lopez et al. 2006 [JoC] ) apply the Tebaldi et al. 2005 [JoC, MoS, SRC] ) method to a 15-member multi-model ensemble to predict future changes in global surface temperature under a 1% yr^–1 increase in atmospheric CO₂.They compare it with the method developed by Allen et al. 2000 [JoC, MoS, SRC] and Stott and Kettleborough 2002 [JoC, MoS, SRC] ) (ASK), which aims to provide relatively model independent probabilities consistent with observed changes (see Section 10.5.4.5 ). The Bayesian method predicts a much narrower uncertainty range than ASK. However its results depend on choices made in its design, particularly the convergence criterion for up-weighting models close to the ensemble mean, relaxation of which substantially reduces the discrepancy with ASK.

Another method by Furrer et al. 2007 [NPR, MoS, SRC, 2007] ) employs a hierarchical Bayesian model to construct PDFs of temperature change at each grid point from a multi-model ensemble. The main assumptions are that the true climate change signal is a common large-scale structure represented to some degree in each of the model simulations, and that the signal unexplained by climate change is AOGCM-specific in terms of small-scale structure, but can be regarded as noise when averaged over all AOGCMs. In this method, spatial fields of future minus present temperature difference from each ensemble member are regressed upon basis functions. One of the basis functions is a map of differences of observed temperatures from late- minus mid-20th century, and others are spherical harmonics. The statistical model then estimates the regression coefficients and their associated errors, which account for the deviation in each AOGCM from the (assumed) true pattern of change. By recombining the coefficients with the basis functions, an estimate is derived of the true climate change field and its associated uncertainty, thus providing joint probabilities for climate change at all grid points around the globe.

Estimates of uncertainty derived from multi-model ensembles of 10 to 20 members are potentially sensitive to outliers ( Räisänen, 2001 [JoC, SRC] Harris et al. 2006 [JoC, MoS, SRC] ) therefore augment a 17-member ensemble of AOGCM transient simulations by scaling the equilibrium response patterns of a large perturbed physics ensemble. Transient responses are emulated by scaling equilibrium response patterns according to global temperature (predicted from an energy balance model tuned to the relevant climate sensitivities). For surface temperature, the scaled equilibrium patterns correspond well to the transient response patterns, while scaling errors for precipitation vary more widely with location. A correction field is added to account for ensemble-mean differences between the equilibrium and transient patterns, and uncertainty is allowed for in the emulated result. The correction field and emulation errors are determined by comparing the responses of model versions for which both transient and equilibrium simulations exist. Results are used to obtain frequency distributions of transient regional changes in surface temperature and precipitation in response to increasing atmospheric CO₂,arising from the combined effects of atmospheric parameter perturbations and internal variability in UKMO-HadCM3.

Figure 10.30 shows probabilities of a temperature change larger than 2°C by the end of the 21st century under the A1B scenario, comparing values estimated from the 21-member AR4 multi-model ensemble ( Furrer et al., 2007 [NPR, MoS, SRC, 2007] ) against values estimated by combining transient and equilibrium perturbed physics ensembles of 17 and 128 members, respectively ( Harris et al., 2006 [JoC, MoS, SRC] ). Although the methods use different ensembles and different statistical approaches, the large-scale patterns are similar in many respects. Both methods show larger probabilities (typically 80% or more) over land, and at high latitudes in the winter hemisphere, with relatively low values (typically less than 50%) over the southern oceans. However, the plots also reveal some substantial differences at a regional level, notably over the North Atlantic Ocean, the sub-tropical Atlantic and Pacific Oceans in the SH, and at high northern latitudes during June to August.

Figure 10.30. Estimated probabilities for a mean surface temperature change exceeding 2°C in 2080 to 2099 relative to 1980 to 1999 under the SRES A1B scenario. Results obtained from a perturbed physics ensemble of a single model (a, c), based on Harris et al. 2006 [JoC, MoS, SRC] ), are compared with results from the AR4 multi-model ensemble (b, d), based on Furrer et al. 2007 [NPR, MoS, SRC, 2007] ), for December to February (DJF, a, b) and June to August (JJA, c, d).

10.5.4.8 Summary

Significant progress has been made since the TAR in exploring ensemble approaches to provide uncertainty ranges and probabilities for global and regional climate change. Different methods show consistency in some aspects of their results, but differ significantly in others (see Box 10.2 ;Figures 10.28 and 10.30 ), because they depend to varying degrees on the nature and use of observational constraints, the nature and design of model ensembles and the specification of prior distributions for uncertain inputs (see, e.g., Table 11.3 ). A preferred method cannot yet be recommended, but the assumptions and limitations underlying the various approaches, and the sensitivity of the results to them, should be communicated to users. A good example concerns the treatment of model error in Bayesian methods, the uncertainty in which affects the calculation of the likelihood of different model versions, but is difficult to specify ( Rougier, 2007 [JoC, MoS, 2007] ). Awareness of this issue is growing in the field of climate prediction ( Annan et al., 2005b [MoS, SRC] ; Knutti et al., 2006 [JoC, SRC] ), however, it is yet to be thoroughly addressed. Probabilistic depictions, particularly at the regional level, are new to climate change science and are being facilitated by the recently available multi-model ensembles. These are discussed further in Section 11.10.2 .

