We analyze four CMIP6 Earth System Models in the historical scenario (Eyring et al., 2016) and in the high CO₂ emission scenario, SSP5-8.5 (Wilbanks and Ebi, 2013; Riahi et al., 2017). These ten Earth System Models are listed in Table 1 and includes CESM2 (Danabasoglu et al., 2020), NorESM2-LM (Seland et al., 2020), ACCESS-ESM1-5 (Ziehn et al., 2020), MPI-ESM1-2-LR (Mauritsen et al., 2019), CMCC-ESM2 (Cherchi et al., 2019), CNRM-ESM2-1 (Séférian et al., 2019), UKESM1-0-LL (Sellar et al., 2019), MIROC-ES2L (Hajima et al., 2020), CanESM5 (Swart et al., 2019), and IPSL-CM6A-LR (Boucher et al., 2020). They have been selected for the availability of the physical and chemical variables ( Table 2 ) necessary for our analysis under future scenarios. The climate model simulations are obtained via the Earth System Grid Federation (ESGF) archive of CMIP6 data (https://esgf-node.llnl.gov). The partial pressure of atmospheric CO₂ is obtained from the input and forcing datasets for the Model Intercomparison Projects (input4MIPs, https://esgf-node.llnl.gov/search/input4mips/) in CMIP6 (Eyring et al., 2016; Meinshausen et al., 2017, 2020). The historical simulations cover the 1850 to 2014 period, and the future scenario simulation covers the period 2015 to 2100 with high radiative forcing under SSP5-8.5 (Meinshausen et al., 2020). Under the historical simulations, our analysis focuses on the period from 1950 to 2014. For all spatially-varying variables in Table 2 , the time series are computed as the area-weighted means within the defined sector of the subpolar North Atlantic basin, which encompasses the area from 40°N to 65°N latitude and from 55°W to 15°W longitude ( Figure 1 ).

As an indicator of simulated AMOC intensity, we use the definition of maximum meridional overturning stream function at 40°N and analyze its evolution over time, as commonly done in modeling studies (Knight et al., 2005; Zhang, 2008). The choice of 40° N is based on the fact that variations in AMOC exhibit their highest anomalies near this latitude (Buckley and Marshall, 2016). We verified that this is indeed the case in all models considered ( Supplementary Figure 1 ). Kinetic Energy is calculated based on velocities in the horizontal and vertical planes ( Table 2 ).

Variations in the ocean surface pCO₂ can be described as a combined effect of alterations in sea surface DIC, and alkalinity, as well as changes in SST and SSS (Sarmiento and Gruber, 2006), as follows:

where δ represents change over time and ∂ p C O 2 / ∂ X are partial derivative of surface ocean pCO₂ with respect to X. SST is sea surface temperature (°C), SSS is sea surface salinity (‰), DIC and ALK are ocean surface DIC and alkalinity (µmol kg^-1), respectively. This decomposition can be applied to understand the trends of surface ocean pCO₂. The temporal change of air-sea CO₂ flux critically depends on the air-sea difference in pCO₂. There would be a decline (increase) in air-sea CO₂ flux if ocean pCO₂ increases faster (slower) than atmospheric pCO₂. This analysis uses the CO2SYS program (Lewis and Wallace, 1998) to calculate each term in Equations 1, 2. The partial derivatives are calculated by making finite perturbations, and they quantify the sensitivity of ocean pCO2 to SST, SSS, DIC or ALK, while the terms ∂ p C O 2 ∂ X δ X are the contributions of SST, SSS, DIC or ALK to pCO₂ changes.

In order to emphasize the salinity impact on ALK and DIC, we separate ALK and DIC into salinity normalized (sALK and sDIC, affected by dynamics, biology and air-sea exchange), and diluted parts (dALK and dDIC, affected by freshwater) (Brix et al., 2013; Chikamoto and DiNezio, 2021; Chikamoto et al., 2023):

where 35 is the standard salinity value which normalizes DIC and ALK. Equation 2 includes three terms multiplied by δSSS. The first term is the direct effect of salinity, and the second and third terms are the dilution components for DIC and ALK. By design, these two terms tend to cancel out each other. In this study we pay attention to sALK and dALK terms to assess their relative importance in driving pCO₂ changes.

