Leading role of outer-Arctic circulation transport in AMOC response to global warming over a century

Using the Alfred Wegener Institute Climate Model (AWI-CM 1.1 LR), we explored how Arctic and extra-Arctic warming affect the response of Atlantic meridional overturning circulation (AMOC) to quadruple carbon dioxide (4 × CO2) forcing. The results suggest that AMOC weakening is mainly affected by circulation adjustment caused by extra-Arctic warming, while Arctic warming has a limited local impact and a relatively small contribution to AMOC weakening. Due to the warming outside the Arctic, the increase in northward advective heat transport dominates the weakening of deep convection in Nordic Seas. While in the Labrador Sea, the decrease in advection heat transport is compensated by a more significant decrease in ocean heat loss to the atmosphere, leading to an enhancement of the upper ocean stratification. Besides, the weakening of deep convection associated with AMOC response under global warming is more pronounced in Nordic Seas than in Labrador Sea.


Introduction
Atlantic meridional overturning circulation (AMOC) plays a critical role in climate systems (McManus et al 2004, Zhao et al 2018, Liu et al 2020).The AMOC delivers warm and salty water in its upper limb into the subpolar region of the North Atlantic (Chafik et al 2014), where they are transformed into dense deepwater in the lower limb via buoyancy loss (Medhaug et al 2012, Suzuki et al 2022).The adjustment of the AMOC has substantial implications for ocean heat exchange, carbon uptake, and global ecosystems (e.g.Li et al 2020, Suzuki et al 2022).Extensive studies have focused on describing and understanding changes in the AMOC and its response to anthropogenic warming (e.g.Lique and Thomas 2018, Levang and Schmitt 2020).Despite the inter-model discrepancy, prevailing observational and modeling evidence has pointed out that AMOC has weakened or will soon weaken in response to global warming (e.g.Chen andTung 2018, Liu et al 2020).However, due to the lack of long-term observational data in the Atlantic, the mechanism of AMOC changes over a century remains uncertain (Kilbourne et al 2022).
The AMOC response to anthropogenic warming is complex as it involves different ocean processes and series of feedback.Based on millennial-length simulations, Bonan et al (2022) found that most models consistently simulate an AMOC weakening in the early centuries in 4 × CO 2 experiments, but exhibit diverse recovery behaviors afterward.General simulation results suggest that centennial-scale AMOC variability in warming climates can be attributed to changes in the North Atlantic, while changes on longer timescales are more related to the Southern Ocean (Jansen et al 2018, Thomas andFedorov 2019).
On a century scale, the production of dense water in the North Atlantic is largely associated with the strength of AMOC (Lozier et al 2019).Previous views considered the Nordic Seas and Labrador Sea as the two main areas of deep convection (Medhaug et al 2012).Petit et al (2020)  Here, we aim to explore the different influences of ocean processes over the Arctic and extra-Arctic on the AMOC response in a century following the sudden onset of forcing.To distinguish between the effects of the Arctic and extra-Arctic processes, this study adopts a set of sensitivity experiments with perturbation in coupled models via separating the Arctic and extra-Arctic warming forcing to examine the AMOC response to anthropogenic warming.This research may help better understand the AMOC response to anthropogenic warming and provide a reference for the deployment and implementation of our future observation program.