10.6.3.1 Mass Balance Sensitivity to Temperature and Precipitation

Since G&IC mass balance depends strongly on their altitude and aspect, use of data from climate models to make projections requires a method of downscaling, because individual G&IC are much smaller than typical AOGCM grid boxes. Statistical relations for meteorological quantities can be developed between the GCM and local scales ( Reichert et al., 2002 [JoC, SRC] ), but they may not continue to hold in future climates. Hence, for projections the approach usually adopted is to use GCM simulations of changes in climate parameters to perturb the observed climatology or mass balance ( Gregory and Oerlemans, 1998 [JoC, MoS, SRC] ; Schneeberger et al., 2003 [MoS, SRC] ).

Change in ablation (mostly melting) of a glacier or ice cap is modelled using b_T (in m yr^–1 °C^–1), the sensitivity of the mean specific surface mass balance to temperature (refer to Section 4.5 for a discussion of the relation of mass balance to climate). One approach determines b_T by energy balance modelling, including evolution of albedo and refreezing of melt water within the firn ( Zuo and Oerlemans, 1997 [JoC, SRC] Oerlemans and Reichert 2000 [SRC] Oerlemans 2001 [NPR, SRC] and Oerlemans et al. 2006 [MoS, SRC] ) refine this approach to include dependence on monthly temperature and precipitation changes. Another approach uses a degree-day method, in which ablation is proportional to the integral of mean daily temperature above the freezing point ( Braithwaite et al., 2003 [SRC] Braithwaite and Raper 2002 [SRC] ) show that there is excellent consistency between the two approaches, which indicates a similar relationship between b_T and climatological precipitation. Schneeberger et al. 2000 [JoC] , 2003 ) use a degree-day method for ablation modified to include incident solar radiation, again obtaining similar results. de Woul and Hock 2006 [ARC] ) find somewhat larger sensitivities for arctic G&IC from the degree-day method than the energy balance method. Calculations of b_T are estimated to have an uncertainty of ±15% (standard deviation) ( Gregory and Oerlemans, 1998 [JoC, MoS, SRC] ; Raper and Braithwaite, 2006 [JoC, MoS, SRC] ).

The global average sensitivity of G&IC surface mass balance to temperature is estimated by weighting the local sensitivities by land ice area in various regions. For a geographically and seasonally uniform rise in global temperature, Oerlemans and Fortuin 1992 [JoC, SRC] ) derive a global average G&IC surface mass balance sensitivity of –0.40 m yr^–1 °C^–1 ,Dyurgerov and Meier 2000 [JoC, SRC] ) –0.37 m yr^–1 °C^–1 (from observations), Braithwaite and Raper 2002 [SRC] ) –0.41 m yr^–1 °C^–1 and Raper and Braithwaite 2005 [JoC, MoS, SRC] ) –0.35 m yr^–1 °C^–1.Applying the scheme of Oerlemans 2001 [NPR, SRC] and Oerlemans et al. 2006 [MoS, SRC] ) worldwide gives a smaller value of –0.32 m yr^–1 °C^–1,the reduction being due to the modified treatment of albedo by Oerlemans 2001 [NPR, SRC] ).

These global average sensitivities for uniform temperature change are given only for scenario-independent comparison of the various methods; they cannot be used for projections, which require regional and seasonal temperature changes ( Gregory and Oerlemans, 1998 [JoC, MoS, SRC] ; van de Wal and Wild, 2001 [JoC, MoS, SRC] ). Using monthly temperature changes simulated in G&IC regions by 17 AR4 AOGCMs for scenarios A1B, A2 and B1, the global total surface mass balance sensitivity to global average temperature change for all G&IC outside Greenland and Antarctica is 0.61 ± 0.12 mm yr^–1 °C^–1 (sea level equivalent) with the b_T of Zuo and Oerlemans 1997 [JoC, SRC] ) or 0.49 ± 0.13 mm yr^–1 °C^–1 with those of Oerlemans 2001 [NPR, SRC] and Oerlemans et al. 2006 [MoS, SRC] ), subject to uncertainty in G&IC area (see Section 4.5.2 and Table 4.4 ).

Hansen and Nazarenko 2004 [JoC, MoS, SRC] ) collate measurements of soot (fossil fuel black carbon) in snow and estimate consequent reductions in snow and ice albedo of between 0.001 for the pristine conditions of Antarctica and over 0.10 for polluted NH land areas. They argue that glacial ablation would be increased by this effect. While it is true that soot has not been explicitly considered in existing sensitivity estimates, it may already be included because the albedo and degree-day parametrizations have been empirically derived from data collected in affected regions.

For seasonally uniform temperature rise, Oerlemans et al. 1998 [JoC, MoS, SRC] ) find that an increase in precipitation of 20 to 50% °C^–1 is required to balance increased ablation, while Braithwaite et al. 2003 [SRC] ) report a required precipitation increase of 29 to 41% °C^–1,in both cases for a sample of G&IC representing a variety of climatic regimes. Oerlemans et al. 2006 [MoS, SRC] ) require a precipitation increase of 20 to 43% °C^–1 to balance ablation increase, and de Woul and Hock 2006 [ARC] ) approximately 20% °C^–1 for Arctic G&IC. Although AOGCMs generally project larger than average precipitation change in northern mid- and high-latitude regions, the global average is 1 to 2% °C^–1 ( Section 10.3.1 ), so ablation increases would be expected to dominate worldwide. However, precipitation changes may sometimes dominate locally (see Section 4.5.3 ).

