9.2.1.1 Summary of ‘Forward’ Estimates of Forcing for the Instrumental Period

Estimates of the radiative forcing (see Section 2.2 for a definition) since 1750 from forward model calculations and observations are reviewed in detail in Chapter2 and provided in Table 2.12. Chapter2 describes estimated forcing resulting from increases in long-lived greenhouse gases (carbon dioxide (CO₂), methane, nitrous oxide, halocarbons), decreases in stratospheric ozone, increases in tropospheric ozone, sulphate aerosols, nitrate aerosols, black carbon and organic matter from fossil fuel burning, biomass burning aerosols, mineral dust aerosols, land use change, indirect aerosol effects on clouds, aircraft cloud effects, solar variability, and stratospheric and tropospheric water vapour increases from methane and irrigation. An example of one model’s implemented set of forcings is given in Figure 2.23 .While some members of the MMD at PCMDI have included a nearly complete list of these forcings for the purpose of simulating the 20th-century climate (see Supplementary Material, Table S9.1), most detection studies to date have used model runs with a more limited set of forcings. The combined anthropogenic forcing from the estimates in Section 2.9.2 since 1750 is 1.6 Wm^–2,with a 90% range of 0.6 to 2.4 Wm^–2,indicating that it is extremely likely that humans have exerted a substantial warming influence on climate over that time period. The combined forcing by greenhouse gases plus ozone is 2.9 ± 0.3 Wm^–2 and the total aerosol forcing (combined direct and indirect ‘cloud albedo’ effect) is virtually certain to be negative and estimated to be –1.3 (90% uncertainty range of –2.2 to –0.5 Wm^–2;see Section 2.9 ). In contrast, the direct radiative forcing due to increases in solar irradiance is estimated to be +0.12 (90% range from 0.06 to 0.3) Wm^–2.In addition, Chapter2 concludes that it is exceptionally unlikely that the combined natural (solar and volcanic) radiative forcing has had a warming influence comparable to that of the combined anthropogenic forcing over the period 1950 to 2005 . As noted in Chapter2 ,the estimated global average surface temperature response from these forcings may differ for a particular magnitude of forcing since all forcings do not have the same ‘efficacy’ (i.e., effectiveness at changing the surface temperature compared to CO₂;see Section 2.8 ). Thus, summing these forcings does not necessarily give an adequate estimate of the response in global average surface temperature.

9.2.1.2 Summary of ‘Inverse’ Estimates of Net Aerosol Forcing

Forward model approaches to estimating aerosol forcing are based on estimates of emissions and models of aerosol physics and chemistry. They directly resolve the separate contributions by various aerosol components and forcing mechanisms. This must be borne in mind when comparing results to those from inverse calculations (see Section 9.6 and Supplementary Material, Appendix 9.B for details), which, for example, infer the net aerosol forcing required to match climate model simulations with observations. These methods can be applied using a global average forcing and response, or using the spatial and temporal patterns of the climate response in order to increase the ability to distinguish between responses to different external forcings. Inverse methods have been used to constrain one or several uncertain radiative forcings (e.g., by aerosols), as well as climate sensitivity( Section 9.6 )and other uncertain climate parameters ( Wigley, 1989 [NPR, PoC, ARC] ; Schlesinger and Ramankutty, 1992 [JoC, ARC] ; Wigley et al., 1997 [PoC, JoC, ARC] ; Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ; Forest et al., 2001 [JoC, MoS, SRC] , 2002; Harvey and Kaufmann, 2002 [JoC, MoS] ; Knutti et al., 2002 [PoC, JoC, MoS, SRC] , 2003; Andronova et al., 2007 [NPR, SRC, 2007] ; Forest et al., 2006 [JoC, MoS, ARC] ; see Table 9.1 – Stott et al., 2006c [JoC, MoS, SRC] ). The reliability of the spatial and temporal patterns used is discussed in Sections 9.2.2.1 and 9.2.2.2 .

In the past, forward calculations have been unable to rule out a total net negative radiative forcing over the 20th century ( Boucher and Haywood, 2001 [JoC, MoS, ARC] ). However, Section 2.9 updates the Boucher and Haywood analysis for current radiative forcing estimates since 1750 and shows that it is extemely likely that the combined anthropogenic RF is both positive and substantial (best estimate: +1.6 Wm^–2). A net forcing close to zero would imply a very high value of climate sensitivity, and would be very difficult to reconcile with the observed increase in temperature (Sections 9.6 and 9.7 ). Inverse calculations yield only the ‘net forcing’, which includes all forcings that project on the fingerprint of the forcing that is estimated. For example, the response to tropospheric ozone forcing could project onto that for sulphate aerosol forcing. Therefore, differences between forward estimates and inverse estimates may have one of several causes, including (1) the magnitude of the forward model calculation is incorrect due to inadequate physics and/or chemistry, (2) the forward calculation has not evaluated all forcings and feedbacks or (3) other forcings project on the fingerprint of the forcing that is estimated in the inverse calculation.

Studies providing inverse estimates of aerosol forcing are compared in Table 9.1 .One type of inverse method uses the ranges of climate change fingerprint scaling factors derived from detection and attribution analyses that attempt to separate the climate response to greenhouse gas forcing from the response to aerosol forcing and often from natural forcing as well ( Gregory et al., 2002a [JoC, MoS, SRC] ; Stott et al., 2006c [JoC, MoS, SRC] ; see also Section 9.4.1.4 ). These provide the range of fingerprint magnitudes (e.g., for the combined temperature response to different aerosol forcings) that are consistent with observed climate change, and can therefore be used to infer the likely range of forcing that is consistent with the observed record. The separation between greenhouse gas and aerosol fingerprints exploits the fact that the forcing from well-mixed greenhouse gases is well known, and that errors in the model’s transient sensitivity can therefore be separated from errors in aerosol forcing in the model (assuming that there are similar errors in a model’s sensitivity to greenhouse gas and aerosol forcing; see Gregory et al., 2002a [JoC, MoS, SRC] ; Table 9.1 ). By scaling spatio-temporal patterns of response up or down, this technique takes account of gross model errors in climate sensitivity and net aerosol forcing but does not fully account for modelling uncertainty in the patterns of temperature response to uncertain forcings.

Another approach uses the response of climate models, most often simple climate models or Earth System Models of Intermediate Complexity (EMICs, Table 8.3 )to explore the range of forcings and climate parameters that yield results consistent with observations ( Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ; Forest et al., 2002 [JoC, SRC] ; Harvey and Kaufmann, 2002 [JoC, MoS] ; Knutti et al., 2002 [PoC, JoC, MoS, SRC] , 2003; Forest et al., 2006 [JoC, MoS, ARC] ). Like detection methods, these approaches seek to fit the space-time patterns, or spatial means in time, of observed surface, atmospheric or ocean temperatures. They determine the probability of combinations of climate sensitivity and net aerosol forcing based on the fit between simulations and observations (see Section 9.6 and Supplementary Material, Appendix 9.B for further discussion). These are often based on Bayesian approaches, where prior assumptions about ranges of external forcing are used to constrain the estimated net aerosol forcing and climate sensitivity. Some of these studies use the difference between Northern and Southern Hemisphere mean temperature to separate the greenhouse gas and aerosol forcing effects (e.g., Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ; Harvey and Kaufmann, 2002 [JoC, MoS] ). In these analyses, it is necessary to accurately account for hemispheric asymmetry in tropospheric ozone forcing in order to infer the hemispheric aerosol forcing. Additionally, aerosols from biomass burning could cause an important fraction of the total aerosol forcing although this forcing shows little hemispheric asymmetry. Since it therefore projects on the greenhouse gas forcing, it is difficult to separate in an inverse calculation. Overall, results will be only as good as the spatial or temporal pattern that is assumed in the analysis. Missing forcings or lack of knowledge about uncertainties, and the highly parametrized spatial distribution of response in some of these models may hamper the interpretation of results.

Aerosol forcing appears to have grown rapidly during the period from 1945 to 1980, while greenhouse gas forcing grew more slowly ( Ramaswamy et al., 2001 [NPR, MoS, ARC] ). Global sulphur emissions (and thus sulphate aerosol forcing) appear to have decreased after 1980 ( (Stern, 2005 ) ), further rendering the temporal evolution of aerosols and greenhouse gases distinct. As long as the temporal pattern of variation in aerosol forcing is approximately correct, the need to achieve a reasonable fit to the temporal variation in global mean temperature and the difference between Northern and Southern Hemisphere temperatures can provide a useful constraint on the net aerosol radiative forcing (as demonstrated, e.g., by Harvey and Kaufmann, 2002 [JoC, MoS] ; Stott et al., 2006c [JoC, MoS, SRC] ).

The inverse estimates summarised in Table 9.1 suggest that to be consistent with observed warming, the net aerosol forcing over the 20th century should be negative with likely ranges between –1.7 and –0.1 Wm^–2.This assessment accounts for the probability of other forcings projecting onto the fingerprints. These results typically provide a somewhat smaller upper limit for the total aerosol forcing than the estimates given in Chapter2 ,which are derived from forward calculations and range between –2.2 and –0.5 Wm^–2 (5 to 95% range, median –1.3 Wm^–2). Note that the uncertainty ranges from inverse and forward calculations are different due to the use of different information, and that they are affected by different uncertainties. Nevertheless, the similarity between results from inverse and forward estimates of aerosol forcing strengthens confidence in estimates of total aerosol forcing, despite remaining uncertainties. Harvey and Kaufmann 2002 [JoC, MoS] ), who use an approach that focuses on the TAR range of climate sensitivity, further conclude that global mean forcing from fossil-fuel related aerosols was probably less than –1.0 Wm^–2 in 1990 and that global mean forcing from biomass burning and anthropogenically enhanced soil dust aerosols is ‘unlikely’ to have exceeded –0.5 Wm^–2 in 1990 .

9.2.1.3 Radiative Forcing of Pre-Industrial Climate Change

Here we briefly discuss the radiative forcing estimates used for understanding climate during the last millennium, the mid-Holocene and the Last Glacial Maximum (LGM)( Section 9.3 )and in estimates of climate sensitivity based on palaeoclimatic records( Section 9.6.3 ).

Regular variation in the Earth’s orbital parameters has been identified as the pacemaker of climate change on the glacial to interglacial time scale (see Berger, 1988 [JoC, ARC] for a review). These orbital variations, which can be calculated from astronomical laws ( Berger, 1978 [ARC] ), force climate variations by changing the seasonal and latitudinal distribution of solar radiation( Chapter6 ).

Insolation at the time of the LGM (21 ka) was similar to today. Nonetheless, the LGM climate remained cold due to the presence of large ice sheets in the Northern Hemisphere ( Peltier, 1994 [JoC] , 2004 ) and reduced atmospheric CO₂ concentration (185 ppm according to recent ice core estimates, see Monnin et al., 2001 [JoC, ARC] ). Most modelling studies of this period do not treat ice sheet extent and elevation or CO₂ concentration prognostically, but specify them as boundary conditions. The LGM radiative forcing from the reduced atmospheric concentrations of well-mixed greenhouse gases is likely to have been about –2.8 Wm^–2 (see Figure 6.5 ). Ice sheet albedo forcing is estimated to have caused a global mean forcing of about –3.2 Wm^–2 (based on a range of several LGM simulations) and radiative forcing from increased atmospheric aerosols (primarily dust and vegetation) is estimated to have been about –1 Wm^–2 each. Therefore, the total annual and global mean radiative forcing during the LGM is likely to have been approximately –8 Wm^–2 relative to 1750, with large seasonal and geographical variations and significant uncertainties (see Section 6.4.1 ).

The major mid-Holocene forcing relative to the present was due to orbital perturbations that led to large changes in the seasonal cycle of insolation. The Northern Hemisphere (NH) seasonal cycle was about 27 Wm^–2 greater, whereas there was only a negligible change in NH annual mean solar forcing. For the Southern Hemisphere (SH), the seasonal forcing was –6.5 Wm^–2.In contrast, the global and annual mean net forcing was only 0.011 Wm^–2.

Changes in the Earth’s orbit have had little impact on annual mean insolation over the past millennium. Summer insolation decreased by 0.33 Wm^–2 at 45°N over the millennium, winter insolation increased by 0.83 Wm^–2 ( Goosse et al., 2005 [JoC, MoS, SRC] ), and the magnitude of the mean seasonal cycle of insolation in the NH decreased by 0.4 Wm^–2.Changes in insolation are also thought to have arisen from small variations in solar irradiance, although both timing and magnitude of past solar radiation fluctuations are highly uncertain (see Chapters 2 and 6 ;Lean et al., 2002 [JoC, MoS, ARC] ; Gray et al., 2005 [NPR, ARC] ; Foukal et al., 2006 [PoC, JoC, ARC] ). For example, sunspots were generally missing from approximately 1675 to 1715 (the so-called Maunder Minimum) and thus solar irradiance is thought to have been reduced during this period. The estimated difference between the present-day solar irradiance cycle mean and the Maunder Minimum is 0.08% (see Section 2.7.1.2.2), which corresponds to a radiative forcing of about 0.2 Wm^–2,which is substantially lower than estimates used in the TAR( Chapter2 ).

Natural external forcing also results from explosive volcanism that introduces aerosols into the stratosphere( Section 2.7.2 ), leading to a global negative forcing during the year following the eruption. Several reconstructions are available for the last two millennia and have been used to force climate models( Section 6.6.3 ). There is close agreement on the timing of large eruptions in the various compilations of historic volcanic activity, but large uncertainty in the magnitude of individual eruptions( Figure 6.13 ). Different reconstructions identify similar periods when eruptions happened more frequently. The uncertainty in the overall amplitude of the reconstruction of volcanic forcing is also important for quantifying the influence of volcanism on temperature reconstructions over longer periods, but is difficult to quantify and may be a substantial fraction of the best estimate (e.g., Hegerl et al., 2006a [NPR, PoC, JoC, SRC] ).

9.2.2.1 Spatial and Temporal Patterns of Response

The ability to distinguish between climate responses to different external forcing factors in observations depends on the extent to which those responses are distinct (see, e.g., Section 9.4.1.4 and Appendix 9.A). Figure 9.1 illustrates the zonal average temperature response in the PCM model (see Table 8.1 for model details) to several different forcing agents over the last 100 years, while Figure 9.2 illustrates the zonal average temperature response in the Commonwealth Scientific and Industrial Research Organisation (CSIRO) atmospheric model (when coupled to a simple mixed layer ocean model) to fossil fuel black carbon and organic matter, and to the combined effect of these forcings together with biomass burning aerosols ( Penner et al., 2007 [NPR, SRC, 2007] ). These figures indicate that the modelled vertical and zonal average signature of the temperature response should depend on the forcings. The major features shown in Figure 9.1 are robust to using different climate models. On the other hand, the response to black carbon forcing has not been widely examined and therefore the features in Figure 9.2 may be model dependent. Nevertheless, the response to black carbon forcings appears to be small.

Figure 9.1. Zonal mean atmospheric temperature change from 1890 to 1999 (°C per century) as simulated by the PCM model from (a) solar forcing, (b) volcanoes, (c) well-mixed greenhouse gases, (d) tropospheric and stratospheric ozone changes, (e) direct sulphate aerosol forcing and (f) the sum of all forcings. Plot is from 1,000 hPa to 10 hPa (shown on left scale) and from 0 km to 30 km (shown on right). See Appendix 9.C for additional information. Based on Santer et al. 2003a [PoC, JoC, MoS, SRC] ).

Figure 9.2. The zonal mean equilibrium temperature change (°C) between a present day minus a pre-industrial simulation by the CSIRO atmospheric model coupled to a mixed-layer ocean model from (a) direct forcing from fossil fuel black carbon and organic matter (BC+OM) and (b) the sum of fossil fuel BC+OM and biomass burning. Plot is from 1,000 hPa to 10 hPa (shown on left scale) and from 0 km to 30 km (shown on right). Note the difference in colour scale from Figure 9.1. See Supplementary Material, Appendix 9.C for additional information. Based on Penner et al. 2007 [NPR, SRC, 2007] ).

Greenhouse gas forcing is expected to produce warming in the troposphere, cooling in the stratosphere, and, for transient simulations, somewhat more warming near the surface in the NH due to its larger land fraction, which has a shorter surface response time to the warming than do ocean regions( Figure 9.1 c). The spatial pattern of the transient surface temperature response to greenhouse gas forcing also typically exhibits a land-sea pattern of stronger warming over land, for the same reason (e.g., Cubasch et al., 2001 [NPR, MoS, ARC] ). Sulphate aerosol forcing results in cooling throughout most of the globe, with greater cooling in the NH due to its higher aerosol loading( Figure 9.1 e; see Chapter2 ), thereby partially offsetting the greater NH greenhouse-gas induced warming. The combined effect of tropospheric and stratospheric ozone forcing( Figure 9.1 d) is expected to warm the troposphere, due to increases in tropospheric ozone, and cool the stratosphere, particularly at high latitudes where stratospheric ozone loss has been greatest. Greenhouse gas forcing is also expected to change the hydrological cycle worldwide, leading to disproportionately greater increases in heavy precipitation( Chapter 10 and Section 9.5.4 ), while aerosol forcing can influence rainfall regionally( Section 9.5.4 ).

The simulated responses to natural forcing are distinct from those due to the anthropogenic forcings described above. Solar forcing results in a general warming of the atmosphere( Figure 9.1 a) with a pattern of surface warming that is similar to that expected from greenhouse gas warming, but in contrast to the response to greenhouse warming, the simulated solar-forced warming extends throughout the atmosphere (see, e.g., Cubasch et al., 1997 [JoC, MoS, ARC] ). A number of independent analyses have identified tropospheric changes that appear to be associated with the solar cycle ( van Loon and Shea, 2000 [JoC, ARC] ; Gleisner and Thejll, 2003 [JoC] ; Haigh, 2003 [JoC, ARC] ; White et al., 2003 [JoC, ARC] ; Coughlin and Tung, 2004 [JoC] ;( Labitzke, 2004; ) Crooks and Gray, 2005 [JoC, SRC] ), suggesting an overall warmer and moister troposphere during solar maximum. The peak-to-trough amplitude of the response to the solar cycle globally is estimated to be approximately 0.1°C near the surface. Such variations over the 11-year solar cycle make it is necessary to use several decades of data in detection and attribution studies. The solar cycle also affects atmospheric ozone concentrations with possible impacts on temperatures and winds in the stratosphere, and has been hypothesised to influence clouds through cosmic rays( Section 2.7.1.3 ). Note that there is substantial uncertainty in the identification of climate response to solar cycle variations because the satellite period is short relative to the solar cycle length, and because the response is difficult to separate from internal climate variations and the response to volcanic eruptions ( Gray et al., 2005 [NPR, ARC] ).

Volcanic sulphur dioxide (SO₂)emissions ejected into the stratosphere form sulphate aerosols and lead to a forcing that causes a surface and tropospheric cooling and a stratospheric warming that peak several months after a volcanic eruption and last for several years. Volcanic forcing also likely leads to a response in the atmospheric circulation in boreal winter (discussed below) and a reduction in land precipitation ( Robock and Liu, 1994 [JoC, MoS] ; Broccoli et al., 2003 [JoC, MoS, ARC] ; Gillett et al., 2004b [JoC, SRC] ). The response to volcanic forcing causes a net cooling over the 20th century because of variations in the frequency and intensity of volcanic eruptions. This results in stronger volcanic forcing towards the end of the 20th century than early in the 20th century. In the PCM, this increase results in a small warming in the lower stratosphere and near the surface at high latitudes, with cooling elsewhere( Figure 9.1 b).

The net effect of all forcings combined is a pattern of NH temperature change near the surface that is dominated by the positive forcings (primarily greenhouse gases), and cooling in the stratosphere that results predominantly from greenhouse gas and stratospheric ozone forcing( Figure 9.1 f). Results obtained with the CSIRO model( Figure 9.2 )suggest that black carbon, organic matter and biomass aerosols would slightly enhance the NH warming shown in Figure 9.1 f. On the other hand, indirect aerosol forcing from fossil fuel aerosols may be larger than the direct effects that are represented in the CSIRO and PCM models, in which case the NH warming could be somewhat diminished. Also, while land use change may cause substantial forcing regionally and seasonally, its forcing and response are expected to have only a small impact at large spatial scales (Sections 9.3.3.3 and 7.2.2 ;Figures 2.20 and 2.23 ).

The spatial signature of a climate model’s response is seldom very similar to that of the forcing, due in part to the strength of the feedbacks relative to the initial forcing. This comes about because climate system feedbacks vary spatially and because the atmospheric and ocean circulation cause a redistribution of energy over the globe. For example, sea ice albedo feedbacks tend to enhance the high-latitude response of both a positive forcing, such as that of CO₂,and a negative forcing such as that of sulphate aerosol (e.g., Mitchell et al., 2001 [NPR] ; Rotstayn and Penner, 2001 [JoC, MoS, SRC] ). Cloud feedbacks can affect both the spatial signature of the response to a given forcing and the sign of the change in temperature relative to the sign of the radiative forcing( Section 8.6 ). Heating by black carbon, for example, can decrease cloudiness ( Ackerman et al., 2000 [JoC] ). If the black carbon is near the surface, it may increase surface temperatures, while at higher altitudes it may reduce surface temperatures ( Hansen et al., 1997 [JoC, MoS, ARC] ; Penner et al., 2003 [JoC, SRC] ). Feedbacks can also lead to differences in the response of different models to a given forcing agent, since the spatial response of a climate model to forcing depends on its representation of these feedbacks and processes. Additional factors that affect the spatial pattern of response include differences in thermal inertia between land and sea areas, and the lifetimes of the various forcing agents. Shorter-lived agents, such as aerosols, tend to have a more distinct spatial pattern of forcing, and can therefore be expected to have some locally distinct response features.

The pattern of response to a radiative forcing can also be altered quite substantially if the atmospheric circulation is affected by the forcing. Modelling studies and data comparisons suggest that volcanic aerosols (e.g., Kirchner et al., 1999 [JoC, MoS] ; Shindell et al., 1999 [PoC, JoC, MoS, ARC] ; Yang and Schlesinger, 2001 [JoC] ; Stenchikov et al., 2006 [JoC, MoS, ARC] ) and greenhouse gas changes (e.g., Fyfe et al., 1999 [JoC, ARC] ; Shindell et al., 1999 [PoC, JoC, MoS, ARC] ; Rauthe et al., 2004 [JoC, MoS] ) can alter the North Atlantic Oscillation (NAO) or the Northern Annular Mode (NAM). For example, volcanic eruptions, with the exception of high-latitude eruptions, are often followed by a positive phase of the NAM or NAO (e.g., Stenchikov et al., 2006 [JoC, MoS, ARC] ) leading to Eurasian winter warming that may reduce the overall cooling effect of volcanic eruptions on annual averages, particularly over Eurasia ( Perlwitz and Graf, 2001 [JoC] ; Stenchikov et al., 2002 [JoC, ARC] ; Shindell et al., 2003 [PoC, JoC, MoS, ARC] ; Stenchikov et al., 2004 [JoC, ARC] ; Oman et al., 2005 [JoC] ; Rind et al., 2005a [JoC, SRC] ; Miller et al., 2006 [PoC, JoC, MoS, ARC] ; Stenchikov et al., 2006 [JoC, MoS, ARC] ). In contrast, NAM or NAO responses to solar forcing vary between studies, some indicating a response, perhaps with dependence of the response on season or other conditions, and some finding no changes ( Shindell et al., 2001a [PoC, JoC, MoS, SRC] ,b; Ruzmaikin and Feynman, 2002 [JoC] ; Tourpali et al., 2003 [JoC, MoS] ; Egorova et al., 2004 [JoC, MoS] ; Palmer et al., 2004 [MoS, SRC] ; Stendel et al., 2006 [JoC, MoS, ARC] ; see also review in Gray et al., 2005 [NPR, ARC] ).