Method
The simulation employs the Alfred Wegener Institute Climate Model version 1.1 (AWI-CM 1.1) which incorporates the atmosphere component ECHAM 6.3 and ocean component Finite Element Sea Ice-Ocean Model (FESOM) 1.4 (Semmler et al 2020a).The ocean model FESOM 1.4 employs horizontal global scale unstructured triangular meshes, with the flexibility allowing for increased resolution in dynamically active places while maintaining relatively coarse resolution elsewhere (Wang et al 2014, Semmler et al 2021).The setup of the horizontal grid is shown in figure S1.In this study, we adopt sensitivity experiments with perturbations in coupled models to investigate the AMOC response to the Arctic and extra-Arctic warming forcing.The AMOC strength is defined as the maximum value of zonally and depth-integrated transport along 26 • N in the Atlantic Ocean.The same as most models in CMIP6, the simulations do not include the process of Greenland ice sheet and its melting in warming scenarios, which may affect local dynamics somewhat (Semmler et al 2020b).We conduct a control (CTRL) run with CO 2 concentration kept at 1950 level (313 ppm) and three sensitivity runs with CO 2 concentration quadrupled at different latitudes.The 4 × CO 2 forcing of sensitivity experiments consistently maintained north of 60 • N, south of 60 • N, and globally called 60NN, 60NS, and GLOB, respectively.The 60NN and 60NS are set up to decompose the oceanic response to warming at different regions.The CTRL runs for 200 years, with the initial 50 years for spin-up, and each sensitivity experiment runs thereafter for 150 years.We chose last 110 years of the simulation (a relatively stable phase of slow change) to analyze the AMOC response to the regional 4 × CO 2 forcing (figure S2).In all experiments, the long-term average (the last 110 years) of full depth ocean heat content variability is nearly zero, also indicating an equilibrium state of the simulation.A detailed description of the model and experiment setup is provided in the supplementary text S1 and table S1.
The average stratification is calculated to investigate the stability of deep convection areas.The regionally averaged stratification (N 2 ) is defined as the weighted average squared buoyancy frequency of the water column over deep convection regions (Thomas and Ferrari 2008): where g is the gravitational acceleration, ρ is the potential density, and '⟨ ⟩' denotes the laterally weighted average of the region.According to the deepest convection of modern record (1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994) at Labrador Sea since 1983 (Yashayaev and Loder 2017), we define the main deep convection zone as the area where the mixed-layer depth (MLD) in March exceeds 2500 m in the subpolar North Atlantic (figure 2).The MLD is defined on density profiles with a fixed threshold of 0.03 kg m −3 (de Boyer Montégut et al 2004).We chose March to represent the late winter months when maximum deep convection occurs at high latitudes in the Northern Hemisphere (Levang and Schmitt 2020).
To better assess the relative contribution of temperature and salinity to the stratification changes, we decompose the total stratification of the ocean into thermal and saline components based on the linearized state equation with time-depth-varying coefficients (Fofonoff andMillard 1983, McDougall 1987) where

AMOC response to warming forcing
Under the anthropogenic warming scenario simulated by quadrupling atmospheric CO 2 concentrations, the global ocean warms on a large scale (especially in the North Atlantic) and is accompanied by a retreat of sea ice extent (figures S3 and S4).
We are concerned with how the AMOC adjusts under these ocean changes.Therefore, we investigate changes in AMOC strength during the first century when 4 × CO 2 forcing is applied to explore the response of the large-scale circulation to ocean warming.All sensitivity experiments exhibit a shoaling of AMOC depth, defined as the lower boundary of the upper overturning cell where stream function nearly equals zero (the gray bold line in figure 1).The depth of the maximum AMOC is also shallower under 4 × CO 2 forcing prescribed at different latitudes, decreasing from 1200 m (in CTRL) to about 800 m (in GLOB).The climatological AMOC strength in CTRL is 16.9 ± 1.0 Sv.Compared to CTRL, the AMOC strength in GLOB decreased by 4.6 Sv, which is nearly equivalent to the sum of the reduction in 60NN (1.9 Sv) and 60NS (3 Sv) (figures 1(e) and (f)).These results indicate that the ocean response to different regional warmings is less nonlinear and affirms the reasonability of our experimental setup.The degree of AMOC weakening is highly dependent on the location of warming forcing, where warming outside the Arctic can account for more than 61%.
March MLD variability over the deep convection zone has been identified as a good indicator of deep convection (Thomas et al 2015).The most dramatic reduction in MLD under 4 × CO 2 forcing occurs in the central Nordic Seas and Labrador Sea deep convection zone in the North Atlantic (figure 2).March convection in CTRL exceeds 2500 m at the central Nordic Seas and Labrador Sea.When the CO 2 forcing is prescribed in the Arctic (60NN), deep convection weakens slightly compared to CTRL.In contrast, the March MLD over the deep convection zone reduces dramatically in 60NS and GLOB to less than 1000 m (figure 2), indicating a general weakening and shoaling of AMOC in the global warming scenario.Hereafter, we therefore focus on the deep convection changes in the Nordic Seas and Labrador Sea that associated with the weakening of AMOC over a century.