Regressing observed global total mass balance changes of all G&IC outside Greenland and Antarctica against global average surface temperature change gives a global total mass balance sensitivity which is greater than model results (see Appendix 10 .A). The current state of knowledge does not permit a satisfactory explanation of the difference. Giving more weight to the observational record but enlarging the uncertainty to allow for systematic error, a value of 0.80 ± 0.33 mm yr^–1 °C^–1 (5 to 95% range) is adopted for projections. The regression indicates that the climate of 1865 to 1895 was 0.13°C warmer globally than the climate that gives a steady state for G&IC (cf., Zuo and Oerlemans, 1997 [JoC, SRC] ; Gregory et al., 2006 [Ambiguous] ). Model results for the 20th century are sensitive to this value, but the projected temperature change in the 21st century is large by comparison, making the effect relatively less important for projections (see Appendix 10 .A).

10.6.3.2 Dynamic Response and Feedback on Mass Balance

As glacier volume is lost, glacier area declines so the ablation decreases. Oerlemans et al. 1998 [JoC, MoS, SRC] ) calculate that omitting this effect leads to overestimates of ablation of about 25% by 2100 Church et al. 2001 [NPR, ARC] ), following Bahr et al. 1997 [JoC] and van de Wal and Wild 2001 [JoC, MoS, SRC] ), make some allowance for it by diminishing the area A of a glacier of volume V according to V ∝ A^1.375.This is a scaling relation derived for glaciers in a steady state, which may hold only approximately during retreat. For example, thinning in the ablation zone will steepen the surface slope and tend to increase the flow. Comparison with a simple flow model suggests the deviations do not exceed 20% ( van de Wal and Wild, 2001 [JoC, MoS, SRC] Schneeberger et al. 2003 [MoS, SRC] ) find that the scaling relation produced a mixture of over- and underestimates of volume loss for their sample of glaciers compared with more detailed dynamic modelling. In some regions where G&IC flow into the sea or lakes there is accelerated dynamic discharge ( Rignot et al., 2003 [JoC, ARC] ) that is not included in currently available glacier models, leading to an underestimate of G&IC mass loss.

The mean specific surface mass balance of the glacier or ice cap will change as volume is lost: lowering the ice surface as the ice thins will tend to make it more negative, but the predominant loss of area at lower altitude in the ablation zone will tend to make it less negative ( Braithwaite and Raper, 2002 [SRC] ). For rapid thinning rates in the ablation zone, of several metres per year, lowering the surface will give enhanced local warmings comparable to the rate of projected climatic warming. However, those areas of the ablation zone of valley glaciers that thin most rapidly will soon be removed altogether, resulting in retreat of the glacier. The enhancement of ablation by surface lowering can only be sustained in glaciers with a relatively large, thick and flat ablation area. On multi-decadal time scales, for the majority of G&IC, the loss of area is more important than lowering of the surface ( Schneeberger et al., 2003 [MoS, SRC] ).

The dynamical approach ( Oerlemans et al., 1998 [JoC, MoS, SRC] ; Schneeberger et al., 2003 [MoS, SRC] ) cannot be applied to all the world’s glaciers individually as the required data are unknown for the vast majority of them. Instead, it might be applied to a representative ensemble derived from statistics of size distributions of G&IC. Raper et al. 2000 [MoS, SRC] ) developed a geometrical approach, in which the width, thickness and length of a glacier are reduced as its volume and area declines. When applied statistically to the world population of glaciers and individually to ice caps, this approach shows that the reduction of area of glaciers strongly reduces the ablation during the 21st century ( Raper and Braithwaite, 2006 [JoC, MoS, SRC] ), by about 45% under scenario SRES A1B for the GFDL-CM2.0 and PCM AOGCMs (see Table 8.1 for model details). For the same cases, using the mass-balance sensitivities to temperature of Oerlemans 2001 [NPR, SRC] and Oerlemans et al. 2006 [MoS, SRC] ), G&IC mass loss is reduced by about 35% following the area scaling of van de Wal and Wild 2001 [JoC, MoS, SRC] ), suggesting that the area scaling and the geometrical model have a similar effect in reducing estimated ablation for the 21st century. The effect is greater when using the observationally derived mass balance sensitivity( Section 10.6.3.1 ), which is larger, implying faster mass loss for fixed area. The uncertainty in present-day glacier volume( Table 4.4 )introduces a 5 to 10% uncertainty into the results of area scaling. For projections, the area scaling of van de Wal and Wild 2001 [JoC, MoS, SRC] ) is applied, using three estimates of world glacier volume (see Table 4.4 and Appendix 10 .A). The scaling reduces the projections of the G&IC contribution up to the mid-21st century by 25% and over the whole century by 40 to 50% with respect to fixed G&IC area.

10.6.3.3 Glaciers and Ice Caps on Greenland and Antarctica

The G&IC on Greenland and Antarctica (apart from the ice sheets) have been less studied and projections for them are consequently more uncertain. A model estimate for the G&IC on Greenland indicates an addition of about 6% to the G&IC sea level contribution in the 21st century ( van de Wal and Wild, 2001 [JoC, MoS, SRC] ). Using a degree-day scheme, Vaughan 2006 [SRC] ) estimates that ablation of glaciers in the Antarctic Peninsula presently amounts to 0.008 to 0.055 mm yr^–1 of sea level, 1 to 9% of the contribution from G&IC outside Greenland and Antarctica( Table 4.4 ( Morris and Mulvaney 2004 ) ) find that accumulation increases on the Antarctic Peninsula were larger than ablation increases during 1972 to 1998, giving a small net negative sea level contribution from the region. However, because ablation increases nonlinearly with temperature, they estimate that for future warming the contribution would become positive, with a sensitivity of 0.07 ± 0.03 mm yr^–1 °C^–1 to uniform temperature change in Antarctica, that is, about 10% of the global sensitivity of G&IC outside Greenland and Antarctica( Section 10.6.3.1 ).

