The Atlantic Meridional Overturning Circulation (AMOC) refers to a three-dimensional circulation present in the Atlantic (Figure 1.4.3A) whereby warm upper ocean waters (‘upper branch’) move northward from the tip of Southern Africa to the northern North Atlantic, where they cool, sink and return southwards as cold deep waters (‘lower branch’). The AMOC moves heat from the South to the North Atlantic, helping to maintain the mild climate of western and northern Europe. Thereby it shapes the climate of the whole Earth, influencing, for example, the 1-2°C temperature difference between the Northern and Southern hemispheres, and the location and strength of rainfall across all tropical regions (Buckley and Marshall, 2016; Feulner et al., 2013; Marshall et al., 2014).
Fresh, warm water is less dense than cold, salty water. In the future, surface waters in the northern North Atlantic may become less dense. This will make it harder for the water in that region to sink, which will disrupt the connection between the upper and lower branches of the AMOC, causing it to weaken significantly or even collapse completely. Therefore, we need to monitor the processes which tend to warm and freshen the upper ocean at high latitudes. AMOC strength has only been observed directly since 2004 (Srokosz and Bryden, 2013), with more uncertain reconstructions based on observations such as surface temperature, which extend back in time before 2004 (‘observational proxies’), or from palaeoclimate archives such as ocean sediment cores which extend back to prehistoric times (‘palaeoclimate proxies’) (Caesar et al,. 2018, 2021; Moffa-Sánchez et al., 2019). The lack of a sufficiently long observational record is a major issue for robust understanding of the AMOC.
The North Atlantic Ocean is freshening at subpolar latitudes (50-65°N), most strongly in the upper 100m, and warming, most strongly between 100-500 m water depth (IPCC, 2021). Both trends act to reduce AMOC strength. Greenland Ice Sheet melt is accelerating and releasing extra fresh water into the North Atlantic (Shepherd et al., 2020). In addition, Arctic sea ice is reducing in surface extent and thickness (Serreze and Meier, 2019) and overall Arctic river discharge is increasing (Druckenmiller et al., 2021), adding fresh water to the Arctic continental shelves and the high Arctic, and this riverine fresh water is potentially leaking into the North Atlantic from the Arctic. The North Atlantic is a region of high variability on interannual to decadal timescales (Boer 2000) and therefore subject to substantial climatic perturbations with the potential to trigger any underlying instability if a tipping point is approached.
Limited direct observations of AMOC strength make current trends uncertain, but there are some signs of ongoing weakening. Observational and palaeoclimate proxies suggest the AMOC may have weakened by around 15 per cent over the past 50 years (Caesar et al,. 2018) and may be at its weakest in 1,000 years (Caesar et al., 2021). However, the proxy data used in these studies have large uncertainties, and some other reconstructions show little evidence of decline (Moffa-Sanchez et al., 2019, Kilbourne et al., 2022). It is therefore difficult to confidently discern potential recent trends from natural variability, due to disagreement between published studies (Bonnet et al,. 2021; Latif et al., 2022; versus Qasmi, 2022).
The IPCC’s most recent assessment is that the AMOC has weakened relative to 1850-1900, but with low confidence due to disagreement among reconstructions (Moffa-Sanchez et al., 2019; Kilbourne et al., 2022) and models (Fox-Kemper et al., 2021). For the future, the IPCC projects that it is very likely that the AMOC will decline in the 21st century (however with low confidence on timings and magnitude) (Figure 1.4.4A).
The AMOC has been proposed as a ‘global core’ tipping system of the climate system with medium confidence by Armstrong McKay et al,. (2022). Palaeorecords indicate it has abruptly switched between stronger and weaker modes during recent glacial cycles (Figure 1.4.5). Most of the time (including the warm Holocene of the past 12,000 years) the AMOC is in a strong, warm mode, but during peak glacials it sometimes shifted to a weak, cold mode instead (Böhm et al., 2014). It also occasionally collapsed entirely to an ‘off’ mode during ‘Heinrich’ events, in which iceberg outbursts from the North American Laurentide Ice Sheet temporarily blocked Atlantic overturning for several centuries.