In addition to the spatial pattern, the temporal evolution of the different forcings( Figure 2.23 )generally helps to distinguish between the responses to different forcings. For example, Santer et al. 1996b [PoC, JoC, SRC] ,c) point out that a temporal pattern in the hemispheric temperature contrast would be expected in the second half of the 20th century with the SH warming more than the NH for the first two decades of this period and the NH subsequently warming more than the SH, as a result of changes in the relative strengths of the greenhouse gas and aerosol forcings. However, it should be noted that the integrating effect of the oceans ( Hasselmann, 1976 [JoC, MoS] ) results in climate responses that are more similar in time between different forcings than the forcings are to each other, and that there are substantial uncertainties in the evolution of the hemispheric temperature contrasts associated with sulphate aerosol forcing.

9.2.2.2 Aerosol Scattering and Cloud Feedback in Models and Observations

One line of observational evidence that reflective aerosol forcing has been changing over time comes from satellite observations of changes in top-of-atmosphere outgoing shortwave radiation flux. Increases in the outgoing shortwave radiation flux can be caused by increases in reflecting aerosols, increases in clouds or a change in the vertical distribution of clouds and water vapour, or increases in surface albedo. Increases in aerosols and clouds can cause decreases in surface radiation fluxes and decreases in surface warming. There has been continuing interest in this possibility ( Gilgen et al., 1998 [JoC, MoS, ARC] ;( Stanhill and Cohen, 2001; ) Liepert, 2002 [JoC, ARC] ). Sometimes called ‘global dimming’, this phenomena has reversed since about 1990 ( Pinker et al., 2005 [JoC] ; Wielicki et al., 2005 [JoC, SRC] ; Wild et al., 2005 [JoC, ARC] ; Section 3.4.3 ), but over the entire period from 1984 to 2001, surface solar radiation has increased by about 0.16 Wm^–2 yr^–1 on average ( Pinker et al., 2005 [JoC] ). Figure 9.3 shows the top-of-atmosphere outgoing shortwave radiation flux anomalies from the MMD at PCMDI, compared to that measured by the Earth Radiation Budget Satellite (ERBS; Wong et al., 2006 [JoC, SRC] ) and inferred from International Satellite Cloud Climatology Project (ISCCP) flux data (FD) ( Zhang et al., 2004 [JoC, MoS] ). The downward trend in outgoing solar radiation is consistent with the long-term upward trend in surface radiation found by Pinker et al. 2005 [JoC] ). The effect of the eruption of Mt. Pinatubo in 1991 results in an increase in the outgoing shortwave radiation flux (and a corresponding dimming at the surface) and its effect has been included in most (but not all) models in the MMD. The ISCCP flux anomaly for the Mt. Pinatubo signal is almost 2 Wm^–2 larger than that for ERBS, possibly due to the aliasing of the stratospheric aerosol signal into the ISCCP cloud properties. Overall, the trends from the ISCCP FD (–0.18 with 95% confidence limits of ±0.11 Wm^–2 yr^–1)and the ERBS data (–0.13 ± 0.08 Wm^–2 yr^–1)from 1984 to 1999 are not significantly different from each other at the 5% significance level, and are in even better agreement if only tropical latitudes are considered ( Wong et al., 2006 [JoC, SRC] ). These observations suggest an overall decrease in aerosols and/or clouds, while estimates of changes in cloudiness are uncertain (see Section 3.4.3 ). The model-predicted trends are also negative over this time period, but are smaller in most models than in the ERBS observations (which are considered more accurate than the ISCCP FD). Wielicki et al. 2002 [JoC, SRC] ) explain the observed downward trend by decreases in cloudiness, which are not well represented in the models on these decadal time scales ( Chen et al., 2002 [JoC] ; Wielicki et al., 2002 [JoC, SRC] ).

Figure 9.3. Comparison of outgoing shortwave radiation flux anomalies (in Wm^–2,calculated relative to the entire time period) from several models in the MMD archive at PCMDI (coloured curves) with ERBS satellite data (black with stars; Wong et al., 2006 [JoC, SRC] ) and with the ISCCP flux data set (black with squares; Zhang et al., 2004 [JoC, MoS] ). Models shown are CCSM3, CGCM3.1(T47), CGCM3.1(T63), CNRM-CM3, CSIRO-MK3.0, FGOALS-g1.0, GFDL-CM2.0, GFDL-CM2.1, GISS-AOM, GISS-EH, GISS-ER, INM-CM3.0, IPSL-CM4, and MRI-CGCM2.3.2 (see Table 8.1 for model details). The comparison is restricted to 60°S to 60°N because the ERBS data are considered more accurate in this region. Note that not all models included the volcanic forcing from Mt. Pinatubo ( 1991 – 1993 ) and so do not predict the observed increase in outgoing solar radiation. See Supplementary Material, Appendix 9.C for additional information.

9.2.2.3 Uncertainty in the Spatial Pattern of Response

Most detection methods identify the magnitude of the space-time patterns of response to forcing (sometimes called ‘fingerprints’) that provide the best fit to the observations. The fingerprints are typically estimated from ensembles of climate model simulations forced with reconstructions of past forcing. Using different forcing reconstructions and climate models in such studies provides some indication of forcing and model uncertainty. However, few studies have examined how uncertainties in the spatial pattern of forcing explicitly contribute to uncertainties in the spatial pattern of the response. For short-lived components, uncertainties in the spatial pattern of forcing are related to uncertainties in emissions patterns, uncertainties in the transport within the climate model or chemical transport model and, especially for aerosols, uncertainties in the representation of relative humidities or clouds. These uncertainties affect the spatial pattern of the forcing. For example, the ratio of the SH to NH indirect aerosol forcing associated with the total aerosol forcing ranges from –0.12 to 0.63 (best guess 0.29) in different studies, and that between ocean and land forcing ranges from 0.03 to 1.85 (see Figure 7.21 ;Rotstayn and Penner, 2001 [JoC, MoS, SRC] ; Chuang et al., 2002 [JoC, MoS] ; Kristjansson, 2002 [NPR, JoC] ; Lohmann and Lesins, 2002 [JoC, ARC] ; Menon et al., 2002a [JoC, MoS, ARC] ; Rotstayn and Liu, 2003 [JoC, SRC] ; Lohmann and Feichter, 2005 [ARC] ).

There are also large uncertainties in the magnitude of low-frequency changes in forcing associated with changes in total solar radiation as well as its spectral variation, particularly on time scales longer than the 11-year cycle. Previous estimates of change in total solar radiation have used sunspot numbers to calculate these slow changes in solar irradiance over the last few centuries, but these earlier estimates are not necessarily supported by current understanding and the estimated magnitude of low-frequency changes has been substantially reduced since the TAR ( Lean et al., 2002 [JoC, MoS, ARC] ; Foukal et al., 2004 [PoC, JoC, ARC] , 2006; Sections 6.6.3.1 and 2.7.1.2 ). In addition, the magnitude of radiative forcing associated with major volcanic eruptions is uncertain and differs between reconstructions ( Sato et al., 1993 [JoC, ARC] ; Andronova et al., 1999 [JoC, MoS, SRC] ; Ammann et al., 2003 [PoC, JoC, MoS, SRC] ), although the timing of the eruptions is well documented.

9.2.2.4 Uncertainty in the Temporal Pattern of Response

Climate model studies have also not systematically explored the effect of uncertainties in the temporal evolution of forcings. These uncertainties depend mainly on the uncertainty in the spatio-temporal expression of emissions, and, for some forcings, fundamental understanding of the possible change over time.

The increasing forcing by greenhouse gases is relatively well known. In addition, the global temporal history of SO₂ emissions, which have a larger overall forcing than the other short-lived aerosol components, is quite well constrained. Seven different reconstructions of the temporal history of global anthropogenic sulphur emissions up to 1990 have a relative standard deviation of less than 20% between 1890 and 1990, with better agreement in more recent years. This robust temporal history increases confidence in results from detection and attribution studies that attempt to separate the effects of sulphate aerosol and greenhouse gas forcing( Section 9.4.1 ).

In contrast, there are large uncertainties related to the anthropogenic emissions of other short-lived compounds and their effects on forcing. For example, estimates of historical emissions from fossil fuel combustion do not account for changes in emission factors (the ratio of the emitted gas or aerosol to the fuel burned) of short-lived species associated with concerns over urban air pollution (e.g., van Aardenne et al., 2001 [JoC] ). Changes in these emission factors would have slowed the emissions of nitrogen oxides as well as carbon monoxide after about 1970 and slowed the accompanying increase in tropospheric ozone compared to that represented by a single emission factor for fossil fuel use. In addition, changes in the height of SO₂ emissions associated with the implementation of tall stacks would have changed the lifetime of sulphate aerosols and the relationship between emissions and effects. Another example relates to the emissions of black carbon associated with the burning of fossil fuels. The spatial and temporal emissions of black carbon by continent reconstructed by Ito and Penner 2005 [JoC, SRC] ) are significantly different from those reconstructed using the methodology of Novakov et al. 2003 [JoC] ). For example, the emissions in Asia grow significantly faster in the inventory based on Novakov et al. 2003 [JoC] ) compared to those based on Ito and Penner 2005 [JoC, SRC] ). In addition, before 1988 the growth in emissions in Eastern Europe using the Ito and Penner 2005 [JoC, SRC] ) inventory is faster than the growth based on the methodology of Novakov et al. 2003 [JoC] ). Such spatial and temporal uncertainties will contribute to both spatial and temporal uncertainties in the net forcing and to spatial and temporal uncertainties in the distribution of forcing and response.

9.3.3.1 Evidence of External Influence on the Climate Over the Past 1,000 Years

Asubstantial number of proxy reconstructions of annual or decadal NH mean surface temperature are now available (see Figure 6.11 ,and the reviews by Jones et al., 2001 [PoC, JoC, ARC] and Jones and Mann, 2004 [PoC, JoC, ARC] ). Several new reconstructions have been published, some of which suggest larger variations over the last millennium than assessed in the TAR, but uncertainty remains in the magnitude of inter-decadal to inter-centennial variability. This uncertainty arises because different studies rely on different proxy data or use different reconstruction methods( Section 6.6.1 ). Nonetheless, NH mean temperatures in the second half of the 20th century were likely warmer than in any other 50-year period in the last 1.3 kyr( Chapter6 ), and very likely warmer than any such period in the last 500 years. Temperatures subsequently decreased, and then rose rapidly during the most recent 100 years. This long-term tendency is punctuated by substantial shorter-term variability( Figure 6.10 ). For example, cooler conditions with temperatures 0.5°C to 1°C below the 20th-century mean value are found in the 17th and early 18th centuries.

Anumber of simulations of the last millennium( Figure 6.13 )have been performed using a range of models, including some simulations with AOGCMs (e.g., Crowley, 2000 [PoC, JoC] ; Goosse and Renssen, 2001 [JoC, SRC] ; Bertrand et al., 2002 [JoC, ARC] ; Bauer et al., 2003 [JoC, MoS, ARC] ; Gerber et al., 2003 [JoC, MoS] ; see also Gonzalez-Rouco et al., 2003 [JoC, MoS, SRC] ; Jones and Mann, 2004 [PoC, JoC, ARC] ; Zorita et al., 2004 [MoS, SRC] ; Weber, 2005 [MoS] ; Tett et al., 2007 [JoC, MoS, SRC, 2007] ). These simulations use different reconstructions of external forcing, particularly solar, volcanic and greenhouse gas forcing, and often include land use changes (e.g., Bertrand et al., 2002 [JoC, ARC] ; Stendel et al., 2006 [JoC, MoS, ARC] ; Tett et al., 2007 [JoC, MoS, SRC, 2007] ). While the use of different models and forcing reconstructions leads to differences, the simulated evolution of the NH annual mean surface temperature displays some common characteristics between models that are consistent with the broad features of the data (Figures 6.13 and 9.4 ). For example, all simulations show relatively cold conditions during the period around 1675 to 1715 in response to natural forcing, which is in qualitative agreement with the proxy reconstructions. In all simulations shown in Figure 6.13 ,the late 20th century is warmer than any other multi-decadal period during the last millennium. In addition, there is significant correlation between simulated and reconstructed variability (e.g., Yoshimori et al., 2005 [JoC, MoS, ARC] ). By comparing simulated and observed atmospheric CO₂ concentration during the last 1 kyr, Gerber et al. 2003 [JoC, MoS] ) suggest that the amplitude of the temperature evolution simulated by simple climate models and EMICs is consistent with the observed evolution of CO₂.Since reconstructions of external forcing are virtually independent from the reconstructions of past temperatures, this broad consistency increases confidence in the broad features of the reconstructions and the understanding of the role of external forcing in recent climate variability. The simulations also show that it is not possible to reproduce the large 20th-century warming without anthropogenic forcing regardless of which solar or volcanic forcing reconstruction is used ( Crowley, 2000 [PoC, JoC] ; Bertrand et al., 2002 [JoC, ARC] ; Bauer et al., 2003 [JoC, MoS, ARC] ; Hegerl et al., 2003 [JoC, SRC] , 2007 ), stressing the impact of human activity on the recent warming.

While there is broad qualitative agreement between simulated and reconstructed temperatures, it is difficult to fully assess model-simulated variability because of uncertainty in the magnitude of historical variations in the reconstructions and differences in the sensitivity to external forcing( Table 8.2 ). The role of internal variability has been found to be smaller than that of the forced variability for hemispheric temperature means at decadal or longer time scales ( Crowley, 2000 [PoC, JoC] ; Hegerl et al., 2003 [JoC, SRC] ; Goosse et al., 2004 [JoC, MoS, SRC] ; Weber et al., 2004 [PoC, JoC, MoS] ; Hegerl et al., 2007 [JoC, SRC, 2007] ; Tett et al., 2007 [JoC, MoS, SRC, 2007] ), and thus internal variability is a relatively small contributor to differences between different simulations of NH mean temperature. Other sources of uncertainty in simulations include model ocean initial conditions, which, for example, explain the warm conditions found in the Zorita et al. 2004 [MoS, SRC] ) simulation during the first part of the millennium ( Goosse et al., 2005 [JoC, MoS, SRC] ; Osborn et al., 2006 [Ambiguous] ).

9.3.3.2 Role of Volcanism and Solar Irradiance

Volcanic eruptions cause rapid decreases in hemispheric and global mean temperatures followed by gradual recovery over several years( Section 9.2.2.1 )in climate simulations driven by volcanic forcing( Figure 6.13 ;Crowley, 2000 [PoC, JoC] ; Bertrand et al., 2002 [JoC, ARC] ; Weber, 2005 [MoS] ; Yoshimori et al., 2005 [JoC, MoS, ARC] ; Tett et al., 2007 [JoC, MoS, SRC, 2007] ). These simulated changes appear to correspond to cool episodes in proxy reconstructions( Figure 6.13 ). This suggestive correspondence has been confirmed in comparisons between composites of temperatures following multiple volcanic eruptions in simulations and reconstructions ( Hegerl et al., 2003 [JoC, SRC] ; Weber, 2005 [MoS] ). In addition, changes in the frequency of large eruptions result in climate variability on decadal and possibly longer time scales ( Crowley, 2000 [PoC, JoC] ; Briffa et al., 2001 [PoC, JoC, ARC] ; Bertrand et al., 2002 [JoC, ARC] ; Bauer et al., 2003 [JoC, MoS, ARC] ; Weber, 2005 [MoS] Hegerl et al. 2003 [JoC, SRC] ; 2007 ), using a multi-regression approach based on Energy Balance Model (EBM) simulated fingerprints of solar, volcanic and greenhouse gas forcing (Appendix 9.A.1; see also Section 9.4.1.4 for the 20th century), simultaneously detect the responses to volcanic and greenhouse gas forcing in a number of proxy reconstructions of average NH mean annual and growing season temperatures( Figure 9.4 )with high significance. They find that a high percentage of decadal variance in the reconstructions used can be explained by external forcing (between 49 and 70% of decadal variance depending upon the reconstruction).

Figure 9.4. Contribution of external forcing to several high-variance reconstructions of NH temperature anomalies, ( Esper et al., 2002 [PoC, JoC, ARC] ; Briffa et al., 2001 [PoC, JoC, ARC] ; Hegerl et al., 2007 [JoC, SRC, 2007] , termed CH-blend and CH-blend long; and Moberg et al., 2005 [JoC] ). The top panel compares reconstructions to an EBM simulation (equilibrium climate sensitivity of 2.5°C) of NH 30°N to 90°N average temperature, forced with volcanic, solar and anthropogenic forcing. All timeseries are centered on the 1500 - 1925 average. Instrumental temperature data are shown by a green line (centered to agree with CH-blend average over the period 1880 - 1960 ). The displayed data are low-pass filtered (20-year cutoff) for clarity. The bottom panel shows the estimated contribution of the response to volcanic (blue lines with blue uncertainty shade), solar (green) and greenhouse gas (GHG) and aerosol forcing (red line with yellow shades, aerosol only in 20th century) to each reconstruction (all timeseries are centered over the analysis period). The estimates are based on multiple regression of the reconstructions on fingerprints for individual forcings. The contributions to different reconstructions are indicated by different line styles (Briffa et al.: solid, fat; Esper et al.: dotted; Moberg: dashed; CH-blend: solid, thin; with shaded 90% confidence limits around best estimates for each detectable signal). All reconstructions show a highly significant volcanic signal, and all but Moberg et al. (which ends in 1925 ) show a detectable greenhouse gas signal at the 5% significance level. The latter shows a detectable greenhouse gas signal with less significance. Only Moberg et al. contains a detectable solar signal (only shown for these data and CH-blend, where it is not detectable). All data are decadally averaged. The reconstructions represent slightly different regions and seasons: Esper et al. 2002 [PoC, JoC, ARC] ) is calibrated to 30°N to 90°N land temperature, CH-blend and CH-blend long ( Hegerl et al., 2007 [JoC, SRC, 2007] ) to 30°N to 90°N mean temperature and Moberg et al. 2005 [JoC] ) to 0° to 90°N temperature. From Hegerl et al. 2007 [JoC, SRC, 2007] ).

There is more uncertainty regarding the influence of solar forcing. In addition to substantial uncertainty in the timing and amplitude of solar variations on time scales of several decades to centuries, which has increased since the TAR although the estimate of solar forcing has been revised downwards (Sections 9.2.1.3 and 2.7.1 ), uncertainty also arises because the spatial response of surface temperature to solar forcing resembles that due to greenhouse gas forcing( Section 9.2.3 ). Analyses that make use of differences in the temporal evolution of solar and volcanic forcings are better able to distinguish between the two( Section 9.2.3 ;see also Section 9.4.1.5 for the 20th century). In such an analysis, solar forcing can only be detected and distinguished from the effect of volcanic and greenhouse gas forcing over some periods in some reconstructions ( Hegerl et al., 2003 [JoC, SRC] , 2007 ), although the effect of solar forcing has been detected over parts of the 20th century in some time-space analyses( Section 9.4.1.5 )and there are similarities between regressions of solar forcing on model simulations and several proxy reconstructions ( Weber, 2005 [MoS] ; see also Waple, 2002 [PoC, JoC, MoS] ). A model simulation ( Shindell et al., 2003 [PoC, JoC, MoS, ARC] ) suggests that solar forcing may play a substantial role in regional anomalies due to dynamical feedbacks. These uncertainties in the contribution of different forcings to climatic events during the last millennium reflect substantial uncertainty in knowledge about past solar and volcanic forcing, as well as differences in the way these effects are taken into account in model simulations.

Overall, modelling and detection and attribution studies confirm a role of volcanic, greenhouse gas and probably solar forcing in explaining the broad temperature evolution of the last millennium, although the role of solar forcing has recently been questioned ( Foukal et al., 2006 [PoC, JoC, ARC] ). The variability that remains in proxy reconstructions after estimates of the responses to external forcing have been removed is broadly consistent with AOGCM-simulated internal variability (e.g., Hegerl et al., 2003 [JoC, SRC] , 2007 ), providing a useful check on AOGCMs even though uncertainties are large. Such studies also help to explain episodes during the climate of the last millennium. For example, several modelling studies suggest that volcanic activity has a dominant role in explaining the cold conditions that prevailed from 1675 to 1715 ( Andronova et al., 2007 [NPR, SRC, 2007] ; Yoshimori et al., 2005 [JoC, MoS, ARC] ). In contrast, Rind et al. 2004 [JoC, MoS, SRC] ) estimate from model simulations that the cooling relative to today was primarily associated with reduced greenhouse gas forcing, with a substantial contribution from solar forcing.

There is also some evidence from proxy data that the response to external forcing may influence modes of climate variability. For example, Cobb et al. 2003 [JoC] ), using fossil corals, attempt to extend the ENSO record back through the last millennium. They find that ENSO events may have been as frequent and intense during the mid-17th century as during the instrumental period, with events possibly rivalling the strong 1997 – 1998 event. On the other hand, there are periods during the 12th and 14th centuries when there may have been significantly less ENSO variability, a period during which there were also cooler conditions in the northeast Pacific ( MacDonald and Case, 2005 [JoC] ) and evidence of droughts in central North America ( Cook et al., 2004 [PoC, JoC, ARC] Cobb et al. 2003 [JoC] ) find that fluctuations in reconstructed ENSO variability do not appear to be correlated in an obvious way with mean state changes in the tropical Pacific or global mean climate, while Adams et al. 2003 [PoC, JoC, MoS, SRC] ) find statistical evidence for an El Niño-like anomaly during the first few years following explosive tropical volcanic eruptions. The Cane-Zebiak model simulates changes similar to those in the Cobb et al. 2003 [JoC] ) data when volcanism and solar forcing are accounted for, supporting the link with volcanic forcing over the past millennium ( Mann et al., 2005 [PoC, JoC, MoS, SRC] ). However, additional studies with different models are needed to fully assess this relationship, since previous work was less conclusive ( Robock, 2000 [JoC] ).

Extratropical variability also appears to respond to volcanic forcing. During the winter following a large volcanic eruption, the zonal circulation may be more intense, causing a relative warming over the continents during the cold season that could partly offset the direct cooling due to the volcanic aerosols (Sections 9.2.2.1 and 8.4.1 ;Robock, 2000 [JoC] ; Shindell et al., 2003 [PoC, JoC, MoS, ARC] ). A tendency towards the negative NAO state during periods of reduced solar input is found in some reconstructions of this pattern for the NH ( Shindell et al., 2001b [JoC, MoS] ; Luterbacher et al., 2002 [ARC] , 2004; Stendel et al., 2006 [JoC, MoS, ARC] ), possibly implying a solar forcing role in some long-term regional changes, such as the cooling over the NH continents around 1700 ( Shindell et al., 2001b [JoC, MoS] ; Section 9.2.2 ). Indications of changes in ENSO variability during the low solar irradiance period of the 17th to early 18th centuries are controversial (e.g., D’Arrigo et al., 2005 [JoC] ).

9.3.3.3 Other Forcings and Sources of Uncertainties

In addition to forcing uncertainties discussed above, a number of other uncertainties affect the understanding of pre-industrial climate change. For example, land cover change may have influenced the pre-industrial climate ( Bertrand et al., 2002 [JoC, ARC] ; Bauer et al., 2003 [JoC, MoS, ARC] ), leading to a regional cooling of 1°C to 2°C in winter and spring over the major agricultural regions of North America and Eurasia in some model simulations, when pre-agriculture vegetation was replaced by present-day vegetation ( Betts, 2001 [MoS, SRC] ). The largest anthropogenic land cover changes involve deforestation( Chapter2 ). The greatest proportion of deforestation has occurred in the temperate regions of the NH ( Ramankutty and Foley, 1999 [JoC, MoS, ARC] ; Goldewijk, 2001 [JoC, MoS] ). Europe had cleared about 80% of its agricultural area by 1860, but over half of the forest removal in North America took place after 1860 ( Betts, 2001 [MoS, SRC] ), mainly in the late 19th century ( Stendel et al., 2006 [JoC, MoS, ARC] ). During the past two decades, the CO₂ flux caused by land use changes has been dominated by tropical deforestation( Section 7.3.2.1.2 ). Climate model simulations suggest that the effect of land use change was likely small at hemispheric and global scales, estimated variously as –0.02°C relative to natural pre-agricultural vegetation ( Betts, 2001 [MoS, SRC] ), less than –0.1°C since 1700 ( Stendel et al., 2006 [JoC, MoS, ARC] ) and about –0.05°C over the 20th century and too small to be detected statistically in observed trends ( Matthews et al., 2004 [JoC, ARC] ). However, the latter authors did find a larger cooling effect since 1700 of between –0.06°C and –0.22°C when they explored the sensitivity to different representations of land cover change.