These results suggest that the Antarctic and Greenland G&IC will together give 10 to 20% of the sea level contribution of other G&IC in future decades. In recent decades, the G&IC on Greenland and Antarctica have together made a contribution of about 20% of the total of other G&IC (see Section 4.5.2 ). On these grounds, the global G&IC sea level contribution is increased by a factor of 1.2 to include those in Greenland and Antarctica in projections for the 21st century (see Section 10.6.5 and Table 10.7 ). Dynamical acceleration of glaciers in Greenland and Antarctica following removal of ice shelves, as has recently happened on the Antarctic Peninsula (Sections 4.6.2.2 and 10.6.4.2 ), would add further to this, and is included in projections of that effect (Section 10.6.4.3).

10.6.4.1 Surface Mass Balance

Surface mass balance (SMB) is immediately influenced by climate change. A good simulation of the ice sheet SMB requires a resolution exceeding that of AGCMs used for long climate experiments, because of the steep slopes at the margins of the ice sheet, where the majority of the precipitation and all of the ablation occur. Precipitation over ice sheets is typically overestimated by AGCMs, because their smooth topography does not present a sufficient barrier to inland penetration ( Ohmura et al., 1996 [MoS, SRC] ; Glover, 1999 [JoC, MoS] ; Murphy et al., 2002 [JoC, MoS] ). Ablation also tends to be overestimated because the area at low altitude around the margins of the ice sheet, where melting preferentially occurs, is exaggerated ( Glover, 1999 [JoC, MoS] ; Wild et al., 2003 [JoC, SRC] ). In addition, AGCMs do not generally have a representation of the refreezing of surface melt water within the snowpack and may not include albedo variations dependent on snow ageing and its conversion to ice.

To address these issues, several groups have computed SMB at resolutions of tens of kilometres or less, with results that compare acceptably well with observations (e.g., van Lipzig et al., 2002 [JoC, MoS, SRC] ; Wild et al., 2003 [JoC, SRC] ). Ablation is calculated either by schemes based on temperature (degree-day or other temperature index methods) or by energy balance modelling. In the studies listed in Table 10.6 ,changes in SMB have been calculated from climate change simulations with high-resolution AGCMs or by perturbing a high-resolution observational climatology with climate model output, rather than by direct use of low-resolution GCM results. The models used for projected SMB changes are similar in kind to those used to study recent SMB changes( Section 4.6.3.1 ).

All the models show an increase in accumulation, but there is considerable uncertainty in its size( Table 10.6 ;van de Wal et al., 2001 [Ambiguous] ; Huybrechts et al., 2004 [JoC, MoS, SRC] ). Precipitation increase could be determined by atmospheric radiative balance, increase in saturation specific humidity with temperature, circulation changes, retreat of sea ice permitting greater evaporation or a combination of these ( van Lipzig et al., 2002 [JoC, MoS, SRC] ). Accumulation also depends on change in local temperature, which strongly affects whether precipitation is solid or liquid ( Janssens and Huybrechts, 2000 [SRC] ), tending to make the accumulation increase smaller than the precipitation increase for a given temperature rise. For Antarctica, accumulation increases by 6 to 9% °C^–1 in the high-resolution AGCMs. Precipitation increases somewhat less in AR4 AOGCMs (typically of lower resolution), by 3 to 8% °C^–1.For Greenland, accumulation derived from the high-resolution AGCMs increases by 5 to 9% °C^–1.Precipitation increases by 4 to 7% °C^–1 in the AR4 AOGCMs.

Kapsner et al. 1995 [JoC] ) do not find a relationship between precipitation and temperature variability inferred from Greenland ice cores for the Holocene, although both show large changes from the Last Glacial Maximum (LGM) to the Holocene. In the UKMO-HadCM3 AOGCM, the relationship is strong for climate change forced by greenhouse gases and the glacial-interglacial transition, but weaker for naturally forced variability ( Gregory et al., 2006 [Ambiguous] ). Increasing precipitation in conjunction with warming has been observed in recent years in Greenland( Section 4.6.3.1 ).

All studies for the 21st century project that antarctic SMB changes will contribute negatively to sea level, owing to increasing accumulation exceeding any ablation increase (see Table 10.6 ). This tendency has not been observed in the average over Antarctica in reanalysis products for the last two decades (see Section 4.6.3.1 ), but during this period Antarctica as a whole has not warmed; on the other hand, precipitation has increased on the Antarctic Peninsula, where there has been strong warming.

In projections for Greenland, ablation increase is important but uncertain, being particularly sensitive to temperature change around the margins. Climate models project less warming in these low-altitude regions than the Greenland average, and less warming in summer (when ablation occurs) than the annual average, but greater warming in Greenland than the global average ( Church et al., 2001 [NPR, ARC] ; Huybrechts et al., 2004 [JoC, MoS, SRC] ; Chylek and Lohmann, 2005 [JoC, MoS, ARC] ; Gregory and Huybrechts, 2006 [JoC, MoS, SRC] ). In most studies, Greenland SMB changes represent a net positive contribution to sea level in the 21st century( Table 10.6 ;Kiilsholm et al., 2003 [JoC, MoS, SRC] ) because the ablation increase is larger than the precipitation increase. Only Wild et al. 2003 [JoC, SRC] ) find the opposite, so that the net SMB change contributes negatively to sea level in the 21st century. Wild et al. 2003 [JoC, SRC] ) attribute this difference to the reduced ablation area in their higher-resolution grid. A positive SMB change is not consistent with analyses of recent changes in Greenland SMB (see Section 4.6.3.1 ).

