Sensitivity of AMOC Fingerprints Under Future Anthropogenic Warming

Detecting the response of the Atlantic meridional overturning circulation (AMOC) to anthropogenic warming can only be made with fingerprints indirectly because of the lack of sufficiently long direct measurements. However, whether the relationship between the AMOC and its fingerprints is stationary is rarely examined. This study uses coupled and ocean‐alone model simulations to investigate the sensitivity of two typical AMOC fingerprints under future anthropogenic warming. We found a lower sensitivity of the North Atlantic warming hole fingerprint in future warming scenarios associated with the differing vulnerability of deep‐water origins to external forcing and climate feedback. In contrast, the remote South Atlantic salinity pile‐up fingerprint is relatively insensitive to variations in AMOC sources, and its sensitivity to the AMOC is slightly enhanced by an intensified hydrological cycle. Our study implies that fingerprints outside the northern deep convection region may become more suitable in representing the response of AMOC to future warming.


Introduction
As the Earth's most extensive oceanic circulation system, the Atlantic meridional overturning circulation (AMOC) profoundly impacts regional and global climates (e.g., Buckley & Marshall, 2016).In climate change scenarios, the AMOC is projected to weaken under global warming induced by anthropogenic greenhouse gasses (GHGs) (Gregory et al., 2005;Gulev et al., 2021;Rhein et al., 2013).This slowdown, however, is difficult to detect directly from the short observational record because the latter contains substantial interdecadal climate variability (Bryden et al., 2005;Robson et al., 2014;Smeed et al., 2014;Srokosz & Bryden, 2015;Zhang, 2007Zhang, , 2008)).Any long-term AMOC change in the real world has to be examined indirectly from the AMOC fingerprints.In the past two decades, multiple AMOC fingerprints have been proposed.The classical warming hole fingerprint, which is characterized by a surface cooling over the North Atlantic (NA) subpolar gyre (SPG) region relative to background global warming, has been used as an "all-climate" AMOC fingerprint (Caesar et al., 2018;Rahmstorf et al., 2015).Different from most fingerprints that located locally in the subpolar NA (Caesar et al., 2021), a unique AMOC fingerprint from the South Atlantic with a higher signal-to-noise ratio was proposed recently, featuring a salinity pile-up over the subtropical South Atlantic (STSA) relative to the subtropical South Indo-Pacific (STSIP; Zhu & Liu, 2020;Zhu et al., 2023).
Based on model simulations, the relationship between these two fingerprints and the AMOC change has been quantified as 3.8 ± 0.5 Sv/K (Caesar et al., 2018) and 2.3 ± 0.34 Sv/0.1 psu (Zhu & Liu, 2020), respectively.However, whether such a scaling relationship or, more generally, the sensitivity of these fingerprints to AMOC, is stationary across time and warming scenarios remains to be explored.Modeling studies suggest that the warming hole and AMOC's relationship can depend on the forcing and processes that drive AMOC changes

Supporting Information:
Supporting Information may be found in the online version of this article.(Jackson & Wood, 2020;Little et al., 2020;Moffa-Sánchez et al., 2019).Studies also show that under future warming, the AMOC source regions will be shifted from the subpolar region to the Arctic Basin and subtropical gyre (Lique & Thomas, 2018;Lique et al., 2018), degrading the suitability of SPG warming hole fingerprint in representing long-term AMOC change.Actually, a weaker warming hole over the SPG has been projected under a more strongly forced scenario run (Drijfhout et al., 2012), yet the exact mechanism at play remains to be clarified.In contrast, the salinity pile-up fingerprint is dynamically controlled by AMOC perturbation as a whole on the mean climatological meridional salinity gradients (Zhu & Liu, 2020;Zhu et al., 2023) and thus would be independent of the driver and sources of AMOC change.Whether the relationship between AMOC and its fingerprints is stationary raises concerns, especially when these fingerprints are used to infer AMOC changes in a changing climate.Moreover, investigating the fingerprint mechanism will improve our knowledge of the interaction between the AMOC and climate system.
By analyzing coupled model simulations and conducting idealized experiments, this study aims to investigate the suitability of two typical AMOC fingerprints (with a focus on their sensitivity) across a range of warming scenarios.The paper is organized as follows.Model simulations and definitions of crucial AMOC indices are described in Section 2 and Section 3, respectively.Section 4 investigates the sensitivity of warming hole and salinity pile-up fingerprints in both CMIP6 simulations and ocean-alone sensitivity experiments.A summary and discussion are given in Section 5.