Stratification adjustment over deep convection regions
The vertical stratification changes provide insights into the deep convection changes and help to better understand the AMOC weakening (Haskins et al 2020).Consequently, we calculated the squared buoyancy frequency (N 2 ) (equation ( 1)) of the water column to examine the deep convection intensity and assess the relative contribution of temperature and salinity (equations ( 2) and ( 3)).The results show that the upper-ocean stratification in the deep convective zones of the Nordic Seas and Labrador Sea is significantly enhanced by climate warming (figures 3(a) and (d)).A layer of strong N 2 increase appears at 200-1000 m with a maximum value of around 300 m, manifesting as a positive buoyancy anomaly.The stratification layer acts as an effective barrier and greatly inhibits the intensity of convective mixing by restricting the sinking of warm water from the surface to the deep layer.In 60NN, the deep convection can still be maintained as the upper-ocean stratification exhibits small variations with 4 × CO 2 forcing applied in the Arctic (figures 3(a) and (b)).When warming forcing is applied outside the Arctic, the stratification enhancement is far stronger in 60NS than in 60NN, suggesting that the anomalous changes in the deep convection zones are dominated by extra-Arctic warming.
The interplay of temperature and salinity changes in the upper layer sets the magnitude of deep-water formation and AMOC weakening.In the Nordic Seas, the warming effect is mainly manifested in the upper ocean, with relatively few changes in the deep ocean.In 60NN, the temperature and salinity changes are much smaller and the vertical gradient is largely t ) in 60NS and GLOB could be even twice as large as the total N 2 , some of which is offset by the salinity stratification (N 2 s ).The enhanced stratification in the upper deep convection zone is largely temperaturedriven while the effect of salinization is compensated by warming (N 2 t > 0, N 2 s < 0) (figures 3(a)-(c)).We also examine the changes in freshwater transport from the north and south of the deep convection zones under the warming scenario.Although warming has led to a widespread shrinkage of Arctic seaice (figure S4), the freshwater impacts from the Arctic are relatively small.In three sensitivity experiments, anomalies in total southward freshwater transport from the Arctic in the range of 0.005-0.02Sv can be seen.Sea ice melting leads to an increase in southward freshwater transport through the Fram Strait, which is compensated by a decrease in sea ice volume exports that leads to a reduction in sea ice freshwater flux (not shown).However, the northward freshwater transport outside the Arctic decreased by 0.16-0.38Sv, which is closely related to the salinization of the upper ocean.The anomalously strong stratification of the subsurface and intermediate layers rather than the surface layer in 60NS and GLOB also reflects a more significant response of AMOC to the warm and salty advection in the upper limb of the North Atlantic.Therefore, although the simulations do not include Greenland ice sheet melting (which would underestimate the freshwater effect), the contribution of Arctic changes to the local stratification enhancement in the deep convective zone is much smaller than in the outer-Arctic regions.In other words, changes in Arctic outflows arising from sea-ice melting (e.g. the East Greenland Current along the continental shelf of Greenland), may have relatively small effects on mixing in the deep convection zones.
In the deep convection zone of the Labrador Sea, a more complex vertical stratification emerges in 60NS and GLOB, with a structure of double maximum of N 2 (figure 3(d)).Warming and salinization extend from the surface to around 2500 m (insets in figures 3(e) and (f)), dramatically altering the vertical gradients of temperature and salinity and resulting in strong stratification adjustments.In the upper 1000 m temperature and salinity gradients combine to enhance the upper-ocean stratification (N 2 t > 0, N 2 s < 0).While in the deep layer, despite salinity changes working to weaken the stratification, the temperature change dominates the buoyancy increase, with the total buoyancy change producing a moderate net increase (figures 3(d)-(f)).In 60NS and GLOB, the weakening of deep convection makes it difficult for the upper water to mix into the bottom as in CTRL and 60NN, and the reduced mixing depth leads to strong temperature and salinity gradients below 1900 m, consequently enhancing deep buoyancy stratification.However, connections between the deep layer and changes in the subpolar upper ocean are complex and require longer simulations for further investigation (Levang and Schmitt 2020).In sensitivity experiments, especially for 60NS and GLOB, the depth of convective mixing in the Nordic Seas drops dramatically, while in the Labrador Sea, it can still be maintained at around 1000 m (figures 3(a) and (d)).Comparing the two main regions, deep convection activities in Nordic Seas appear to be more sensitive to greenhouse gas forcing than that in the Labrador Sea.Lique and Thomas (2018) have also predicted that the deep convection in the Nordic Sea may disappear or even undergo a latitudinal shift under 4 × CO 2 forcing.