We investigate the difference between the CO₂ flux (positive from air to sea) among models by considering the spatial patterns of two 31-year periods: 1970-2000 within the historical time and 2070-2100 in the SSP5-8.5 scenario as representative for a high emission future ( Figure 2 ). In most models, the CO₂ flux shows a decrease in the western part of the SPNA while increasing in the eastern and northern parts ( Figure 2 ). One primary factor is the weakening of the AMOC ( Figure 3 ), which has been linked to strengthening stratification and shoaling mixing between density layers, both observed in recent decades and predicted for the future (Kim et al., 2023). There are, however, significant inter-model differences in the pattern of air-sea CO₂ flux change, some of which may be driven by the thermodynamic change due to the different SST responses to climate warming and AMOC slowdown. In the eastern and northern parts of SPNA, there are patches of cooling known as the North Atlantic Warming Hole (NAWH) (Drijfhout et al., 2012; Rahmstorf et al., 2015; Gervais et al., 2018; Menary et al., 2018; Keil et al., 2020; Oudar et al., 2020; Bellomo et al., 2021). AMOC weakening changes the relative proportions of warm subtropical waters and cold subpolar waters being transported into the eastern and western SPNA respectively. The weakening Labrador Current and the shift of the Gulf Stream result in colder and fresher waters south of Greenland due to the reduced transport of warm, salty water from the tropics and much lower SSS and higher sea surface height in the subpolar region (Claret et al., 2018; Bellomo et al., 2021; Gonçalves Neto et al., 2021). This alteration in water mass distribution can lead to significant CO₂ uptake changes.

Figure 3 presents the differences in AMOC stream functions from the CMIP6 models. Here we provide a comparative analysis of the mean AMOC magnitude between two timeframes: the historical period of 1970-2000 and the projected period of 2070-2100 in SSP5-8.5. All models exhibit a weakening of the AMOC by the end of the 21^st century. The magnitudes of AMOC decrease vary among the models, but broadly correlates with the initial AMOC strength ( Supplementary Figure 1 ): models with a stronger AMOC in historical times show a more pronounced decrease in the projections, especially in CESM2 and NorESM2-LM ( Figures 3A,B ; Supplementary Figures 1A, B ). While in models with a weak AMOC in the historical period, such as in IPSL-CM6A-LR ( Supplementary Figure 1J ), the reduction is moderate in the future ( Figure 3J ). The AMOC exhibits its maximum strength at around 40°N both during the historical period (1970-2000) ( Supplementary Figure 1 ) and in the projected future period (2070-2100), and the greatest reduction in AMOC strength occurs at this latitude according to the model simulations ( Figure 3 ). These findings about the AMOC trends in CMIP6 are consistent with the earlier study by Weijer et al. (2020).

Temporal trajectories of the physical variables from the ten CMIP6 Earth System Models are displayed in Figure 4 . Robust changes across all ten models in the high CO₂ emission scenario include AMOC weakening, surface freshening, and increasing SST, DIC and sea surface pCO₂ ( Figures 4A–E, G ). These changes can be attributed to the increasing atmospheric pCO₂ and the greenhouse effect but the rates of warming and DIC increases are model-dependent. Surface DIC generally increases due to the absorption of anthropogenic CO₂, but this is not the only factor controlling its growth. Warming can reduce the solubility of gases, leading to higher pCO₂ at the sea surface. Additionally, increased ocean stratification can impact the circulation, contributing to AMOC weakening.

