IPCC Fourth Assessment Report: Climate Change 2007
Climate Change 2007: Working Group I: The Physical Science Basis

10.3.4 Changes in the Atlantic Meridional Overturning Circulation

A feature common to all climate model projections is the increase in high-latitude temperature as well as an increase in high-latitude precipitation. This was reported in the TAR and is confirmed by the projections using the latest versions of comprehensive climate models (see Section 10.3.2). Both of these effects tend to make the high-latitude surface waters less dense and hence increase their stability, thereby inhibiting convective processes. As more coupled models have become available since the TAR, the evolution of the Atlantic Meridional Overturning Circulation (MOC) can be more thoroughly assessed. Figure 10.15 shows simulations from 19 coupled models integrated from 1850 to 2100 under SRES A1B atmospheric CO2 and aerosol scenarios up to year 2100, and constant concentrations thereafter (see Figure 10.5). All of the models, except CGCM3.1, INM-CM3.0 and MRI-CGCM2.3.2, were run without flux adjustments (see Table 8.1). The MOC is influenced by the density structure of the Atlantic Ocean, small-scale mixing and the surface momentum and buoyancy fluxes. Some models simulate a MOC strength that is inconsistent with the range of present-day estimates (Smethie and Fine, 2001; Ganachaud, 2003; Lumpkin and Speer, 2003; Talley, 2003). The MOC for these models is shown for completeness but is not used in assessing potential future changes in the MOC in response to various emissions scenarios.

Figure 10.15

Figure 10.15. Evolution of the Atlantic meridional overturning circulation (MOC) at 30°N in simulations with the suite of comprehensive coupled climate models (see Table 8.1 for model details) from 1850 to 2100 using 20th Century Climate in Coupled Models (20C3M) simulations for 1850 to 1999 and the SRES A1B emissions scenario for 1999 to 2100. Some of the models continue the integration to year 2200 with the forcing held constant at the values of year 2100. Observationally based estimates of late-20th century MOC are shown as vertical bars on the left. Three simulations show a steady or rapid slow down of the MOC that is unrelated to the forcing; a few others have late-20th century simulated values that are inconsistent with observational estimates. Of the model simulations consistent with the late-20th century observational estimates, no simulation shows an increase in the MOC during the 21st century; reductions range from indistinguishable within the simulated natural variability to over 50% relative to the 1960 to 1990 mean; and none of the models projects an abrupt transition to an off state of the MOC. Adapted from Schmittner et al. (2005) with additions.

Fewer studies have focused on projected changes in the Southern Ocean resulting from future climate warming. A common feature of coupled model simulations is the projected poleward shift and strengthening of the SH westerlies (Yin, 2005; Fyfe and Saenko, 2006). This in turn leads to a strengthening, poleward shift and narrowing of the Antarctic Circumpolar Current. Fyfe and Saenko (2006) further note that the enhanced equatorward surface Ekman transport, associated with the intensified westerlies, is balanced by an enhanced deep geostrophic poleward return flow below 2,000 m.

Generally, the simulated late-20th century Atlantic MOC shows a spread ranging from a weak MOC of about 12 Sv to over 20 Sv (Figure 10.15; Schmittner et al., 2005). When forced with the SRES A1B scenario, the models show a reduction in the MOC of up to 50% or more, but in one model, the changes are not distinguishable from the simulated natural variability. The reduction in the MOC proceeds on the time scale of the simulated warming because it is a direct response to the increase in buoyancy at the ocean surface. A positive North Atlantic Oscillation (NAO) trend might delay this response by a few decades but not prevent it (Delworth and Dixon, 2000). Such a weakening of the MOC in future climate causes reduced sea surface temperature (SST) and salinity in the region of the Gulf Stream and North Atlantic Current (Dai et al., 2005). This can produce a decrease in northward heat transport south of 60°N, but increased northward heat transport north of 60°N (A. Hu et al., 2004). No model shows an increase in the MOC in response to the increase in greenhouse gases, and no model simulates an abrupt shut-down of the MOC within the 21st century. One study suggests that inherent low-frequency variability in the Atlantic region, the Atlantic Multidecadal Oscillation, may produce a natural weakening of the MOC over the next few decades that could further accentuate the decrease due to anthropogenic climate change (Knight et al., 2005; see Section 8.4.6).

