Editing IPCC:AR6/SROCC/Chapter-6 (section)

=== 6.7.1 Key Processes and Feedbacks, Observations, Detection and Attribution, Projections ===

<div id="section-6-7-1-1observational-and-model-understanding-of-atlantic-ocean-circulation-changes"></div>

<span id="observational-and-model-understanding-of-atlantic-ocean-circulation-changes"></span>
==== 6.7.1.1 Observational and Model Understanding of Atlantic Ocean Circulation Changes ====

<div id="section-6-7-1-1observational-and-model-understanding-of-atlantic-ocean-circulation-changes-block-1"></div>

Palaeo-reconstructions indicate that the North Atlantic is a region where rapid climatic variations can occur (IPCC, 2013 <sup>[[#fn:r668|668]]</sup> ). Deep waters formed in the northern North Atlantic induces a large-scale AMOC which transports large amounts of heat northward across the hemispheres, explaining part of the difference in temperature between the two hemispheres, as well as the northward location of the ITCZ (e.g., Buckley and Marshall, 2016). This circulation system is believed to be a key tipping point of the Earth’s climate system (IPCC, 2013 <sup>[[#fn:r669|669]]</sup> ).

Considerable effort has been dedicated in the last decades to improve the observation system of the large-scale ocean circulation (e.g., Argo and its array of about 3,800 free-drifting profiling floats), including the AMOC through dedicated large-scale observing arrays (at 16°N (Send et al., 2011) and 26°N (McCarthy et al., 2015b <sup>[[#fn:r670|670]]</sup> ), in the subpolar gyre (SPG) (Lozier et al., 2017 <sup>[[#fn:r671|671]]</sup> ), between Portugal and the tip of Greenland (Mercier et al., 2015 <sup>[[#fn:r672|672]]</sup> ), at 34.5°S (Meinen et al., 2013 <sup>[[#fn:r673|673]]</sup> ), among others). The strength of the AMOC at 26°N has been continuously estimated since 2004 with an annual mean estimate of 17 ± 1.9 x 10 6 m 3 s –1 over the 2004–2017 period (Smeed et al., 2018 <sup>[[#fn:r674|674]]</sup> ). The AMOC at 26°N has been 2.7 x 10 6 m 3 s –1 weaker in 2008-2017 than in the first four years of measurement (Smeed et al., 2018 <sup>[[#fn:r675|675]]</sup> ). However, the record is not yet long enough to determine if there is a long-term decline of the AMOC. McCarthy et al. (2012) reported a 30% reduction in the AMOC in 2009–2010, followed by a weaker minimum a year later. Analysis of forced ocean models suggests such events may occur once every two or three decades (Blaker et al., 2015 <sup>[[#fn:r676|676]]</sup> ). At 34.5°S, the mean AMOC is estimated as 14.7 ± 8.3 x 10 6 m 3 s –1 over the period 2009–2017 (Meinen et al., 2018 <sup>[[#fn:r677|677]]</sup> ) also with large interannual variability, while no trend has been identified at this latitude. Estimates based on ocean reanalyses show considerable diversity in their AMOC mean state, and its evolution over the last 50 years (Karspeck et al., 2017 <sup>[[#fn:r678|678]]</sup> ; Menary and Hermanson, 2018 <sup>[[#fn:r679|679]]</sup> ), because only very few deep ocean observations before the Argo era, starting around 2004, are available. During the Argo era, the reanalyses agree better with each other (Jackson et al., 2016 <sup>[[#fn:r680|680]]</sup> ).