In terms of AMOC, all models but MIROC-ES2L show a moderate increase in AMOC strength in the ~1950-1980 period, likely due to the aerosol forcing (Menary et al., 2020), followed by a decrease after the 1980s ( Figure 4A ). In the multi-model mean sense, AMOC weakens by about 9 Sv by the end of the 21^st century ( Figure 4A ). Surface freshening (salinity decrease) may be driven by the freshwater fluxes including precipitation, ice melt, and runoff as well as internal dynamics including changing transport of salty subtropical waters to the SPNA by the weakened AMOC. A significant decline of surface alkalinity ( Figure 4E ) is linked to the reduction in SSS and AMOC ( Figures 4A, C ) in all models. The reduction in alkalinity accelerates the acidification of the surface water and decrease the buffering capacity. In the SPNA, all models show some decline of surface alkalinity associated with freshening ( Figures 4C, E ). Under the warming climate, the reduction in AMOC transport and the stronger freshwater fluxes together act to decrease the surface alkalinity by freshening or dilution, which tends to increase surface ocean pCO₂. On interannual and longer timescale, surface ocean pCO₂ increases, following the rising atmospheric pCO₂. All models exhibit increasing ocean pCO₂ of about 800 µatm at the end of the 21^st century in SSP5-8.5 ( Figure 4G ). Air-sea CO₂ exchange adjusts surface ocean pCO₂ to the atmospheric pCO₂ with some lag as determined by the air-sea equilibration timescale. Carbonate chemistry dictates that the increase must be achieved by the combination of the four factors in Equation 2. The increase in surface ocean pCO₂ induced by alkalinity decrease reduces the regional carbon sink by limiting the ocean pCO₂ growth by the air-sea equilibration and resultant DIC increase. This effect is especially strong in models with significant freshening (SSS decline) which is mechanistically linked to the AMOC slowdown. The pattern of temporal changes in the air-sea pCO₂ difference (dpCO₂) generally matches that of air-sea CO₂ flux (fgCO₂) because dpCO₂ drives fgCO₂ ( Figures 4H, I ). Thus, any factor that increases the dpCO₂ gradient can also increase the magnitude of fgCO₂, assuming other conditions remain the same. The trends in fgCO₂ are influenced by several factors beyond the dpCO₂ due to the dependence of air-sea gas transfer on wind speed and seaice cover in polar regions. Hence, there are some small discrepancies between the trends and variability of fgCO₂ and that of dpCO₂ (Jing et al., 2022).

CESM2 and NorESM2-LM exhibit similar trends in terms of AMOC and SST ( Figures 4A, B ). The strong AMOC weakening (~18 Sv) in CESM2 and NorESM2-LM is associated with the formation of a strong NAWH by the end of 2050. However, despite their similar model components, trends in SSS, DIC, and alkalinity differ between CESM2 and NorESM2-LM. Those differences could be attributed to how each model simulates ocean biogeochemical processes, freshwater inputs, and atmospheric CO₂ exchange dynamics ( Figure 4 ). The less pronounced increase in DIC and the strong decrease in alkalinity in CESM2 might suggest a different representation of the freshwater fluxes, ocean circulation and mixing patterns, which can affect the regional carbon cycle. Even with the similar SST trends, the stronger freshening and alkalinity loss would weaken the regional carbon uptake, explaining the relatively weaker growth of surface simulated DIC and weaker carbon uptake (fgCO₂) in CESM2 compared to NorESM2-LM ( Figures 4B–E, I ). Differences in precipitation, evaporation, river runoff, and ice melt could cause variations in SSS and consequently influence CO₂ uptake and alkalinity levels. While CESM2 and NorESM2-LM exhibit similar AMOC and kinetic energy (KE) changes, there are indeed some minor differences in the evolution of kinetic energy ( Figures 4A, F ). Here, the KE is calculated as a measure of ocean transport strength in addition to AMOC intensity, and it includes other aspects of the ocean circulation system such as boundary currents (Irminger and Labrador Current), wind-driven subpolar gyre, as well as AMOC. The KE decrease in CESM2 is less pronounced after 2060s in comparison to NorESM2-LM indicating somewhat different circulation changes in the late 21^st century. This could reflect the potential regional variations in freshwater fluxes and wind stress, and different parameterizations choices in the two models.

CanESM5 exhibits both the lowest dpCO₂ trend for most of the 21^st century, and the most significant SST rise ( Figures 4B, H ) thus reduced CO₂ solubility. This model exhibits some reduction in fgCO₂, though it is not the most substantial, and the fgCO₂ decrease occurs in conjunction with an increase in KE in the projections ( Figures 4F, I ).

Conversely, ACCESS-ESM1-5 shows the most considerable increase in fgCO₂, consistent with minimal projected change in SSS ( Figures 4C, I ). This model’s moderate AMOC attenuation (~7 Sv) and limited ocean freshening suggest that atmospheric pCO₂ increase predominantly drives the ocean CO₂ uptake with no opposing effects from alkalinity.