Heat budget of the upper-ocean
To assess the impact of different thermodynamic and dynamic processes on deep convection, we analyze the source of ocean heat content (OHC) variability in subpolar regions.Our main concern is how the Arctic and extra-Arctic warming affects the horizontal thermal advection transport (HT) and local air-sea heat fluxes (HFs), and in turn modulates the weakening of deep convection activities in the subpolar North Atlantic and the weakening of AMOC.Two closed areas are therefore selected in the upper ocean (i.e. the upper limb of overturning circulation) of the Nordic Seas and Labrador Sea, respectively (figure 4, shaded area in the inset).Considering the circulation and vertical topographic constraints, we choose an integration depth of 410 m (details in supplementary text S2).Note that positive values of HT and HF are defined as the heat input to the ocean box through the cross-sections and air-sea interface.Heat budget analysis in our results reveal that the OHC calculated from temperature is mainly influenced by the integrated effect of anomalous HT and HF.In CTRL, the upper-ocean OHC varies at a small annual rate in the Nordic Seas (1.8 × 10 6 W) and Labrador Sea (−3.3 × 10 6 W), which are basically balanced in by HT and HF (figure 4), and the residual is negligible.In GLOB, the annual rate of OHC change in the Nordic Seas increases to 2.4 × 10 7 W, more than ten times that of CTRL.In the Labrador Sea, the annual rate of OHC change is 6.9 × 10 6 W, a small increase compared to CTRL.
In the Nordic Seas, both the increase in advective HT and the decrease in HF are no more than 30 TW in 60NN.However, in 60NS and GLOB, the HT and HF vary considerably compared to CTRL and the variation in HT is even more than twice that of HF, suggesting that the Arctic warming effects have relatively limited impacts on the deep convection zone.As the ocean warms (inset in figure 3(b)), the decrease in air-sea temperature difference leads to a weakening sensible heat exchange.Opposite variations of latent and sensible HFs and almost negligible variations in radiation (not shown) make the HF in the Nordic Seas insensitive to regional forcing.In fact, surface wind stress over the subpolar North Atlantic is also weakening in a global warming scenario (figure S5), which may partly contribute to the enhanced stratification of the upper ocean in the deep convection zone by weakening local air-sea HF and vertical mixing.The HT variation is much stronger than that of the local HF.Positive temperature flux through the Denmark Strait and Iceland-Faroe Ridge (southern transect N2 and N3) enables large amounts of heat transport to the closed region.Examination of the circulation transport reveals that the AMOC decrease is accompanied by a weakening of the upper ocean circulation (figure S6).The upper limb of AMOC (i.e.The North Atlantic current (NAC)) (Chafik et al 2014) carries warm and salty water northward into the Arctic.Despite the weakened circulation acting to decrease the northward volume transport into the subpolar region (figure S6), it is apparent that the advection heat transport by the AMOC upper limb becomes more efficient as the upper ocean warms.More heat and salt are trapped in the Nordic Seas in terms of circulation transport, which further reinforces the stratification of the upper ocean (figures 3(b) and (c)) and inhibits the deep-water formation in this region.
As is the case in the Nordic Seas, the impact of Arctic warming (60NN) on deep convection in the Labrador Sea is relatively slight compared to extra-Arctic warming (60NS).However, in 60NS and GLOB, the local HF variation in the Labrador Sea is roughly comparable to that of the HT.In GLOB, both advective HT and ocean heat loss are reduced (figure 4(b)), with the HT and local heat loss decreasing to one-third of the CTRL.Even though the weakening of advection heat transport facilitates deep convection, the reduction of air-sea HF into the atmosphere in the winter leads to ocean heat gain and contributes to enhanced stratification.By the time NAC enters the North Atlantic, the subpolar gyre weakens and the net volume transport of another mainstream into the Nordic Seas also weakens, which further leads to the weakening of the branches that form in the Irminger Sea and Iceland Basin toward the Labrador Sea (L3 transect) (figure S6).Owing to the relatively long distance from the east to the west subpolar region and the ocean heat loss along the way (figure S9), the accumulation of advection heat transport into the Labrador Sea is reduced and does not result in dramatic increase in the stratification of the upper ocean (figure 3).This also explains the weaker increase of the upper ocean heat content in the Labrador Sea than in the Nordic Seas (figure S8).It has been recognized that convective processes in the Labrador Sea are highly correlated with AMOC (Pickart and Spall 2007, Yeager et al 2021), but recent observations suggest that dense water formation in the eastern subpolar region plays a more dominant role in driving AMOC (Lozier et al 2019).However, due to the complexity of the regional circulation in the North Atlantic and the potential connections between sea areas, the source region of AMOC and its changes under global warming are still controversial.Although the upper ocean processes in the Nordic Sea and Labrador Sea exhibit different dynamics in the weakening of deep convection, identifying the relative contribution of these two regions to AMOC weakening requires further research work.