Oceanic processes and ocean-atmosphere interaction may also have played a role in the climate evolution during the last millennium ( Delworth and Knutson, 2000 [JoC, MoS, SRC] ; Weber et al., 2004 [PoC, JoC, MoS] ; van der Schrier and Barkmeijer, 2005 [JoC] ). Climate models generally simulate a weak to moderate increase in the intensity of the oceanic meridional overturning circulation in response to a decrease in solar irradiance ( Cubasch et al., 1997 [JoC, MoS, ARC] ; Goosse and Renssen, 2004 [JoC, MoS, SRC] ; Weber et al., 2004 [PoC, JoC, MoS] ). A delayed response to natural forcing due to the storage and transport of heat anomalies by the deep ocean has been proposed to explain the warm Southern Ocean around the 14th to 15th centuries ( Goosse et al., 2004 [JoC, MoS, SRC] ).

9.4.1.1 Observed Changes

Six additional years of observations since the TAR( Chapter3 )show that temperatures are continuing to warm near the surface of the planet. The annual global mean temperature for every year since the TAR has been among the 10 warmest years since the beginning of the instrumental record. The global mean temperature averaged over land and ocean surfaces warmed by 0.76°C ± 0.19°C between the first 50 years of the instrumental record ( 1850 – 1899 ) and the last 5 years ( 2001 – 2005 )( Chapter3 ;with a linear warming trend of 0.74°C ± 0.18°C over the last 100 years ( 1906 – 2005 )). The rate of warming over the last 50 years is almost double that over the last 100 years (0.13°C ± 0.03°C vs 0.07°C ± 0.02°C per decade; Chapter3 ). The larger number of proxy reconstructions from palaeodata than were available for the TAR indicate that it is very likely that average NH temperatures during the second half of the 20th century were warmer than any other 50-year period in the last 500 years and it is likely that this was the warmest period in the past 1.3 kyr( Chapter6 ). Global mean temperature has not increased smoothly since 1900 as would be expected if it were influenced only by forcing from increasing greenhouse gas concentrations (i.e., if natural variability and other forcings did not have a role; see Section 9.2.1 ; Chapter2 ). A rise in near-surface temperatures also occurred over several decades during the first half of the 20th century, followed by a period of more than three decades when temperatures showed no pronounced trend( Figure 3.6 ). Since the mid- 1970 s, land regions have warmed at a faster rate than oceans in both hemispheres( Figure 3.8 )and warming over the SH was smaller than that over the NH during this period( Figure 3.6 ), while warming rates during the early 20th century were similar over land and ocean.

9.4.1.2 Simulations of the 20th Century

There are now a greater number of climate simulations from AOGCMs for the period of the global surface instrumental record than were available for the TAR, including a greater variety of forcings in a greater variety of combinations. These simulations used models with different climate sensitivities, rates of ocean heat uptake and magnitudes and types of forcings (Supplementary Material, Table S9.1). Figure 9.5 shows that simulations that incorporate anthropogenic forcings, including increasing greenhouse gas concentrations and the effects of aerosols, and that also incorporate natural external forcings provide a consistent explanation of the observed temperature record, whereas simulations that include only natural forcings do not simulate the warming observed over the last three decades. A variety of different forcings is used in these simulations. For example, some anthropogenically forced simulations include both the direct and indirect effects of sulphate aerosols whereas others include just the direct effect, and the aerosol forcing that is calculated within models differs due to differences in the representation of physics. Similarly, the effects of tropospheric and stratospheric ozone changes are included in some simulations but not others, and a few simulations include the effects of carbonaceous aerosols and land use changes, while the naturally forced simulations include different representations of changing solar and volcanic forcing. Despite this additional uncertainty, there is a clear separation in Figure 9.5 between the simulations with anthropogenic forcings and those without.

Figure 9.5. Comparison between global mean surface temperature anomalies (°C) from observations (black) and AOGCM simulations forced with (a) both anthropogenic and natural forcings and (b) natural forcings only. All data are shown as global mean temperature anomalies relative to the period 1901 to 1950, as observed (black, Hadley Centre/Climatic Research Unit gridded surface temperature data set (HadCRUT3); Brohan et al., 2006 [JoC, MoS] ) and, in (a) as obtained from 58 simulations produced by 14 models with both anthropogenic and natural forcings. The multi-model ensemble mean is shown as a thick red curve and individual simulations are shown as thin yellow curves. Vertical grey lines indicate the timing of major volcanic events. Those simulations that ended before 2005 were extended to 2005 by using the first few years of the IPCC Special Report on Emission Scenarios (SRES) A1B scenario simulations that continued from the respective 20th-century simulations, where available. The simulated global mean temperature anomalies in (b) are from 19 simulations produced by five models with natural forcings only. The multi-model ensemble mean is shown as a thick blue curve and individual simulations are shown as thin blue curves. Simulations are selected that do not exhibit excessive drift in their control simulations (no more than 0.2°C per century). Each simulation was sampled so that coverage corresponds to that of the observations. Further details of the models included and the methodology for producing this figure are given in the Supplementary Material, Appendix 9.C .After Stott et al. 2006b [JoC, MoS, SRC] ).

Global mean and hemispheric-scale temperatures on multi-decadal time scales are largely controlled by external forcings ( Stott et al., 2000 [JoC, MoS, SRC] ). This external control is demonstrated by ensembles of model simulations with identical forcings (whether anthropogenic or natural) whose members exhibit very similar simulations of global mean temperature on multi-decadal time scales (e.g., Stott et al., 2000 [JoC, MoS, SRC] ; Broccoli et al., 2003 [JoC, MoS, ARC] ; Meehl et al., 2004 [JoC, MoS, ARC] ). Larger interannual variations are seen in the observations than in the ensemble mean model simulation of the 20th century because the ensemble averaging process filters out much of the natural internal interannual variability that is simulated by the models. The interannual variability in the individual simulations that is evident in Figure 9.5 suggests that current models generally simulate large-scale natural internal variability quite well, and also capture the cooling associated with volcanic eruptions on shorter time scales. Section 9.4.1.3 assesses the variability of near surface temperature observations and simulations.

The fact that climate models are only able to reproduce observed global mean temperature changes over the 20th century when they include anthropogenic forcings, and that they fail to do so when they exclude anthropogenic forcings, is evidence for the influence of humans on global climate. Further evidence is provided by spatial patterns of temperature change. Figure 9.6 compares observed near-surface temperature trends over the globe (top row) with those simulated by climate models when they include anthropogenic and natural forcing (second row) and the same trends simulated by climate models when only natural forcings are included (third row). The observed trend over the entire 20th century( Figure 9.6 ,top left panel) shows warming almost everywhere with the exception of the southeastern USA, northern North Atlantic, and isolated grid boxes in Africa and South America (see also Figure 3.9 ). Such a pattern of warming is not associated with known modes of internal climate variability. For example, while El Niño or El Niño-like decadal variability results in unusually warm annual temperatures, the spatial pattern associated with such a warming is more structured, with cooling in the North Pacific and South Pacific (see, e.g., Zhang et al., 1997 [JoC] ). In contrast, the trends in climate model simulations that include anthropogenic and natural forcing( Figure 9.6 ,second row) show a pattern of spatially near-uniform warming similar to that observed. There is much greater similarity between the general evolution of the warming in observations and that simulated by models when anthropogenic and natural forcings are included than when only natural forcing is included( Figure 9.6 ,third row). Figure 9.6 (fourth row) shows that climate models are only able to reproduce the observed patterns of zonal mean near-surface temperature trends over the 1901 to 2005 and 1979 to 2005 periods when they include anthropogenic forcings and fail to do so when they exclude anthropogenic forcings. Although there is less warming at low latitudes than at high northern latitudes, there is also less internal variability at low latitudes, which results in a greater separation of the climate simulations with and without anthropogenic forcings.

Figure 9.6. Trends in observed and simulated temperature changes (°C) over the 1901 to 2005 (left column) and 1979 to 2005 (right column) periods. First row: trends in observed temperature changes (Hadley Centre/Climatic Research Unit gridded surface temperature data set (HadCRUT3), Brohan et al., 2006 [JoC, MoS] ). Second row: average trends in 58 historical simulations from 14 climate models including both anthropogenic and natural forcings. Third row: average trends in 19 historical simulations from five climate models including natural forcings only. Grey shading in top three rows indicates regions where there are insufficient observed data to calculate a trend for that grid box (see Supplementary Material, Appendix 9.C for further details of data exclusion criteria). Fourth row: average trends for each latitude; observed trends are indicated by solid black curves. Red shading indicates the middle 90% range of trend estimates from the 58 simulations including both anthropogenic and natural forcings (estimated as the range between 4th and 55th of the 58 ranked simulations); blue shading indicates the middle 90% range of trend estimates from the 19 simulations with natural forcings only (estimated as the range between 2nd and 18th of the 19 ranked simulations); for comparison, the dotted black curve in the right-hand plot shows the observed 1901 to 2005 trend. Note that scales are different between columns. The ‘ALL’ simulations were extended to 2005 by adding their IPCC Special Report on Emission Scenarios (SRES) A1B continuation runs where available. Where not available, and in the case of the ‘NAT’ simulations, the mean for the 1996 to 2005 decade was estimated using model output from 1996 to the end of the available runs. In all plots, each climate simulation was sampled so that coverage corresponds to that of the observations. Further details of the models included and the methodology for producing this figure are given in the Supplementary Material, Appendix 9.C .

Climate simulations are consistent in showing that the global mean warming observed since 1970 can only be reproduced when models are forced with combinations of external forcings that include anthropogenic forcings( Figure 9.5 ). This conclusion holds despite a variety of different anthropogenic forcings and processes being included in these models (e.g., Tett et al., 2002 [JoC, MoS, SRC] ; Broccoli et al., 2003 [JoC, MoS, ARC] ; Meehl et al., 2004 [JoC, MoS, ARC] ; Knutson et al., 2006 [JoC, MoS, ARC] ). In all cases, the response to forcing from well-mixed greenhouse gases dominates the anthropogenic warming in the model. No climate model using natural forcings alone has reproduced the observed global warming trend in the second half of the 20th century. Therefore, modelling studies suggest that late 20th-century warming is much more likely to be anthropogenic than natural in origin, a finding which is confirmed by studies relying on formal detection and attribution methods( Section 9.4.1.4 ).

Modelling studies are also in moderately good agreement with observations during the first half of the 20th century when both anthropogenic and natural forcings are considered, although assessments of which forcings are important differ, with some studies finding that solar forcing is more important ( Meehl et al., 2004 [JoC, MoS, ARC] ) while other studies find that volcanic forcing ( Broccoli et al., 2003 [JoC, MoS, ARC] ) or internal variability ( Delworth and Knutson, 2000 [JoC, MoS, SRC] ) could be more important. Differences between simulations including greenhouse gas forcing only and those that also include the cooling effects of sulphate aerosols (e.g., Tett et al., 2002 [JoC, MoS, SRC] ) indicate that the cooling effects of sulphate aerosols may account for some of the lack of observational warming between 1950 and 1970, despite increasing greenhouse gas concentrations, as was proposed by Schwartz 1993 [ARC] ). In contrast, Nagashima et al. 2006 [JoC] ) find that carbonaceous aerosols are required for the MIROC model (see Table 8.1 for a description) to provide a statistically consistent representation of observed changes in near-surface temperature in the middle part of the 20th century. The mid-century cooling that the model simulates in some regions is also observed, and is caused in the model by regional negative surface forcing from organic and black carbon associated with biomass burning. Variations in the Atlantic Multi-decadal Oscillation (see Section 3.6.6 for a more detailed discussion) could account for some of the evolution of global and hemispheric mean temperatures during the instrumental period ( Schlesinger and Ramankutty, 1994 [JoC, ARC] ; Andronova and Schlesinger, 2000 [JoC, SRC] ; Delworth and Mann, 2000 [PoC, JoC, MoS, SRC] Knight et al. 2005 [JoC] ) estimate that variations in the Atlantic Multi-decadal Oscillation could account for up to 0.2°C peak-to-trough variability in NH mean decadal temperatures.

9.4.1.3 Variability of Temperature from Observations and Models

Year-to-year variability of global mean temperatures simulated by the most recent models compares reasonably well with that of observations, as can be seen by comparing observed and modelled variations in Figure 9.5 a. A more quantitative evaluation of modelled variability can be carried out by comparing the power spectra of observed and modelled global mean temperatures. Figure 9.7 compares the power spectrum of observations with the power spectra of transient simulations of the instrumental period. This avoids the need to compare variability estimated from long control runs of models with observed variability, which is difficult because observations are likely to contain a response to external forcings that cannot be reliably removed by subtracting a simple linear trend. The simulations considered contain both anthropogenic and natural forcings, and include most 20th Century Climate in Coupled Models (20C3M) simulations in the MMD at PCMDI. Figure 9.7 shows that the models have variance at global scales that is consistent with the observed variance at the 5% significance level on the decadal to inter-decadal time scales important for detection and attribution. Figure 9.8 shows that this is also generally the case at continental scales, although model uncertainty is larger at smaller scales( Section 9.4.2.2 ).

Figure 9.7. Comparison of variability as a function of time scale of annual global mean temperatures (°C² yr^–1)from the observed record (Hadley Centre/Climatic Research Unit gridded surface temperature data set (HadCRUT3), Brohan et al., 2006 [JoC, MoS] ) and from AOGCM simulations including both anthropogenic and natural forcings. All power spectra are estimated using a Tukey-Hanning filter of width 97 years. The model spectra displayed are the averages of the individual spectra estimated from individual ensemble members. The same 58 simulations and 14 models are used as in Figure 9.5a. All models simulate variability on decadal time scales and longer that is consistent with observations at the 10% significance level. Further details of the method of calculating the spectra are given in the Supplementary Material, Appendix 9.C .

Figure 9.8. As Figure 9.7, except for continental mean temperature. Spectra are calculated in the same manner as Figure 9.7. See the Supplementary Material, Appendix 9.C for a description of the regions and for details of the method used. Models simulate variability on decadal time scales and longer that is consistent with observations in all cases except two models over South America, five models over Asia and two models over Australia (at the 10% significance level).

Detection and attribution studies routinely assess if the residual variability unexplained by forcing is consistent with the estimate of internal variability (e.g., Allen and Tett, 1999 [JoC, MoS, SRC] ; Tett et al., 1999 [JoC, SRC] ; Stott et al., 2001 [JoC, SRC] ; Zwiers and Zhang, 2003 [JoC, SRC] ). Furthermore, there is no evidence that the variability in palaeoclimatic reconstructions that is not explained by forcing is stronger than that in models, and simulations of the last 1 kyr show similar variability to reconstructions( Section 9.3.3.2 ). Chapter8 discusses the simulation of major modes of variability and the extent to which they are simulated by models (including on decadal to inter-decadal time scales).

9.4.1.4 The Influence of Greenhouse Gas and Total Anthropogenic Forcing on Global Surface Temperature

Since the TAR, a large number of studies based on the longer observational record, improved models and stronger signal-to-noise ratio have increased confidence in the detection of an anthropogenic signal in the instrumental record (see, e.g., the recent review by IDAG, 2005 [JoC] ). Many more detection and attribution studies are now available than were available for the TAR, and these have used more recent climate data than previous studies and a much greater variety of climate simulations with more sophisticated treatments of a greater number of both anthropogenic and natural forcings.

Fingerprint studies that use climate change signals estimated from an array of climate models indicate that detection of an anthropogenic contribution to the observed warming is a result that is robust to a wide range of model uncertainty, forcing uncertainties and analysis techniques ( Hegerl et al., 2001 [JoC, SRC] ; Gillett et al., 2002c [NPR, JoC, MoS, SRC] ; Tett et al., 2002 [JoC, MoS, SRC] ; Zwiers and Zhang, 2003 [JoC, SRC] ; IDAG, 2005 [JoC] ; Stone and Allen, 2005b [JoC, MoS, SRC] ; Stone et al., 2007a [JoC, MoS, SRC, 2007] ,b; Stott et al., 2006b [JoC, MoS, SRC] ,c; Zhang et al., 2006 [JoC, MoS, SRC] ). These studies account for the possibility that the agreement between simulated and observed global mean temperature changes could be fortuitous as a result of, for example, balancing too great (or too small) a model sensitivity with a too large (or too small) negative aerosol forcing ( Schwartz, 2004 [MoS, ARC] ; Hansen et al., 2005 [JoC, ARC] ) or a too small (or too large) warming due to solar changes. Multi-signal detection and attribution analyses do not rely on such agreement because they seek to explain the observed temperature changes in terms of the responses to individual forcings, using model-derived patterns of response and a noise-reducing metric (Appendix 9.A) but determining their amplitudes from observations. As discussed in Section 9.2.2.1 ,these approaches make use of differences in the temporal and spatial responses to forcings to separate their effect in observations.

Since the TAR, there has also been an increased emphasis on quantifying the greenhouse gas contribution to observed warming, and distinguishing this contribution from other factors, both anthropogenic, such as the cooling effects of aerosols, and natural, such as from volcanic eruptions and changes in solar radiation.

Acomparison of results using four different models( Figure 9.9 )shows that there is a robust identification of a significant greenhouse warming contribution to observed warming that is likely greater than the observed warming over the last 50 years with a significant net cooling from other anthropogenic forcings over that period, dominated by aerosols. Stott et al. 2006c [JoC, MoS, SRC] ) compare results over the 20th century obtained using the UKMO-HadCM3, PCM (see Table 8.1 for model descriptions) and Geophysical Fluid Dynamics Laboratory (GFDL) R30 models. They find consistent estimates for the greenhouse gas attributable warming over the century, expressed as the difference between temperatures in the last and first decades of the century, of 0.6°C to 1.3°C (5 to 95%) offset by cooling from other anthropogenic factors associated mainly with cooling from aerosols of 0.1°C to 0.7°C and a small net contribution from natural factors over the century of –0.1°C to 0.1°C( Figure 9.9 b). Scaling factors for the model response to three forcings are shown in Figure 9.9 a. A similar analysis for the MIROC3.2 model (see Table 8.1 for a description) finds a somewhat larger warming contribution from greenhouse gases of 1.2°C to 1.5°C offset by a cooling of 0.6°C to 0.8°C from other anthropogenic factors and a very small net natural contribution( Figure 9.9 b). In all cases, the fifth percentile of the warming attributable to greenhouse gases is greater than the observed warming over the last 50 years of the 20th century( Figure 9.9 c).

The detection and estimation of a greenhouse gas signal is also robust to accounting more fully for model uncertainty. An analysis that combines results from three climate models and thereby incorporates uncertainty in the response of these three models (by including an estimate of the inter-model covariance structure in the regression method; Huntingford et al., 2006 [JoC, MoS, SRC] ), supports the results from each of the models individually that it is likely that greenhouse gases would have caused more warming than was observed over the 1950 to 1999 period( Figure 9.9 ,results labelled ‘EIV’). These results are consistent with the results of an earlier analysis, which calculated the mean response patterns from five models and included a simpler estimate of model uncertainty (obtained by a simple rescaling of the variability estimated from a long control run, thereby assuming that inter-model uncertainty has the same covariance structure as internal variability; Gillett et al., 2002c [NPR, JoC, MoS, SRC] ). Both the results of Gillett et al. 2002c [NPR, JoC, MoS, SRC] and Huntingford et al. 2006 [JoC, MoS, SRC] ) indicate that inter-model differences do not greatly increase detection and attribution uncertainties and that averaging fingerprints improves detection results.

Figure 9.9. Estimated contribution from greenhouse gas (red), other anthropogenic (green) and natural (blue) components to observed global mean surface temperature changes, based on ‘optimal’ detection analyses (Appendix 9.A). (a) 5 to 95% uncertainty limits on scaling factors (dimensionless) based on an analysis over the 20th century, (b) the estimated contribution of forced changes to temperature changes over the 20th century, expressed as the difference between 1990 to 1999 mean temperature and 1900 to 1909 mean temperature (°C) and (c) estimated contribution to temperature trends over 1950 to 1999 (°C per 50 years). The horizontal black lines in (b) and (c) show the observed temperature changes from the Hadley Centre/Climatic Research Unit gridded surface temperature data set (HadCRUT2v; Parker et al., 2004 [ARC] ). The results of full space-time optimal detection analyses ( Nozawa et al., 2005 [JoC, SRC] ; Stott et al., 2006c [JoC, MoS, SRC] ) using a total least squares algorithm ( Allen and Stott, 2003 [JoC, MoS, SRC] ) from ensembles of simulations containing each set of forcings separately are shown for four models, MIROC3.2(medres), PCM, UKMO-HadCM3 and GFDL-R30. Also shown, labelled ‘EIV’, is an optimal detection analysis using the combined spatio-temporal patterns of response from three models (PCM, UKMO-HadCM3 and GFDL-R30) for each of the three forcings separately, thus incorporating inter-model uncertainty ( Huntingford et al., 2006 [JoC, MoS, SRC] ).

Arobust anthropogenic signal is also found in a wide range of climate models that do not have the full range of simulations required to directly estimate the responses to individual forcings required for the full multi-signal detection and attribution analyses ( Stone et al., 2007a [JoC, MoS, SRC, 2007] ,b). In these cases, an estimate of the model’s pattern of response to each individual forcing can be diagnosed by fitting a series of EBMs, one for each forcing, to the mean coupled model response to all the forcings to diagnose the time-dependent response in the global mean for each individual forcing. The magnitude of these time-only signals can then be inferred from observations using detection methods ( Stone et al., 2007a [JoC, MoS, SRC, 2007] ,b). When applied to 13 different climate models that had transient simulations of 1901 to 2005 temperature change, Stone et al. 2007a [JoC, MoS, SRC, 2007] ) find a robust detection across the models of greenhouse gas warming over this period, although uncertainties in attributable temperature changes due to the different forcings are larger than when considering spatio-temporal patterns. By tuning an EBM to the observations, and using an AOGCM solely to estimate internal variability, Stone and Allen 2005b [JoC, MoS, SRC] ) detect the effects of greenhouse gases and tropospheric sulphate aerosols in the observed 1900 to 2004 record, but not the effects of volcanic and solar forcing.

The detection of an anthropogenic signal is also robust to using different methods. For example, Bayesian detection analyses (Appendix 9.A.2) robustly detect anthropogenic influence on near-surface temperature changes ( Smith et al., 2003 [PoC, JoC, SRC] ; Schnur and Hasselmann, 2005 [JoC] ; Min and Hense, 2006a [JoC, MoS] ,b). In these studies, Bayes Factors (ratios of posterior to prior odds) are used to assess evidence supporting competing hypotheses ( (Kass and Raftery, 1995; ) see Appendix 9.A.2). A Bayesian analysis of seven climate models (Schnur and Hasselman, 2005 [NotFound] ) and Bayesian analyses of MMD 20C3M simulations ( Min and Hense, 2006a [JoC, MoS] ,b) find decisive evidence for the influence of anthropogenic forcings. Lee et al. 2005 [JoC, SRC] ), using an approach suggested by Berliner et al. 2000 [JoC, ARC] ), evaluate the evidence for the presence of the combined greenhouse gas and sulphate aerosol (GS) signal, estimated from CGCM1 and CGCM2( Table 8.1 ;McAvaney et al., 2001 [NPR, MoS, ARC] ), in observations for several five-decade windows, beginning with 1900 to 1949 and ending with 1950 to 1999 . Very strong evidence was found in support of detection of the forced response during both halves of the 20th century regardless of the choice of prior distribution. However, evidence for attribution in that approach is based on the extent to which observed data narrow the prior uncertainty on the size of the anthropogenic signal. That evidence was not found to be very strong, although Lee et al. 2005 [JoC, SRC] ) estimate that strong evidence for attribution as defined in their approach may emerge within the next two decades as the anthropogenic signal strengthens.