In two previous censuses of climate model projections, a shut-down of the AMOC over many decades was observed in a small minority of simulations (Drijfhout et al., 2015; Sgubin et al., 2017). This shut-down was preceded by decreases in subpolar surface air and ocean temperature and increased sea ice cover. Ultimately, deep mixing ceased to occur, destroying the connection between the surface and the deep ocean. There are, however, concerns that the AMOC may be too stable in CMIP-type climate models (Mecking et al., 2017, Liu et al., 2017), which suggests the CMIP multimodel ensembles may underestimate the likelihood of AMOC collapse (Fox-Kemper et al., 2021).
Some recent studies have suggested that ‘early warning signals’ indicating destabilisation (see Chapter 1.6) can be detected in reconstructed ‘fingerprints’ of AMOC strength over the 20th Century (Boers, 2021), and if a tipping point is assumed then the collapse threshold could be reached during the 21st Century (Ditlevsen & Ditlevsen, 2023). These studies used observational proxies for temperature and salinity from the Northeastern subpolar North Atlantic, which are used as indirect AMOC fingerprints rather than direct measurements of AMOC strength. This gives long enough data to analyse for early warning signals, but using indirect proxies adds uncertainty. The model used by Ditlevsen & Ditlevsen (2023) to project collapse is also highly simplified with a tipping point assumed, and does not take into account the low-frequency variability of the AMOC, nor the presence of external forcings such as increasing greenhouse gases. So while signals in this dataset are consistent with approaching a tipping point, there are substantial uncertainties with this methodology (see also Michel et al., 2023 highlighting potential false warnings). Further potential early warning signals have been found from analysis of Northern Hemisphere palaeoproxies (Michel et al., 2022). Despite the caveats mentioned above, these results amount to a serious warning that the AMOC might be en route to tipping. However, the claim that we might expect tipping in a few decades is – in the view of the present authors – not substantiated enough.
AMOC stability is strongly linked to the ‘salt-advection feedback’ (Stommel, 1961, see Figure 1.4.3B). The AMOC imports salt into the Atlantic and transports it from the South Atlantic to the northern North Atlantic. If the AMOC weakens then less salt is transported to the northern North Atlantic, the surface waters freshen, which inhibits sinking, and the AMOC weakens further. The AMOC collapses seen in models (Drijfhout et al., 2015; Sgubin et al., 2017) were driven by this salt-advection feedback. However, the strength of this feedback, and the timescale over which it operates are governed by processes whose effects are quite uncertain. Although Figure 1.4.3A shows typical pathways of surface and deep water through the Atlantic, these are an average picture over many decades. Individual water parcels may get caught up in basin-scale surface or deep recirculations, smaller-scale eddies and meandering currents. There is no definitive evidence though from models or observations that these systematically impact the salt advection feedback.
Additionally, changes in the AMOC have other impacts on salinity – for instance through affecting evaporation and precipitation patterns (Jackson, 2013, Weijer et al, 2019). These other feedbacks can temporarily mask, and may even overcome, the salt advection feedback, potentially changing the stability of the AMOC (Jackson, 2013; Gent, 2018). It is difficult to characterise these processes and feedbacks from observations alone due to insufficient data coverage both in time and space, so we are dependent on numerical models. However, many studies have used reduced complexity models, which may not capture all the potential feedbacks, and even the current generation of climate models have quite low spatial resolution and do not well characterise narrow currents, eddies and processes such as horizontal and vertical mixing (Swingedouw et al,. 2022).
Armstrong McKay et al (2022) estimated with low confidence a global warming threshold for AMOC collapse of ~4°C (1.4-8°C). In our view, the range is a better indication of the uncertainty in the different model responses rather than a relationship to global warming, as the likelihood is probably less dependent on temperature, but strongly depends on salinity changes and the strength of opposing feedbacks on the freshwater budget. Studies with climate models have found that adding freshwater can cause the AMOC to collapse and not recover in some models. Since many climate models might be biased towards stability, however, these studies use an unphysically large amount of freshwater to explore the sensitivity (Jackson et al., 2023). Although adding freshwater causes a collapse, they show the threshold is dependent on the strength of the AMOC and deep convection, rather than on the amount of freshwater added (Jackson and Wood 2018; Jackson et al, 2023). AMOC collapse may also be more sensitive to the rate of freshwater forcing than the total magnitude (Lohmann & Ditlevsen, 2021).