For an average temperature change of 3°C over each ice sheet, a combination of four high-resolution AGCM simulations and 18 AR4 AOGCMs ( Huybrechts et al., 2004 [JoC, MoS, SRC] ; Gregory and Huybrechts, 2006 [JoC, MoS, SRC] ) gives SMB changes of 0.3 ± 0.3 mm yr^–1 for Greenland and –0.9 ± 0.5 mm yr^–1 for Antarctica (sea level equivalent), that is, sensitivities of 0.11 ± 0.09 mm yr^–1 °C^–1 for Greenland and –0.29 ± 0.18 mm yr^–1 °C^–1 for Antarctica. These results generally cover the range shown in Table 10.6 ,but tend to give more positive (Greenland) or less negative (Antarctica) sea level rise because of the smaller precipitation increases projected by the AOGCMs than by the high-resolution AGCMs. The uncertainties are from the spatial and seasonal patterns of precipitation and temperature change over the ice sheets, and from the ablation calculation. Projections under SRES scenarios for the 21st century are shown in Table 10.7 .

10.6.4.2 Dynamics

Ice sheet flow reacts to changes in topography produced by SMB change. Projections for the 21st century are given in Section 10.6.5 and Table 10.7 ,based on the discussion in this section. In Antarctica, topographic change tends to increase ice flow and discharge. In Greenland, lowering of the surface tends to increase the ablation, while a steepening slope in the ablation zone opposes the lowering, and thinning of outlet glaciers reduces discharge. Topographic and dynamic changes simulated by ice flow models ( Huybrechts and de Wolde, 1999 [JoC, SRC] ; van de Wal et al., 2001 [Ambiguous] ; Huybrechts et al., 2002 [MoS, SRC] , 2004; Gregory and Huybrechts, 2006 [JoC, MoS, SRC] ) can be roughly represented as modifying the sea level changes due to SMB change with fixed topography by –5% ± 5% from Antarctica, and 0% ±10% from Greenland (± one standard deviation) during the 21st century.

The TAR concluded that accelerated sea level rise caused by rapid dynamic response of the ice sheets to climate change is very unlikely during the 21st century ( Church et al., 2001 [NPR, ARC] ). However, new evidence of recent rapid changes in the Antarctic Peninsula, West Antarctica and Greenland (see Section 4.6.3.3 )has again raised the possibility of larger dynamical changes in the future than are projected by state-of-the-art continental models, such as cited above, because these models do not incorporate all the processes responsible for the rapid marginal thinning currently taking place( Box 4.1 ;Alley et al., 2005a [JoC, SRC] ; Vaughan, 2007 [NPR, JoC, SRC, 2007] ).

The main uncertainty is the degree to which the presence of ice shelves affects the flow of inland ice across the grounding line. A strong argument for enhanced flow when the ice shelf is removed is yielded by the acceleration of Jakobshavn Glacier (Greenland) following the loss of its floating tongue, and of the glaciers supplying the Larsen B Ice Shelf (Antarctic Peninsula) after it collapsed (see Section 4.6.3.3 ). The onset of disintegration of the Larsen B Ice Shelf has been attributed to enhanced fracturing by crevasses promoted by surface melt water ( (Scambos et al., 2000 ) ). Large portions of the Ross and Filchner-Ronne Ice Shelves (West Antarctica) currently have mean summer surface temperatures of around –5°C ( Comiso, 2000 [JoC, ARC] , updated). Four high-resolution GCMs ( Gregory and Huybrechts, 2006 [JoC, MoS, SRC] ) project summer surface warming in these major ice shelf regions of between 0.2 and 1.3 times the antarctic annual average warming, which in turn will be a factor 1.1 ± 0.3 greater than global average warming according to AOGCM simulations using SRES scenarios. These figures indicate that a local mean summer warming of 5°C is unlikely for a global warming of less than 5°C (see Appendix 10 .A). This suggests that ice shelf collapse due to surface melting is unlikely under most SRES scenarios during the 21st century, but we have low confidence in the inference because there is evidently large systematic uncertainty in the regional climate projections, and it is not known whether episodic surface melting might initiate disintegration in a warmer climate while mean summer temperatures remain below freezing.

In the Amundsen Sea sector of West Antarctica, ice shelves are not so extensive and the cause of ice shelf thinning is not surface melting, but bottom melting at the grounding line ( Rignot and Jacobs, 2002 [JoC, ARC] Shepherd et al. 2004 [JoC, SRC] ) find an average ice-shelf thinning rate of 1.5 ± 0.5 m yr^–1.At the same time as the basal melting, accelerated inland flow has been observed for Pine Island, Thwaites and other glaciers in the sector ( Rignot, 1998 [JoC, ARC] , 2001; Thomas et al., 2004 [JoC, ARC] ). The synchronicity of these changes strongly implies that their cause lies in oceanographic change in the Amundsen Sea, but this has not been attributed to anthropogenic climate change and could be connected with variability in the SAM.