In a further study, Lee et al. 2006 [JoC, MoS, SRC] ) assess whether anthropogenic forcing has enhanced the predictability of decadal global-scale temperature changes; a forcing-related enhancement in predictability would give a further indication of its role in the evolution of the 20th-century climate. Using an ensemble of simulations of the 20th century with GS forcing, they use Bayesian tools similar to those of Lee et al. 2005 [JoC, SRC] ) to produce, for each decade beginning with 1930 to 1939, a forecast of the probability of above-normal temperatures where ‘normal’ is defined as the mean temperature of the preceding three decades. These hindcasts become skilful during the last two decades of the 20th century as indicated both by their Brier skill scores, a standard measure of the skill of probabilistic forecasts, and by the confidence bounds on hindcasts of global mean temperature anomalies( Figure 9.10 ). This indicates that greenhouse gas forcing contributes to predictability of decadal temperature changes during the latter part of the 20th century.

Another type of analysis is a Granger causality analysis of the lagged covariance structure of observed hemispheric temperatures ( Kaufmann and Stern, 2002 [JoC] ), which also provides evidence for an anthropogenic signal, although such evidence may not be conclusive on its own without additional information from climate models ( (Triacca, 2001 ) ). Consistently, a neural network model is unable to reconstruct the observed global temperature record from 1860 to 2000 if anthropogenic forcings are not taken into account ( Pasini et al., 2006 [MoS] ). Further, an assessment of recent climate change relative to the long-term persistence of NH mean temperature as diagnosed from a range of reconstructed temperature records ( Rybski et al., 2006 [JoC] ) suggests that the recent warming cannot be explained solely in terms of natural factors, regardless of the reconstruction used. Similarly, Fomby and Vogelsang 2002 [JoC] ), using a test of trend that accounts for the effects of serial correlation, find that the increase in global mean temperature over the 20th century is statistically significant even if it is assumed that natural climate variability has strong serial correlation.

Figure 9.10. Observed and hindcast decadal mean surface temperature anomalies (°C) expressed, for each decade, relative to the preceding three decades. Observed anomalies are represented by horizontal black lines. Hindcast decadal anomalies and their uncertainties (5 to 95% confidence bounds) are displayed as vertical bars. Hindcasts are based on a Bayesian detection analysis using the estimated response to historical external forcing. Hindcasts made with CGCM2, HadCM2 (see Table 8.1 of the TAR) and HadCM3 (see Table 8.1 ,this report) use the estimated response to anthropogenic forcing only (left hand column of legend) while those made with selected MMD 20C3M models used anthropogenic and natural forcings (centre column of legend; see Table 8.1 for model descriptions). Hindcasts made with the ensemble mean of the selected 20C3M models are indicated by the thick green line. A hindcast based on persisting anomalies from the previous decade is also shown. The hindcasts agree well with observations from the 1950 s onward. Hindcasts for the decades of the 1930 s and 1940 s are sensitive to the details of the hindcast procedure. A forecast for the decadal global mean anomaly for the decade 2000 to 2009, relative to the 1970 to 1999 climatology, based on simulations performed with the Canadian Centre for Climate Modelling and Analysis Coupled Global Climate Model (CGCM2) is also displayed. From Lee et al. 2006 [JoC, MoS, SRC] ).

9.4.1.5 The Influence of Other Anthropogenic and Natural Forcings

Asignificant cooling due to other anthropogenic factors, dominated by aerosols, is a robust feature of a wide range of detection analyses. These analyses indicate that it is likely that greenhouse gases alone would have caused more than the observed warming over the last 50 years of the 20th century, with some warming offset by cooling from natural and other anthropogenic factors, notably aerosols, which have a very short residence time in the atmosphere relative to that of well-mixed greenhouse gases ( Schwartz, 1993 [ARC] ). A key factor in identifying the aerosol fingerprint, and therefore the amount of aerosol cooling counteracting greenhouse warming, is the change through time of the hemispheric temperature contrast, which is affected by the different evolution of aerosol forcing in the two hemispheres as well as the greater thermal inertia of the larger ocean area in the SH ( Santer et al., 1996b [PoC, JoC, SRC] ,c; Hegerl et al., 2001 [JoC, SRC] ; Stott et al., 2006c [JoC, MoS, SRC] ). Regional and seasonal aspects of the temperature response may help to distinguish further the response to greenhouse gas increases from the response to aerosols (e.g., Ramanathan et al., 2005 [JoC, ARC] ; Nagashima et al., 2006 [JoC] ).

Results on the importance and contribution from anthropogenic forcings other than greenhouse gases vary more between different approaches. For example, Bayesian analyses differ in the strength of evidence they find for an aerosol effect. Schnur and Hasselman 2005 [NotFound] ), for example, fail to find decisive evidence for the influence of aerosols. They postulate that this could be due to taking account of modelling uncertainty in the response to aerosols. However, two other studies using frequentist methods that also include modelling uncertainty find a clear detection of sulphate aerosols, suggesting that the use of multiple models helps to reduce uncertainties and improves detection of a sulphate aerosol effect ( Gillett et al., 2002c [NPR, JoC, MoS, SRC] ; Huntingford et al., 2006 [JoC, MoS, SRC] ). Similarly, a Bayesian study of hemispheric mean temperatures from 1900 to 1996 finds decisive evidence for an aerosol cooling effect ( Smith et al., 2003 [PoC, JoC, SRC] ). Differences in the separate detection of sulphate aerosol influences in multi-signal approaches can also reflect differences in the diagnostics applied (e.g., the space-time analysis of Tett et al. 1999 [JoC, SRC] ) versus the space-only analysis of Hegerl et al. 1997 [JoC, SRC] , 2000 )) as was shown by Gillett et al. 2002a [JoC, SRC] ).

Recent estimates( Figure 9.9 )indicate a relatively small combined effect of natural forcings on the global mean temperature evolution of the second half of the 20th century, with a small net cooling from the combined effects of solar and volcanic forcings. Coupled models simulate much less warming over the 20th century in response to solar forcing alone than to greenhouse gas forcing ( Cubasch et al., 1997 [JoC, MoS, ARC] ; Broccoli et al., 2003 [JoC, MoS, ARC] ; Meehl et al., 2004 [JoC, MoS, ARC] ), independent of which solar forcing reconstruction is used( Chapter2 ). Several studies have attempted to estimate the individual contributions from solar and volcanic forcings separately, thus allowing for the possibility of enhancement of the solar response in observations due to processes not represented in models. Optimal detection studies that attempt to separate the responses to solar and other forcings in observations can also account for gross errors in the overall magnitude of past solar forcing, which remains uncertain( Chapter2 ), by scaling the space-time patterns of response( Section 9.2.2.1 ). Using such a method, Tett et al. 1999 [JoC, SRC] ) estimate that the net anthropogenic warming in the second half of the 20th century was much greater than any possible solar warming, even when using the solar forcing reconstruction by Hoyt and Schatten 1993 [JoC] ), which indicates larger solar forcing and a different evolution over time than more recent reconstructions( Section 2.7.1 ). However, Stott et al. 2003b [JoC, MoS, SRC] ), using the same solar reconstruction but a different model, are not able to completely rule out the possibility that solar forcing might have caused more warming than greenhouse gas forcing over the 20th century due to difficulties in distinguishing between the patterns of response to solar and greenhouse forcing. This was not the case when using the response to solar forcing based on the alternative reconstruction of Lean et al. 1995 [PoC, JoC, ARC] ), in which case they find a very small likelihood (less than 1%, as opposed to approximately 10%) that solar warming could be greater than greenhouse warming since 1950 . Note that recent solar forcing reconstructions show a substantially decreased magnitude of low-frequency variations in solar forcing( Section 2.7.1 )compared to Lean et al. 1995 [PoC, JoC, ARC] ) and particularly Hoyt and Schatten 1993 [JoC] ).

The conclusion that greenhouse warming dominates over solar warming is supported further by a detection and attribution analysis using 13 models from the MMD at PCMDI ( Stone et al., 2007a [JoC, MoS, SRC, 2007] ) and an analysis of the National Center for Atmospheric Research (NCAR) Community Climate System Model (CCSM1.4; Stone et al., 2007b [JoC, SRC, 2007] ). In both these analyses, the response to solar forcing in the model was inferred by fitting a series of EBMs to the mean coupled model response to the combined effects of anthropogenic and natural forcings. In addition, a combined analysis of the response at the surface and through the depth of the atmosphere using HadCM3 and the solar reconstruction of Lean et al. 1995 [PoC, JoC, ARC] ) concluded that the near-surface temperature response to solar forcing over 1960 to 1999 is much smaller than the response to greenhouse gases ( Jones et al., 2003 [JoC, SRC] ). This conclusion is also supported by the vertical pattern of climate change, which is more consistent with the response to greenhouse gas than to solar forcing( Figure 9.1 ). Further evidence against a dominant solar role arises from older analyses targeted at detecting the solar response (e.g., North and Stevens, 1998 [JoC] ). Based on these detection results, which allow for possible amplification of the solar influence by processes not represented in climate models, we conclude that it is very likely that greenhouse gases caused more global warming over the last 50 years than changes in solar irradiance.

Detection and attribution as well as modelling studies indicate more uncertainty regarding the causes of early 20th-century warming than the recent warming. A number of studies detect a significant natural contribution to early 20th-century warming ( Tett et al., 2002 [JoC, MoS, SRC] ; Stott et al., 2003b [JoC, MoS, SRC] ; Nozawa et al., 2005 [JoC, SRC] ; Shiogama et al., 2006 [JoC] ). Some studies find a greater role for solar forcing than other forcings before 1950 ( Stott et al., 2003b [JoC, MoS, SRC] ), although one detection study finds a roughly equal role for solar and volcanic forcing ( Shiogama et al., 2006 [JoC] ), and others find that volcanic forcing ( Hegerl et al., 2003 [JoC, SRC] , 2007 ) or a substantial contribution from natural internal variability ( Tett et al., 2002 [JoC, MoS, SRC] ; Hegerl et al., 2007 [JoC, SRC, 2007] ) could be important. There could also be an early expression of greenhouse warming in the early 20th century ( Tett et al., 2002 [JoC, MoS, SRC] ; Hegerl et al., 2003 [JoC, SRC] , 2007 ).

9.4.1.6 Implications for Transient Climate Response

Quantification of the likely contributions of greenhouse gases and other forcing factors to past temperature change( Section 9.4.1.4 )in turn provides observational constraints on the transient climate response, which determines the rapidity and strength of a global temperature response to external forcing (see Glossary and Sections 9.6.2.3 and 8.6.2.1 for detailed definitions) and therefore helps to constrain likely future rates of warming. Scaling factors derived from detection analyses can be used to scale predictions of future change by assuming that the fractional error in model predictions of global mean temperature change is constant ( Allen et al., 2000 [JoC, MoS, SRC] , 2002; Allen and Stainforth, 2002 [JoC, MoS, SRC] ; Stott and Kettleborough, 2002 [JoC, MoS, SRC] ). This linear relationship between past and future fractional error in temperature change has been found to be sufficiently robust over a number of realistic forcing scenarios to introduce little additional uncertainty ( Kettleborough et al., 2007 [JoC, MoS, SRC, 2007] ). In this approach based on detection and attribution methods, which is compared with other approaches for producing probabilistic projections in Section 10.5.4.5 ,different scaling factors are applied to the greenhouse gases and to the response to other anthropogenic forcings (notably aerosols); these separate scaling factors are used to account for possible errors in the models and aerosol forcing. Uncertainties calculated in this way are likely to be more reliable than uncertainty ranges derived from simulations by coupled AOGCMs that happen to be available. Such ensembles could provide a misleading estimate of forecast uncertainty because they do not systematically explore modelling uncertainty ( Allen et al., 2002 [Ambiguous] ; Allen and Stainforth, 2002 [JoC, MoS, SRC] Stott et al. 2006c [JoC, MoS, SRC] ) compare observationally constrained predictions from three coupled climate models with a range of sensitivities and show that predictions made in this way are relatively insensitive to the particular choice of model used to produce them. The robustness to choice of model of such observationally constrained predictions was also demonstrated by Stone et al. 2007a [JoC, MoS, SRC, 2007] ) for the MMD ensemble. The observationally constrained transient climate response at the time of doubling of atmospheric CO₂ following a 1% per year increase in CO₂ was estimated by Stott et al. 2006c [JoC, MoS, SRC] ) to lie between 1.5°C and 2.8°C( Section 9.6.2 , Figure 9.21 ). Such approaches have also been used to provide observationally constrained predictions of global mean ( Stott and Kettleborough, 2002 [JoC, MoS, SRC] ; Stone et al., 2007a [JoC, MoS, SRC, 2007] ) and continental-scale temperatures ( Stott et al., 2006a [JoC, MoS, SRC] ) following the IPCC Special Report on Emission Scenarios (SRES) emissions scenarios, and these are discussed in Sections 10.5.4.5 and 11.10 .

Diurnal temperature range (DTR) has decreased over land by about 0.4°C over the last 50 years, with most of that change occurring prior to 1980 ( Section 3.2.2.1 ). This decreasing trend has been shown to be outside the range of natural internal variability estimated from models. Hansen et al. 1995 [ARC] ) demonstrate that tropospheric aerosols plus increases in continental cloud cover, possibly associated with aerosols, could account for the observed decrease in DTR. However, although models simulate a decrease in DTR when they include anthropogenic changes in greenhouse gases and aerosols, the observed decrease is larger than the model-simulated decrease ( Stone and Weaver, 2002 [JoC, MoS, SRC] , 2003; Braganza et al., 2004 [JoC] ). This discrepancy is associated with simulated increases in daily maximum temperature being larger than observed, and could be associated with simulated increases in cloud cover being smaller than observed ( Braganza et al., 2004 [JoC] ; see Section 3.4.3.1 for observations), a result supported by other analyses ( Dai et al., 1999 [PoC, JoC, ARC] ; Stone and Weaver, 2002 [JoC, MoS, SRC] , 2003 ).

9.4.1.7 Studies of Indices of Temperature Change

Another method for identifying fingerprints of climate change in the observational record is to use simple indices of surface air temperature patterns that reflect features of the anticipated response to anthropogenic forcing ( Karoly and Braganza, 2001 [JoC, SRC] ; Braganza et al., 2003 [JoC] ). By comparing modelled and observed changes in such indices, which include the global mean surface temperature, the land-ocean temperature contrast, the temperature contrast between the NH and SH, the mean magnitude of the annual cycle in temperature over land and the mean meridional temperature gradient in the NH mid-latitudes, Braganza et al. 2004 [JoC] ) estimate that anthropogenic forcing accounts for almost all of the warming observed between 1946 and 1995 whereas warming between 1896 and 1945 is explained by a combination of anthropogenic and natural forcing and internal variability. These results are consistent with the results from studies using space-time detection techniques( Section 9.4.1.4 ).

9.4.1.8 Remaining Uncertainties

Amuch larger range of forcing combinations and climate model simulations has been analysed in detection studies than was available for the TAR (Supplementary Material, Table S9.1). Detection and attribution analyses show robust evidence for an anthropogenic influence on climate. However, some forcings are still omitted by many models and uncertainties remain in the treatment of those forcings that are included by the majority of models.

Most studies omit two forcings that could have significant effects, particularly at regional scales, namely carbonaceous aerosols and land use changes. However, detection and attribution analyses based on climate simulations that include these forcings, (e.g., Stott et al., 2006b [JoC, MoS, SRC] ), continue to detect a significant anthropogenic influence in 20th-century temperature observations even though the near-surface patterns of response to black carbon aerosols and sulphate aerosols could be so similar at large spatial scales (although opposite in sign) that detection analyses may be unable to distinguish between them ( Jones et al., 2005 [JoC, SRC] ). Forcing from surface albedo changes due to land use change is expected to be negative globally (Sections 2.5.3 , 7.3.3 and 9.3.3.3 )although tropical deforestation could increase evaporation and warm the climate( Section 2.5.5 ), counteracting cooling from albedo change. However, the albedo-induced cooling effect is expected to be small and was not detected in observed trends in the study by Matthews et al. 2004 [JoC, ARC] ).

For those forcings that have been included in attribution analyses, uncertainties associated with the temporal and spatial pattern of the forcing and the modelled response can affect the results. Large uncertainties associated with estimates of past solar forcing( Section 2.7.1 )and omission of some chemical and dynamical response mechanisms ( Gray et al., 2005 [NPR, ARC] ) make it difficult to reliably estimate the contribution of solar forcing to warming over the 20th century. Nevertheless, as discussed above, results generally indicate that the contribution is small even if allowance is made for amplification of the response in observations, and simulations used in attribution analyses use several different estimates of solar forcing changes over the 20th century (Supplementary Material, Table S9.1). A number of different volcanic reconstructions are included in the modelling studies described in Section 9.4.1.2 (e.g., Sato et al., 1993 [JoC, ARC] ; Andronova et al., 1999 [JoC, MoS, SRC] ; Ammann et al., 2003 [PoC, JoC, MoS, SRC] ; Supplementary Material, Table S9.1). Some models include volcanic effects by simply perturbing the incoming shortwave radiation at the top of the atmosphere, while others simulate explicitly the radiative effects of the aerosols in the stratosphere. In addition, some models include the indirect effects of tropospheric sulphate aerosols on clouds (e.g., Tett et al., 2002 [JoC, MoS, SRC] ), whereas others consider only the direct radiative effect (e.g., Meehl et al., 2004 [JoC, MoS, ARC] ). In models that include indirect effects, different treatments of the indirect effect are used, including changing the albedo of clouds according to an off-line calculation (e.g., Tett et al., 2002 [JoC, MoS, SRC] ) and a fully interactive treatment of the effects of aerosols on clouds (e.g., Stott et al., 2006b [JoC, MoS, SRC] ). The overall level of consistency between attribution results derived from different models (as shown in Figure 9.9 ), and the ability of climate models to simulate large-scale temperature changes during the 20th century (Figures 9.5 and 9.6 ), indicate that such model differences are likely to have a relatively small impact on attribution results of large-scale temperature change at the surface.

There have also been methodological developments that have resulted in attribution analyses taking uncertainties more fully into account. Attribution analyses normally directly account for errors in the magnitude of the model’s pattern of response to different forcings by the inclusion of factors that scale the model responses up or down to best match observed climate changes. These scaling factors compensate for under- or overestimates of the amplitude of the model response to forcing that may result from factors such as errors in the model’s climate sensitivity, ocean heat uptake efficiency or errors in the imposed external forcing. Older analyses (e.g., Tett et al., 2002 [JoC, MoS, SRC] ) did not take account of uncertainty due to sampling signal estimates from finite-member ensembles. This can lead to a low bias, particularly for weak forcings, in the scaling factor estimates (Appendix 9.A.1; Allen and Stott, 2003 [JoC, MoS, SRC] ; Stott et al., 2003a [NPR, JoC, MoS, SRC] ). However, taking account of sampling uncertainty (as most more recent detection and attribution studies do, including those shown in Figure 9.9 )makes relatively little difference to estimates of attributable warming rates, particularly those due to greenhouse gases; the largest differences occur in estimates of upper bounds for small signals, such as the response to solar forcing ( Allen and Stott, 2003 [JoC, MoS, SRC] ; Stott et al., 2003a [NPR, JoC, MoS, SRC] ). Studies that compare results between models and analysis techniques (e.g., Hegerl et al., 2000 [JoC, MoS, SRC] ; Gillett et al., 2002a [JoC, SRC] ; Hegerl and Allen, 2002 [JoC, MoS, SRC] ), and more recently, that use multiple models to determine fingerprints of climate change ( Gillett et al., 2002c [NPR, JoC, MoS, SRC] ; Huntingford et al., 2006 [JoC, MoS, SRC] ; Stott et al., 2006c [JoC, MoS, SRC] ; Zhang et al., 2006 [JoC, MoS, SRC] ) find a robust detection of an anthropogenic signal in past temperature change.

Acommon aspect of detection analyses is that they assume the response in models to combinations of forcings to be additive. This was shown to be the case for near-surface temperatures in the PCM ( Meehl et al., 2004 [JoC, MoS, ARC] ), in the Hadley Centre Climate Model version 2 (HadCM2; Gillett et al., 2004c [JoC, MoS, SRC] ) and in the GFDL CM2.1 (see Table 8.1 )model ( Knutson et al., 2006 [JoC, MoS, ARC] ), although none of these studies considered the indirect effects of sulphate aerosols. Sexton et al. 2003 [PoC, JoC, MoS, SRC] ) did find some evidence for a nonlinear interaction between the effects of greenhouse gases and the indirect effect of sulphate aerosols in the atmosphere-only version of HadCM3 forced by observed SSTs; the additional effect of combining greenhouse gases and indirect aerosol effects together was much smaller than each term separately but was found to be comparable to the warming due to increasing tropospheric ozone. In addition, Meehl et al. 2003 [JoC, MoS, ARC] ) found that additivity does not hold so well for regional responses to solar and greenhouse forcing in the PCM. Linear additivity was found to hold in the PCM model for changes in tropopause height and synthetic satellite-borne Microwave Sounding Unit (MSU) temperatures ( Christy et al., 2000 [ARC] ; Mears et al., 2003 [JoC, ARC] ; Santer et al., 2003b [PoC, JoC, MoS, SRC] ).

Afurther source of uncertainty derives from the estimates of internal variability that are required for all detection analyses. These estimates are generally model-based because of difficulties in obtaining reliable internal variability estimates from the observational record on the spatial and temporal scales considered in detection studies. However, models would need to underestimate variability by factors of over two in their standard deviation to nullify detection of greenhouse gases in near-surface temperature data ( Tett et al., 2002 [JoC, MoS, SRC] ), which appears unlikely given the quality of agreement between models and observations at global and continental scales (Figures 9.7 and 9.8 )and agreement with inferences on temperature variability from NH temperature reconstructions of the last millennium. The detection of the effects of other forcings, including aerosols, is likely to be more sensitive (e.g., an increase of 40% in the estimate of internal variability is enough to nullify detection of aerosol and natural forcings in HadCM3; Tett et al., 2002 [JoC, MoS, SRC] )

Few detection studies have explicitly considered the influence of observational uncertainty on near-surface temperature changes. However, Hegerl et al. 2001 [JoC, SRC] ) show that inclusion of observational sampling uncertainty has relatively little effect on detection results and that random instrumental error has even less effect. Systematic instrumental errors, such as changes in measurement practices or urbanisation, could be more important, especially earlier in the record( Chapter3 ), although these errors are calculated to be relatively small at large spatial scales. Urbanisation effects appear to have negligible effects on continental and hemispheric average temperatures( Chapter3 ). Observational uncertainties are likely to be more important for surface temperature changes averaged over small regions( Section 9.4.2 )and for analyses of free atmosphere temperature changes( Section 9.4.4 ).

9.4.2.1 Observed Changes

Over the 1901 to 2005 period there has been warming over most of the Earth’s surface with the exception of an area south of Greenland and parts of North and South America( Figure 3.9 and Section 3.2.2.7 ,see also Figure 9.6 ). Warming has been strongest over the continental interiors of Asia and north-western North America and some mid-latitude ocean regions of the SH as well as south-eastern Brazil. Since 1979, almost all land areas with observational data coverage show warming( Figure 9.6 ). Warming is smaller in the SH than in the NH, with cooling over parts of the mid-latitude oceans. There have been widespread decreases in continental DTR since the 1950 s which coincide with increases in cloud amounts( Section 3.4.3.1 ).