CMIP6 Simulations
Coupled Modeling Intercomparison Project Phase 6 (CMIP6) employs the latest generation of fully-coupled climate models.In this study, we use the model output of 8 CMIP6 models with a total of 152 ensemble members to evaluate the response of AMOC and its indices under historical forcing and in two shared socioeconomic pathways (SSPs), namely SSP245 (an intermediate level of GHGs emissions) and SSP585 (a high level of GHGs emissions).Details on models and ensemble members used for each forcing run are listed in Table S1 of Supporting Information S1.

Ocean-Alone Model Simulations
The ocean model used in this study is POP2 (Danabasoglu et al., 2012;Smith et al., 2010), with a nominally 3-degree resolution in the horizontal direction and 60 levels in the vertical direction.The control run is forced by the normal year forcing from the Coordinated Ocean-Ice Reference Experiments (CORE) dataset (Large & Yeager, 2004), using the CORE experimental design as outlined in Griffies et al. (2009).
Due to a fresh bias in the subpolar NA, a common bias also in coupled simulations (Sgubin et al., 2017), the default POP2 shows an unrealistically shallow late-winter mixed layer depth (MLD) around the Labrador Sea and Irminger Sea, where the observations show a deepest MLD of over 1,000 m (Lique & Thomas, 2018;Sgubin et al., 2017).Consequently, the simulated AMOC has a lower intensity of 14 Sv compared with observations (17.9 ± 3.2 Sv; Mielke et al., 2013;Talley, 2013), and the warming hole was shifted from the SPG region into the Greenland, Iceland, and Norwegian Seas (GIN Seas).To correct this model bias, we first performed a flux adjustment on surface salinity.Interested readers are referred to Zhu and Liu (2020) for details on the adjustment process.We note here that by applying flux adjustment, a realistic deep mixed layer (>1,000 m in late winter; Figure S1 in Supporting Information S1) is situated in the Labrador-Irminger Sea and the Nordic Seas and the AMOC intensity is increased to ∼20 Sv, all consistent with the observations (Lique & Thomas, 2018;McCarthy et al., 2015;Mielke et al., 2013;Sgubin et al., 2017;Talley, 2013).This flux adjustment run is our new control run from which the sensitivity experiments are launched.
Three idealized heat flux forcing experiments are designed to investigate the response of AMOC fingerprints under a range of warming scenarios.We apply a globally uniform heat flux anomaly over the ocean with the magnitude increasing linearly from 0 to 2 W/m 2 (Case 2W), 5 W/m 2 (Case 5W), and 10 W/m 2 (Case 10W) in 100 years, mimicking the radiative forcing associated with different levels of GHGs emissions.All three experiments are integrated for 100 years, and by the end of year-100, the AMOC has been reduced by 8 Sv (40%), 10 Sv (50%), and 12 Sv (60%), respectively.

Definitions
AMOC intensity is defined as the maximum overturning stream function below 300 m over 20-50°N in the Atlantic.The SPG warming hole index (SPG SST index or simply the SST index), is defined as annual mean, subpolar NA mean (15-50°W, 50-65°N) SST minus annual mean, global mean SST T N = T SPG T Gloal .Another domain choice of (20-55°W, 55-65°N) is also used to investigate local SST response over the SPG convection sites (SPG convection SST index).The salinity pileup index (SSS index), is defined as the difference between the annual mean, STSA mean (averaged over 10-34°S) SSS, and annual mean SSS at the same latitude band in the STSIP S S = (S STSA S STSIP ), with the minus sign to ensure the index change consistent in sign with the AMOC change.The sensitivity of AMOC fingerprints is quantified as the change in their respective index per unit AMOC change, that is ∆T N ∆AMOC and ∆S S ∆AMOC , respectively.The contribution of Nordic Seas convection to AMOC intensity is diagnosed as the maximum overturning stream function below 300 m at the southern boundary of Nordic Seas (∼65°N) in the Atlantic following Cheng et al. (2011).By this definition, the deep-water formation by Nordic Seas convection is 6-7 Sv in coupled historical simulations and our new POP2 control run, consistent with ORAS5 reanalysis and available observations (Årthun, 2023).The SPG circulation index is defined as the local minimum of the barotropic stream function within the subpolar NA.