Because the acceleration took place in only a few years ( Rignot et al., 2002 [ARC] ; Joughin et al., 2003 [JoC, ARC] ) but appears up to about 150 km inland, it implies that the dynamical response to changes in the ice shelf can propagate rapidly up the ice stream. This conclusion is supported by modelling studies of Pine Island Glacier by Payne et al. 2004 [JoC, SRC] and Dupont and Alley 2005 [JoC, SRC] ), in which a single and instantaneous reduction of the basal or lateral drag at the ice front is imposed in idealised ways, such as a step retreat of the grounding line. The simulated acceleration and inland thinning are rapid but transient; the rate of contribution to sea level declines as a new steady state is reached over a few decades. In the study of Payne et al. 2004 [JoC, SRC] ) the imposed perturbations were designed to resemble loss of drag in the ‘ice plain’, a partially grounded region near the ice front, and produced a velocity increase of about 1 km yr^–1 there. Thomas et al. 2005 [ARC] ) suggest the ice plain will become ungrounded during the next decade and obtain a similar velocity increase using a simplified approach.

Most of inland ice of West Antarctica is grounded below sea level and so it could float if it thinned sufficiently; discharge therefore promotes inland retreat of the grounding line, which represents a positive feedback by further reducing basal traction. Unlike the one-time change in the idealised studies, this would represent a sustained dynamical forcing that would prolong the contribution to sea level rise. Grounding line retreat of the ice streams has been observed recently at rates of up to about 1 km yr^–1 ( Rignot, 1998 [JoC, ARC] , 2001; Shepherd et al., 2002 [JoC, SRC] ), but a numerical model formulation is difficult to construct ( Vieli and Payne, 2005 [JoC, MoS, SRC] ).

The majority of West Antarctic ice discharge is through the ice streams that feed the Ross and Ronne-Filchner ice shelves, but in these regions no accelerated flow causing thinning is currently observed; on the contrary, they are thickening or near balance ( (Zwally et al., 2005 ) ). Excluding these regions, and likewise those parts of the East Antarctic Ice Sheet that drain into the large Amery ice shelf, the total area of ice streams (areas flowing faster than 100 m yr^–1)discharging directly into the sea or via a small ice shelf is 270,000 km².If all these areas thinned at 2 m yr^–1,the order of magnitude of the larger rates observed in fast-flowing areas of the Amundsen Sea sector ( Shepherd et al., 2001 [JoC, SRC] , 2002 ), the contribution to sea level rise would be about 1.5 mm yr^–1.This would require sustained retreat simultaneously on many fronts, and should be taken as an indicative upper limit for the 21st century (see also Section 10.6.5 ).

The observation in west-central Greenland of seasonal variation in ice flow rate and of a correlation with summer temperature variation ( Zwally et al., 2002 [JoC] ) suggest that surface melt water may join a sub-glacially routed drainage system lubricating the ice flow (although this implies that it penetrates more than 1,200 m of subfreezing ice). By this mechanism, increased surface melting during the 21st century could cause acceleration of ice flow and discharge; a sensitivity study ( Parizek and Alley, 2004 [JoC, MoS, SRC] ) indicated that this might increase the sea level contribution from the Greenland Ice Sheet during the 21st century by up to 0.2 m, depending on the warming and other assumptions. However, other studies ( (Echelmeyer and Harrison, 1990; ) Joughin et al., 2004 [JoC, ARC] ) found no evidence of seasonal fluctuations in the flow rate of nearby Jakobshavn Glacier despite a substantial supply of surface melt water.

10.7.4.1 Thermal Expansion

The sea level rise commitment due to thermal expansion has much longer time scales than the surface warming commitment, owing to the slow processes that mix heat into the deep ocean ( Church et al., 2001 [NPR, ARC] ). If atmospheric composition were stabilised at A1B levels in 2100, thermal expansion in the 22nd century would be similar to in the 21st (see, e.g., Section 10.6.1 ;Meehl et al., 2005c [JoC, SRC] ), reaching 0.3 to 0.8 m by 2300 ( Figure 10.37 ). The ranges of thermal expansion overlap substantially for stabilisation at different levels, since model uncertainty is dominant; A1B is given here because results are available from more models for this scenario than for other scenarios. Thermal expansion would continue over many centuries at a gradually decreasing rate( Figure 10.34 ). There is a wide spread among the models for the thermal expansion commitment at constant composition due partly to climate sensitivity, and partly to differences in the parametrization of vertical mixing affecting ocean heat uptake (e.g., Weaver and Wiebe, 1999 [JoC, SRC] ). If there is deep-water formation in the final steady state as in the present day, the ocean will eventually warm up fairly uniformly by the amount of the global average surface temperature change ( Stouffer and Manabe, 2003 [JoC, SRC] ), which would result in about 0.5 m of thermal expansion per degree celsius of warming, calculated from observed climatology; the EMICs in Figure 10.34 indicate 0.2 to 0.6 m °C^–1 for their final steady state (year 3000) relative to 2000 . If deep-water formation is weakened or suppressed, the deep ocean will warm up more ( Knutti and Stocker, 2000 [JoC, SRC] ). For instance, in the 3 × CO₂ experiment of Bi et al. 2001 [JoC, MoS] ) with the CSIRO AOGCM, both North Atlantic Deep Water and Antarctic Bottom Water formation cease, and the steady-state thermal expansion is 4.5 m. Although these commitments to sea level rise are large compared with 21st-century changes, the eventual contributions from the ice sheets could be larger still.

Figure 10.37. Globally averaged sea level rise from thermal expansion relative to the period 1980 to 1999 for the A1B commitment experiment calculated from AOGCMs. See Table 8.1 for model details.