9.4.2.3 Studies Based on Indices of Temperature Change and Temperature-Precipitation Relationships

Studies based on indices of temperature change support the robust detection of human influence on continental-scale land areas. Observed trends in indices of North American continental-scale temperature change, (including the regional mean, the mean land-ocean temperature contrast and the annual cycle) were found by Karoly et al. 2003 [Ambiguous] ) to be generally consistent with simulated trends under historical forcing from greenhouse gases and sulphate aerosols during the second half of the 20th century. In contrast, they find only a small likelihood of agreement with trends driven by natural forcing only during this period. An analysis of changes in Australian mean, daily maximum and daily minimum temperatures and diurnal temperature range using six coupled climate models showed that it is likely that there has been a significant contribution to observed warming in Australia from increasing greenhouse gases and sulphate aerosols ( Karoly and Braganza, 2005a [JoC, SRC] ). An anomalous warming has been found over all Australia ( Nicholls, 2003 [NPR, JoC, SRC] ) and in New South Wales ( Nicholls et al., 2005 [SRC] ) since the early 1970 s, associated with a changed relationship between annual mean maximum temperature and rainfall. Whereas interannual rainfall and temperature variations in this region are strongly inversely correlated, in recent decades temperatures have tended to be higher for a given rainfall than in previous decades. By removing the rainfall-related component of Australian temperature variations, thereby enhancing the signal-to-noise ratio, Karoly and Braganza 2005b [SRC] ) detect an anthropogenic warming signal in south-eastern Australia, although their results are affected by some uncertainty associated with their removal of rainfall-related temperature variability. A similar technique applied to the Sudan and Sahel region improved the agreement between model simulations and observations of temperature change over the last 60 years in this region ( Douville, 2006 [JoC] ) and could improve the detectability of regional temperature signals over other regions where precipitation is likely to affect the surface energy budget ( Trenberth and Shea, 2005 [PoC, JoC, ARC] ).

Observed changes in temperature extremes are consistent with the observed warming of the climate ( Alexander et al., 2006 [JoC, ARC] ) and are summarised in Section 3.8.2.1 .There has been a widespread reduction in the number of frost days in mid-latitude regions in recent decades, an increase in the number of warm extremes, particularly warm nights, and a reduction in the number of cold extremes, particularly cold nights. A number of regional studies all show patterns of changes in extremes consistent with a general warming, although the observed changes in the tails of the temperature distributions are generally not consistent with a simple shift of the entire distribution alone.

9.4.3.1 Observed Changes

9.4.3.2 Global Assessments

Evidence for observed changes in short-duration extremes generally depends on the region considered and the analysis method ( IPCC, 2001 [NPR] ). Global analyses have been restricted by the limited availability of quality-controlled and homogenised daily station data. Indices of temperature extremes have been calculated from station data, including some indices from regions where daily station data are not released ( (Frich et al., 2002; ) Klein Tank and Können, 2003 [JoC] ; Alexander et al., 2006 [JoC, ARC] Kiktev et al. 2003 [PoC, JoC, MoS, SRC] ) analyse a subset of such indices by using fingerprints from atmospheric model simulations driven by prescribed SSTs. They find significant decreases in the number of frost days and increases in the number of very warm nights over much of the NH. Comparisons of observed and modelled trend estimates show that inclusion of anthropogenic effects in the model integrations improves the simulation of these changing temperature extremes, indicating that human influences are probably an important contributor to changes in the number of frost days and warm nights. Tebaldi et al. 2006 [JoC, MoS, ARC] ) find that changes simulated by eight MMD models agreed well with observed trends in heat waves, warm nights and frost days over the last four decades.

Christidis et al. 2005 [JoC] ) analyse a new gridded data set of daily temperature data ( Caesar et al., 2006 [JoC, ARC] ) using the indices shown by Hegerl et al. 2004 [JoC, SRC] ) to have a potential for attribution, namely the average temperature of the most extreme 1, 5, 10 and 30 days of the year. Christidis et al. 2005 [JoC] ) detect robust anthropogenic changes in indices of extremely warm nights using signals estimated with the HadCM3 model, although with some indications that the model overestimates the observed warming of warm nights. They also detect human influence on cold days and nights, but in this case the model underestimates the observed changes, significantly so in the case of the coldest day of the year. Anthropogenic influence was not detected in observed changes in extremely warm days.

9.4.3.3 Attributable Changes in the Risk of Extremes

Many important impacts of climate change may manifest themselves through a change in the frequency or likelihood of occurrence of extreme events. While individual extreme events cannot be attributed to external influences, a change in the probability of such events might be attributable to external influences ( Palmer, 1999 [MoS, ARC] ; Palmer and Räisänen, 2002 [JoC, ARC] ). One study estimates that anthropogenic forcings have significantly increased the risk of extremely warm summer conditions over southern Europe, as was observed during the 2003 European heat wave. Stott et al. 2004 [JoC, SRC] ) apply a methodology for making quantitative statements about change in the likelihood of such specific types of climatic events ( Allen, 2003 [JoC, SRC] ; Stone and Allen, 2005a [JoC, SRC] ), by expressing the contribution of external forcing to the risk of an event exceeding a specific magnitude. If P1 is the probability of a climatic event (such as a heat wave) occurring in the presence of anthropogenic forcing of the climate system, and P0 is the probability of it occurring if anthropogenic forcing had not been present, then the fraction of the current risk that is attributable to past greenhouse gas emissions (fraction of attributable risk; FAR) is given by FAR = 1 – P0 / P1 ( Allen, 2003 [JoC, SRC] Stott et al. 2004 [JoC, SRC] ) apply the FAR concept to mean summer temperatures of a large part of continental Europe and the Mediterranean. Using a detection and attribution analysis, they determine that regional summer mean temperature has likely increased due to anthropogenic forcing, and that the observed change is inconsistent with natural forcing. They then use the HadCM3 model to estimate the FAR associated with a particular extreme threshold of regional summer mean temperature that was exceeded in 2003, but in no other year since the beginning of the record in 1851 Stott et al. 2004 [JoC, SRC] ) estimate that it is very likely that human influence has more than doubled the risk of the regional summer mean temperature exceeding this threshold( Figure 9.13 ).

Figure 9.13. Change in risk of mean European summer temperatures exceeding a threshold of 1.6°C above 1961 to 1990 mean temperatures, a threshold that was exceeded in 2003 but in no other year since the start of the instrumental record in 1851 . (Top) Frequency histograms of the estimated likelihood of the risk (probability) of exceeding a 1.6°C threshold (relative to the 1961 – 1990 mean) in the 1990 s in the presence (red curve) and absence (green curve) of anthropogenic change, expressed as an occurrence rate. (Bottom) Fraction of attributable risk (FAR). The vertical line indicates the ‘best estimate’ FAR, the mean risk attributable to anthropogenic factors averaged over the distribution. The alternation between grey and white bands indicates the deciles of the estimated FAR distribution. The shift from the green to the red distribution in (a) implies a FAR distribution with mean 0.75, corresponding to a four-fold increase in the risk of such an event (b). From Stott et al. 2004 [JoC, SRC] ).

This study considered only continental mean seasonally averaged temperatures. Consideration of shorter-term and smaller-scale heat waves will require higher resolution modelling and will need to take complexities such as land surface processes into account ( Schär and Jendritzky, 2004 [JoC, ARC] ). Also, Stott et al. 2004 [JoC, SRC] ) assume no change in internal variability in the region they consider (which was the case in HadCM3 21st-century climate projections for summer mean temperatures in the region they consider), thereby ascribing the increase in risk only to an increase in mean temperatures (i.e., as shown in Box TS.5, Figure 1, which illustrates how a shift in the mean of a distribution can cause a large increase in the frequency of extremes). However, there is some evidence for a weak increase in European temperature variability in summer (and a decrease in winter) for the period 1961 to 2004 ( Scherrer et al., 2005 [JoC, ARC] ), which could contribute to an increase in the likelihood of extremes. Schär et al. 2004 [JoC, ARC] ) show that the central European heat wave of 2003 could also be consistent with model-predicted increases in temperature variability due to soil moisture and vegetation feedbacks. In addition, multi-decadal scale variability, associated with basin-scale changes in the Atlantic Ocean related to the Meridional Overturning Circulation (MOC) could have contributed to changes in European summer temperatures ( Sutton and Hodson, 2005 [JoC, MoS] ), although Klein Tank et al. 2005 [JoC] ) show evidence that patterns of change in European temperature variance in spring and summer are not consistent with patterns of change in temperature variance expected from natural variability. Meteorological aspects of the summer 2003 European heat wave are discussed in Box 3.6 .

9.4.4.1 Observed Changes

Observed free atmosphere temperature changes are discussed in Section 3.4.1 and Karl et al. 2006 [NPR, PoC, ARC] ) provide a comprehensive review. Radiosonde-based observations (with near global coverage since 1958 ) and satellite-based temperature measurements (beginning in late 1978 ) show warming trends in the troposphere and cooling trends in the stratosphere. All data sets show that the global mean and tropical troposphere has warmed from 1958 to the present, with the warming trend in the troposphere slightly greater than at the surface. Since 1979, it is likely that there is slightly greater warming in the troposphere than at the surface, although uncertainties remain in observed tropospheric warming trends and whether these are greater or less than the surface trend. The range (due to different data sets) of the global mean tropospheric temperature trend since 1979 is 0.12°C to 0.19°C per decade based on satellite-based estimates( Chapter3 )compared to a range of 0.16°C to 0.18°C per decade for the global surface warming. While all data sets show that the stratosphere has cooled considerably from 1958 and from 1979 to present, there are large differences in the linear trends estimated from different data sets. However, a linear trend is a poor fit to the data in the stratosphere and the tropics at all levels( Section 3.4.1 ). The uncertainties in the observational records are discussed in detail in Section 3.4.1 and by Karl et al. 2006 [NPR, PoC, ARC] ). Uncertainties remain in homogenised radiosonde data sets which could result in a spurious inference of net cooling in the tropical troposphere. Differences between temperature trends measured from different versions of tropospheric satellite data result primarily from differences in how data from different satellites are merged.

9.4.4.2 Changes in Tropopause Height

The height of the lapse rate tropopause (the boundary between the stratosphere and the troposphere) is sensitive to bulk changes in the thermal structure of the stratosphere and the troposphere, and may also be affected by changes in surface temperature gradients ( Schneider, 2004 [JoC] ). Analyses of radiosonde data have documented increases in tropopause height over the past 3 to 4 decades ( Highwood et al., 2000 [JoC, ARC] ; Seidel et al., 2001 [JoC, ARC] ). Similar increases have been inferred from three different reanalysis products, the European Centre for Medium Range Weather Forecasts (ECMWF) 15- and 40-year reanalyses (ERA-15 and ERA-40) and the NCAR- National Center for Environmental Prediction (NCEP) reanalysis ( Kalnay et al., 1996 [JoC] ; Gibson et al., 1997 [NPR, MoS] ; Simmons and Gibson, 2000 [NPR, MoS, ARC] ; Kistler et al., 2001 [JoC] ), and from model simulations with combined anthropogenic and natural forcing ( Santer et al., 2003a [PoC, JoC, MoS, SRC] ,b, 2004; see Figure 9.14 ). In both models and reanalyses, changes in tropopause height over the satellite and radiosonde eras are smallest in the tropics and largest over Antarctica ( Santer et al., 2003a [PoC, JoC, MoS, SRC] ,b, 2004 ). Model simulations with individual forcings indicate that the major drivers of the model tropopause height increases are ozone-induced stratospheric cooling and the tropospheric warming caused by greenhouse gas increases ( Santer et al., 2003a [PoC, JoC, MoS, SRC] ). However, earlier model studies have found that it is difficult to alter tropopause height through stratospheric ozone changes alone ( Thuburn and Craig, 2000 [JoC] Santer et al. 2003c [PoC, JoC, SRC] ) found that the model-simulated response to combined anthropogenic and natural forcing is robustly detectable in different reanalysis products, and that solar and volcanic forcing alone could not explain the tropopause height increases( Figure 9.14 ). Climate data from reanalyses, especially the ‘first generation’ reanalysis analysed by Santer et al. 2003a [PoC, JoC, MoS, SRC] ), are subject to some deficiencies, notably inhomogeneities related to changes over time in the availability and quality of input data, and are subject to a number of specific technical choices in the reanalysis scheme (see Santer et al., 2004 [PoC, JoC, SRC] , for a discussion). Also, the NCEP reanalysis detection results could be due to compensating errors because of excessive stratospheric cooling in the reanalysis ( Santer et al., 2004 [PoC, JoC, SRC] ), since the stratosphere cools more relative to the troposphere in the NCEP reanalysis while models warm the troposphere. In contrast, the finding of a significant anthropogenic influence on tropopause height in the ‘second generation’ ERA-40 reanalysis is driven by similar large-scale changes in both models and the reanalysis. Detection results there are robust to removing global mean tropopause height increases.

Figure 9.14. Comparison between reanalysis and climate-model simulated global monthly mean anomalies in tropopause height. Model results are from two different PCM( Table 8.1 )ensemble experiments using either natural forcings, or natural and anthropogenic forcings (ALL). There are four realisations of each experiment. Both the low-pass filtered ensemble mean and the unfiltered range between the highest and lowest values of the realisations are shown. All model anomalies are defined relative to climatological monthly means computed over 1890 to 1999 . Reanalysis-based tropopause height anomalies estimated from ERA-40 were filtered in the same way as model data. The ERA-40 record spans 1957 to 2002 and was forced to have the same mean as ALL over 1960 to 1999 . After Santer et al. 2003a [PoC, JoC, MoS, SRC] and Santer et al. 2004 [PoC, JoC, SRC] ).

9.4.4.3 Overall Atmospheric Temperature Change

Anthropogenic influence on free atmosphere temperatures has been detected in analyses of satellite data since 1979, although this finding has been found to be sensitive to which analysis of satellite data is used. Satellite-borne MSUs, beginning in 1978, estimate the temperature of thick layers of the atmosphere. The main layers represent the lower troposphere (T2_LT), the mid-troposphere (T2) and the lower stratosphere (T4)( Section 3.4.1.2.1 Santer et al. 2003c [PoC, JoC, SRC] ) compare T2 and T4 temperature changes simulated by the PCM model including anthropogenic and natural forcings with the University of Alabama in Huntsville (UAH; Christy et al., 2000 [ARC] ) and Remote Sensing Systems (RSS; Mears and Wentz, 2005 [JoC, ARC] ) satellite data sets( Section 3.4.1.2.2 ). They find that the model fingerprint of the T4 response to combined anthropogenic and natural forcing is consistently detected in both satellite data sets, whereas the T2 response is detected only in the RSS data set. However, when the global mean changes are removed, the T2 fingerprint is detected in both data sets, suggesting a common spatial pattern of response overlain by a systematic global mean difference.

Anthropogenic influence on free atmosphere temperatures has been robustly detected in a number of different studies analysing various versions of the Hadley Centre Radiosonde Temperature (HadRT2) data set ( Parker et al., 1997 [JoC] ) by means of a variety of different diagnostics and fingerprints estimated with the HadCM2 and HadCM3 models ( Tett et al., 2002 [JoC, MoS, SRC] ; Thorne et al., 2002 [NPR, PoC, JoC, SRC] , 2003; Jones et al., 2003 [JoC, SRC] ). Whereas an analysis of spatial patterns of zonal mean free atmosphere temperature changes was unable to detect the response to natural forcings ( Tett et al., 2002 [JoC, MoS, SRC] ), an analysis of spatio-temporal patterns detected the influence of volcanic aerosols and (less convincingly) solar irradiance changes, in addition to detecting the effects of greenhouse gases and sulphate aerosols ( Jones et al., 2003 [JoC, SRC] ). In addition, Crooks 2004 [NPR, SRC] ) detects a solar signal in atmospheric temperature changes as seen in the HadRT2.1s radiosonde data set when a diagnostic chosen to extract the solar signal from other signals is used. The models used in these studies have poor vertical resolution in the stratosphere and they significantly underestimate stratospheric variability, thus possibly overestimating the significance of these detected signals ( Tett et al., 2002 [JoC, MoS, SRC] ). However, a sensitivity study ( Thorne et al., 2002 [NPR, PoC, JoC, SRC] ) showed that detection of human influence on free atmosphere temperature changes does not depend on the inclusion of stratospheric temperatures. An analysis of spatial patterns of temperature change, represented by large-scale area averages at the surface, in broad atmospheric layers and in lapse rates between layers, showed robust detection of an anthropogenic influence on climate when a range of uncertainties were explored relating to the choice of fingerprints and the radiosonde and model data sets ( Thorne et al., 2003 [PoC, JoC, SRC] ). However, Thorne et al. were not able to attribute recent observed tropospheric temperature changes to any particular combination of external forcing influences because the models analysed (HadCM2 and HadCM3) overestimate free atmosphere warming as estimated by the radiosonde data sets, an effect also seen by Douglass et al. 2004 [NPR, JoC, MoS] ) during the satellite era. However, there is evidence that radiosonde data during the satellite era are contaminated by spurious cooling trends ( Sherwood et al., 2005 [JoC, ARC] ; Randel and Wu, 2006 [JoC] ; Section 3.4.1 ), and since structural uncertainty arising from the choice of techniques used to analyse radiosonde data has not yet been quantified ( Thorne et al., 2005 [PoC, JoC, SRC] ), it is difficult to assess, based on these analyses alone, whether model-data discrepancies are due to model or observational deficiencies. However further information is provided by an analysis of modelled and observed tropospheric lapse rates, discussed in Section 9.4.4.4 .

Adifferent approach is to assess detectability of observed temperature changes through the depth of the atmosphere with AGCM simulations forced with observed SSTs, although the vertical profile of the atmospheric temperature change signal estimated in this way can be quite different from the same signal estimated by coupled models with the same external forcings ( Hansen et al., 2002 [JoC, MoS, ARC] ; Sun and Hansen, 2003 [JoC, MoS, ARC] ; Santer et al., 2005 [PoC, JoC, SRC] Sexton et al. 2001 [JoC, SRC] ) find that inclusion of anthropogenic effects improves the simulation of zonally averaged upper air temperature changes from the HadRTt1.2 data set such that an anthropogenic signal is detected at the 5% significance level in patterns of seasonal mean temperature change calculated as overlapping eight-year means over the 1976 to 1994 period and expressed as anomalies relative to the 1961 to 1975 base period. In addition, analysing patterns of annual mean temperature change for individual years shows that an anthropogenic signal is also detected on interannual time scales for a number of years towards the end of the analysis period.

9.4.4.4 Differential Temperature Trends

Subtracting temperature trends at the surface from those in the free atmosphere removes much of the common variability between these layers and tests whether the model-predicted trends in tropospheric lapse rate are consistent with those observed by radiosondes and satellites ( Karl et al., 2006 [NPR, PoC, ARC] ). Since 1979, globally averaged modelled trends in tropospheric lapse rates are consistent with those observed. However, this is not the case in the tropics, where most models have more warming aloft than at the surface while most observational estimates show more warming at the surface than in the troposphere ( Karl et al., 2006 [NPR, PoC, ARC] Karl et al. 2006 [NPR, PoC, ARC] ) carried out a systematic review of this issue. There is greater consistency between simulated and observed differential warming in the tropics in some satellite measurements of tropospheric temperature change, particularly when the effect of the cooling stratosphere on tropospheric retrievals is taken into account ( Karl et al., 2006 [NPR, PoC, ARC] ). External forcing other than greenhouse gas changes can also help to reconcile some of the differential warming, since both volcanic eruptions and stratospheric ozone depletion are expected to have cooled the troposphere more than the surface over the last several decades ( Santer et al., 2000 [PoC, JoC, SRC] , 2001; IPCC, 2001 [NPR] ; Free and Angell, 2002 [NPR, JoC, ARC] ; Karl et al., 2006 [NPR, PoC, ARC] ). There are, however, uncertainties in quantifying the differential cooling caused by these forcings, both in models and observations, arising from uncertainties in the forcings and model response to the forcings. Differential effects of natural modes of variability, such as ENSO and the NAM, on observed surface and tropospheric temperatures, which arise from differences in the amplitudes and spatial expression of these modes at the surface and in the troposphere, make only minor contributions to the overall differences in observed surface and tropospheric warming rates ( Santer et al., 2001 [PoC, JoC, MoS, SRC] ; Hegerl and Wallace, 2002 [JoC, SRC] ; Karl et al., 2006 [NPR, PoC, ARC] ).

Asystematic intercomparison between radiosonde-based (Radiosonde Atmospheric Temperature Products for Assessing Climate (RATPAC); Free et al., 2005 [JoC, ARC] , and Hadley Centre Atmospheric Temperature (HadAT), Thorne et al., 2005 [PoC, JoC, SRC] ) and satellite-based (RSS, UAH) observational estimates of tropical lapse rate trends with those simulated by 19 MMD models shows that on monthly and annual time scales, variations in temperature at the surface are amplified aloft in both models and observations by consistent amounts ( Santer et al., 2005 [PoC, JoC, SRC] ; Karl et al., 2006 [NPR, PoC, ARC] ). It is only on longer time scales that disagreement between modelled and observed lapse rates arises ( Hegerl and Wallace, 2002 [JoC, SRC] ), that is, on the time scales over which discrepancies would arise from inhomogeneities in the observational record. Only one observational data set (RSS) was found to be consistent with the models on both short and long time scales. While Vinnikov et al. 2006 [JoC] ) have not produced a lower-tropospheric retrieval, their estimate of the T2 temperature trend( Figure 3.18 )is consistent with model simulations ( Karl et al., 2006 [NPR, PoC, ARC] ). One possibility is that amplification effects are controlled by different physical mechanisms on short and long time scales, although a more probable explanation is that some observational records are contaminated by errors that affect their long-term trends( Section 3.4.1 ;Karl et al., 2006 [NPR, PoC, ARC] ).

9.5.1.1 Ocean Heat Content Changes

Since the TAR, evidence of climate change has accumulated within the ocean, both at regional and global scales( Chapter5 ). The overall heat content in the World Ocean is estimated to have increased by 14.2 × 10²² J during the period 1961 to 2003 ( Section 5.2.2 ). This overall increase has been superimposed on strong interannual and inter-decadal variations. The fact that the entire ocean, which is by far the system’s largest heat reservoir ( Levitus et al., 2005 [JoC, ARC] ; see also Figure 5.4 )gained heat during the latter half of the 20th century is consistent with a net positive radiative forcing of the climate system. Late 20th-century ocean heat content changes were at least one order of magnitude larger than the increase in energy content of any other component of the Earth’s ocean-atmosphere-cryosphere system( Figure 5.4 ;Levitus et al., 2005 [JoC, ARC] ).

All analyses indicate a large anthropogenic component of the positive trend in global ocean heat content. Levitus et al. 2001 [JoC, ARC] and Gregory et al 2004 [JoC, MoS, SRC] ) analyse simulations from the GFDL R30 and HadCM3 models respectively and show that climate simulations agree best with observed changes when the models include anthropogenic forcings from increasing greenhouse gas concentrations and sulphate aerosols. Gent and Danabasoglu 2004 [JoC, MoS, ARC] ) show that the observed trend cannot be explained by natural internal variability as simulated by a long control run of the Community Climate System Model (CCSM2). Barnett et al. 2001 [JoC, SRC] and Reichert et al. 2002b [JoC, MoS] ) use detection analyses similar to those described in Section 9.4 to detect model-simulated ocean climate change signals in the observed spatio-temporal patterns of ocean heat content across the ocean basins.

Barnett et al. 2005 [JoC] ) extend previous detection and attribution analyses of ocean heat content changes to a basin by basin analysis of the temporal evolution of temperature changes in the upper 700 m of the ocean (see also Pierce et al., 2006 [JoC, MoS, SRC] ). They report that whereas the observed change is not consistent with internal variability and the response to natural external forcing as simulated by two climate models (PCM and HadCM3), the simulated ocean warming due to anthropogenic factors (including well-mixed greenhouse gases and sulphate aerosols) is consistent with the observed changes and reproduces many of the different responses seen in the individual ocean basins( Figure 9.15 ), indicating a human-induced warming of the world’s oceans with a complex vertical and geographical structure that is simulated quite well by the two AOGCMs. Barnett et al. 2005 [JoC] ) find that the earlier conclusions of Barnett et al. 2001 [JoC, SRC] ) were not affected by the Levitus et al. 2005 [JoC, ARC] ) revisions to the Levitus et al. 2000 [JoC, ARC] ) ocean heat content data.