In some of the older models (e.g., Dixon et al., 1999), increased high-latitude precipitation dominates over increased high-latitude warming in causing the weakening, while in others (e.g., Mikolajewicz and Voss, 2000), the opposite is found. In a recent model intercomparison, Gregory et al. (2005) find that for all 11 models analysed, the MOC reduction is caused more by changes in surface heat flux than changes in surface freshwater flux. In addition, simulations using models of varying complexity (Stocker et al., 1992b; Saenko et al., 2003; Weaver et al., 2003) show that freshening or warming in the Southern Ocean acts to increase or stabilise the Atlantic MOC. This is likely a consequence of the complex coupling of Southern Ocean processes with North Atlantic Deep Water production.

A few simulations using coupled models are available that permit the assessment of the long-term stability of the MOC (Stouffer and Manabe, 1999; Voss and Mikolajewicz, 2001; Stouffer and Manabe, 2003; Wood et al., 2003; Yoshida et al., 2005; Bryan et al., 2006). Most of these simulations assume an idealised increase in atmospheric CO2 by 1% yr–1 to various levels ranging from two to four times pre-industrial levels. One study also considers slower increases (Stouffer and Manabe, 1999), or a reduction in CO2 (Stouffer and Manabe, 2003). The more recent models are not flux adjusted and have higher resolution (about 1.0°) (Yoshida et al., 2005; Bryan et al., 2006). A common feature of all simulations is a reduction in the MOC in response to the warming and a stabilisation or recovery of the MOC when the concentration is kept constant after achieving a level of two to four times the pre-industrial atmospheric CO2 concentration. None of these models shows a shutdown of the MOC that continues after the forcing is kept constant. But such a long-term shutdown cannot be excluded if the amount of warming and its rate exceed certain thresholds as shown using an EMIC (Stocker and Schmittner, 1997). Complete shut-downs, although not permanent, were also simulated by a flux-adjusted coupled model (Manabe and Stouffer, 1994; Stouffer and Manabe, 2003; see also Chan and Motoi, 2005). In none of these AOGCM simulations were the thresholds, as determined by the EMIC, passed (Stocker and Schmittner, 1997). As such, the long-term stability of the MOC found in the present AOGCM simulations is consistent with the results from the simpler models.

The reduction in MOC strength associated with increasing greenhouse gases represents a negative feedback for the warming in and around the North Atlantic. That is, through reducing the transport of heat from low to high latitudes, SSTs are cooler than they would otherwise be if the MOC was unchanged. As such, warming is reduced over and downstream of the North Atlantic. It is important to note that in models where the MOC weakens, warming still occurs downstream over Europe due to the overall dominant role of the radiative forcing associated with increasing greenhouse gases (Gregory et al., 2005). Many future projections show that once the radiative forcing is held fixed, re-establishment of the MOC occurs to a state similar to that of the present day. The partial or complete re-establishment of the MOC is slow and causes additional warming in and around the North Atlantic. While the oceanic meridional heat flux at low latitudes is reduced upon a slowdown of the MOC, many simulations show increasing meridional heat flux into the Arctic which contributes to accelerated warming and sea ice melting there. This is due to both the advection of warmer water and an intensification of the influx of North Atlantic water into the Arctic (A. Hu et al., 2004).

Climate models that simulated a complete shutdown of the MOC in response to sustained warming were flux-adjusted coupled GCMs or EMICs. A robust result from such simulations is that the shutdown of the MOC takes several centuries after the forcing is kept fixed (e.g., at 4 × atmospheric CO2 concentration). Besides the forcing amplitude and rate (Stocker and Schmittner, 1997), the amount of mixing in the ocean also appears to determine the stability of the MOC: increased vertical and horizontal mixing tends to stabilise the MOC and to eliminate the possibility of a second equilibrium state (Manabe and Stouffer, 1999; Knutti and Stocker, 2000; Longworth et al., 2005). Random internal variability or noise, often not present in simpler models, may also be important in determining the effective MOC stability (Knutti and Stocker, 2002; Monahan, 2002).