10.7.4.2 Glaciers and Ice Caps

Steady-state projections for G&IC require a model that evolves their area-altitude distribution (see, e.g., Section 10.6.3.3 ). Little information is available on this. A comparative study including seven GCM simulations at 2 × CO₂ conditions inferred that many glaciers may disappear completely due to an increase of the equilibrium line altitude ( Bradley et al., 2004 [PoC, JoC] ), but even in a warmer climate, some glacier volume may persist at high altitude. With a geographically uniform warming relative to 1900 of 4°C maintained after 2100, about 60% of G&IC volume would vanish by 2200 and practically all by 3000 ( Raper and Braithwaite, 2006 [JoC, MoS, SRC] ). Nonetheless, this commitment to sea level rise is relatively small (<1 m; Table 4.4 )compared with those from thermal expansion and ice sheets.

10.7.4.3 Greenland Ice Sheet

The present SMB of Greenland is a net accumulation estimated as 0.6 mm yr^–1 of sea level equivalent from a compilation of studies ( Church et al., 2001 [NPR, ARC] ) and 0.47 mm yr^–1 for 1988 to 2004 ( Box et al., 2006 [JoC, ARC] ). In a steady state, the net accumulation would be balanced by calving of icebergs. General Circulation Models suggest that ablation increases more rapidly than accumulation with temperature ( van de Wal et al., 2001 [Ambiguous] ; Gregory and Huybrechts, 2006 [JoC, MoS, SRC] ), so warming will tend to reduce the SMB, as has been observed in recent years (see Section 4.6.3 ), and is projected for the 21st century( Section 10.6.4.1 ). Sufficient warming will reduce the SMB to zero. This gives a threshold for the long-term viability of the ice sheet because negative SMB means that the ice sheet must contract even if ice discharge has ceased owing to retreat from the coast. If a warmer climate is maintained, the ice sheet will eventually be eliminated, except perhaps for remnant glaciers in the mountains, raising sea level by about 7 m (see Table 4.1 Huybrechts et al. 1991 [SRC] ) evaluated the threshold as 2.7°C of seasonally and geographically uniform warming over Greenland relative to a steady state (i.e. pre-industrial temperature). Gregory et al. 2004a [JoC, SRC] ) examine the probability of this threshold being reached under various CO₂ stabilisation scenarios for 450 to 1000 ppm using TAR projections, and find that it was exceeded in 34 out of 35 combinations of AOGCM and CO₂ concentration considering seasonally uniform warming, and 24 out of 35 considering summer warming and using an upper bound on the threshold.

Assuming the warming to be uniform underestimates the threshold, because warming is projected by GCMs to be weaker in the ablation area and in summer, when ablation occurs. Using geographical and seasonal patterns of simulated temperature change derived from a combination of four high-resolution AGCM simulations and 18 AR4 AOGCMs raises the threshold to 3.2°C to 6.2°C in annual- and area-average warming in Greenland, and 1.9°C to 4.6°C in the global average ( Gregory and Huybrechts, 2006 [JoC, MoS, SRC] ), relative to pre-industrial temperatures. This is likely to be reached by 2100 under the SRES A1B scenario, for instance( Figure 10.29 ). These results are supported by evidence from the last interglacial, when the temperature in Greenland was 3°C to 5°C warmer than today and the ice sheet survived, but may have been smaller by 2 to 4 m in sea level equivalent (including contributions from arctic ice caps, see Section 6.4.3 ). However, a lower threshold of 1°C ( Hansen, 2005 [JoC, SRC] ) in global warming above present-day temperatures has also been suggested, on the basis that global mean (rather than Greenland) temperatures during previous interglacials exceeded today’s temperatures by no more than that.

For stabilisation in 2100 with SRES A1B atmospheric composition, Greenland would initially contribute 0.3 to 2.1 mm yr^–1 to sea level( Table 10.7 ). The greater the warming, the faster the loss of mass. Ablation would be further enhanced by the lowering of the surface, which is not included in the calculations in Table 10.7 .To include this and other climate feedbacks in calculating long-term rates of sea level rise requires coupling an ice sheet model to a climate model. Ridley et al. 2005 [JoC, SRC] ) couple the Greenland Ice Sheet model of Huybrechts and de Wolde 1999 [JoC, SRC] ) to the UKMO-HadCM3 AOGCM. Under constant 4 × CO₂,the sea level contribution is 5.5 mm yr^–1 over the first 300 years and declines as the ice sheet contracts; after 1 kyr only about 40% of the original volume remains and after 3 kyr only 4%( Figure 10.38 ). The rate of deglaciation would increase if ice flow accelerated, as in recent years( Section 4.6.3.3 ). Basal lubrication due to surface melt water might cause such an effect (see Section 10.6.4.2 ). The best estimate of Parizek and Alley 2004 [JoC, MoS, SRC] ) is that this could add an extra 0.15 to 0.40 m to sea level by 2500, compared with 0.4 to 3.2 m calculated by Huybrechts and de Wolde 1999 [JoC, SRC] ) without this effect. The processes whereby melt water might penetrate through subfreezing ice to the bed are unclear and only conceptual models exist at present ( Alley et al., 2005b [SRC] ).

Under pre-industrial or present-day atmospheric CO₂ concentrations, the climate of Greenland would be much warmer without the ice sheet, because of lower surface altitude and albedo, so it is possible that Greenland deglaciation and the resulting sea level rise would be irreversible. Toniazzo et al. 2004 [JoC, SRC] ) find that snow does not accumulate anywhere on an ice-free Greenland with pre-industrial atmospheric CO₂,whereas Lunt et al. 2004 [JoC] ) obtain a substantial regenerated ice sheet in east and central Greenland using a higher-resolution model.