So far, our discussion focused on interpreting different evolution of physical and biogeochemical variables. Here we decompose ocean pCO₂ and quantify the effect of SST, SSS, DIC and ALK on the change of dpCO₂ with the purpose of quantitatively assessing different drivers of pCO₂ increase on centennial timescale ( Figure 5F ). pCO₂ decomposition indicates that the overall change in pCO₂ can be expressed as the sum of four factors (SST, SSS, DIC, ALK) where each factor is multiplied by its respective sensitivity (Equations 1, 2). Figure 6A shows the trends in oceanic pCO₂ over time in the SPNA, alongside contributions from various factors that influence pCO₂ levels. The increasing trend in pCO₂ signifies a growing partial pressure of CO₂ in the ocean’s surface. ALK contributes to this increasing trend and, by the end of the 21^st century, ALK becomes the most important contributor to pCO₂ change in the multi-model mean sense. The influence of sALK is close to zero, suggesting that changes in salinity-normalized ALK alone do not account for the change in pCO₂ trend. ALK affects pCO₂ due to dilution effects from freshwater fluxes and the weakened ocean salinity transport. The SST contribution to the increasing pCO₂ trend shows a moderate contribution by the warming-induced solubility decrease. The DIC contribution indicates that increases in DIC are driving pCO₂ upward, as more CO₂ is being added to the ocean. The SSS contribution is less significant but slightly negative, implying that a decrease in salinity impacts carbonate chemistry and CO₂ solubility. Overall, the increasing pCO₂ trend appears to be driven by ALK, DIC and SST changes, with SSS playing a minimal role.

By examining the temporal evolution of pCO₂ sensitivity to all components, the sensitivity increases over time due to the changing background states of carbonate chemistry over the centennial timescale considered ( Figure 6B ). The nearly opposite sensitivities to DIC and ALK indicate that mutually compensating pCO₂ response for synchronous changes in DIC and ALK. This can be easily understood considering the changes induced by a pulse of freshwater flux. An input of freshwater can decrease DIC and ALK at similar rates, leading to a muted response of surface ocean pCO₂ over a relatively short timescale. On the interannual and longer timescales, the gradual increase in atmospheric pCO₂ and the resultant ocean carbon uptake increase DIC but not ALK ( Figure 6B ). Thus, increasing DIC and decreasing ALK together increase the surface ocean pCO₂ on longer timescales. The increasing acidity of seawater increases the buffer factor which is approximately equal to [H⁺]/K₂ where K₂ is the equilibrium constant for the dissolution reaction of bicarbonate ion into carbonate ion. The acidification of seawater (increase in [H⁺]) directly causes the increase in buffer factor and the sensitivity of pCO₂ to DIC on centennial timescales.

We further investigate the difference of pCO₂ components between two 31-year periods: 1970-2000 as representative of the late 20^th century and 2070-2100 in the SSP5-8.5 scenario as representative of the future towards the end of the 21^st century ( Figure 7 ). Over these two periods, all models show AMOC slowdown in the future scenario, together with increasing pCO₂ in the atmosphere and in the surface ocean ( Figure 7 ). The contribution of SST to the surface ocean pCO₂ change is relatively small in comparison to alkalinity and DIC. ALK contributes the most to pCO₂ increase in CESM2, NorESM2-LM, CanESM5, MIROC-ES2L, UKESM1-0-LL, and CNRM-ESM2-1 while in the other four models’ the pCO₂ increase is mainly driven by DIC contributions ( Figure 7A ).

The pCO₂ trends associated with DIC and ALK are further decomposed using the salinity-normalized and the dilution (freshwater) components. The DIC contribution of pCO₂ trends can be understood as a residual between the increase in salinity-normalized DIC (affected by air-sea CO₂ exchange, dynamics and biology) and the decrease in the dilution component. For most models, the salinity-normalized DIC dominates except for CESM2 ( Figure 7B ).