Sensitivity of AMOC Fingerprints in CMIP6 Simulations
The evolution of AMOC and its fingerprints from the year 1850 to 2100 are shown in Figure 1.AMOC intensity in the multi-model ensemble mean (MMEM) averaged over the historical period 1980-2014 is ∼20 Sv, consistent with observations (McCarthy et al., 2015;Talley, 2013).Due to the combined forcing of increasing GHGs and decreasing anthropogenic aerosols, the AMOC has been significantly weakened since the 1980s in models (Figure 1a) and likely also in the real world (Zhu et al., 2023).The weakening trend continues in SSP245 and SSP585 runs.By the end of the 21st century, the AMOC has been reduced by 7 Sv (35%) and 11 Sv (55%), as projected in SSP245 and SSP585, respectively (Figure 1a).In the following, we focus on the sensitivity of AMOC fingerprints under future anthropogenic warming in these two SSP scenarios.
In response to a weakening AMOC, both SST and SSS indices are reduced but with changing sensitivity.The sensitivity change is most evident across different warming scenarios (Figures 1 and 2).The scaling relationship derived from a regression between the MMEM SSS index and MMEM AMOC over 2015-2100 is 2.4 Sv/0.1 psu in SSP245 and 2.0 Sv/0.1 psu in SSP585 (Figure 1b), respectively, both falling into the range of 2.3 ± 0.34 Sv/0.1 psu derived across a range of forcing scenarios (Zhu & Liu, 2020), suggesting a stationary relationship between the salinity pile-up fingerprint and the AMOC.Compared with SSP245 (0.042 psu/Sv), the sensitivity of the SSS index in SSP585 (0.050 psu/Sv) shows a slight increase (less than ∼20%), which can also be seen from individual models (except MPI-ESM1-2-LR; Figure S2a in Supporting Information S1).This enhancement in sensitivity is likely related to the intensified atmospheric hydrological cycle under global warming (Held & Soden, 2006).Consistent with the hydrological cycle change, the global SSS response features "salty-getting-saltier" pattern, as identified in observations and models (Figures 2b and 2d; Boyer et al., 2005;Durack et al., 2012;Lago et al., 2016).Therefore, the mean salinity gradients between the southern and northern sides of the STSA will be slightly increased (larger SSS increase in the northern side; Figures 2b and 2d).The larger salinity gradients in a more strongly forced scenario enhance the efficiency of unit AMOC variation in producing a stronger salinity convergence over the STSA.
In contrast, the sensitivity of NA SPG warming hole to AMOC change is largely reduced in warmer scenario run, an exception to the overall enhanced warming (scaled by AMOC change) over the Northern Hemisphere (Figures 2a and 2c).It is worth noting that there is notable cooling (relative to the global mean) around the Antarctic continent in models (Figures 2a and 2c) and observations (Haumann et al., 2020;Zhu & Liu, 2020), which can be linked to an increase in sea ice (Haumann et al., 2020).For the SPG warming hole, the sensitivity in SSP245 and SSP585 is 0.19 K/Sv and 0.13 K/Sv, respectively (Figure 1c), reduced by ∼30% and 50% respectively from the historical value (0.26 K/Sv; Caesar et al., 2018).For individual models, two models, namely CanESM5 and IPSL-CM6A-LR, cannot produce a warming hole in the SPG region due to the absence of Labrador Sea convection (Figure S3 in Supporting Information S1).The bias in simulating SPG convection makes the warming hole in these two models situated around Iceland, like the default POP2 model (without adjustment).The other four models that reproduce the SPG warming hole all witness significantly reduced sensitivity (Figure S2b in Supporting Information S1) under stronger warming, questioning the reliability of the warming hole under future warming.
The reduced warming hole sensitivity can be understood from the characteristics of AMOC sources.The AMOC deep-water formation system consists of deep convection in the Labrador-Irminger Sea (SPG convection) and the Nordic Seas.Modeling efforts show that SPG convection is more vulnerable to changes in external forcing (Drijfhout et al., 2012).This means that SPG convection will be reduced first in response to global warming.Reduction in SPG convection prohibits local heat exchange between cooling surface water and relatively warmer subsurface water, overcompensating the warming due to the radiative forcing and finally leading to surface cooling (Sgubin et al., 2017;Zhu & Liu, 2020).Additional cooling can also be contributed by weakened northward heat transport associated with a weakened AMOC.In contrast to the relatively quick response of the SPG convection, the northern convection area over the Nordic Seas is less vulnerable and can remain active over a longer time (Drijfhout et al., 2012;Sgubin et al., 2017).Therefore, we speculate that the SPG warming hole evolves in concert with initial AMOC weakening associated with SPG convection but is less responsive to further AMOC weakening associated with Nordic Seas convection, making it less sensitive in the longer term.
The above mechanism is well supported by CESM2 simulations which perfectly reproduce the AMOC intensity and MLD distributions as well as the pattern of SPG warming hole (Figure 3).In CESM2, as a response to increasing radiative forcing, nearly 90% of the MLD shoaling and SST cooling (relative to the global mean) over the SPG region happens during the first 50 years and remains relatively unchanged thereafter (Figures 3a-3e), opposing to the fairly linear reduction of AMOC for the whole period 2015-2100 (Figure 3g).It is shown that the AMOC weakening before the 2060s is dominated by the reduction in SPG convection (Figure 3e).At the same time, its further weakening afterward is contributed dominantly (>90%) by reduced Nordic Seas convection (Figures 3f and 3h).The differing vulnerability between deep-water formation sources is likely related to the background stratification.In the mean climatology, the stratification in the SPG region is generally weaker than that in the Nordic Seas, evident from the deeper mixed layer there (Lique & Thomas, 2018;Sallée et al., 2021), making it less stable in a warming climate (Figures 3a and 3b).Given that the SPG SST is more sensitive to SPG convection than the Nordic Seas convection, the sensitivity of SPG warming hole to AMOC change will be reduced with time (Figures 3c, 3d, and 3g).In stronger warming scenario SSP585, the warming hole, SPG convection, and the AMOC all evolved similarly to that of SSP245 during 2015-2060 (Figure 3).After the 2060s, the warming hole is relatively stable in SSP245 while it reverses to a warming (i.e., exceeding the global mean SST increase; Figures 3c and 3d) in SSP585 albeit with stronger AMOC reduction associated with stronger decrease in the Nordic Seas convection (Figures 3g and 3h).The contrast between stronger AMOC reduction and weaker warming hole under a more strongly forced scenario can also be seen from the CMIP6 MMEM (Figure 1c).These results suggest that the warming due to the radiative forcing and related climate feedback under stronger warming scenario will overcompensate for the cooling effects of AMOC weakening, leading to an anticorrelation relationship between the warming hole and the AMOC later this century.
We note that unlike its relatively slow response in CESM2, the SPG convection in models MIROC6 and MPI-ESM1-2-LR collapses abruptly in one to two decades (especially in the Labrador Sea; not shown).In CMIP5 RCP scenarios, Sgubin et al. ( 2017) also identified a model ensemble in which the SPG convection and SST show abrupt change while the AMOC evolves much more slowly.The abrupt change in SPG is associated with less stratified SPG in these models, while non-abrupt models generally overestimate the present-day SPG stratification (Sgubin et al., 2017).With a correction on the surface fresh bias, the stratification over the SPG region in our flux-adjusted POP2 has been reduced and is close to the real world.Using sensitivity experiments conducted with this more realistic ocean model, we investigated the behavior of SPG convection, AMOC and its fingerprints from the perspective of ocean dynamics.