During the last interglacial warm period, palaeo-data suggest that the AMOC may have been weaker (Govin et al., 2012 <sup>[[#fn:r681|681]]</sup> ) and also show proxy record evidences of instabilities (Galaasen et al., 2014 <sup>[[#fn:r682|682]]</sup> ). Based on an AMOC reconstruction using SST fingerprints, it has been suggested that the AMOC may have experienced around 3 ± 1 x 10 6 m 3 s –1 of weakening (about 15% decrease) since the mid-20th century (Caesar et al., 2018 <sup>[[#fn:r683|683]]</sup> ). Such a trend in AMOC was also suspected in a former study using Principal Component Analysis of SST (Dima and Lohmann, 2010 <sup>[[#fn:r684|684]]</sup> ). Palaeo-proxies also highlight that the historical era may exhibit an unprecedented low AMOC over the last 1,600 years (Sherwood et al., 2011 <sup>[[#fn:r685|685]]</sup> ; Rahmstorf et al., 2015 <sup>[[#fn:r686|686]]</sup> ; Thibodeau et al., 2018 <sup>[[#fn:r687|687]]</sup> ; Thornalley et al., 2018 <sup>[[#fn:r688|688]]</sup> ). Nevertheless, these proxy records are indirect measurements of the AMOC so that considerable uncertainty remains concerning these results. Moreover, the exact mechanisms to explain such a long-term weakening are not fully understood and some reconstructions show a weakening starting very early in the historical era, when the level of anthropogenic perturbation and warming was very low. Climate model simulations (Figure 6.8) do show a weakening over the historical era, but this weakening is mainly occurring over the recent decades. Climate projections exhibit a weakening of around 1.4 ± 1.4 x 10 6 m 3 s –1 for present day (2006–2015) minus pre-industrial (1850–1900), highlighting that anthropogenic warming may have already forced an AMOC weakening. Nevertheless, no proper detection and attribution of the on-going changes has been led so far due to still limited observational evidences. Thus, we conclude that there is ''medium confidence'' that the AMOC has weakened over the historical era but there is insufficient evidence to quantify a ''likely'' range of the magnitude of the change.

Examination of 14 models from the CMIP5 archive, which do not take into account the melting (either from runoff, basal melting or icebergs) from the GIS (cf. Section 6.7.1.2), led to the assessment that the AMOC is ''very unlikely'' to collapse in the 21st century in response to increasing GHG concentrations (IPCC, 2013 <sup>[[#fn:r689|689]]</sup> ). Nonetheless, the CMIP5 models agree that a weakening of the AMOC into the 21st century will lead to localised cooling (relative to the global mean) centred in the North Atlantic SPG (Menary and Wood, 2018 <sup>[[#fn:r690|690]]</sup> ), although the precise location as well as the extension of this cooling patch, notably towards Europe, remains uncertain (Sgubin et al., 2017 <sup>[[#fn:r691|691]]</sup> ; Menary and Wood, 2018 <sup>[[#fn:r692|692]]</sup> ).

Abrupt variations in SST or sea ice cover have been found in 19 out of the 40 models of the CMIP5 archive (Drijfhout et al., 2015 <sup>[[#fn:r693|693]]</sup> ). Large cooling trends, which can occur in a decade, are found in the subpolar North Atlantic in 9 out of 40 models. Results show that the heat transport in the AMOC plays a role in explaining such a rapid cooling, but other processes are also key for setting the rapid (decadal-scale) timeframe of SPG cooling, notably vertical heat transport in the ocean and interactions with sea ice and the atmosphere (Sgubin et al., 2017 <sup>[[#fn:r694|694]]</sup> ). Using the representation of stratification as an emergent constraint, rapid changes in subpolar convection and associated cooling are occurring in the 21st century in 5 of the 11 best models (Sgubin et al., 2017 <sup>[[#fn:r695|695]]</sup> ). The poor representation of ocean deep convection in most CMIP5 models has been confirmed in Heuze (2017), which can notably limit a key feedback mechanism related with warm summer in the North Atlantic and its impact on oceanic convection in winter (Oltmanns et al., 2018 <sup>[[#fn:r697|697]]</sup> ). Thus, there is ''low confidence'' in the projections of SPG fate. Increasing the horizontal resolution of the ocean in next generation climate models might be a way to increase confidence in ocean convection future changes.