The MOC is not necessarily a comprehensive indicator of ocean circulation changes in response to global warming. In a transient 2 × atmospheric CO2 experiment using a coupled AOGCM, the MOC changes were small, but convection in the Labrador Sea stopped due to warmer and hence less dense waters that inflow from the Greenland-Iceland-Norwegian Sea (GIN Sea) (Wood et al., 1999; Stouffer et al., 2006a). Similar results were found by A. Hu et al. (2004), who also report an increase in convection in the GIN Sea due to the influx of more saline waters from the North Atlantic. Various simulations using coupled models of different complexity find significant reductions in convection in the GIN Sea in response to warming (Schaeffer et al., 2004; Bryan et al., 2006). Presumably, a delicate balance exists in the GIN Sea between the circum-arctic river runoff, sea ice production and advection of saline waters from the North Atlantic, and on a longer time scale, the inflow of freshwater through Bering Strait. The projected increases in circum-arctic river runoff (Wu et al., 2005) may enhance the tendency towards a reduction in GIN Sea convection (Stocker and Raible, 2005; Wu et al., 2005). Cessation of convection in the Labrador Sea in the next few decades is also simulated in a high-resolution model of the Atlantic Ocean driven by surface fluxes from two AOGCMs (Schweckendiek and Willebrand, 2005). The large-scale responses of the high-resolution ocean model (e.g., MOC, Labrador Seas) agree with those from the AOGCMs. The grid resolution of the ocean components in the coupled AOGCMs has significantly increased since the TAR, and some consistent patterns of changes in convection and water mass properties in the Atlantic Ocean emerge in response to the warming, but models still show a variety of responses in the details.

The best estimate of sea level from 1993 to 2003 (see Section associated with the slight net negative mass balance from Greenland is 0.1 to 0.3 mm yr–1 over the total ocean surface. This converts to only about 0.002 to 0.003 Sv of freshwater forcing. Such an amount, even when added directly and exclusively to the North Atlantic, has been suggested to be too small to affect the North Atlantic MOC (see Weaver and Hillaire-Marcel, 2004a). While one model exhibits a MOC weakening in the later part of the 21st century due to Greenland Ice Sheet melting (Fichefet et al., 2003), this same model had a very large downward drift of its overturning in the control climate, making it difficult to actually attribute the model MOC changes to the ice sheet melting. As noted in Section, Ridley et al. (2005) find the peak rate of Greenland Ice Sheet melting is about 0.1 Sv when they instantaneously elevate greenhouse gas levels in UKMO-HadCM3. They further note that this has little effect on the North Atlantic meridional overturning, although 0.1 Sv is sufficiently large to cause more dramatic transient changes in the strength of the MOC in other models (Stouffer et al., 2006b).

Taken together, it is very likely that the MOC, based on currently available simulations, will decrease, perhaps associated with a significant reduction in Labrador Sea Water formation, but very unlikely that the MOC will undergo an abrupt transition during the course of the 21st century. At this stage, it is too early to assess the likelihood of an abrupt change of the MOC beyond the end of the 21st century, but the possibility cannot be excluded (see Box 10.1). The few available simulations with models of different complexity instead suggest a centennial slowdown. Recovery of the MOC is simulated in some models if the radiative forcing is stabilised but would take several centuries; in other models, the reduction persists.

Box 10.1: Future Abrupt Climate Change, ‘Climate Surprises’, and Irreversible Changes

Theory, models and palaeoclimatic reconstructions (see Chapter 6) have established the fact that changes in the climate system can be abrupt and widespread. A working definition of ‘abrupt climate change’ is given in Alley et al. (2002): ‘Technically, an abrupt climate change occurs when the climate system is forced to cross some threshold, triggering a transition to a new state at a rate determined by the climate system itself and faster than the cause’. More generally, a gradual change in some determining quantity of the climate system (e.g., radiation balance, land surface properties, sea ice, etc.) can cause a variety of structurally different responses (Box 10.1, Figure 1). The response of a purely linear system scales with the forcing, and at stabilisation of the forcing, a new equilibrium is achieved which is structurally similar, but not necessarily close to the original state. However, if the system contains more than one equilibrium state, transitions to structurally different states are possible. Upon the crossing of a tipping point (bifurcation point), the evolution of the system is no longer controlled by the time scale of the forcing, but rather determined by its internal dynamics, which can either be much faster than the forcing, or significantly slower. Only the former case would be termed ‘abrupt climate change’, but the latter case is of equal importance. For the long-term evolution of a climate variable one must distinguish between reversible and irreversible changes. The notion of ‘climate surprises’ usually refers to abrupt transitions and temporary or permanent transitions to a different state in parts of the climate system such as, for example, the 8.2 kyr event (see Section