Sensitivity of AMOC Fingerprints in Ocean-Alone Experiments
Similar to coupled simulations, our ocean-alone sensitivity experiments reproduce the reduced (relatively stable) sensitivity of warming hole (salinity pile-up) to the AMOC (Figures 4a-4f).We note that by default, the evaporation in the ocean-alone model adjusts with changing SST in response to surface heat flux forcing, leading to a change in surface freshwater flux (SFWF; Figures 4a-4c).However, the contribution of SFWF change to SA salinity change is neglectable compared with the dynamic effect associated with the AMOC (see the Extended Data Figure 9 in Zhu and Liu (2020)).Therefore, the STSA salinity response in our experiments is primarily an AMOC imprint, which is not the same as the coupled simulations (see Figures 1 and 4 in Zhu and Liu (2020)) with the latter also subject to an intensified hydrological cycle.This may partly explain the smaller sensitivity of SSS index in our experiments (0.01 psu/Sv).Albeit with this discrepancy, the sensitivity of SSS index to AMOC is relatively stable (Figure 4g), confirming the stationary dynamical control of AMOC on salinity pile-up fingerprint in a warming climate.
For the warming hole, a significantly reduced sensitivity is seen in the NA SPG region with the largest reduction located over the Labrador-Irminger Sea convection sites (Figures 4d-4g).The collapse of these convection sites occurs within one decade with the timing (defined as the time when an 80% reduction in MLD is reached) being later under more strongly forced cases, and the Irminger Sea lagging the Labrador Sea by 5-10 years (Figures 4h-4i).The consequently abrupt SPG cooling is also consistent with coupled simulations with realistic SPG stratification (Sgubin et al., 2017), especially for the Labrador Sea convection region (Figure S4a in Supporting Information S1).Such abrupt cooling is predominantly driven by suppressed vertical mixing associated with reduced deep convections.The reduction of SPG convection also reduces the oceanic heat loss to the atmosphere (Figures 4d-4f), further heating the SPG region and degrading the warming hole fingerprint.After the collapse of Labrador Sea convection, the cooling of SSTs over the SPG convection sites is reduced, resulting in a relatively stable warming hole with similar intensity in the three experiments (Figure S5a in Supporting Information S1).
In contrast to the fast and strong reduction in the SPG convection, the Nordic Seas convection is reduced more slowly (Figures 4j-4k) and does not collapse in the three experiments.As such, the reduction of the AMOC, the integral result of convection in the Labrador-Irminger Sea and the Nordic Seas, is characterized by a relatively linear behavior with larger reduction in more strongly forced cases (Figure S5b in Supporting Information S1).The inconsistency between the SPG SST and AMOC dynamics lowers the sensitivity of the SPG warming hole, making it less suitable in longer-term and stronger warming scenarios (Figures 4d-4g).
Compared with the abrupt cooling over the Labrador Sea convection sites, the cooling in the central SPG is gentler .The coincidence of the warming hole and subpolar gyre indicates that cooling that emerged from the convection sites would spread downstream along the local cyclonic route.Under global warming, the intensity of SPG circulation (quantified by the SPG index) is reduced gradually (Figure S5c in Supporting Information S1), resembling coupled warming simulations (e.g., Lique & Thomas, 2018).This slowdown can be linked to reduced density gradients between the less dense SPG core and its surrounding waters, which is thereby modulated by the AMOC transport (Sgubin et al., 2017).The cooling originating from the SPG convection sites is transmitted by a gradually weakening SPG circulation, contributing to a more gradual and weaker cooling in the SPG interior but still with reduced sensitivity (Figures 3a-3d and 4d-4f).