The SPG dynamical system has been identified as a tipping element of the climate system (Mengel et al., 2012 <sup>[[#fn:r699|699]]</sup> ; Born et al., 2013 <sup>[[#fn:r700|700]]</sup> ). If this element reaches its tipping point, the SPG circulation can change very abruptly between different stable steady states, due to positive feedback between convective activity and salinity transport within the gyre (Born et al., 2016). It has been argued that a transition between two SPG stable states can explain the onset of the Little Ice Age that may have occurred around the 14–15th century (Lehner et al., 2013 <sup>[[#fn:r701|701]]</sup> ; Schleussner et al., 2015 <sup>[[#fn:r702|702]]</sup> ; Moreno-Chamarro et al., 2017 <sup>[[#fn:r703|703]]</sup> ) possibly triggered by large volcanic eruption (Schleussner and Feulner, 2013 <sup>[[#fn:r704|704]]</sup> ). Furthermore a few CMIP5 climate models also showed a rapid cooling in the SPG within the 1970s cooling events, as a nonlinear response to aerosols (Bellucci et al., 2017 <sup>[[#fn:r705|705]]</sup> ). The SPG therefore appears as a tipping element in the climate system, with a faster (decade) response than the AMOC (century), but with lower induced SST cooling. Thus, the SPG system can cross a threshold in climate projections when surface water in the subpolar becomes lighter due to increase in temperature and decrease in salinity related with changes in radiative forcing (Sgubin et al., 2017 <sup>[[#fn:r706|706]]</sup> ).

<div id="section-6-7-1-1observational-and-model-understanding-of-atlantic-ocean-circulation-changes-block-2"></div>

<span id="figure-6.8"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 6.8'''

<span id="figure-6.8-atlantic-meridional-overturning-circulation-amoc-changes-at-26on-as-simulated-by-27-models-only-14-were-shown-in-the-ipcc-5th-assessment-report-ar5-ipcc-2013.-the-dotted-line-shows-the-observation-based-estimate-at-26on-mccarthy-et-al.-2015b-and-the-thick-greybluered-lines-the-multi-model-ensemble-mean.-values-of-amoc-maximum"></span>
<!-- IMG CAPTION -->
'''Figure 6.8 | Atlantic Meridional Overturning Circulation (AMOC) changes at 26oN as simulated by 27 models (only 14 were shown in the IPCC 5th Assessment Report (AR5); IPCC, 2013). The dotted line shows the observation-based estimate at 26oN (McCarthy et al. 2015b) and the thick grey/blue/red lines the multi-model ensemble mean. Values of AMOC maximum […]'''
<!-- IMG FILE -->
[[File:f23977ff8e3f6de41f0463557bc54805 IPCC-SROCC-CH_6_8.jpg]]

Figure 6.8 | Atlantic Meridional Overturning Circulation (AMOC) changes at 26oN as simulated by 27 models (only 14 were shown in the IPCC 5th Assessment Report (AR5); IPCC, 2013). The dotted line shows the observation-based estimate at 26oN (McCarthy et al. 2015b) and the thick grey/blue/red lines the multi-model ensemble mean. Values of AMOC maximum at 26oN (in units 106 m3 s–1) are shown in historical simulations (most of the time 1850–2005) followed for 2006–2100 by a) Representative Concentration Pathway (RCP)2.6 simulations and b) RCP8.5 simulations. In c) and d), the time series show the AMOC strength relative to the value during 2006–2015, a period over which observations are available. c) shows historical followed by RCP2.6 simulations and d) shows historical followed by RCP8.5 simulations. The 66% and 100% ranges of all-available CMIP5 simulations are shown in grey for historical, blue for RCP2.6 scenario and red for RCP8.5 scenario.

<!-- END IMG -->
<div id="section-6-7-1-1observational-and-model-understanding-of-atlantic-ocean-circulation-changes-block-3"></div>