ALK contribution to pCO₂ changes relies on the freshwater (dilution) component of ALK decrease except for IPSL-CM6A-LR. In models with stronger SSS decrease, ALK contributes more to increased surface ocean pCO₂ and weakened ocean CO₂ uptake. Thus, the mechanisms behind the representation of weakened CO₂ uptake in the late 21^st century are model dependent. In ACCESS-ESM1-5, the ALK and dALK contributions are lower in 2070-2100 than that in 1970-2000. In this particular model, ALK makes a small contribution to pCO₂ change due to the sudden shift in SSS, DIC and ALK in the last 5 years of the 21^st century (2096-2100, Figures 4C–E ). If we exclude this event by adjusting the analysis window to 2065-2095 in ACCESS-ESM1-5, the dilution component (dALK) clearly controls the pCO₂ change in this model, and its fgCO₂ decreases significantly ( Supplementary Figure 3 ). CESM2 and NorESM2-LM both exhibit a pronounced AMOC slowdown, consistent with strong fgCO₂ decreasing trends ( Figure 7A ). What distinguish CESM2 and NorESM2-LM are the magnitude of the AMOC reduction, and the resultant ALK-driven pCO₂ increase is much greater than that of DIC in comparison to other models ( Figure 7 ). IPSL-CM6A-LR and MPI-ESM1-2-LR, on the other hand, show the weakest AMOC slowdown associated with the least fgCO₂ decrease ( Figure 7A ). In these models, the ALK contributions to pCO₂ trend are also the weakest ( Figure 7B , Supplementary Figure 3 ). The correlation between AMOC changes and ALK contribution to the fgCO₂ trends can be explained through salinity changes.

To understand the relationship between ALK and AMOC with carbon uptake, we examine the differences between the two 31-year periods (1970-2000 vs 2070-2100) among the ten models. Figure 5A illustrates a strong positive correlation between changes in ALK and SSS across climate models, indicating that a decrease in ocean SSS is typically accompanied by a decrease in ALK. This means that the ALK change is mainly caused by the freshwater effect and by changes in the northward transport of salinity by the AMOC. This is reinforced by a notable positive correlation between ALK differences and changes in AMOC strength, suggesting that the decrease in ALK is associated with the reduction in AMOC ( Figure 5B ). Furthermore, a moderate positive correlation is observed between ALK differences and oceanic carbon uptake, implying that an increase (or less dilution) in ALK could increase the ocean’s capacity to sequester carbon ( Figure 5C ). However, this correlation is relatively small (R² = 0.23) indicating that there are other mechanisms at play. Indeed, Goris et al. (2018) demonstrated that the vertical transport of DIC plays an important role, which is reflected in the seasonal amplitude of surface ocean pCO₂. This factor alone, however, cannot explain the future evolution of carbon uptake in the North Atlantic in the models. Furthermore, the relationship between SST differences and carbon uptake is weak in the models considered, indicating that SST variations unlikely control carbon uptake changes ( Figure 5D ).

Our analysis identified the inter-model relationship modulating the centennial changes in ALK and AMOC, which are quite strongly correlated (R² = 0.65). AMOC changes may have a profound impact on ALK in the SPNA, through changes in freshwater fluxes and in the transport of subtropical water masses to SPNA ( Figure 5B ). AMOC-induced changes in ALK in turn influence the ocean’s ability to absorb CO₂, despite the moderate direct correlation between ALK and carbon uptake. Considering the above pCO₂ decomposition analysis, ALK changes can affect the regional pCO₂ significantly, and more so in the future than in historical times ( Figure 7 ). Furthermore, the AMOC changes are moderately correlated (R² = 0.43) with trend in fgCO₂ indicating that a diminished convective mixing further decreases the carbon uptake ( Figure 5E ). This analysis highlights the importance of ALK reduction, which increases surface ocean pCO₂ and reduces the ocean’s capacity to absorb CO₂. These correlations underscore the interconnectedness of ocean circulation, chemical composition, and the carbon cycle in the Earth’s climate system.