Summary and Discussion
Using fully-coupled simulations and idealized ocean-alone experiments, this study investigates the sensitivity of two typical AMOC fingerprints in a range of warming scenarios.It is shown that the warming hole fingerprint will become less sensitive to AMOC change, especially under high GHGs emission scenario.In contrast, the salinity fingerprint will retain or even strengthen its sensitivity in response to future warming.The mechanism behind the changing sensitivity is further examined.For salinity fingerprint, the dynamical relationship between it and the AMOC is more straightforward as it is mainly determined by the AMOC perturbation as a whole on mean meridional salinity gradients over the STSA, regardless of the variations in specific integral parts of AMOC.Under global warming, the meridional salinity gradients over the STSA are projected to be slightly strengthened, which may lead to an enhanced sensitivity of this salinity fingerprint.Such intensified sensitivity of AMOC fingerprints is preferred in the early warning of potential critical transition of AMOC.
As to the warming hole fingerprint, its connection with AMOC is complicated by several factors.The warming hole as an "all-climate" AMOC fingerprint is established upon a strong assumption of the stationary relationship between SPG SST and AMOC under an evolving climate.Our study provides modeling evidence that the warming hole cannot reliably represent AMOC in a longer-term and a strong warming climate.This is because the warming hole mainly reflects a change to SPG convection, while AMOC is influenced by multiple deep-water formation sources within and beyond the SPG region.Under short-term radiative forcing when the SPG convection is still active, the warming hole can largely reflect the AMOC change as SPG convection is more vulnerable to external forcing.Later, when external forcing is sufficiently strong, the SPG convection will collapse, and the warming hole will reach its maximum.In contrast, the AMOC can be continually weakened by