Atlantic Meridional Overturning Circulation and other ocean circulation changes:

The best-documented type of abrupt climate change in the palaeoclimatic archives is that associated with changes in the ocean circulation (Stocker, 2000). Since the TAR, many new results from climate models of different complexity have provided a more detailed view on the anticipated changes in the Atlantic MOC in response to global warming. Most models agree that the MOC weakens over the next 100 years and that this reduction ranges from indistinguishable from natural variability to over 50% by 2100 (Figure 10.15). None of the AOGCM simulations shows an abrupt change when forced with the SRES emissions scenarios until 2100, but some long-term model simulations suggest that a complete cessation can result for large forcings (Stouffer and Manabe, 2003). Models of intermediate complexity indicate that thresholds in the MOC may be present but that they depend on the amount and rate of warming for a given model (Stocker and Schmittner, 1997). The few long-term simulations from AOGCMs indicate that even complete shutdowns of the MOC may be reversible (Stouffer and Manabe, 2003; Yoshida et al., 2005; Stouffer et al., 2006b). However, until millennial simulations with AOGCMs are available, the important question of potential irreversibility of an MOC shutdown remains unanswered. Both simplified models and AOGCMs agree, however, that a potentially complete shut-down of the MOC, induced by global warming, would take many decades to more than a century. There is no direct model evidence that the MOC could collapse within a few decades in response to global warming. However, a few studies do show the potential for rapid changes in the MOC (Manabe and Stouffer, 1999), and the processes concerned are poorly understood (see Section 8.7). This is not inconsistent with the palaeoclimate records. The cooling events during the last ice ages registered in the Greenland ice cores developed over a couple of centuries to millennia. In contrast, there were also a number of very rapid warmings, the so-called Dansgaard-Oeschger events (NorthGRIP Members, 2004), or rapid cooling (LeGrande et al., 2006), which evolved over decades or less, most probably associated with rapid latitudinal shifts in ocean convection sites and changes in strength of the MOC (see Section 6.3.2).

Recent simulations with models with ocean components that resolve topography in sufficient detail obtain a consistent pattern of a strong to complete reduction of convection in the Labrador Sea (Wood et al., 1999; Schweckendiek and Willebrand, 2005). Such changes in the convection, with implications for the atmospheric circulation, can develop within a few years (Schaeffer et al., 2002). The long-term and regional-to-hemispheric scale effects of such changes in water mass properties have not yet been investigated.

With a reduction in the MOC, the meridional heat flux also decreases in the subtropical and mid-latitudes with large-scale effects on the atmospheric circulation. In consequence, the warming of the North Atlantic surface proceeds more slowly. Even for strong reductions in MOC towards the end of the 21st century, no cooling is observed in the regions around the North Atlantic because it is overcompensated by the radiative forcing that caused the ocean response in the first place.

At high latitudes, an increase in the oceanic meridional heat flux is simulated by these models. This increase is due to both an increase in the overturning circulation in the Arctic and the advection of warmer waters from lower latitudes and thus contributes significantly to continuing sea ice reduction in the Atlantic sector of the Arctic (A. Hu et al., 2004). Few simulations have also addressed the changes in overturning in the South Atlantic and Southern Ocean. In addition to water mass modifications, this also has an effect on the transport by the Antarctic Circumpolar Current, but results are not yet conclusive.

Current understanding of the processes responsible for the initiation of an ice age indicate that a reduction or collapse of the MOC in response to global warming could not start an ice age (Berger and Loutre, 2002; Crucifix and Loutre, 2002; Yoshimori et al., 2002; Weaver and Hillaire-Marcel, 2004b).