Hysteresis and bistability both refer to systems which can adopt one of two or more states for the same external forcing, such as CO2 concentration (see Figure 1.4.4B & Glossary; Boucher et al., 2012). Commonly, this is explored by approaching the same external conditions with different trajectories in model simulations, e.g. increasing and reversing the forcing to study reversibility. Bistability involving a full collapse of the AMOC by artificially flooding the North Atlantic with freshwater has been demonstrated (or strongly implied) in theoretical models (Stommel, 1961) and climate models of reduced complexity (Rahmstorf et al., 2005; Hawkins et al., 2011). These types of numerical experiments study bistability through forcings that change slowly enough for the system to equilibrate, typically requiring long simulations and thus coarse model resolution for reasonable computational performance. In more complex models it is not possible to conduct experiments for long enough to demonstrate bistability or hysteresis, however weak states have been shown to be stable for at least 100 years in about half of a test group of CMIP6-type models (Jackson et al., 2023) and in a high-resolution ocean-atmosphere coupled climate model (Mecking et al., 2016). A recent study finds AMOC tipping in a CMIP-type model in response to gradually increasing freshwater release in the North Atlantic (Van Westen et al., 2023). AMOC bistability is model-dependent though, controlled by the balance of the positive and negative feedbacks that determine the salinity of the subpolar North Atlantic. It is not yet understood why the bistability occurs in some models and not others (Jackson et al., 2023). However, as previously mentioned, there is evidence that the present generation of climate models is too stable due to model biases in the distribution of ocean salinity (Liu et al., 2017; Mecking et al 2017).
Not only is it difficult to prove system bistability, the complexity of the system and interaction with multiple drivers make it hard to assess collapse thresholds. It may be that realistic freshwater input is not sufficient to cause the transition, or that changing CO2 alters the underlying system stability, thus increasing the critical freshwater threshold (Wood et al., 2019). Nevertheless, overshoot scenarios, where the CO2 trend is assumed to reverse at some point in the future, provide some useful information about reversibility of the AMOC on human timescales. Figure 1.4.4B shows how the AMOC changes in the UKESM climate model under an overshoot emission scenario exceeding and returning to 500 ppm. Even if CO2 concentrations return to 500 ppm by 2100 the AMOC is still only 77 per cent of the strength it was in 2050 also at 500 ppm. Although the AMOC does not collapse in this model, it seems unlikely that it will recover its former strength on human timescales.
The timescale of AMOC tipping was estimated by Armstrong McKay et al (2022) to be 15-300 years, however this range is very dependent on the strength of the freshwater forcing applied in experiments, which in many cases is unrealistically large as compared to projected melting of the Greenland Ice Sheet and increase in precipitation and river runoff. Moreover, the assessment is also potentially impacted by the models being unrealistically stable. With a realistic forcing scenario, the timescale will depend on the feedbacks. A basin-wide salt advection feedback may have a century timescale, while if it is preceded by a local North Atlantic salt advection feedback it may be reduced to a few decades. Even faster timescales are possible when deep mixing is capped off by sudden increases in sea ice cover in all convective regions (Rahmstorf et al., 2001; Kuhlbrodt et al., 2001).
AMOC collapse would lead to cooling over most of the Northern Hemisphere, particularly strong (up to 10°C relative to preindustrial) over Western and Northern Europe. In addition, a southward shift of the Intertropical Convergence Zone would occur, impacting monsoon systems globally and causing large changes in storminess and rainfall patterns (Jackson et al, 2015). A collapse of the AMOC would influence sea level rise along the boundaries of the North Atlantic, modify Arctic sea ice and permafrost distribution (Schwinger et al., 2022; Bulgin et al, 2023), reduce oceanic carbon uptake (Rhein et al., 2017) and potentially lead to ocean deoxygenation (Kwiatkowski et al., 2020) and severe disruption of marine ecosystems (including changes in the North Atlantic Subpolar Gyre, see below), impacting North Atlantic fish stocks. See Chapter 2.2 for more discussion on impacts.