Geophysical Research Letters
10.1029/2023GL107170 other convection sites (e.g., Nordic Seas convection).Therefore, the sensitivity of the warming hole fingerprint to AMOC weakening will be reduced with time.
Meanwhile, enhanced high-latitude warming associated with sea ice-albedo feedback and reduced oceanic heat loss associated with depressed deep convection can also contribute to the degradation of the warming hole.Such results may have implications for AMOC reconstructions.Our results imply that fingerprints or proxies over a single subpolar NA convection region may not fully represent long-term AMOC changes.In contrast, remote fingerprints outside the subpolar convection sites, such as the salinity pile-up fingerprint, may become an optimal choice as it responds to the full AMOC rather than any individual integral parts.
Finally, we note that there are also caveats in the present study.The model resolution used in this study is too low to explicitly resolve the mesoscale eddies and dense overflows (parameterized in a low-resolution model), which may alter the response of AMOC to climate change (Marzocchi et al., 2015).Also, the reaction of the warming hole, including its strength and geographical location, is strongly model-dependent (Menary & Wood, 2018).For example, in some models, the warming hole is situated southeast of the convection sites in the Labrador and Irminger Seas (Drijfhout et al., 2012;Menary & Wood, 2018).This mismatch between the warming hole and deep convection sites may indicate an adjustment of the gyre circulation (Drijfhout et al., 2012).Nevertheless, the SPG warming hole in models is always related to the reduction of SPG convection, and the proposed mechanism above can still be valid.Our study, combined with recent analysis that reduced deep-water formation over the SPG region would be more or less compensated by strengthening deep-water formation in other sources (Lique & Thomas, 2018;Årthun et al., 2023), may suggest a delayed collapse of the AMOC system under anthropogenic warming.Accordingly, any detection of critical slowing down in AMOC based on SPG warming hole needs to be interpreted with caution.Future studies are needed to better understand the interplay between the AMOC system and Earth's climate using longer-term observations and higher-resolution models with more complexity.

•
Relationship between Atlantic meridional overturning circulation (AMOC) fingerprints and AMOC can be nonstationary in a changing climate • Warming hole shows reduced sensitivity to AMOC in a warmer climate while the sensitivity of salinity pile-up is slightly strengthened • Differing vulnerability of AMOC sources and changed hydrological cycle influence the sensitivity of the two fingerprints, respectively

Figure 1 .
Figure 1.(a) Atlantic meridional overturning circulation (AMOC) evolution simulated in CMIP6 historical run (purple), ssp245 scenario (pink) and ssp585 scenario (red) shown in multi-model ensemble mean.(b) Same as (a) but for SSS index.(c) Same as (a) but for SST index.The scaling relationship between the two indices and the AMOC for period 2015-2100 are shown in (b) and (c), respectively.

Figure 2 .
Figure 2. (a) Pattern of sensitivity of SST change (relative to global mean change) to Atlantic meridional overturning circulation (AMOC) reduction (units: K/Sv; for period 2015-2100) in ssp245 scenario.(b) Pattern of sensitivity of SSS change to AMOC reduction (units: psu/Sv; for period 2015-2100) in ssp245 scenario.(c) Same as (a) but for ssp585 scenario.(d) Same as (b) but for ssp585 scenario.

Figure 3 .
Figure 3. (a) CESM2 simulated sensitivity of SST change (relative to global mean change) to Atlantic meridional overturning circulation (AMOC) reduction (units: K/ Sv; for period 2015-2100) and reduction rate of March-time mixed layer depth (MLD) (units: %) from 2015 to 2060 in ssp245 scenario.(b) Same as (a) but for ssp585 scenario.(c) CESM2 simulated SPG SST index (units: K) in ssp245 (orchid) and ssp585 (red).(d) Same as (c) but for SPG convection SST index.(e) Same as (c) but for March-time mean MLD (units: m) in the Labrador-Irminger Sea.(f) Same as (e) but for the Nordic Seas.(g) CESM2 simulated AMOC intensity in ssp245 (orchid) and ssp585 (red) (units: Sv).(h) Same as (g) but for contribution from the Nordic Seas convection.

Figure 4 .
Figure 4. (a-c) POP2 simulated sensitivity of SSS change to Atlantic meridional overturning circulation (AMOC) reduction (shading; units: psu/Sv) and simulated surface freshwater flux (SFWF) change (contours; units: 10 6 kg/m 2 /s) in (a) Case 2W, (b) Case 5W, and (c) Case 10W.(d-f) Same as (a-c) but for simulated sensitivity of SST change (relative to global mean change) to AMOC reduction (shading; units: K/Sv) and simulated surface heat flux change (contours: units: W/m 2 ).(g) Scatter diagram between change in AMOC and its indices in Case 2W (circle), Case 5W (triangle), and Case 10W (square).Also shown in (g) is the values of sensitivity of the two fingerprints in the three experiments with units in psu/Sv (red) for the SSS index and K/Sv for the SST index (blue), respectively.(h-k) March-time mixed layer depth (MLD) change in different convection sites in the three experiments.