Figure 10.38. Evolution of Greenland surface elevation and ice sheet volume versus time in the experiment of Ridley et al. 2005 [JoC, SRC] ) with the UKMO-HadCM3 AOGCM coupled to the Greenland Ice Sheet model of Huybrechts and de Wolde 1999 [JoC, SRC] ) under a climate of constant quadrupled pre-industrial atmospheric CO₂.

10.7.4.4 Antarctic Ice Sheet

With rising global temperature, GCMs indicate increasingly positive SMB for the Antarctic Ice Sheet as a whole because of greater accumulation( Section 10.6.4.1 ). For stabilisation in 2100 with SRES A1B atmospheric composition, antarctic SMB would contribute 0.4 to 2.0 mm yr^–1 of sea level fall( Table 10.7 ). Continental ice sheet models indicate that this would be offset by tens of percent by increased ice discharge( Section 10.6.4.2 ), but still give a negative contribution to sea level, of –0.8 m by 3000 in one simulation with antarctic warming of about 4.5°C ( Huybrechts and de Wolde, 1999 [JoC, SRC] ).

However, discharge could increase substantially if buttressing due to the major West Antarctic ice shelves were reduced (see Sections 4.6.3.3 and 10.6.4.2 ), and could outweigh the accumulation increase, leading to a net positive antarctic sea level contribution in the long term. If the Amundsen Sea sector were eventually deglaciated, it would add about 1.5 m to sea level, while the entire West Antarctic Ice Sheet (WAIS) would account for about 5 m ( Vaughan, 2007 [NPR, JoC, SRC, 2007] ). Contributions could also come in this manner from the limited marine-based portions of East Antarctica that discharge into large ice shelves.

Weakening or collapse of the ice shelves could be caused either by surface melting or by thinning due to basal melting. In equilibrium experiments with mixed-layer ocean models, the ratio of antarctic to global annual warming is 1.4 ± 0.3. Following reasoning in Section 10.6.4.2 and Appendix 10 .A, it appears that mean summer temperatures over the major West Antarctic ice shelves are about as likely as not to pass the melting point if global warming exceeds 5°C, and disintegration might be initiated earlier by surface melting. Observational and modelling studies indicate that basal melt rates depend on water temperature near to the base, with a constant of proportionality of about 10 m yr^–1 °C^–1 indicated for the Amundsen Sea ice shelves ( Rignot and Jacobs, 2002 [JoC, ARC] ; Shepherd et al., 2004 [JoC, SRC] ) and 0.5 to 10 m yr^–1 °C^–1 for the Amery ice shelf ( Williams et al., 2002 [JoC] ). If this order of magnitude applies to future changes, a warming of about 1°C under the major ice shelves would eliminate them within centuries. We are not able to relate this quantitatively to global warming with any confidence, because the issue has so far received little attention, and current models may be inadequate to treat it because of limited resolution and poorly understood processes. Nonetheless, it is reasonable to suppose that sustained global warming would eventually lead to warming in the seawater circulating beneath the ice shelves.

Because the available models do not include all relevant processes, there is much uncertainty and no consensus about what dynamical changes could occur in the Antarctic Ice Sheet (see, e.g., Vaughan and Spouge, 2002 [JoC, MoS, SRC] ; Alley et al., 2005a [JoC, SRC] ). One line of argument is to consider an analogy with palaeoclimate (see Box 4.1 ). Palaeoclimatic evidence that sea level was 4 to 6 m above present during the last interglacial may not all be explained by reduction in the Greenland Ice Sheet, implying a contribution from the Antarctic Ice Sheet (see Section 6.4.3 ). On this basis, using the limited available evidence, sustained global warming of 2°C ( Oppenheimer and Alley, 2005 [JoC, SRC] ) above present-day temperatures has been suggested as a threshold beyond which there will be a commitment to a large sea level contribution from the WAIS. The maximum rates of sea level rise during previous glacial terminations were of the order of 10 mm yr^–1 ( Church et al., 2001 [NPR, ARC] ). We can be confident that future accelerated discharge from WAIS will not exceed this size, which is roughly an order of magnitude increase in present-day WAIS discharge, since no observed recent acceleration has exceeded a factor of ten.

Another line of argument is that there is insufficient evidence that rates of dynamical discharge of this magnitude could be sustained over long periods. The WAIS is 20 times smaller than the LGM NH ice sheets that contributed most of the melt water during the last deglaciation at rates that can be explained by surface melting alone ( Zweck and Huybrechts, 2005 [JoC, MoS, SRC] ). In the study of Huybrechts and de Wolde 1999 [JoC, SRC] ), the largest simulated rate of sea level rise from the Antarctic Ice Sheet over the next 1 kyr is 2.5 mm yr^–1.This is dominated by dynamical discharge associated with grounding line retreat. The model did not simulate ice streams, for which widespread acceleration would give larger rates. However, the maximum loss of ice possible from rapid discharge of existing ice streams is the volume in excess of flotation in the regions occupied by these ice streams (defined as regions of flow exceeding 100 m yr^–1;see Section 10.6.4.2 ). This volume (in both West and East Antarctica) is 230,000 km³,equivalent to about 0.6 m of sea level, or about 1% of the mass of the Antarctic Ice Sheet, most of which does not flow in ice streams. Loss of ice affecting larger portions of the ice sheet could be sustained at rapid rates only if new ice streams developed in currently slow-moving ice. The possible extent and rate of such changes cannot presently be estimated, since there is only very limited understanding of controls on the development and variability of ice streams. In this argument, rapid discharge may be transient and the long-term sign of the antarctic contribution to sea level depends on whether increased accumulation is more important than large-scale retreat of the grounding line.