Figure 9.15. Strength of observed and model-simulated warming signal by depth for the World Ocean and for each ocean basin individually (in ^oC, see Barnett et al., 2005 [JoC] and Pierce et al., 2006 [JoC, MoS, SRC] for calculation of signal strength). For ocean basins, the signal is estimated from PCM( Table 8.1 )while for the World Ocean it is estimated from both PCM and HadCM3( Table 8.1 ). Red dots represent the projection of the observed temperature changes onto the normalised model-based pattern of warming. They show substantial basin-to-basin differences in how the oceans have warmed over the past 40 years, although all oceans have experienced net warming over that interval. The red bars represent the ±2 standard deviation limits associated with sampling uncertainty. The blue crosshatched swaths represent the 90% confidence limits of the natural internal variability strength. The green crosshatched swaths represent the range of the anthropogenically forced signal estimates from different realisations of identically forced simulations with the PCM model for each ocean basin (the smaller dots within the green swaths are the individual realisations) and the green shaded regions represent the range of anthropogenically forced signal estimates from different realisations of identically forced simulations with the PCM and HadCM3 models for the World Ocean (note that PCM and HadCM3 use different representations of anthropogenic forcing). The ensemble-averaged strength of the warming signal in four PCM simulations with solar and volcanic forcing is also shown (grey triangles). From Barnett et al. 2005 [JoC] and Pierce et al. 2006 [JoC, MoS, SRC] ).

In contrast, changes in solar forcing can potentially explain only a small fraction of the observationally based estimates of the increase in ocean heat content ( Crowley et al., 2003 [PoC, JoC, MoS] ), and the cooling influence of natural (volcanic) and anthropogenic aerosols would have slowed ocean warming over the last half century. Delworth et al. 2005 [JoC, MoS, SRC] ) find a delay of several decades and a reduction in the magnitude of the warming of approximately two-thirds in simulations with the GFDL-CM2 model that included these forcings compared to the response to increasing greenhouse gases alone, consistent with results based on an upwelling diffusion EBM ( Crowley et al., 2003 [PoC, JoC, MoS] ). Reductions in ocean heat content are found following volcanic eruptions in climate simulations ( Church et al., 2005 [JoC, ARC] ), including a persistent centennial time-scale signal of ocean cooling at depth following the eruption of Krakatoa ( Gleckler et al., 2006 [JoC, ARC] ).

Although the heat uptake in the ocean cannot be explained without invoking anthropogenic forcing, there is some evidence that the models have overestimated how rapidly heat has penetrated below the ocean’s mixed layer ( Forest et al., 2006 [JoC, MoS, ARC] ; see also Figure 9.15 ). In simulations that include natural forcings in addition to anthropogenic forcings, eight coupled climate models simulate heat uptake of 0.26 ± 0.06 Wm^–2 (±1 standard deviation) for 1961 to 2003, whereas observations of ocean temperature changes indicate a heat uptake of 0.21 ± 0.04 Wm^–2 ( Section 5.2.2.1 ). These could be consistent within their uncertainties but might indicate a tendency of climate models to overestimate ocean heat uptake.

In addition, the interannual to decadal variability seen in Levitus et al. 2000 [JoC, ARC] , 2005 )( Section 5.2.2 )is underestimated by models; Gregory et al. 2004 [JoC, MoS, SRC] ) show significant differences between observed and modelled interannual deviations from a linear trend in five-year running means of world ocean heat content above 3,000 m for 1957 to 1994 . While some studies note the potential importance of the choice of infilling method in poorly sampled regions ( Gregory et al., 2004 [JoC, MoS, SRC] ; AchutaRao et al., 2006 [JoC, MoS, ARC] ), the consistency of the differently processed data from the Levitus et al. 2005 [JoC, ARC] Ishii et al. 2006 [MoS, ARC] and Willis et al. 2004 [JoC, ARC] ) analyses adds confidence to their use for analysing trends in climate change studies( Chapter5 Gregory et al. 2004 [JoC, MoS, SRC] ) show that agreement between models and observations is better in the well-observed upper ocean (above 300 m) in the NH and that there is large sensitivity to the method of infilling the observational data set outside this well-observed region. They find a strong maximum in variability in the Levitus data set at around 500 m depth that is not seen in HadCM3 simulations, a possible indication of model deficiency or an artefact in the Levitus data. AchutaRao et al. 2006 [JoC, MoS, ARC] ) also find that observational estimates of temperature variability over much of the oceans may be substantially affected by sparse observational coverage and the method of infilling.

9.5.1.2 Water Mass Properties

Interior water masses, which are directly ventilated at the ocean surface, act to integrate highly variable surface changes in heat and freshwater, and could therefore provide indicators of global change ( Stark et al., 2006 [JoC, SRC] ). Some studies have attempted to investigate changes in three-dimensional water mass properties( Section 5.3 ). Sub-Antarctic Mode Water (SAMW) and the subtropical gyres have warmed in the Indian and Pacific basins since the 1960 s, waters at high latitudes have freshened in the upper 500 m and salinity has increased in some of the subtropical gyres. These changes are consistent with an increase in meridional moisture flux over the oceans over the last 50 years leading to increased precipitation at high latitudes( Section 5.2.3 ;Wong et al., 1999 [JoC, ARC] ) and a reduction in the difference between precipitation and evaporation at mid-latitudes( Section 5.6 ). This suggests that the ocean might integrate rainfall changes to produce detectable salinity changes. Boyer et al. 2005 [JoC, ARC] ) estimated linear trends in salinity for the global ocean from 1955 to 1998 that indicate salinification in the Antarctic Polar Frontal Zone around 40°S and in the subtropical North Atlantic, and freshening in the sub-polar Atlantic (Figures 5.5 and 5.7 ). However, variations in other terms (e.g., ocean freshwater transport) may be contributing substantially to the observed salinity changes and have not been quantified.

An observed freshening of SAMW in the South Indian Ocean between the 1960 s and 1990 s has been shown to be consistent with anthropogenically forced simulations from HadCM3 ( Banks et al., 2000 [JoC] ) but care should be taken in interpreting sparse hydrographic data, since apparent trends could reflect natural variability or the aliased effect of changing observational coverage. Although SAMW was fresher along isopycnals in 1987 than in the 1960 s, in 2002 the salinity was again near the 1960 s values ( Bindoff and McDougall, 2000 [ARC] ; Bryden et al., 2003 [JoC, ARC] ). An analysis of an ocean model forced by observed atmospheric fluxes and SSTs indicates that this is likely associated with natural variability ( Murray et al., 2007 [NPR, JoC, MoS, ARC, 2007] ), a result supported by an analysis of 20th-century simulations with HadCM3 which shows that it is not possible to reject the null hypothesis that the observed differences are due to internal variability ( Stark et al., 2006 [JoC, SRC] ), although this model does project a long-term freshening trend in the 21st century due to the large-scale response to surface heating and hydrological changes ( Banks et al., 2000 [JoC] ).

9.5.1.3 Changes in the Meridional Overturning Circulation

It is possible that anthropogenic and natural forcing may have influenced the MOC in the Atlantic (see also Box 5.1 ). One possible oceanic consequence of climate change is a slowing down or even halting of the MOC. An estimate of the overturning circulation and associated heat transport based on a trans-Atlantic section along latitude 25°N indicates that the Atlantic MOC has slowed by about 30% over five samples taken between 1957 and 2004 ( Bryden et al., 2005 [JoC] ), although given the infrequent sampling and considerable variability it is not clear whether the trend estimate is robust( Box 5.1 ). Freshening of North East Atlantic Deep Water has been observed ( Dickson et al., 2002 [JoC] ; Curry et al., 2003 [JoC, ARC] ; Figure 5.6 )and has been interpreted as being consistent with an enhanced difference between precipitation and evaporation at high latitudes and a possible slowing down of the MOC. Wu et al. 2004 [NPR, JoC, SRC] ) show that the observed freshening trend is well reproduced by an ensemble of HadCM3 simulations that includes both anthropogenic and natural forcings, but this freshening coincides with a strengthening rather than a weakening trend in the MOC. Therefore, this analysis is not consistent with an interpretation of the observed freshening trends in the North Atlantic as an early signal of a slowdown of the thermohaline circulation. Dickson et al. 2002 [JoC] ) propose a possible role for the Arctic in driving the observed freshening of the subpolar North Atlantic. Wu et al. 2005 [JoC, SRC] ) show that observed increases in arctic river flow ( Peterson et al., 2002 [JoC] ) are well simulated by HadCM3 including anthropogenic and natural forcings and propose that this increase is anthropogenic, since it is not seen in HadCM3 simulations including just natural forcing factors. However, the relationship between this increased source of freshwater and freshening in the Labrador Sea is not clear in the HadCM3 simulations, since Wu et al. 2007 [NPR, MoS, SRC, 2007] ) find that recent freshening in the Labrador Sea is simulated by the model when it is driven by natural rather than anthropogenic forcings. Importantly, freshening is also associated with decadal and multi-decadal variability, with links to the NAO( Box 5.1 )and the AMO( Box 5.1 ;Vellinga and Wu, 2004 [JoC] ; Knight et al., 2005 [JoC] ).

9.5.3.1 El Niño-Southern Oscillation/Pacific Decadal Oscillation

The El Niño-Southern Oscillation is the leading mode of variability in the tropical Pacific, and it has impacts on climate around the globe( Section 3.6.2 ). There have been multi-decadal oscillations in the ENSO index (conventionally defined as a mean SST anomaly in the eastern equatorial Pacific) throughout the 20th century, with more intense El Niño events since the late 1970 s, which may reflect in part a mean warming of the eastern equatorial Pacific ( Mendelssohn et al., 2005 [JoC] ). Model projections of future climate change generally show a mean state shift towards more El-Niño-like conditions, with enhanced warming in the eastern tropical Pacific and a weakened Walker Circulation( Section 10.3.5.3 ); there is some evidence that such a weakening has been observed over the past 140 years ( Vecchi et al., 2006 [JoC, MoS] ). While some simulations of the response to anthropogenic influence have shown an increase in ENSO variability in response to greenhouse gas increases ( Timmermann, 1999 [Ambiguous] ; Timmermann et al., 1999 [Ambiguous] ; Collins, 2000b [JoC, ARC] ), others have shown no change (e.g., Collins, 2000a [JoC, MoS, ARC] ) or a decrease in variability ( Knutson et al., 1997 [JoC, MoS, ARC] ). A recent survey of the simulated response to atmospheric CO₂ doubling in 15 MMD AOGCMs ( Merryfield, 2006 [JoC, MoS] ) finds that three of the models exhibited significant increases in ENSO variability, five exhibited significant decreases and seven exhibited no significant change. Thus, as yet there is no detectable change in ENSO variability in the observations, and no consistent picture of how it might be expected to change in response to anthropogenic forcing( Section 10.3.5.3 ).

Decadal variability in the North Pacific is characterised by variations in the strength of the Aleutian Low coupled to changes in North Pacific SST (Sections 3.6.3 and 8.4.2 ). The leading mode of decadal variability in the North Pacific is usually referred to as the PDO, and has a spatial structure in the atmosphere and upper North Pacific Ocean similar to the pattern that is associated with ENSO. One recent study showed a consistent tendency towards the positive phase of the PDO in observations and simulations with the MIROC model that included anthropogenic forcing ( Shiogama et al., 2005 [JoC, MoS, SRC] ), although differences between the observed and simulated PDO patterns, and the lack of additional studies, limit confidence in these findings.

9.5.3.2 North Atlantic Oscillation/Northern Annular Mode

The NAM is an approximately zonally symmetric mode of variability in the NH ( Thompson and Wallace, 1998 [JoC] ), and the NAO ( Hurrell, 1996 [JoC, ARC] ) may be viewed as its Atlantic counterpart( Section 3.6.4 ). The NAM index exhibited a pronounced trend towards its positive phase between the 1960 s and the 1990 s, corresponding to a decrease in surface pressure over the Arctic and an increase over the subtropical North Atlantic (see Section 3.6.4 ;see also Hurrell, 1996 [JoC, ARC] ; Thompson et al., 2000 [Ambiguous] ; Gillett et al., 2003a [NPR, PoC, SRC] ). Several studies have shown this trend to be inconsistent with simulated internal variability ( Osborn et al., 1999 [PoC, JoC, MoS, ARC] ; Gillett et al., 2000 [JoC, SRC] , 2002b; Osborn, 2004 [PoC, JoC, MoS, ARC] ; Gillett, 2005 [JoC, SRC] ). Although the NAM index has decreased somewhat since its peak in the mid- 1990 s, the trend calculated over recent decades remains significant at the 5% significance level compared to simulated internal variability in most models ( Osborn, 2004 [PoC, JoC, MoS, ARC] ; Gillett, 2005 [JoC, SRC] ), although one study found that the NAO index trend was marginally consistent with internal variability in one model ( Selten et al., 2004 [JoC, MoS] ).

Most climate models simulate some increase in the NAM index in response to increased concentrations of greenhouse gases ( Fyfe et al., 1999 [JoC, ARC] ; Paeth et al., 1999 [JoC] ; Shindell et al., 1999 [PoC, JoC, MoS, ARC] ; Gillett et al., 2003a [NPR, PoC, SRC] ,b; Osborn, 2004 [PoC, JoC, MoS, ARC] ; Rauthe et al., 2004 [JoC, MoS] ), although the simulated trend is generally smaller than that observed ( Gillett et al., 2002b [NPR, JoC, SRC] , 2003b; Osborn, 2004 [PoC, JoC, MoS, ARC] ; Gillett, 2005 [JoC, SRC] ; and see Figure 9.16 ). Simulated sea level pressure changes are generally found to project more strongly onto the hemispheric NAM index than onto a two-station NAO index ( Gillett et al., 2002b [NPR, JoC, SRC] ; Osborn, 2004 [PoC, JoC, MoS, ARC] ; Rauthe et al., 2004 [JoC, MoS] ). Some studies have postulated an influence of ozone depletion ( Volodin and Galin, 1999 [JoC, ARC] ; Shindell et al., 2001a [PoC, JoC, MoS, SRC] ), changes in solar irradiance ( Shindell et al., 2001a [PoC, JoC, MoS, SRC] ) and volcanic eruptions ( Kirchner et al., 1999 [JoC, MoS] ; Shindell et al., 2001a [PoC, JoC, MoS, SRC] ; Stenchikov et al., 2006 [JoC, MoS, ARC] ) on the NAM. Stenchikov et al. 2006 [JoC, MoS, ARC] ) examine changes in sea level pressure following nine volcanic eruptions in the MMD 20C3M ensemble of 20th-century simulations, and find that the models simulated a positive NAM response to the volcanoes, albeit one that was smaller than that observed. Nevertheless, ozone, solar and volcanic forcing changes are generally not found to have made a large contribution to the observed NAM trend over recent decades ( Shindell et al., 2001a [PoC, JoC, MoS, SRC] ; Gillett et al., 2003a [NPR, PoC, SRC] ). Simulations incorporating all the major anthropogenic and natural forcings from the MMD 20C3M ensemble generally showed some increase in the NAM over the latter part of the 20th century ( Gillett, 2005 [JoC, SRC] ; Miller et al., 2006 [PoC, JoC, MoS, ARC] ; and see Figure 9.16 ), although the simulated trend is in all cases smaller than that observed, indicating inconsistency between simulated and observed trends at the 5% significance level ( Gillett, 2005 [JoC, SRC] ).

Figure 9.16. Comparison between observed (top) and model-simulated (bottom) December to February sea-level pressure trends (hPa per 50 years) in the NH (left panels) and SH (right panels) based on decadal means for the period 1955 to 2005 . Observed trends are based on the Hadley Centre Mean Sea Level Pressure data set (HadSLP2r, an infilled observational data set; Allan and Ansell, 2006 [JoC, ARC] ). Model-simulated trends are the mean simulated response to greenhouse gas, sulphate aerosol, stratospheric ozone, volcanic aerosol and solar irradiance changes from eight coupled models (CCSM3, GFDL-CM2.0, GFDL-CM2.1, GISS-EH, GISS-ER, MIROC3.2(medres), PCM, UKMO-HadCM3; see Table 8.1 for model descriptions). Streamlines indicate the direction of the trends (ms^–1 per 50 years) in the geostrophic wind velocity derived from the trends in sea level pressure, and the shading of the streamlines indicates the magnitude of the change, with darker streamlines corresponding to larger changes in geostrophic wind. White areas in all panels indicate regions with insufficient station-based measurements to constrain analysis. Further explanation of the construction of this figure is provided in the Supplementary Material, Appendix 9.C .Updated after Gillett et al. 2005 [JoC, SRC] ).

The mechanisms underlying NH circulation changes remain open to debate. Simulations in which observed SST changes, which may in part be externally forced, were prescribed either globally or in the tropics alone were able to capture around half of the recent trend towards the positive phase of the NAO ( Hoerling et al., 2005 [JoC] ; Hurrell et al., 2005 [JoC, ARC] ), suggesting that the trend may in part relate to SST changes, particularly over the Indian Ocean ( Hoerling et al., 2005 [JoC] ). Another simulation in which a realistic trend in stratospheric winds was prescribed was able to reproduce the observed trend in the NAO ( Scaife et al., 2005 [PoC, JoC, ARC] Rind et al. 2005a [JoC, SRC] ,b) find that both stratospheric changes and changes in SST can force changes in the NAM and NAO, with changes in SSTs being the dominant forcing mechanism.

Over the period 1968 to 1997, the trend in the NAM was associated with approximately 50% of the winter surface warming in Eurasia, due to increased advection of maritime air onto the continent, but only a small fraction (16%) of the NH extratropical annual mean warming trend ( Thompson et al., 2000 [Ambiguous] ; Section 3.6.4 and Figure 3.30 ). It was also associated with a decrease in winter precipitation over southern Europe and an increase over northern Europe, due the northward displacement of the storm track ( Thompson et al., 2000 [Ambiguous] ).

9.5.3.3 Southern Annular Mode

The SAM is more zonally symmetric than its NH counterpart ( Thompson and Wallace, 2000 [JoC] ; Section 3.6.5 ). It too has exhibited a pronounced upward trend over the past 30 years, corresponding to a decrease in surface pressure over the Antarctic and an increase over the southern mid-latitudes( Figure 9.16 ), although the mean SAM index since 2000 has been below the mean in the late 1990 s, but above the long term mean( Figure 3.32 ). An upward trend in the SAM has occurred in all seasons, but the largest trend has been observed during the southern summer ( Thompson et al., 2000 [Ambiguous] ; Marshall, 2003 [JoC, ARC] Marshall et al. 2004 [JoC, ARC] ) show that observed trends in the SAM are not consistent with simulated internal variability in HadCM3, suggesting an external cause. On the other hand, Jones and Widmann 2004 [JoC, ARC] ) develop a 95-year reconstruction of the summer SAM index based largely on mid-latitude pressure measurements, and find that their reconstructed SAM index was as high in the early 1960 s as in the late 1990 s. However, a more reliable reconstruction from 1958, using more Antarctic data and a different method, indicates that the summer SAM index was higher at the end of the 1990 s than at any other time in the observed record ( Marshall et al., 2004 [JoC, ARC] ).

Based on an analysis of the structure and seasonality of the observed trends in SH circulation, Thompson and Solomon 2002 [PoC, JoC, ARC] ) suggest that they have been largely induced by stratospheric ozone depletion. Several modelling studies simulate an upward trend in the SAM in response to stratospheric ozone depletion ( Sexton, 2001 [JoC, SRC] ; Gillett and Thompson, 2003 [JoC, MoS, SRC] ; Marshall et al., 2004 [JoC, ARC] ; Shindell and Schmidt, 2004 [PoC, JoC, ARC] ; Arblaster and Meehl, 2006 [JoC, MoS, ARC] ; Miller et al., 2006 [PoC, JoC, MoS, ARC] ), particularly in the southern summer. Stratospheric ozone depletion cools and strengthens the antarctic stratospheric vortex in spring, and observations and models indicate that this strengthening of the stratospheric westerlies can be communicated downwards into the troposphere ( Thompson and Solomon, 2002 [PoC, JoC, ARC] ; Gillett and Thompson, 2003 [JoC, MoS, SRC] ). While ozone depletion may be the dominant cause of the trends, other studies have indicated that greenhouse gas increases have also likely contributed ( Fyfe et al., 1999 [JoC, ARC] ; Kushner et al., 2001 [JoC, SRC] ; Stone et al., 2001 [JoC, MoS, SRC] ; Cai et al., 2003 [JoC, SRC] ; Marshall et al., 2004 [JoC, ARC] ; Shindell and Schmidt, 2004 [PoC, JoC, ARC] ; Stone and Fyfe, 2005 [JoC, SRC] ; Arblaster and Meehl, 2006 [JoC, MoS, ARC] ). During the southern summer, the trend in the SAM has been associated with the observed increase of about 3 ms^–1 in the circumpolar westerly winds over the Southern Ocean. This circulation change is estimated to explain most of the summer surface cooling over the Antarctic Plateau, and about one-third to one-half of the warming of the Antarctic Peninsula ( Thompson and Solomon, 2002 [PoC, JoC, ARC] ; Carril et al., 2005 [JoC, MoS] ; Section 3.6.5 ), with the largest influence on the eastern side of the Peninsula ( Marshall et al., 2006 [JoC, ARC] ), although other factors are also likely to have contributed to this warming ( Vaughan et al., 2001 [JoC, ARC] ).

9.5.3.4 Sea Level Pressure Detection and Attribution

Global December to February sea level pressure changes observed over the past 50 years have been shown to be inconsistent with simulated internal variability ( Gillett et al., 2003b [JoC, SRC] , 2005 ), but are consistent with the simulated response to greenhouse gas, stratospheric ozone, sulphate aerosol, volcanic aerosol and solar irradiance changes based on 20C3M simulations by eight MMD coupled models ( Gillett et al., 2005 [JoC, SRC] ; Figure 9.16 ). This result is dominated by the SH, where the inclusion of stratospheric ozone depletion leads to consistency between simulated and observed sea level pressure changes. In the NH, simulated sea level pressure trends are much smaller than those observed ( Gillett, 2005 [JoC, SRC] ). Global mean sea level pressure changes associated with increases in atmospheric water vapour are small in comparison to the spatial variations in the observed change in sea level pressure, and are hard to detect because of large observational uncertainties ( Trenberth and Smith, 2005 [PoC, JoC, ARC] ).

The current understanding of climate change in the monsoon regions remains one of considerable uncertainty with respect to circulation and precipitation (Sections 3.7 , 8.4.10 and 10.3.5.2 ). The Asian monsoon circulation in the MMD models was found to decrease by 15% by the late 21st century under the SRES A1B scenario ( Tanaka et al., 2005 [MoS] ; Ueda et al., 2006 [JoC, MoS] ), but trends during the 20th century were not examined. Ramanathan et al. 2005 [JoC, ARC] ) simulate a pronounced weakening of the Asian monsoon circulation between 1985 and 2000 in response to black carbon aerosol increases. ( Chase et al. 2003 ) ) examine changes in several indices of four major tropical monsoonal circulations (Southeastern Asia, western Africa, eastern Africa and the Australia/Maritime Continent) for the period 1950 to 1998 . They find significantly diminished monsoonal circulation in each region, although this result is uncertain due to changes in the observing system affecting the NCEP reanalysis( Section 3.7 ). These results are consistent with simulations ( Ramanathan et al., 2005 [JoC, ARC] ; Tanaka et al., 2005 [MoS] ) of weakening monsoons due to anthropogenic factors, but further model and empirical studies are required to confirm this.