This study examined how the representation of the AMOC and its future evolution modulates ocean carbon uptake in the SPNA in ten of the CMIP6 Earth System Models through the analysis of physical and biogeochemical changes on centennial timescales. The AMOC slowdown is generally associated with SSS and ALK decrease, along with increased DIC in the multi-model mean sense, consistent with previous studies (Jackson et al., 2015; Schmittner and Lund, 2015; Zhang et al., 2019; Chikamoto et al., 2023; Sinet et al., 2023; van Westen and Dijkstra, 2023). The extent of these changes can vary across models. For example, CESM2 and NorESM2-LM project pronounced AMOC reductions, which are correlated with modest SST increases due to an intense NAWH (Keil et al., 2020; Bellomo et al., 2021; Swingedouw et al., 2021; Noël et al., 2022; Iversen et al., 2023) and more significant reductions in SSS and ALK ( Figure 4 ). Upper ocean currents weaken in these models, and these changes are aligned with a strong decrease in KE ( Figure 4F ) (Gregory and Tailleux, 2010). In contrast, CanESM5 displays weaker AMOC slowdown, notable SST increases and moderate reduction in SSS and ALK ( Figure 4 ). This model shows a drop in dpCO₂, suggesting that the intensified warming raises ocean pCO₂ ( Figure 4H ). The SPNA in CanESM5 continues to absorb CO₂ until the late 21st century while affected by strong warming, moderate reduction in ALK and moderate AMOC weakening ( Figure 4 ).

The models characterized by strong AMOC mean states in the historical period exhibit both a more pronounced slowdown of AMOC (CESM2/NorESM2-LM) ( Supplementary Figure 1 ) (Gregory et al., 2005; Weijer et al., 2020; Bellomo et al., 2021) and a considerably weaker cumulative CO₂ uptake compared to the models with weaker AMOC slowdown (IPSL-CM6A-LR/MPI-ESM1-2-LR) ( Figure 7A ). A recent study suggests that under high CO₂-emission scenarios, the area to the southeast of Greenland is expected to reach the point of anthropogenic CO₂ uptake saturation earlier in the 21^st century, while under low CO₂-emission scenarios, it will transit from a carbon sink to a carbon source for the atmosphere around 2080 (Ridge and McKinley, 2021). Two mechanisms that link AMOC weakening and reduced CO₂ uptake include weakened subduction of surface water (DeGrandpre et al., 2006; Goris et al., 2018) and alkalinity decrease due to freshening of surface waters in the SPNA. These two mechanisms are not exclusive to one another, and both affect the air-sea CO₂ transfer. Their relative importance is model-dependent. Having said this, we can point to the most plausible explanation in the model context considering CESM2, which has the strongest AMOC of the ten models in the historical period, and displays the strongest decline as well. The high correlations between AMOC, SSS and ALK link the AMOC slowdown to the alkalinity decline and resultant weakened ocean carbon uptake ( Figure 5 ), while also explaining the reduced rate of increase or even decline in surface DIC concentrations ( Figure 4D ). Warming-induced solubility decline is also sensitive to the magnitude of AMOC slowdown, but the SST difference has little correlation with carbon uptake in the future scenario ( Figure 5D ). In CESM2 (or NorESM2-LM), the strong AMOC slowdown causes the SPNA to reach the maximum rate of CO₂ uptake earlier than the inflection point of the transient atmospheric CO₂ levels ( Figure 4I ) (Terhaar et al., 2022). In the SSP5-8.5 scenario, where the global CO₂ emission is assumed to increase through the 2080s, their regional ocean carbon sink reaches its peak and then becomes less effective by ~ 2040-2060 ( Figure 4I ). The increasing contribution of ALK changes to the pCO₂ in the SPNA ( Figure 6A ) is unique to this region. The ALK contribution to pCO₂ increase is relatively modest both globally and in the Southern Ocean ( Supplementary Figure 3 ), suggesting a distinct influence to the North Atlantic that may warrant further investigation.

Ocean carbon uptake can reach its peak at an earlier period when the biogeochemical response to physical climate changes (including the AMOC slowdown) further reduces the ocean’s ability to absorb atmospheric CO₂ (Maier-Reimer et al., 1996; Humphreys et al., 2018; Stock et al., 2020). The surface ocean DIC concentrations increased during the 20^th century due to the uptake of anthropogenic carbon ( Figure 4D ) (Wanninkhof et al., 2010; Jiang et al., 2023; Keppler et al., 2023; Ma et al., 2023). In the high emission scenario, the atmospheric pCO₂ levels continue to increase and the surface ocean DIC generally follows the atmospheric forcing trend ( Figures 4D, G ). However, there are significant model differences in the surface DIC future trajectories. CESM2, for example, stands out for its weak DIC increase even in the highest emission scenario during the late 21^st century ( Figure 4D ). A weakened biological pump may potentially explain the reduced ocean carbon uptake (Passow and Carlson, 2012; Barange et al., 2017), as a weakened export of organic matter from the surface ocean would leave behind higher DIC concentrations (Tian et al., 2015).