Although the AMOC does not always behave like a tipping system in many ocean/climate models, palaeoceanographic evidence strongly points to its capability for tipping or at least to shift to another state that can be quasi-stable for many centuries (Figure 1.4.5). Tipping is also suggested in a recent study of several CMIP6 models (Jackson et al, 2023) and in another study which found that removing model salinity biases strongly increased the likelihood of tipping (Liu et al., 2017). This does not necessarily mean that tipping is likely in a future climate, since some of these scenarios specified unrealistic inputs of freshwater or GHG emissions. Nonetheless, although the likelihood for collapse is considered small compared to the likelihood of AMOC decline, the potential impacts of AMOC tipping make it an important risk to consider in framing mitigation targets, for instance.
The latest AR6 assessment states that we have only medium confidence that an AMOC collapse will not happen before 2100 (Fox-Kemper et al., 2021). This uncertainty is due to models having strong ocean salinity biases, absence of meltwater release from the Greenland Ice Sheet in climate change scenarios, and the possible impact of eddies and other unresolved ocean processes on freshwater pathways. However, a recent study with the PAGES2K database of climate reconstructions of the past 2,000 years suggests, using statistical methods based on dynamical systems theory, that we may be close to an AMOC tipping point (Michel et al., 2022), as do the studies of Boers (2021) and Ditlevsen and Ditlevsen (2023) cited above. AR6 also concluded that reported recent weakening in both historical model simulations and observation-based reconstructions of the AMOC have low confidence. Direct AMOC observations have not been made for long enough to separate a long-term weakening from short-term variability. Another recent study suggests that we will need to wait until at least 2028 to obtain a robust statistical signal of AMOC weakening (Lobelle et al,. 2020). Thus, the coming years will be crucial for detection of an AMOC weakening potentially leading to longer-term instability.
There are substantial uncertainties around how the AMOC evolves over long timescales, because of a lack of direct observations. More palaeo-reconstructions of AMOC strength, ocean surface temperature, and other AMOC-related properties with high temporal resolution, using appropriate proxies and careful chronological control performed for key past periods (e.g. last millennium, millennial-scale climate change events, previous interglacials), hold great potential to improve our understanding about the AMOC as a tipping point. Other open issues are to: (i) reconcile disagreements between palaeo-reconstructions and model simulations, and (ii) develop improved metrics for creating historical reconstructions and monitoring the AMOC.
Current climate models suffer from imperfect representation of some important processes (such as eddies and mixing) and from biases which can impact the AMOC response to forcings. Hence we need to assess how important these issues are for representing AMOC stability, in particular, to understand how different feedbacks vary across models and are affected by modelling deficiencies. Given these issues, a robust assessment of the likelihood of an AMOC collapse is difficult, but based on the evidence presented, we assess that the AMOC features tipping dynamics with medium confidence. One potential way forward, given these uncertainties, is in developing observable precursors to a collapse that could be monitored.
The North Atlantic Subpolar Gyre (SPG) is an oceanic cyclonic (counter-clockwise in the northern hemisphere) flow to the south of Greenland (Figure 1.4.6). It is linked to a site of deep ocean convection in the Labrador-Irminger Seas, i.e. sinking of the subsurface ocean waters to great depths, contributing to the AMOC (Figures 1.4.3, 1.4.6-7).
There are indications for change in the SPG, as observations show that Labrador Sea Water (LSW) formed during oceanic deep convection events after 2014 was less dense than the LSW formed between 1987 and 1994 (Yashayaev and Loder, 2016), potentially influencing the AMOC. Moreover, the observed ‘warming hole’ over the North Atlantic can be explained by AMOC slowdown (Drijfhout et al., 2012; Caesar et al., 2018; also see AMOC above) and has also been linked to SPG weakening in CMIP6 models (Sgubin et al 2017; Swingedouw et al 2021). In these models, a collapse of the oceanic convection causes a localised North Atlantic regional surface air temperature drop of ~2-3°C. This cooling moderates warming over north-west Europe and eastern Canada in global warming scenarios, although it is smaller and less widespread than that associated with AMOC collapse.