Evaluation of AMOC variations in the CMIP5 database has been further analysed in this report (Figure 6.8) using almost twice as many models as in AR5 (IPCC, 2013 <sup>[[#fn:r707|707]]</sup> ). The AR5 assessment of a ''very unlikely'' AMOC collapse has been confirmed, although one model (FGOALS-s2) does show such a collapse (e.g., decrease larger than 80% relative to present day) before the end of the century for RCP8.5 scenario (Figure 6.8). Now based on up to 27 model simulations, the decrease of the AMOC is assessed to be of -2.1 '''±''' 2.6 x 10 6 m 3 s –1 (-11 '''±''' 14%, ''likely'' range) in 2081–2100 relative to present day (2006–2015) for RCP2.6 scenario and –5.5  '''±'''  2.7 x 10 6 m 3 s –1 (-32 '''±''' 14%) for RCP8.5 scenario, in line with a process-based probabilistic assessment (Schleussner et al., 2014 <sup>[[#fn:r708|708]]</sup> ). Furthermore, the uncertainty in AMOC changes has been shown to be mainly related to the spread in model responses rather than scenarios (RCP4.5 and RCP8.5) or internal variability uncertainty (Reintges et al., 2017 <sup>[[#fn:r709|709]]</sup> ). This behaviour is very different from the uncertainty in global SST changes, which is mainly driven by emission scenario after a few decades (Frölicher et al., 2016 <sup>[[#fn:r711|711]]</sup> ). To explain the AMOC decline, a new mechanism has been proposed on top of the classical changes in heat and freshwater forcing (Gregory et al., 2016 <sup>[[#fn:r712|712]]</sup> ). A potential role for sea ice decrease has been highlighted (Sevellec et al., 2017 <sup>[[#fn:r713|713]]</sup> ), due to large heat uptake increase in the Arctic leading to a strong warming of the North Atlantic, increasing the vertical stability of the upper ocean, as already observed in the Greenland and Iceland seas (Moore et al., 2015 <sup>[[#fn:r714|714]]</sup> ). It has also been showed that convection sites may move northward in future projections, following the sea ice edge (Lique and Thomas, 2018 <sup>[[#fn:r715|715]]</sup> ).

<div id="section-6-7-1-2role-of-gis-melting-and-their-freshwater-release-sources"></div>

<span id="role-of-gis-melting-and-their-freshwater-release-sources"></span>
==== 6.7.1.2 Role of GIS Melting and their Freshwater Release Sources ====

<div id="section-6-7-1-2role-of-gis-melting-and-their-freshwater-release-sources-block-1"></div>

Satellite data indicate accelerated mass loss from the GIS beginning around 1996, and freshwater contributions to the subpolar North Atlantic from Greenland, Canadian Arctic Archipelago glaciers and sea ice melt totalling around 60,000 m 3 s –1 in 2013, a 50% increase since the mid-1990s (Yang et al., 2016b), in line with more recent estimates (Bamber et al., 2018). This increase in GIS melting is unprecedented over the last 350 years (Trusel et al., 2018). Since the mid-1990s, there has been about a 50% decrease in the thickness of the dense water mass formed in the Labrador Sea, suggesting a possible relationship between enhanced freshwater fluxes and suppressed formation of North Atlantic Deep Water (Yang et al., 2016b). This hypothesis has been further supported by high-resolution ocean-only simulations showing that GIS melting may have affected the Labrador Sea convection since 2010, which may imply an emerging on-going impact of this melting on the SPG but a still non-detectable impact on the AMOC (Boning et al., 2016). Thus, while some studies argue that this melting may have affected the evolution of the AMOC over the 20th century (Rahmstorf et al., 2015; Yang et al., 2016b), considerable variability and limitation in ocean models restrain the full validation of this hypothesis, which remains model dependent (Proshutinsky et al., 2015; Dukhovskoy et al., 2016). Furthermore, some deep convection events resumed since 2014 (Yashayaev and Loder, 2017).

The impact of GIS melting is neglected in AR5 projections (Swingedouw et al., 2013) but has been considered in a recent multi-model study (Bakker et al., 2016; Figure 6.9). The decrease of the AMOC in projections including this melting term is depicted in Figure 6.9. GIS melting estimates added in those simulations were based on the Lenaerts et al. (2015) approach, using a regional atmosphere model to estimate GIS mass balance. Results from eight climate models and an extrapolation by an emulator calibrated on these models showed that GIS melting has an impact on the AMOC, potentially adding up to around 5–10% more AMOC weakening in 2100 under RCP8.5. Based on Figure 6.8 and 6.9, the risk of collapse before the end of the century is ''very unlikely'' , although biases in present-day climate models only provide ''medium confidence'' in this assessment. By 2290–2300, Bakker et al. (2016; Figure 6.9) estimated at 44% the likelihood of an AMOC collapse in RCP8.5 scenario, while the AMOC weakening stabilises in RCP4.5 (37% reduction, (15–65%) ''very likely'' range). This result suggests that an AMOC collapse can be avoided in the long term by mitigation.