Box 10.1 Figure 1

Box 10.1, Figure 1. Schematic illustration of various responses of a climate variable to forcing. The forcing (top panels) reaches a new stable level (left part of figure), and later approaches the original level on very long time scales (right part of the figure). The response of the climate variable (bottom panels) can be smooth (solid line) or cross a tipping point inducing a transition to a structurally different state (dashed lines). That transition can be rapid (abrupt change, long-dashed), or gradual (short-dashed), but is usually dictated by the internal dynamics of the climate system rather than the forcing. The long-term behaviour (right part) also exhibits different possibilities. Changes can be irreversible (dash-dotted) with the system settling at a different stable state, or reversible (solid, dotted) when the forcing is set back to its original value. In the latter case, the transition again can be gradual or abrupt. An example for illustration, but not the only one, is the response of the Atlantic meridional overturning circulation to a gradual change in radiative forcing.

Arctic sea ice:

Arctic sea ice is responding sensitively to global warming. While changes in winter sea ice cover are moderate, late summer sea ice is projected to disappear almost completely towards the end of the 21st century. A number of positive feedbacks in the climate system accelerate the melt back of sea ice. The ice-albedo feedback allows open water to receive more heat from the Sun during summer, and the increase in ocean heat transport to the Arctic through the advection of warmer waters and stronger circulation further reduces ice cover. Minimum arctic sea ice cover is observed in September. Model simulations indicate that the September sea ice cover decreases substantially in response to global warming, generally evolving on the time scale of the warming. With sustained warming, the late summer disappearance of a major fraction of arctic sea ice is permanent.

Glaciers and ice caps:

Glaciers and ice caps are sensitive to changes in temperature and precipitation. Observations point to a reduction in volume over the last 20 years (see Section 4.5.2), with a rate during 1993 to 2003 corresponding to 0.77 ± 0.22 mm yr–1 sea level equivalent, with a larger mean central estimate than that for 1961 to 1998 (corresponding to 0.50 ± 0.18 mm yr–1 sea level equivalent). Rapid changes are therefore already underway and enhanced by positive feedbacks associated with the surface energy balance of shrinking glaciers and newly exposed land surface in periglacial areas. Acceleration of glacier loss over the next few decades is likely (see Section 10.6.3). Based on simulations of 11 glaciers in various regions, a volume loss of 60% of these glaciers is projected by the year 2050 (Schneeberger et al., 2003). Glaciated areas in the Americas are also affected. A comparative study including seven GCM simulations at 2 × atmospheric CO2 conditions inferred that many glaciers may disappear completely due to an increase in the equilibrium line altitude (Bradley et al., 2004). The disappearance of these ice bodies is much faster than a potential re-glaciation several centuries hence, and may in some areas be irreversible.

Greenland and West Antarctic Ice Sheets:

Satellite and in situ measurement networks have demonstrated increasing melting and accelerated ice flow around the periphery of the Greenland Ice Sheet (GIS) over the past 25 years (see Section 4.6.2). The few simulations of long-term ice sheet simulations suggest that the GIS will significantly decrease in volume and area over the coming centuries if a warmer climate is maintained (Gregory et al., 2004a; Huybrechts et al., 2004; Ridley et al., 2005). A threshold of annual mean warming of 1.9°C to 4.6°C in Greenland has been estimated for elimination of the GIS (Gregory and Huybrechts, 2006; see section, a process which would take many centuries to complete. Even if temperatures were to decrease later, the reduction of the GIS to a much smaller extent might be irreversible, because the climate of an ice-free Greenland could be too warm for accumulation; however, this result is model dependent (see Section The positive feedbacks involved here are that once the ice sheet gets thinner, temperatures in the accumulation region are higher, increasing the melting and causing more precipitation to fall as rain rather than snow; that the lower albedo of the exposed ice-free land causes a local climatic warming; and that surface melt water might accelerate ice flow (see Section

A collapse of the West Antarctic Ice Sheet (WAIS) has been discussed as a potential response to global warming for many years (Bindschadler, 1998; Oppenheimer, 1998; Vaughan, 2007). A complete collapse would cause a global sea level rise of about 5 m. The observed acceleration of ice streams in the Amundsen Sea sector of the WAIS, the rapidity of propagation of this signal upstream and the acceleration of glaciers that fed the Larsen B Ice Shelf after its collapse have renewed these concerns (see Section (continued)

It is possible that the presence of ice shelves tends to stabilise the ice sheet, at least regionally. Therefore, a weakening or collapse of ice shelves, caused by melting on the surface or by melting at the bottom by a warmer ocean, might contribute to a potential destabilisation of the WAIS, which could proceed through the positive feedback of grounding-line retreat. Present understanding is insufficient for prediction of the possible speed or extent of such a collapse (see Box 4.1 and Section