Discussion and conclusion
We investigate the AMOC response to anthropogenic warming over a century by separating and quantifying the effects of Arctic and extra-Arctic warming.According to the results of sensitivity experiments, AMOC intensity shows varying degrees of weakening under regional warming forcing.Deep convective activity over the Nordic Seas and Labrador Sea is significantly suppressed.Heat budget analysis in closed regions of both deep convective zones reveals that advection HT from the south flank of the North Atlantic plays an important role in the changes of deep convection.However, the two deep convection zones exhibit different local features.In the global warming scenario, more heat is transported into the Nordic Seas by the northward branch of NAC and dominates the local changes.Another major branch of NAC that enters the Labrador Sea along East Greenland topography weakens, and less heat enters the region in the form of advection transport after a somewhat oceanic heat loss along the way.Deep convection is suppressed but not as strong as in the Nordic Seas, since advection HT and HF variations into this region can partially compensate each other.Overall, deep convection weakening in these two areas eventually contributes to the weakening of AMOC, with a stronger convection response in the Nordic Seas than in the Labrador Sea.
From the perspective of the evolution of deep convection activity in the subpolar North Atlantic, we investigate the possible response mechanism of AMOC in global warming scenarios, which may serve as a reference for climate projection.AMOC is a large-scale ocean circulation that shows the response times on a century and even millennium scale (Lique and Thomas 2018, He et al 2019, Bonan et al 2022).Here, we discuss the response of AMOC to a centurylong anthropogenic warming, a timespan over which most of the GCMs show a decline in AMOC response to abrupt 4 × CO 2 forcing before they diverge to either recovered or further diminished AMOC states (Bonan et al 2022).Both changes in deep water formation in the North Atlantic and surface wind-driven ventilation in the Southern Ocean can lead to adjustments of AMOC.The latter controls the longer-term variability of AMOC, and its remote effect is crucial to the equilibrium of AMOC (Yang et al 2016).Under a warming climate, shifts in surface wind forcing would induce changes in upwelling in the Southern Ocean (Chen et al 2019), and it remains to be further identified how the Southern Ocean effect would potentially cause AMOC changes through deep circulation on millennial scales.Local feedback processes in the North Atlantic are also affected to some extent by the freshwater effect from Greenland ice sheet melting.However, the dynamic variations associated with the Greenland ice sheet melting involve nonlinear processes, and uncertainties remain about the intensity and pathways of the freshwater release.Therefore, the impact of Greenland ice sheet melting on local buoyancy and AMOC needs to be rigorously assessed (Lique and Thomas 2018), which is beyond the scope of our study.More generally, the response of the ocean to the warming forcing results from a complex combination of various mechanisms such as tropical precipitation, local overflow, as well as subpolar circulation changes (Daniault et al 2016, Levang and Schmitt 2020, Yeager et al 2021).The relative contribution of the deep convective zones to the transformation of deep water in the North Atlantic is also a matter that remains to be clarified.Further detailed assessment of high-resolution models and observational support is needed (Lozier et al 2019, Sallée et al 2021).
and S are the potential temperature and salinity, α = − 1 ρ ∂ρ ∂T and β = 1 ρ ∂ρ ∂S are the thermal expansion and haline contraction coefficients of water calculated from corresponding T and S. N 2 t and N 2 s represent the contribution term of temperature and salinity, respectively (N 2 ≈ N 2 t + N 2 s ).