A northward-shift of the atmospheric jet stream, which is predicted to take place with SPG weakening, means more weather extremes in Europe (which may be linked to the unusual cooling and heat waves in recent years) (Osman et al., 2021) and southward shift of the intertropical convergence zone (ITCZ, see Figure 1.4.1) (Sgubin et al,. 2017; Swingedouw et al,. 2021). Models suggest potential impact of the SPG collapse on the European weather, precipitation regime and climate (Swingedouw et al,., 2021) (Figure 1.4.7I,J). These changes in the physical system may trigger changes in the ecosystems with detrimental consequences for the North Atlantic spring bloom and overall Atlantic marine primary productivity. Neither of these return to the preindustrial state even if emissions reverse by 2100 in models (Yool et al., 2015; Heinze et al., 2023). This would impose a strong impact on fisheries and biodiversity, with wide societal implications (see Section 2). Last but not least, a transition between two SPG stable states has been suggested to explain the onset of the so-called ‘Little Ice Age’ in which colder conditions prevailed in Europe during the 16th-19th centuries (Lehner et al,. 2013; Michel et al. 2022).
Ventilation of LSW is accompanied by an uptake of oxygen. Starting in 2014, the convection in the Labrador Sea became more intense and reached depths of 1,500m and below. Consequently, oxygen in LSW is in general increased, but this increase did not penetrate the densest part of this water mass (Rhein et al., 2017). The oxygen concentrations in the deepest part of the LSW (around 2,000m) have decreased in the formation region and along the main export pathways (southward and eastward crossing the Mid-Atlantic Ridge) for more than 20 years. Most of the oxygen from the export of newly formed LSW has been consumed north of the equator (Koelling et al., 2022), and the long-term oxygen decline along the southward LSW pathway might have impacts on ecosystems in the tropics and subtropics over longer timescales (e.g., Heinze et al., 2023).
The potential shutting-down of winter convection in the Labrador Sea (see Figure 1.4.7A,B and Swingedouw et al., 2021) will also stop the production of Labrador Slope Water (LSLW). This water is next to the Labrador Sea continental slope and is lighter and less deep than LSW. It contributes to AMOC and the Gulf Stream and can influence variability of the Atlantic climate system overall (New et al., 2021). The LSLW is rich in nutrients and oxygen too, thereby affecting the ecosystems on the North American continental shelf and shelf slope (e.g. Claret et al., 2018) and might affect tropical and subtropical marine ecosystems on a timescale of several decades. Furthermore, the SPG takes up large amounts of atmospheric carbon and exports it to the deep ocean (Henson et al., 2022).
Shallowing of the SPG (Sgubin et al., 2017; Swingedouw et al., 2021) would directly increase regional CO2 uptake but negatively impact marine biology, for instance threatening the habitat of cold-water corals in the area due to higher acidity with more CO2 dissolved in the water (Fröb et al., 2019; Fontela et al., 2020; García-Ibáñez et al., 2021). Weakening or collapse of the SPG would reduce the amount of carbon-depleted intermediate water being upwelled and newly carbon-enriched water being convected, reducing export of anthropogenic CO2 to the deep ocean (Halloran et al. 2015; Ridge & McKinley 2021), which in turn might lead to an increase of atmospheric CO2 concentration in the long term (Schmittner et al. 2007). Declining SPG strength may also be reducing the currently high phytoplankton productivity in this area (Osman et al., 2019; Henson et al. 2022), reducing the amount of biologically fixed carbon to deeper water too.
Changes in the overall Atlantic ocean circulation (AMOC and SPG) can impact the spread of Atlantic water into the Arctic and affect marine ecosystems there. Summer sea ice decline reduces light limitation, rendering Arctic ecosystems more similar to the present North Atlantic (Yool et al., 2015). Increased seasonal phytoplankton blooms will deplete nutrients in the ocean, but increased inputs from rivers and coastal erosion can alleviate this, with Arctic primary production (i.e. the turnover photosynthesising plankton biomass) projected to increase by about 30-50 per cent in this century. Invasive species can also extend further into the Arctic habitat due to warming and current changes, e.g. in the Barents Sea and from the Pacific (Kelly et al., 2020; Neukermans et al., 2018; Oziel et al., 2020; Terhaar et al., 2021) (see also Chapter 1.3).