Vegetation cover:

Irreversible and relatively rapid changes in vegetation cover and composition have occurred frequently in the past. The most prominent example is the desertification of the Sahara region about 4 to 6 ka (Claussen et al., 1999). The reason for this behaviour is believed to lie in the limits of plant communities with respect to temperature and precipitation. Once critical levels are crossed, certain species can no longer compete within their ecosystem. Areas close to vegetation boundaries will experience particularly large and rapid changes due to the slow migration of these boundaries induced by global warming. A climate model simulation into the future shows that drying and warming in South America leads to a continuous reduction in the forest of Amazonia (Cox et al., 2000, 2004). While evolving continuously over the 21st century, such a change and ultimate disappearance could be irreversible, although this result could be model dependent since an analysis of 11 AOGCMs shows a wide range of future possible rainfall changes over the Amazon (Li et al., 2006).

One of the possible ‘climate surprises’ concerns the role of the soil in the global carbon cycle. As the concentration of CO2 is increasing, the soil is acting, in the global mean, as a carbon sink by assimilating carbon due to accelerated growth of the terrestrial biosphere (see also Section However, by about 2050, a model simulation suggests that the soil changes to a source of carbon by releasing previously accumulated carbon due to increased respiration (Cox et al., 2000) induced by increasing temperature and precipitation. This represents a positive feedback to the increase in atmospheric CO2. While different models agree regarding the sign of the feedback, large uncertainties exist regarding the strength (Cox et al., 2000; Dufresne et al., 2002; Friedlingstein et al., 2006). However, the respiration increase is caused by a warmer and wetter climate. The switch from moderate sink to strong source of atmospheric carbon is rather rapid and occurs within two decades (Cox et al., 2004), but the timing of the onset is uncertain (Huntingford et al., 2004). A model intercomparison reveals that once set in motion, the increase in respiration continues even after the CO2 levels are held constant (Cramer et al., 2001). Although considerable uncertainties still exist, it is clear that feedback mechanisms between the terrestrial biosphere and the physical climate system exist which can qualitatively and quantitatively alter the response to an increase in radiative forcing.

Atmospheric and ocean-atmosphere regimes:

Changes in weather patterns and regimes can be abrupt processes that might occur spontaneously due to dynamical interactions in the atmosphere-ice-ocean system, or manifest as the crossing of a threshold in the system due to slow external forcing. Such shifts have been reported in SST in the tropical Pacific, leading to a more positive ENSO phase (Trenberth, 1990), in the stratospheric polar vortex (Christiansen, 2003), in a shut-down of deep convection in the Greenland Sea (Bönisch et al., 1997; Ronski and Budeus, 2005) and in an abrupt freshening of the Labrador Sea (Dickson et al., 2002). In the latter, the freshening evolved throughout the entire depth but the shift in salinity was particularly rapid: the 34.87 psu isohaline plunged from seasonally surface to 1,600 metres within 2 years with no return since 1973.

In a long, unforced model simulation, a period of a few decades with anomalously cold temperatures (up to 10 standard deviations below average) in the region south of Greenland was found (Hall and Stouffer, 2001). It was caused by persistent winds that changed the stratification of the ocean and inhibited convection, thereby reducing heat transfer from the ocean to the atmosphere. Similar results were found in a different model in which the major convection site in the North Atlantic spontaneously switched to a more southerly location for several decades to centuries (Goosse et al., 2002). Other simulations show that the slowly increasing radiative forcing is able to cause transitions in the convective activity in the Greenland-Iceland-Norwegian Sea that have an influence on the atmospheric circulation over Greenland and Western Europe (Schaeffer et al., 2002). The changes unfold within a few years and indicate that the system has crossed a threshold.

A multi-model analysis of regimes of polar variability (NAO, Arctic and Antarctic Oscillations) reveals that the simulated trends in the 21st century influence the Arctic and Antarctic Oscillations and point towards more zonal circulation (Rauthe et al., 2004). Temperature changes associated with changes in atmospheric circulation regimes such as the NAO can exceed in certain regions (e.g., Northern Europe) the long-term global warming that causes such inter-decadal regime shifts (Dorn et al., 2003).