Concerning the question of the reversibility of the AMOC, a few ramp-up/ramp-down simulations have been performed to evaluate it for transient time scales (a few centuries, while millennia will be necessary for a full steady state). Results usually show a reversibility of the AMOC (Jackson et al., 2014; Sgubin et al., 2015) although the timing and amplitude is highly model dependent (Palter et al., 2018). A hysteresis behaviour of the AMOC in response to freshwater release has been found in a few climate models (Hawkins et al., 2011; Jackson et al., 2017) even at the eddy resolving resolution (Mecking et al., 2016; Jackson and Wood, 2018). This is in line with the possibility of tipping point in the AMOC system. The biases of present-day models in representing the transport at 30°S (Deshayes et al., 2013; Liu et al., 2017a; Mecking et al., 2017) or the salinity in the tropical era (Liu et al., 2014b) may considerably affect the sensitivity of the models to freshwater release, but more on the multi-centennial time scale.

<div id="section-6-7-1-2role-of-gis-melting-and-their-freshwater-release-sources-block-2"></div>

<span id="figure-6.9"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 6.9'''

<span id="figure-6.9-the-changes-in-the-atlantic-meridional-overturning-circulation-amoc-strength-as-a-function-of-transient-changes-in-global-mean-temperature-for-projections-from-rcp4.5-and-rcp8.5-scenario.-this-probabilistic-assessment-of-annual-mean-amoc-strength-changes-at-26on-below-500-m-and-relative-to-18501900-as-a-function-of-global-temperature"></span>
<!-- IMG CAPTION -->
'''Figure 6.9 | The changes in the Atlantic Meridional Overturning Circulation (AMOC) strength as a function of transient changes in global mean temperature for projections from RCP4.5 and RCP8.5 scenario. This probabilistic assessment of annual mean AMOC strength changes (%) at 26oN (below 500 m and relative to 1850–1900) as a function of global temperature […]'''
<!-- IMG FILE -->
[[File:a5ee90ca9c3f611f762e466ea8bde0a2 IPCC-SROCC-CH_6_9.jpg]]

Figure 6.9 | The changes in the Atlantic Meridional Overturning Circulation (AMOC) strength as a function of transient changes in global mean temperature for projections from RCP4.5 and RCP8.5 scenario. This probabilistic assessment of annual mean AMOC strength changes (%) at 26oN (below 500 m and relative to 1850–1900) as a function of global temperature change (degrees Celsius; relative to 1850–1900) results from 10,000 RCP4.5 and 10,000 RCP8.5 experiments over the period 2006–2300, which are derived from an AMOC emulator calibrated with simulations from eight climate models including the Greenland Ice Sheet (GIS) melting (Bakker et al. 2016). The annual mean AMOC strength changes are taken from transient simulations and are therefore not equilibrium values per se. Moreover, it should be stressed that the results stem from future runs, not past or historical runs. Thus, due to internal variability both in the global mean temperature and AMOC in this transient simulation, large weakening can be found even at 0oC global warming. The ranges (66%, 90% and 99%) correspond to the amount of simulations that are within each envelope. The thick black line corresponds to the ensemble mean, while the different colours stand for different probability quantiles. The horizontal black thick line corresponds to the value of 80% of AMOC decrease, which can be seen as an almost total collapse of the AMOC. The horizontal black dashed thick line corresponds to a reduction of 50% of the AMOC, which can be considered as a substantial weakening. The vertical dashed green line stands for the 1.5oC of global warming threshold (relative to 1850–1900). The violet cross stands for the observation-based reduction estimate from Caesar et al. (2018). The size of the cross represents the uncertainty in this estimate.

<!-- END IMG -->
<span id="impacts-on-climate-natural-and-human-systems"></span>