9.5.3.5 Monsoon Circulation

9.5.3.6 Tropical Cyclones

Several recent events, including the active North Atlantic hurricane seasons of 2004 and 2005, the unusual development of a cyclonic system in the subtropical South Atlantic that hit the coast of southern Brazil in March 2004 (e.g., Pezza and Simmonds, 2005 [JoC, ARC] ) and a hurricane close to the Iberian Peninsula in October 2005, have raised public and media interest in the possible effects of climate change on tropical cyclone activity. The TAR concluded that there was ‘no compelling evidence to indicate that the characteristics of tropical and extratropical storms have changed’, but that an increase in tropical peak wind intensities was likely to occur in some areas with an enhanced greenhouse effect (see also Box 3.5 and Trenberth, 2005 [PoC, JoC, ARC] ). The spatial resolution of most climate models limits their ability to realistically simulate tropical cyclones( Section 8.5.3 ), therefore, most studies of projected changes in hurricanes have either used time slice experiments with high-resolution atmosphere models and prescribed SSTs, or embedded hurricane models in lower-resolution General Circulation Models (GCMs)( Section 10.3.6.3 ). While results vary somewhat, these studies generally indicate a reduced frequency of tropical cyclones in response to enhanced greenhouse gas forcing, but an increase in the intensity of the most intense cyclones( Section 10.3.6.3 ). It has been suggested that the simulated frequency reduction may result from a decrease in radiative cooling associated with increased CO₂ concentration ( Sugi and Yoshimura, 2004 [JoC, ARC] ; Yoshimura and Sugi, 2005 [ARC] ; Section 10.3.6.3 ; Box 3.5 ), while the enhanced atmospheric water vapour concentration under greenhouse warming increases available potential energy and thus cyclone intensity ( Trenberth, 2005 [PoC, JoC, ARC] ).

There continues to be little evidence of any trend in the observed total frequency of global tropical cyclones, at least up until the late 1990 s (e.g., Solow and Moore, 2002 [JoC] ; Elsner et al., 2004 [JoC, MoS] ; Pielke et al., 2005 [Ambiguous] ; Webster et al., 2005 [JoC] ). However, there is some evidence that tropical cyclone intensity may have increased. Globally, Webster et al. 2005 [JoC] ) find a strong increase in the number and proportion of the most intense tropical cyclones over the past 35 years. Emanuel 2005 [JoC] ) reports a marked increase since the mid- 1970 s in the Power Dissipation Index (PDI), an index of the destructiveness of tropical cyclones (essentially an integral, over the lifetime of the cyclone, of the cube of the maximum wind speed), in the western North Pacific and North Atlantic, reflecting the apparent increases in both the duration of cyclones and their peak intensity. Several studies have shown that tropical cyclone activity was also high in the 1950 to 1970 period in the North Atlantic ( Landsea, 2005 [JoC] ) and North Pacific ( Chan, 2006 [JoC] ), although recent values of the PDI may be higher than those recorded previously ( Emanuel, 2005 [JoC] ; Section 3.8.3 Emanuel 2005 [JoC] and Elsner et al. 2006 [JoC] ) report a strong correlation between the PDI and tropical Atlantic SSTs, although Chan and Liu 2004 [JoC] ) find no analogous relationship in the western North Pacific. While changes in Atlantic SSTs have been linked in part to the AMO, the recent warming appears to be mainly associated with increasing global temperatures( Section 3.8.3.2 ;Mann and Emanuel, 2006 [PoC, JoC] ; Trenberth and Shea, 2006 [PoC, JoC, ARC] ). Tropical cyclone development is also strongly influenced by vertical wind shear and static stability( Box 3.5 ). While increasing greenhouse gas concentrations have likely contributed to a warming of SSTs, effects on static stability and wind shear may have partly opposed this influence on tropical cyclone formation( Box 3.5 ). Thus, detection and attribution of observed changes in hurricane intensity or frequency due to external influences remains difficult because of deficiencies in theoretical understanding of tropical cyclones, their modelling and their long-term monitoring (e.g., Emanuel, 2005 [JoC] ; Landsea, 2005 [JoC] ; Pielke, 2005 [Ambiguous] ). These deficiencies preclude a stronger conclusion than an assessment that anthropogenic factors more likely than not have contributed to an increase in tropical cyclone intensity.

9.5.3.7 Extratropical Cyclones

Simulations of 21st-century climate change in the MMD 20C3M model ensemble generally exhibit a decrease in the total number of extratropical cyclones in both hemispheres, but an increase in the number of the most intense events ( Lambert and Fyfe, 2006 [JoC, MoS, ARC] ), although this behaviour is not reproduced by all models ( Bengtsson et al., 2006 [JoC, ARC] ; Section 10.3.6.4 ). Many 21st-century simulations also show a poleward shift in the storm tracks in both hemispheres ( Bengtsson et al., 2006 [JoC, ARC] ; Section 10.3.6.4 ). Recent observational studies of winter NH storms have found a poleward shift in storm tracks and increased storm intensity, but a decrease in total storm numbers, in the second half of the 20th century( Section 3.5.3 ). Analysis of observed wind and significant wave height suggests an increase in storm activity in the NH. In the SH, the storm track has also shifted poleward, with increases in the radius and depth of storms, but decreases in their frequency. These features appear to be associated with the observed trends in the SAM and NAM. Thus, simulated and observed changes in extratropical cyclones are broadly consistent, but an anthropogenic influence has not yet been detected, owing to large internal variability and problems due to changes in observing systems( Section 3.5.3 ).

Figure 9.17. Global mean (ocean-only) anomalies relative to 1987 to 2000 in column-integrated water vapour (%) from simulations with the GFDL AM2-LM2 AGCM forced with observed SSTs (red), and satellite observations from SSM/I (black, Wentz and Schabel, 2000 [JoC] ). From Soden et al. 2005 [JoC, ARC] ).

9.5.4.1 Changes in Atmospheric Water Vapour

The amount of moisture in the atmosphere is expected to increase in a warming climate ( Trenberth et al., 2005 [PoC, JoC, ARC] ) because saturation vapour pressure increases with temperature according to the Clausius-Clapeyron equation. Satellite-borne Special Sensor Microwave/Imager (SSM/I) measurements of water vapour since 1988 are of higher quality than either radiosonde or reanalysis data ( Trenberth et al., 2005 [PoC, JoC, ARC] ) and show a statistically significant upward trend in precipitable (column-integrated) water of 1.2 ± 0.3 % per decade averaged over the global oceans( Section 3.4.2.1 Soden et al. 2005 [JoC, ARC] ) demonstrate that the observed changes, including the upward trend, are well simulated in the GFDL atmospheric model when observed SSTs are prescribed( Figure 9.17 ). The simulation and observations show common low-frequency variability, which is largely associated with ENSO. Soden et al. 2005 [JoC, ARC] ) also demonstrate that upper-tropospheric changes in water vapour are realistically simulated by the model. Observed warming over the global oceans is likely largely anthropogenic( Figure 9.12 ), suggesting that anthropogenic influence has contributed to the observed increase in atmospheric water vapour over the oceans.

9.5.4.2 Global Precipitation Changes

The increased atmospheric moisture content associated with warming might be expected to lead to increased global mean precipitation( Section 9.5.4.1 ). Global annual land mean precipitation showed a small, but uncertain, upward trend over the 20th century of approximately 1.1 mm per decade ( Section 3.3.2.1 and Table 3.4 ). However, the record is characterised by large inter-decadal variability, and global annual land mean precipitation shows a non-significant decrease since 1950 ( Figure 9.18 ;see also Table 3.4 ).

Figure 9.18. Simulated and observed anomalies (with respect to 1961 - 1990 ) in terrestrial mean precipitation (a), and zonal mean precipitation trends 1901 - 1998 (b). Observations (thick black line) are based on a gridded data set of terrestrial rain gauge measurements ( Hulme et al., 1998 [PoC, JoC, MoS, ARC] ). Model data are from 20th-century MMD integrations with anthropogenic, solar and volcanic forcing from the following coupled climate models (see Table 8.1 for model details): UKMO-HadCM3 (brown), CCSM3 (dark blue), GFDL-CM2.0 (pale green), GFDL-CM2.1 (pale blue), GISS-EH (red), GISS-ER (thin black), MIROC3.2(medres) (orange), MRI-CGCM2.3.2 (dark green) and PCM (pink). Coloured curves are ensemble means from individual models. In (a), a five-year running mean was applied to suppress other sources of natural variability, such as ENSO. In (b), the grey band indicates the range of trends simulated by individual ensemble members, and the thick dark blue line indicates the multi-model ensemble mean. External influence in observations on global terrestrial mean precipitation is detected with those precipitation simulations shown by continuous lines in the top panel. Adapted from Lambert et al. 2005 [JoC, MoS, SRC] ).

9.5.4.3 Regional Precipitation Changes

Observed trends in annual precipitation during the period 1901 to 2003 are shown in Figure 3.13 for regions in which data is available. Responses to external forcing in regional precipitation trends are expected to exhibit low signal-to-noise ratios and are likely to exhibit strong spatial variations because of the dependence of precipitation on atmospheric circulation and on geographic factors such as orography. There have been some suggestions, for specific regions, of a possible anthropogenic influence on precipitation, which are discussed below.

9.5.4.3.1 Sahel drought

Rainfall decreased substantially across the Sahel from the 1950 s until at least the late 1980 s ( Dai et al., 2004 [JoC, ARC] ; Figure 9.19 ,see also Figure 3.37 ). There has been a partial recovery since about 1990, although rainfall has not returned to levels typical of the period 1920 to 1965 Zeng 2003 [JoC] ) note that two main hypotheses have been proposed as a cause of the extended drought: overgrazing and conversion of woodland to agriculture increasing surface albedo and reducing moisture supply to the atmosphere, and large-scale atmospheric circulation changes related to decadal global SST changes that could be of anthropogenic or natural origin ( (Nicholson, 2001 ) ). Black carbon has also been suggested as a contributor ( Menon et al., 2002b [JoC, SRC] Taylor et al. 2002 [JoC] ) examine the impact of land use change with an atmospheric GCM forced only by estimates of Sahelian land use change since 1961 . They simulate a small decrease in Sahel rainfall (around 5% by 1996 ) and conclude that the impacts of recent land use changes are not large enough to have been the principal cause of the drought.

Figure 9.19. Observed (Climatic Research Unit TS 2.1; Mitchell and Jones, 2005 [JoC, ARC] ) Sahel July to September rainfall for each year (black), compared to an ensemble mean of 10 simulations of the atmospheric/land component of the GFDL-CM2.0 model (see Table 8.1 for model details) forced with observed SSTs (red). Both model and observations are normalized to unit mean over 1950 - 2000 . The grey band represents ±1 standard deviation of intra-ensemble variability. After Held et al. 2005 [JoC, MoS, ARC] ), based on results in Lu and Delworth 2005 [JoC, MoS, SRC] ).

Several recent studies have demonstrated that simulations with a range of atmospheric models using prescribed observed SSTs are able to reproduce observed decadal variations in Sahel rainfall ( Bader and Latif, 2003 [JoC, ARC] ; Giannini et al., 2003 [JoC, MoS] ; Rowell, 2003 [JoC] ; Haarsma et al., 2005 [JoC] ; Held et al., 2005 [JoC, MoS, ARC] ; Lu and Delworth, 2005 [JoC, MoS, SRC] ; see also Figure 9.19 ;Hoerling et al., 2006 [JoC, ARC] ), consistent with earlier findings ( Folland, 1986 [PoC, JoC, ARC] ; Rowell, 1996 [JoC] Hoerling et al. 2006 [JoC, ARC] ) show that AGCMs with observed SST changes typically underestimate the magnitude of the observed precipitation changes, although the models and observations are not inconsistent. These studies differ somewhat in terms of which ocean SSTs they find to be most important: Giannini et al. 2003 [JoC, MoS] and Bader and Latif 2003 [JoC, ARC] ) emphasize the role of tropical Indian Ocean warming, Hoerling et al. 2006 [JoC, ARC] ) attribute the drying trend to a progressive warming of the South Atlantic relative to the North Atlantic, and Rowell 2003 [JoC] ) finds that Mediterranean SSTs are an additional important contributor to decadal variations in Sahel rainfall. Based on a multi-model ensemble of coupled model simulations Hoerling et al. 2006 [JoC, ARC] ) conclude that the observed drying trend in the Sahel is not consistent with simulated internal variability alone.

Thus, recent research indicates that changes in SSTs are probably the dominant influence on rainfall in the Sahel, although land use changes possibly also contribute ( Taylor et al., 2002 [JoC] ). But what has caused the differential SST changes? Rotstayn and Lohmann 2002 [JoC, SRC] ) propose that spatially varying, anthropogenic sulphate aerosol forcing (both direct and indirect) can alter low-latitude atmospheric circulation leading to a decline in Sahel rainfall. They find a southward shift of tropical rainfall due to a hemispheric asymmetry in the SST response to changes in cloud albedo and lifetime in a climate simulation forced with recent anthropogenic changes in sulphate aerosol. Williams et al. 2001 [JoC] ) also find a southward shift of tropical rainfall as a response to the indirect effect of sulphate aerosol. These results suggest that sulphate aerosol changes may have led to reduced warming of the northern tropical oceans, which in turn led to the decrease in Sahel rainfall, possibly enhanced through land-atmosphere interaction, although a full attribution analysis has yet to be conducted. Held et al. 2005 [JoC, MoS, ARC] ) show that historical climate simulations with the both the GFDL-CM2.0 and CM2.1 models (see Table 8.1 for details) exhibit drying trends over the Sahel in the second half of the 20th century, which they ascribe to a combination of greenhouse gas and sulphate aerosol changes. The spatial pattern of the trends in simulated rainfall also shows some agreement with observations. However, Hoerling et al. 2006 [JoC, ARC] ) find that eight other coupled climate models with prescribed anthropogenic forcing do not simulate significant trends in Sahel rainfall over the 1950 to 1999 period.

9.5.5.1 Sea Ice

Widespread warming would, in the absence of other countervailing effects, lead to declines in sea ice, snow, and glacier and ice sheet extent and thickness. The annual mean area of arctic sea ice cover has decreased in recent decades, with stronger declines in summer than in winter, and some thinning( Section 4.4 Gregory et al. 2002b [JoC, MoS, SRC] ) show that a four-member ensemble of HadCM3 integrations with all major anthropogenic and natural forcing factors simulates a decline in arctic sea ice extent of about 2.5% per decade over the period 1970 to 1999, which is close to the observed decline of 2.7% per decade over the satellite period 1978 to 2004 . This decline is inconsistent with simulated internal climate variability and the response to natural forcings alone ( Vinnikov et al., 1999 [JoC] ; Gregory et al., 2002b [JoC, MoS, SRC] ; Johannssen et al., 2004 [JoC, MoS] ), indicating that anthropogenic forcing has likely contributed to the trend in NH sea ice extent. Models such as those described by Rothrock et al. 2003 [JoC, MoS, ARC] ) and references therein are able to reproduce the observed interannual variations in ice thickness, at least when averaged over fairly large regions. Simulations of historical arctic ice thickness or volume ( Goeberle and Gerdes, 2003 [JoC, ARC] ; Rothrock et al., 2003 [JoC, MoS, ARC] ) show a marked reduction in ice thickness starting in the late 1980 s, but disagree somewhat with respect to trends and/or variations earlier in the century. Although some of the dramatic change inferred may be a consequence of a spatial redistribution of ice volume over time (e.g., Holloway and Sou, 2002 [JoC] ), thermodynamic changes are also believed to be important. Low-frequency atmospheric variability (such as interannual changes in circulation connected to the NAM) appears to be important in flushing ice out of the Arctic Basin, thus increasing the amount of summer open water and enhancing thermodynamic thinning through the ice-albedo feedback (e.g., Lindsay and Zhang, 2005 [JoC] ). Large-scale modes of variability affect both wind driving and heat transport in the atmosphere, and therefore contribute to interannual variations in ice formation, growth and melt (e.g., Rigor et al., 2002 [JoC, ARC] ; Dumas et al., 2003 [JoC, MoS, ARC] ). Thus, the decline in arctic sea ice extent and its thinning appears to be largely, but not wholly, due to greenhouse gas forcing.

Unlike in the Arctic, a strong decline in sea ice extent has not been observed in the Antarctic during the period of satellite observations( Section 4.4.2.2 Fichefet et al. 2003 [JoC, MoS, SRC] ) conducted a simulation of Antarctic ice thickness using observationally based atmospheric forcing covering the period 1958 to 1999 . They note pronounced decadal variability, with area average ice thickness varying by ±0.1 m (compared to a mean thickness of roughly 0.9 m), but no long-term trend. However, Gregory et al. 2002b [JoC, MoS, SRC] ) find a decline in antarctic sea ice extent in their model, contrary to observations. They suggest that the lack of consistency between the observed and modelled changes in sea ice extent might reflect an unrealistic simulation of regional warming around Antarctica, rather than a deficiency in the ice model. Holland and Raphael 2006 [JoC, MoS, ARC] ) examine sea ice variability in six MMD 20C3M simulations that include stratospheric ozone depletion. They conclude that the observed weak increase in antarctic sea ice extent is not inconsistent with simulated internal variability, with some simulations reproducing the observed trend over 1979 to 2000, although the models exhibit larger interannual variability in sea ice extent than satellite observations.

9.5.5.2 Snow and Frozen Ground

Snow cover in the NH, as measured from satellites, has declined substantially in the past 30 years, particularly from early spring through summer( Section 4.2 ). Trends in snow depth and cover can be driven by precipitation or temperature trends. The trends in recent decades have generally been driven by warming at lower and middle elevations. Evidence for this includes: (a) interannual variations in NH April snow-covered area are strongly correlated (r = –0.68) with April 40°N to 60°N temperature; (b) interannual variations in snow (water equivalent, depth or duration) are strongly correlated with temperature at lower- and middle-elevation sites in North America ( Mote et al., 2005 [JoC, ARC] ), Switzerland ( Scherrer et al., 2004 [JoC, ARC] ) and Australia ( Nicholls, 2005 [SRC] ); (c) trends in snow water equivalent or snow depth show strong dependence on elevation or equivalently mean winter temperature, both in western North America and Switzerland (with stronger decreases at lower, warmer elevations where a warming is more likely to affect snowfall and snowmelt); and (d) the trends in North America, Switzerland and Australia have been shown to be well explained by warming and cannot be explained by changes in precipitation. In some very cold places, increases in snow depth have been observed and have been linked to higher precipitation.

Widespread permafrost warming and degradation appear to be the result of increased summer air temperatures and changes in the depth and duration of snow cover( Section 4.7.2 ). The thickness of seasonally frozen ground has decreased in response to winter warming and increases in snow depth( Section 4.7.3 ).

9.5.5.3 Glaciers, Ice Sheets and Ice Shelves

During the 20th century, glaciers generally lost mass with the strongest retreats in the 1930 s and 1940 s and after 1990 ( Section 4.5 ). The widespread shrinkage appears to imply widespread warming as the probable cause ( Oerlemans, 2005 [JoC, ARC] ), although in the tropics changes in atmospheric moisture might be contributing( Section 4.5.3 ). Over the last half century, both global mean winter accumulation and summer melting have increased steadily ( Ohmura, 2004 [NPR, ARC] ; Dyurgerov and Meier, 2005 [NPR, ARC] ;( Greene, 2005 ) ), and at least in the NH, winter accumulation and summer melting correlate positively with hemispheric air temperature ( (Greene, 2005 ) ); the negative correlation of net balance with temperature indicates the primary role of temperature in forcing the respective glacier fluctuations.

There have been a few studies for glaciers in specific regions examining likely causes of trends. Mass balances for glaciers in western North America are strongly correlated with global mean winter (October–April) temperatures and the decline in glacier mass balance has paralleled the increase in temperature since 1968 ( Meier et al., 2003 [JoC, ARC] Reichert et al. 2002a [JoC, ARC] ) forced a glacier mass balance model for the Nigardsbreen and Rhône glaciers with downscaled data from an AOGCM control simulation and conclude that the rate of glacier advance during the ‘Little Ice Age’ could be explained by internal climate variability for both glaciers, but that the recent retreat cannot, implying that the recent retreat of both glaciers is probably due to externally forced climate change. As well, the thinning and acceleration of some polar glaciers (e.g., Thomas et al., 2004 [JoC, ARC] ) appear to be the result of ice sheet calving driven by oceanic and atmospheric warming( Section 4.6.3.4 ).

Taken together, the ice sheets of Greenland and Antarctica are shrinking. Slight thickening in inland Greenland is more than compensated for by thinning near the coast( Section 4.6.2.2 ). Warming is expected to increase low-altitude melting and high-altitude precipitation in Greenland; altimetry data suggest that the former effect is dominant. However, because some portions of ice sheets respond only slowly to climate changes, past forcing may be influencing ongoing changes, complicating attribution of recent trends( Section 4.6.3.2 ).

9.6.2.1 Estimates of Climate Sensitivity Based on 20th-Century Warming

Anumber of recent studies have used instrumental records of surface, ocean and atmospheric temperature changes to estimate climate sensitivity. Most studies use the observed surface temperature changes over the 20th century or the last 150 years( Chapter3 ). In addition, some studies also use the estimated ocean heat uptake since 1955 based on Levitus et al. 2000 [JoC, ARC] , 2005 )( Chapter5 ), and temperature changes in the free atmosphere( Chapter3 ;see also Table 9.3 ). For example, Frame et al. 2005 [JoC, MoS] and Andronova and Schlesinger 2000 [JoC, SRC] ) use surface air temperature alone, while Forest et al. 2002 [JoC, SRC] , 2006 Knutti et al. 2002 [PoC, JoC, MoS, SRC] , 2003 and Gregory et al. 2002a [JoC, MoS, SRC] ) use both surface air temperature and ocean temperature change to constrain climate sensitivity. Forest et al. 2002 [JoC, SRC] , 2006 and Lindzen and Giannitsis 2002 [NPR, JoC] ) use free atmospheric temperature data from radiosondes in addition to surface air temperature. Note that studies using radiosonde data may be affected by recently discovered inhomogeneities( Section 3.4.1.1 ), although Forest et al. 2006 [JoC, MoS, ARC] ) illustrate that the impact of the radiosonde atmospheric temperature data on their climate sensitivity estimate is smaller than that of surface and ocean warming data. A further recent study uses Earth Radiation Budget Experiment (ERBE) data ( Forster and Gregory, 2006 [JoC, SRC] ) in addition to surface temperature changes to estimate climate feedbacks (and thus ECS) from observed changes in forcing and climate.

Wigley et al. 1997 [PoC, JoC, ARC] ) pointed out that uncertainties in forcing and response made it impossible to use observed global temperature changes to constrain ECS more tightly than the range explored by climate models at the time (1.5°C to 4.5°C), and particularly the upper end of the range, a conclusion confirmed by subsequent studies. A number of subsequent publications qualitatively describe parameter values that allow models to reproduce features of observed changes, but without directly estimating a climate sensitivity probability density function (PDF). For example, Harvey and Kaufmann 2002 [JoC, MoS] ) find a best-fit ECS of 2.0°C out of a range of 1°C to 5°C, and constrain fossil fuel and biomass aerosol forcing( Section 9.2.1.2 Lindzen and Giannitsis 2002 [NPR, JoC] ) pose the hypothesis that the rapid change in tropospheric (850–300 hPa) temperatures around 1976 triggered a delayed response in surface temperature that is best modelled with a climate sensitivity of less than 1°C. However, their estimate does not account for substantial uncertainties in the analysis of such a short time period, most notably those associated with the role of internal climate variability in the rapid tropospheric warming of 1976 . The 1976 – 1977 climate shift occurred along with a phase shift of the PDO, and a concurrent change in the ocean( Section 3.6.3 )that appears to contradict the Lindzen and Giannitsis 2002 [NPR, JoC] ) assumption that the change was initiated by tropospheric forcing. In addition, the authors do not account for uncertainties in the simple model whose sensitivity is fitted. The finding of Lindzen and Giannitsis is in contrast with that of Forest et al. 2002 [JoC, SRC] , 2006 ) who consider the joint evolution of surface and upper air temperatures on much longer time scales.