Figure 1 .
Figure 1.Climatology (last 110 years of simulation) AMOC stream function (Sv) between 20 • S and 80 • N in different experiments.(a) CTRL, (b) 60NN, (c) 60NS, (d) GLOB.The gray bold line denotes the position where the meridional overturning stream function transport is zero.(e), (f) Comparison of the anomalies of AMOC stream function relative to CTRL run in different experiments.(e) Sum of AMOC stream function anomalies in 60NN and 60NS, (f) AMOC stream function anomalies in GLOB.

Figure 2 .
Figure 2. Climatology (last 110 years of simulation) March mixed-layer depth (km).(a) CTRL, (b) 60NN, (c) 60NS, (d) GLOB.In CTRL run, the magenta line in Nordic Seas and Labrador Sea marks the areas of MLD greater than 2500 m, which are defined as the main area of deep convection in our study.

Figure 3 .
Figure 3. Climatology (last 110 years of simulation) wintertime stratification at deep convection zones.(a) Regionally weighted average squared buoyancy frequency (N 2 , 10 −5 s −2 ) throughout all depths at the Nordic Seas, (b) the temperature component of N 2 , defined as N 2 t , (c) the salinity component of N 2 , defined as N 2 s .The insets in (b) and (c) are the corresponding profiles of potential temperature and salinity.(d)-(f) The same as (a)-(c), but for the Labrador Sea deep convection zone.Note the different ranges of x-axis in the three columns.

Figure 4 .
Figure 4. Two main components of the upper-ocean heat budget at the irregularly closed region in different experiments (detailed information can be found in text S2).(a) Climatology averaged advection heat transport (HT) and net ocean surface heat flux (HF) at Nordic Seas for last 110 years of simulation (positive, inward, heat gain of the closed region).HT-N1, HT-N2 and HT-N3 represent the advection heat transport through the upper ocean of the Fram Strait-Barents Trough (N1), Denmark Strait (N2) and Iceland-Faroe Ridge (N3).HT-net is net advection heat transport for the closed areas.(b) The same as (a), but for the region at Labrador Sea and the sections located as L1, L2 and L3.The inset displays the position of closed areas at the Nordic Seas and Labrador Sea (shaded in red).
revealed that the deep-water formation occurs primarily in the Iceland Basin and Irminger Sea by local buoyancy forcing.How these deep convective regions in the North Atlantic respond to warming is still under debate.The key challenge is to explore the variability of deep convection in the North Atlantic and its modulation of AMOC in global warming scenarios.The North Atlantic Ocean lies at a junction where local processes are influenced by both Arctic and extra-Arctic warming.Changes in large-scale circulation outside the Arctic, freshwater effects from the Arctic and even the local melting of Greenland's ice sheet (Martin et al 2022) can cause dynamic changes in the deep convection region of the North Atlantic (e.g.Semmler et al 2020a, Smedsrud et al 2022).Therefore, it is important to identify the influence and relative contribution of ocean processes at different latitudes to changes in the North Atlantic deep convection regions under climate warming.