We complete this work by presenting the inter-model correlations of variables which reveal diverging behaviors among the ten models ( Figure 5 ). There is generally a strong positive correlation between changes in ALK and SSS which are both declining in tandem, due to freshwater influences and weakened transport of saline subtropical waters to the SPNA region ( Figure 5A ) (Lauvset and Gruber, 2014). The ALK dependence extends to the AMOC, as a decrease in AMOC strength is associated with a reduction in ALK ( Figure 5B ). These alterations in ALK appear to moderately affect the ocean’s capacity for carbon uptake (Takahashi et al., 2014; Middelburg et al., 2020; Chikamoto et al., 2023). The findings suggest that the interplay between AMOC and ALK is significant, as AMOC’s role in transporting chemically distinct water masses directly impacts ALK levels, which in turn, indirectly influence the CO₂ sequestration capabilities of the ocean. Consistent with our work, previous studies have established that models simulating lower present-day SSS in the SPNA tend to reduce future carbon absorption (Estella-Perez et al., 2020; Park et al., 2023). As our results indicate, the likelihood that SSS and ALK will diminish in the future is very high, and consequently, the SPNA will not help mitigate global warming given the substantial reduction in its ability to uptake CO₂.

This work analyzed CMIP6 models with the goal of identifying processes or quantities that could help us distinguish among behaviors that are indicative of the future evolution of ocean circulation and carbon cycling in the SPNA. There is indeed an urgent need to better quantify the carbon dioxide removal potential in the ocean.

In summary, this study underscores the complex relationship among AMOC dynamics, ALK, surface ocean salinity and the ocean’s capacity to uptake atmospheric carbon, offering insights into the intricate interactions between physical variables in SPNA. Despite the different extent of the decline in AMOC in the ten of the CMIP6 models considered, all models agree on the trend direction ( Figure 4A ). A stronger AMOC slowdown correlates with less ocean carbon uptake (Schmittner and Galbraith, 2008; Nielsen et al., 2019; Katavouta and Williams, 2021). This study emphasized the relatively underappreciated role of diminishing ALK and its impact on regional carbon uptake. Despite the complexity of the relationships among the variables analyzed, it clearly emerges that changes in ALK are an important factor in the oceanic absorption of CO₂ (Takahashi et al., 2014; Middelburg et al., 2020; Chikamoto et al., 2023). Moving forward, a deeper dive into the mechanistic links between these variables is imperative to enhance our understanding of their collective impact on the carbon cycle within the Earth’s climate system.

Observations are still limited, while being the best tool to identify specific quantities and relationships in both physics and biogeochemistry. Model simulations will continue to serve as a key tool to understand ocean physics and biogeochemistry under changing climate (Menary et al., 2020; Stock et al., 2020; Weijer et al., 2020; Swingedouw et al., 2021; Robson et al., 2022). The role of the biological pump on carbon uptake and its future evolution remains largely uncertain (Middelburg et al., 2020; Seferian et al., 2020). More work is needed to better understand what drives the observed export of particle organic carbon and why models generally misrepresent both mean and variability. A better characterization of the biological pump and of carbon export will enhance our ability to predict the evolution of the marine ecosystem in the SPNA and is especially important considering the monitoring, reporting, and verification (MRV) required for the deployment of carbon removal technologies.

As final note of caution, we should keep in mind that all climate models present biases that limit their reliability, and the robustness of common behaviors could be tested paying attention to the outliers. For instance, in our analysis ACCESS-ESM1-5 stands out for its strong correlation of carbon uptake with ALK or SST ( Figures 5C, D ), possibly indicating an anomaly or significant deviation in the model’s simulation relative to others. This could be due to unique model parameterizations, sensitivities to specific inputs, or how ACCESS-ESM1-5 simulates the interactions between oceanic processes and the carbon cycle (Ziehn et al., 2020). To understand the cause of this outlier behavior, a detailed examination of the model’s internal processes and assumptions is necessary, which may involve a closer look at the representation of ocean chemistry, circulation patterns, and their response to changing climatic conditions. Further diagnostic studies comparing ACCESS-ESM1-5 with other models and observational data would be beneficial to determine the reasons behind its distinctive response.