Several recent studies have derived probability estimates for ECS using a range of models and diagnostics. The diagnostics, which are used to compare model-simulated and observed changes, are often simple temperature indices such as the global mean surface temperature and ocean mean warming ( Knutti et al., 2002 [PoC, JoC, MoS, SRC] , 2003 ) or the differential warming between the SH and NH (together with the global mean; Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ). Results that use more detailed information about the space-time evolution of climate may be able to provide tighter constraints than those that use simpler indices. Forest et al. 2002 [JoC, SRC] , 2006 ) use a so-called ‘optimal’ detection method( Section 9.4.1.4 and Appendix 9.A.1) to diagnose the fit between model-simulated and observed patterns of zonal mean temperature change. Frame et al. 2005 [JoC, MoS] ) use detection results from an analysis based on several multi-model AOGCM fingerprints( Section 9.4.1.4 )that separate the greenhouse gas response from that to other anthropogenic and natural forcings ( Stott et al., 2006c [JoC, MoS, SRC] ). Similarly, Gregory et al. 2002a [JoC, MoS, SRC] ) apply an inverse estimate of the range of aerosol forcing based on fingerprint detection results. Note that while results from fingerprint detection approaches will be affected by uncertainty in separation between greenhouse gas and aerosol forcing, the resulting uncertainty in estimates of the near-surface temperature response to greenhouse gas forcing is relatively small (Sections 9.2.3 and 9.4.1.4 ).

Afurther consideration in assessing these results is the extent to which realistic forcing estimates were used, and whether forcing uncertainty was included. Most studies consider a range of anthropogenic forcing factors, including greenhouse gases and sulphate aerosol forcing, sometimes directly including the indirect forcing effect, such as Knutti et al. 2002 [PoC, JoC, MoS, SRC] , 2003 ), and sometimes indirectly accounting for the indirect effect by using a wide range of direct forcing (e.g., Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ; Forest et al., 2002 [JoC, SRC] , 2006 ). Many studies also consider tropospheric ozone (e.g., Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ; Knutti et al., 2002 [PoC, JoC, MoS, SRC] , 2003 Forest et al. 2006 [JoC, MoS, ARC] ) demonstrate that the inclusion of natural forcing affects the estimated PDF of climate sensitivity since net negative natural forcing in the second half of the 20th century favours higher sensitivities than earlier results that disregarded natural forcing ( Forest et al., 2002 [JoC, SRC] ; see Figure 9.20 ), particularly if the same ocean warming estimates were used. Note that some of the changes due to inclusion of natural forcing are offset by using recently revised ocean warming data ( Levitus et al., 2005 [JoC, ARC] ), which favour somewhat smaller ocean heat uptakes than earlier data ( Levitus et al., 2001 [JoC, ARC] ; Forest et al., 2006 [JoC, MoS, ARC] ). Only a few estimates account for uncertainty in forcings other than from aerosols (e.g., Gregory et al., 2002a [JoC, MoS, SRC] ; Knutti et al., 2002 [PoC, JoC, MoS, SRC] , 2003 ); some other studies perform some sensitivity testing to assess the effect of forcing uncertainty not accounted for, for example, in natural forcing (e.g., Forest et al., 2006 [JoC, MoS, ARC] ; see Table 9.1 for an overview).

The treatment of uncertainty in the ocean’s uptake of heat varies, from assuming a fixed value for a model’s ocean diffusivity ( Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ) to trying to allow for a wide range of ocean mixing parameters ( Knutti et al., 2002 [PoC, JoC, MoS, SRC] , 2003 ) or systematically varying the ocean’s effective diffusivity (e.g., Forest et al., 2002 [JoC, SRC] , 2006; Frame et al., 2005 [JoC, MoS] ). Furthermore, all approaches that use the climate’s time evolution attempt to account for uncertainty due to internal climate variability, either by bootstrapping ( Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ), by using a noise model in fingerprint studies whose results are used ( Frame et al., 2005 [JoC, MoS] ) or directly ( Forest et al., 2002 [JoC, SRC] , 2006 ).

Figure 9.20. Comparison between different estimates of the PDF (or relative likelihood) for ECS (°C). All PDFs/likelihoods have been scaled to integrate to unity between 0°C and 10°C ECS. The bars show the respective 5 to 95% ranges, dots the median estimate. The PDFs/likelihoods based on instrumental data are from Andronova and Schlesinger 2001 [JoC, MoS, SRC] Forest et al. 2002 [JoC, SRC] ; dashed line, considering anthropogenic forcings only), Forest et al. 2006 [JoC, MoS, ARC] ; solid, anthropogenic and natural forcings), Gregory et al. 2002a [JoC, MoS, SRC] Knutti et al. 2002 [PoC, JoC, MoS, SRC] Frame et al. 2005 [JoC, MoS] and Forster and Gregory 2006 [JoC, SRC] ), transformed to a uniform prior distribution in ECS using the method after Frame et al. 2005 [JoC, MoS] Hegerl et al. 2006a [NPR, PoC, JoC, SRC] ) is based on multiple palaeoclimatic reconstructions of NH mean temperatures over the last 700 years. Also shown are the 5 to 95% approximate ranges for two estimates from the LGM (dashed, Annan et al., 2005 [MoS, ARC] ; solid, Schneider von Deimling et al., 2006 [PoC, JoC, MoS, ARC] ) which are based on models with different structural properties. Note that ranges extending beyond the published range in Annan et al. 2005 [MoS, ARC] ), and beyond that sampled by the climate model used there, are indicated by dots and an arrow, since Annan et al. only provide an upper limit. For details of the likelihood estimates, see Table 9.3 .After Hegerl et al. 2006a [NPR, PoC, JoC, SRC] ).

Figure 9.20 compares results from many of these studies. All PDFs shown are based on a uniform prior distribution of ECS and have been rescaled to integrate to unity for all positive sensitivities up to 10°C to enable comparisons of results using different ranges of uniform prior distributions (this affects both median and upper 95th percentiles if original estimates were based on a wider uniform range). Thus, zero prior probability is assumed for sensitivities exceeding 10°C, since many results do not consider those, and for negative sensitivities. Negative climate sensitivity would lead to cooling in response to a positive forcing and is inconsistent with understanding of the energy balance of the system ( Stouffer et al., 2000 [JoC, MoS, SRC] ; Gregory et al., 2002a [JoC, MoS, SRC] ; Lindzen and Giannitsis, 2002 [NPR, JoC] ). This figure shows that best estimates of the ECS (mode of the estimated PDFs) typically range between 1.2°C and 4°C when inferred from constraints provided by historical instrumental data, in agreement with estimates derived from more comprehensive climate models. Most studies suggest a 5th percentile for climate sensitivity of 1°C or above. The upper 95th percentile is not well constrained, particularly in studies that account conservatively for uncertainty in, for example, 20th-century radiative forcing and ocean heat uptake. The upper tail is particularly long in studies using diagnostics based on large-scale mean data because separation of the greenhouse gas response from that to aerosols or climate variability is more difficult with such diagnostics ( Andronova and Schlesinger, 2001 [JoC, MoS, SRC] ; Gregory et al., 2002a [JoC, MoS, SRC] ; Knutti et al., 2002 [PoC, JoC, MoS, SRC] , 2003 Forest et al. 2006 [JoC, MoS, ARC] ) find a 5 to 95% range of 2.1°C to 8.9°C for climate sensitivity( Table 9.3 ), which is a wider range than their earlier result based on anthropogenic forcing only ( Forest et al., 2002 [JoC, SRC] Frame et al. 2005 [JoC, MoS] ) infer a 5 to 95% uncertainty range for the ECS of 1.2°C to 11.8°C, using a uniform prior distribution that extends well beyond 10°C sensitivity. Studies generally do not find meaningful constraints on the rate at which the climate system mixes heat into the deep ocean (e.g., Forest et al., 2002 [JoC, SRC] , 2006 ). However, Forest et al. 2006 [JoC, MoS, ARC] ) find that many coupled AOGCMs mix heat too rapidly into the deep ocean, which is broadly consistent with comparisons based on heat uptake( Section 9.5.1.1 ,). However the relevance of this finding is unclear because most MMD AOGCMs were not included in the Forest et al. comparison, and because they used a relatively simple ocean model. Knutti et al. 2002 [PoC, JoC, MoS, SRC] ) also determine that strongly negative aerosol forcing, as has been suggested by several observational studies ( Anderson et al., 2003 [JoC, MoS] ), is incompatible with the observed warming trend over the last century( Section 9.2.1.2 and Table 9.1 ).

Some studies have further attempted to use non-uniform prior distributions. Forest et al. 2002 [JoC, SRC] , 2006 ) obtained narrower uncertainty ranges when using expert prior distributions( Table 9.3 ). While they reflect credible prior ranges of ECS, expert priors may also be influenced by knowledge about observed climate change, and thus may yield overly confident estimates when combined with the same data (Supplementary Material, Appendix 9.B Frame et al. 2005 [JoC, MoS] ) find that sampling uniformly in TCR results in an estimated ECS of 1.2°C to 5.2°C with a median value of 2.3°C. In addition, several approaches have been based on a uniform prior distribution of climate feedback. Translating these results into ECS estimates is equivalent to using a prior distribution that favours smaller sensitivities, and hence tends to result in narrower ECS ranges ( Frame et al., 2005 [JoC, MoS] Forster and Gregory 2006 [JoC, SRC] ) estimate ECS based on radiation budget data from the ERBE combined with surface temperature observations based on a regression approach, using the observation that there was little change in aerosol forcing over that time. They find a climate feedback parameter of 2.3 ± 1.4 Wm^–2 °C^–1,which corresponds to a 5 to 95% ECS range of 1.0°C to 4.1°C if using a prior distribution that puts more emphasis on lower sensitivities as discussed above, and a wider range if the prior distribution is reformulated so that it is uniform in sensitivity( Table 9.3 ). The climate feedback parameter estimated from the MMD AOGCMs ranges from about 0.7 to 2.0 Wm^–2 °C^–1 (Supplementary Material, Table S8.1).

9.6.2.2 Estimates Based on Individual Volcanic Eruptions

Some recent analyses have attempted to derive insights into ECS from the well-observed forcing and response to the eruption of Mt. Pinatubo, or from other major eruptions during the 20th century. Such events allow for the study of physical mechanisms and feedbacks and are discussed in detail in Section 8.6 .For example, Soden et al. 2002 [JoC, ARC] ) demonstrate agreement between observed and simulated responses based on an AGCM with a climate sensitivity of 3.0°C coupled to a mixed-layer ocean, and that the agreement breaks down if the water vapour feedback in the model is switched off. Yokohata et al. 2005 [JoC, MoS] ) find that a version of the MIROC climate model with a sensitivity of 4.0°C yields a much better simulation of the Mt. Pinatubo eruption than a model version with sensitivity of 6.3°C, concluding that the cloud feedback in the latter model appears inconsistent with data. Note that both results may be specific to the model analysed.

Constraining ECS from the observed responses to individual volcanic eruptions is difficult because the response to short-term volcanic forcing is strongly nonlinear in ECS, yielding only slightly enhanced peak responses and substantially extended response times for very high sensitivities ( Frame et al., 2005 [JoC, MoS] ; Wigley et al., 2005a [PoC, JoC, MoS, SRC] ). The latter are difficult to distinguish from a noisy background climate. A further difficulty arises from uncertainty in the rate of heat taken up by the ocean in response to a short, strong forcing. Wigley et al. 2005a [PoC, JoC, MoS, SRC] ) find that the lower boundary and best estimate obtained by comparing observed and simulated responses to major eruptions in the 20th century are consistent with the TAR range of 1.5°C to 4.5°C, and that the response to the eruption of Mt. Pinatubo suggests a best fit sensitivity of 3.0°C and an upper 95% limit of 5.2°C. However, as pointed out by the authors, this estimate does not account for forcing uncertainties. In contrast, an analysis by Douglass and Knox 2005 [JoC, MoS] ) based on a box model suggests a very low climate sensitivity (under 1°C) and negative climate feedbacks based on the eruption of Mt. Pinatubo. Wigley et al. 2005b [PoC, JoC, MoS, SRC] ) demonstrate that the analysis method of Douglass and Knox 2005 [JoC, MoS] ) severely underestimates (by a factor of three) climate sensitivity if applied to a model with known sensitivity. Furthermore, as pointed out by Frame et al. 2005 [JoC, MoS] ), the effect of noise on the estimate of the climatic background level can lead to a substantial underestimate of uncertainties if not taken into account.

In summary, the responses to individual volcanic eruptions provide a useful test for feedbacks in climate models( Section 8.6 ). However, due to the physics involved in the response, such individual events cannot provide tight constraints on ECS. Estimates of the most likely sensitivity from most such studies are, however, consistent with those based on other analyses.

9.6.2.3 Constraints on Transient Climate Response

While ECS is the equilibrium global mean temperature change that eventually results from atmospheric CO₂ doubling, the smaller TCR refers to the global mean temperature change that is realised at the time of CO₂ doubling under an idealised scenario in which CO₂ concentrations increase by 1% yr^–1 ( Cubasch et al., 2001 [NPR, MoS, ARC] ; see also Section 8.6.2.1 ). The TCR is therefore indicative of the temperature trend associated with external forcing, and can be constrained by an observable quantity, the observed warming trend that is attributable to greenhouse gas forcing. Since external forcing is likely to continue to increase through the coming century, TCR may be more relevant to determining near-term climate change than ECS.

Stott et al. 2006c [JoC, MoS, SRC] ) estimate TCR based on scaling factors for the response to greenhouse gases only (separated from aerosol and natural forcing in a three-pattern optimal detection analysis) using fingerprints from three different model simulations( Figure 9.21 )and find a relatively tight constraint. Using three model simulations together, their estimated median TCR is 2.1°C at the time of CO₂ doubling (based on a 1% yr^–1 increase in CO₂), with a 5 to 95% range of 1.5°C to 2.8°C. Note that since TCR scales linearly with the errors in the estimated scaling factors, estimates do not show a tendency for a long upper tail, as is the case for ECS. However, the separation of greenhouse gas response from the responses to other external forcing in a multi-fingerprint analysis introduces a small uncertainty, illustrated by small differences in results between three models( Figure 9.21 ). The TCR does not scale linearly with ECS because the transient response is strongly influenced by the speed with which the ocean transports heat into its interior, while the equilibrium sensitivity is governed by feedback strengths (discussion in Frame et al., 2005 [JoC, MoS] ).

Estimates of a likely range for TCR can also be inferred directly from estimates of attributable greenhouse warming obtained in optimal detection analyses since there is a direct linear relationship between the two ( Frame et al., 2005 [JoC, MoS] ). The attributable greenhouse warming rates inferred from Figure 9.9 generally support the TCR range shown in Figure 9.21 ,although the lowest 5th percentile (1.3°C) and the highest 95th percentile (3.3°C) estimated in this way from detection and attribution analyses based on individual models lie outside the 5% to 95% range of 1.5°C to 2.8°C obtained from Figure 9.21 .

Choosing lower and upper limits that encompass the range of these results and deflating significance levels in order to account for structural uncertainty in the estimate leads to the conclusion that it is very unlikely that TCR is less than 1°C and very unlikely that TCR is greater than 3.5°C. Information based on the models discussed in Chapter 10 provides additional information that can help constrain TCR further( Section 10.5.4.5 ).

Figure 9.21. Probability distributions of TCR (expressed as warming at the time of CO₂ doubling), as constrained by observed 20th-century temperature change, for the HadCM3( Table 8.1 ,red), PCM( Table 8.1 ,green) and GFDL R30 ( Delworth et al., 2002 [JoC, MoS, SRC] , blue) models. The average of the PDFs derived from each model is shown in turquoise. Coloured circles show each model’s TCR. (After Stott et al., 2006c [JoC, MoS, SRC] ).

9.6.3.1 Estimates of Climate Sensitivity Based on Data for the Last Millennium

The relationship between forcing and response based on a long time horizon can be studied using palaeoclimatic reconstructions of temperature and radiative forcing, particularly volcanism and solar forcing, for the last millennium. However, both forcing and temperature reconstructions are subject to large uncertainties( Chapter6 ). To account for the uncertainty in reconstructions, Hegerl et al. 2006a [NPR, PoC, JoC, SRC] ) use several proxy data reconstructions of NH extratropical temperature for the past millennium ( Briffa et al., 2001 [PoC, JoC, ARC] ; Esper et al., 2002 [PoC, JoC, ARC] ; Mann and Jones, 2003 [PoC, JoC, ARC] ; Hegerl et al., 2007 [JoC, SRC, 2007] ) to constrain ECS estimates for the pre-industrial period up to 1850 . This study used a large ensemble of simulations of the last millennium performed with an energy balance model forced with reconstructions of volcanic ( Crowley, 2000 [PoC, JoC] , updated), solar ( Lean et al., 2002 [JoC, MoS, ARC] ) and greenhouse gas forcing (see Section 9.3.3 for results on the detection of these external influences). Their estimated PDFs for ECS incorporate an estimate of uncertainty in the overall amplitude (including an attempt to account for uncertainty in efficacy), but not the time evolution, of volcanic and solar forcing. They also attempt to account for uncertainty in the amplitude of reconstructed temperatures in one reconstruction ( Hegerl et al., 2007 [JoC, SRC, 2007] ), and assess the sensitivity of their results to changes in amplitude for others. All reconstructions combined yield a median climate sensitivity of 3.4°C and a 5 to 95% range of 1.2°C to 8.6°C( Figure 9.20 ). Reconstructions with a higher amplitude of past climate variations (e.g., Esper et al., 2002 [PoC, JoC, ARC] ; Hegerl et al., 2007 [JoC, SRC, 2007] ) are found to support higher ECS estimates than reconstructions with lower amplitude (e.g., Mann and Jones, 2003 [PoC, JoC, ARC] ). Note that the constraint on ECS originates mainly from low-frequency temperature variations associated with changes in the frequency and intensity of volcanism which lead to a highly significant detection of volcanic response( Section 9.3.3 )in all records used in the study.

The results of Andronova et al. 2004 [PoC, JoC, SRC] ) are broadly consistent with these estimates. Andronova et al. 2004 [PoC, JoC, SRC] ) demonstrate that climate sensitivities in the range of 2.3°C to 3.4°C yield reasonable simulations of both the NH mean temperature from 1500 onward when compared to the Mann and Jones 2003 [PoC, JoC, ARC] ) reconstruction, and for the instrumental period. The agreement is less good for reconstructed SH temperature, where reconstructions are substantially more uncertain( Chapter6 ).

Rind et al. 2004 [JoC, MoS, SRC] ) studied the period from about 1675 to 1715 to attempt a direct estimate of climate sensitivity. This period has reduced radiative forcing relative to the present due to decreased solar radiation, decreased greenhouse gas and possibly increased volcanic forcing( Section 9.2.1.3 ). Different NH temperature reconstructions( Figure 6.10 )have a wide range of cooling estimates relative to the late 20th century that is broadly reproduced by climate model simulations. While climate in this cold period may have been close to radiative balance ( Rind et al., 2004 [JoC, MoS, SRC] ), some of the forcing during the present period is not yet realised in the system (estimated as 0.85 Wm^–2 ;Hansen et al., 2005 [JoC, ARC] ). Thus, ECS estimates based on a comparison between radiative forcing and climate response are subject to large uncertainties, but are broadly similar to estimates discussed above. Again, reconstructions with stronger cooling in this period imply higher climate sensitivities than those with weaker cooling (results updated from Rind et al., 2004 [JoC, MoS, SRC] ).

9.6.3.2 Inferences About Climate Sensitivity Based on the Last Glacial Maximum

The LGM is one of the key periods used to estimate ECS ( Hansen et al., 1984 [NPR, ARC] ; Lorius et al., 1990 [JoC, MoS] ; Hoffert and Covey, 1992 [JoC, ARC] ), since it represents a quasi-equilibrium climate response to substantially altered boundary conditions. When forced with changes in greenhouse gas concentrations and the extent and height of ice sheet boundary conditions, AOGCMs or EMICs identical or similar to those used for 20th- and 21st-century simulations produce a 3.3°C to 5.1°C cooling for this period in response to radiative perturbations of 4.6 to 7.2 Wm^–2 (Sections 6.4.1.3 ;see also Section 9.3.2 ;see also Masson-Delmotte et al., 2006 [JoC, MoS, SRC] ). The simulated cooling in the tropics ranges from 1.7°C to 2.4°C. The ECS of the models used in PMIP2 ranges from 2.3°C to 3.7°C( Table 8.2 ), and there is some tendency for models with larger sensitivity to produce larger tropical cooling for the LGM, but this relationship is not very tight. Comparison between simulated climate change and reconstructed climate is affected by substantial uncertainties in forcing and data( Chapter6 and Section 9.2.1.3 ). For example, the PMIP2 forcing does not account for changes in mineral dust, since the level of scientific understanding for this forcing is very low( Figure 6.5 ). The range of simulated temperature changes is also affected by differences in the radiative influence of the ice-covered regions in different models ( Taylor et al., 2000 [NPR, MoS, ARC] ). Nevertheless, the PMIP2 models simulate LGM climate changes that are approximately consistent with proxy information( Chapter6 ).

Recent studies ( Annan et al., 2005 [MoS, ARC] ; Schneider von Deimling et al., 2006 [PoC, JoC, MoS, ARC] ) attempt to estimate the PDF of ECS from ensemble simulations of the LGM by systematically exploring model uncertainty. Both studies investigate the relationship between climate sensitivity and LGM tropical SSTs, which are influenced strongly by CO₂ changes. In a perturbed physics ensemble, Schneider von Deimling et al. 2006 [PoC, JoC, MoS, ARC] ) vary 11 ocean and atmospheric parameters in a 1,000-member ensemble simulation of the LGM with the CLIMBER-2 EMIC( Table 8.3 ). They find a close relationship between ECS and tropical SST cooling in their model, implying a 5 to 95% range of ECS of 1.2°C to 4.3°C when attempting to account for model parameter, forcing and palaeoclimate data uncertainties. Similar constraints on climate sensitivity are found when proxy reconstructions of LGM antarctic temperatures are used instead of tropical SSTs ( Schneider von Deimling et al., 2006 [PoC, JoC, MoS, ARC] ). In contrast, Annan et al. 2005 [MoS, ARC] ) use a perturbed physics ensemble based on a low-resolution version of the atmospheric component of the MIROC3.2 model, perturbing a range of model parameters over prior distributions determined from the ability of the model to reproduce seasonal mean climate in a range of climate variables. They find a best-fit sensitivity of about 4.5°C, and their results suggest that sensitivities in excess of 6°C are unlikely given observational estimates of LGM tropical cooling and the relationship between tropical SST and sensitivity in their model. Since the perturbed physics ensemble based on that atmospheric model does not produce sensitivities less than 4°C, this result cannot provide a lower limit or a PDF for ECS.

The discrepancy between the inferred upper limits in the two studies probably arises from both different radiative forcing and structural differences between the models used. Forcing from changes in vegetation cover and dust is not included in the simulations done by Annan et al. 2005 [MoS, ARC] ), which according to Schneider von Deimling et al. 2006 [PoC, JoC, MoS, ARC] ) would reduce the Annan et al. ECS estimates and yield better agreement between the results of the two studies. However, the effect of these forcings and their interaction with other LGM forcings is very uncertain, limiting confidence in such estimates of their effect( Figure 6.5 ). Structural differences in models are also likely to play a role. The Annan et al. 2005 [MoS, ARC] ) estimate shows a weaker association between simulated tropical SST changes and ECS than the Schneider von Deimling et al. 2006 [PoC, JoC, MoS, ARC] ) result. Since Annan et al. use a mixed-layer ocean model, and Schneider von Deimling a simplified ocean model, both models may not capture the full ocean response affecting tropical SSTs. The atmospheric model used in Schneider von Deimling is substantially simpler than that used in the Annan et al. 2005 [MoS, ARC] ) study. Overall, estimates of climate sensitivity from the LGM are broadly consistent with other estimates of climate sensitivity derived, for example, from the instrumental period.