In the North Atlantic, the AMOC can be defined as north-going warm ‘limb’ and saline upper waters and south-going, colder, denser deep water ‘limb’ (Frajka-Williams et al. 2019). In contrast, in the Subpolar North Atlantic and the SPG, the AMOC features a third ‘limb’ of a cold, fresh western boundary current with the origin in the Arctic Ocean and Nordic Seas (Bacon et al., 2003). This is likely linked with the deep convection and winter oceanic mixing in the Labrador, Irminger and Iceland seas, injecting waters into the deep, southward-flowing limb of the AMOC (Bower et al,. 2019). Changes in SPG circulation are associated with the shallowing of the oceanic mixed layer and convection (Figure 1.4.7A,B) in the SPG and link the predicted future weakening of the North Atlantic subtropical gyre and a strengthening of the Nordic Seas gyre, pointing to the influences of the upstream changes in the Arctic on the North Atlantic (Swingedouw et al., 2021).
Potential convection instability in the Labrador and Irminger Seas and the wider SPG is believed to be linked to lightening of the upper ocean waters due to reduced salinity (e,g., due to increased precipitation, Figure 1.4.7d), thus increasing ‘stratification’ – i.e. reduced mixing between layers of the water column. Warming (Figure 1.4.7c) also plays a role and could contribute to convection collapse (Armstrong McKay et al., 2022). Freshening and warming make surface waters more buoyant and thus harder to sink, which, beyond a threshold, can abruptly propel a self-sustained convection collapse (Drijfhout et al,. 2015; Sgubin et al., 2017). This process can result in two alternative stable SPG states (Levermann and Born, 2007), with or without deep convection (Armstrong McKay et al., 2022). Similar to the AMOC, SPG stability is also strongly linked to the salt-advection feedback. When the SPG is ‘on’, it brings dense salty waters from the North Atlantic drift into the Irminger and Labrador Seas, allowing deep sinking and convection to occur (Born & Stocker, 2014; Born et al., 2016). When convection decreases due to stratification, the SPG weakens, less salty North Atlantic water flows eastwards, and the convection is further weakened, which eventually leads to convection collapse in some models. SPG collapse leads to cooling across the SPG region, and so will impact marine biology and bordering regions.
A freshwater anomaly is currently building up in the Beaufort gyre – a pile-up of fresh water at the surface of the Beaufort Sea in the Arctic – due to increased input from rivers, sea ice and snow melting as well as the prevailing clockwise (anticyclonic in the northern hemisphere) winds over the sea (Haine, et al., 2015; Regan et al., 2019; Kelly et al., 2020). There is a considerable risk that this freshwater excess might flush into the SPG, disrupting the AMOC (Zhang et al,. 2021). The most recent changes in Beaufort gyre size and circulation (Lin et al,. 2023) suggest flushing might occur very soon or has already started. The SPG system has recently experienced its largest freshening for the last 120 years in its eastern side due to changes in the atmospheric circulation (Spooner et al,. 2020). In contrast, so far there is only limited evidence of Arctic freshwater fluxes impacting freshwater accumulation in the Labrador Sea (Florindo-Lopez et al., 2020). An increased freshwater input into SPG water mass formation regions from melting of Greenland’s glaciers can also inhibit deep water formation and reduce the SPG and AMOC (Dukhovskoy et al., 2021).
Although SPG changes are apparently linked to the AMOC, SPG collapse can occur much faster than AMOC collapse, on the timescale of only a few decades (Armstrong McKay et al., 2022). Armstrong McKay et al. (2022) estimated global warming threshold of ~1.8°C (1.1 to 3.8°C) for the SPG collapse (high confidence) based on climate models from CMIP5 and CMIP6. Abrupt future SPG collapse is diverse in the CMIP6 models, occurring as early as the 2040s (~1 to 2°C) but in only a subset of models. However, as these models better represent some key processes, the chance of SPG collapse is estimated at 36-44 per cent (Sgubin et al., 2017; Swingedouw et al., 2021).
Similar to Armstrong McKay et al,. (2022), the SPG is classified as a tipping system with medium confidence. A global warming threshold for tipping that could be passed within the next few decades, and an estimated tipping timescale of years to a few decades, raise reasons for concern. Furthermore, cessation of deep water production from other sources in the Labrador and Nordic Seas and the Arctic could also present other potential tipping points in the future North Atlantic (Sgubin et al., 2017).