Introduction

The Southern Ocean (south of 30°S) takes up approximately 40% of ocean anthropogenic CO2 and ~75% of excess heat1,2,3,4, making it one of the most pivotal ocean buffer of climate change. In addition, it supplies 33 – 75% of the nutrients required for new primary production in the global oceans5,6,7. Nevertheless, existing model projections indicate large uncertainty in the future Southern Ocean sink of anthropogenic CO2 emissions8. In recent years, emergent constraints have shown some success in constraining uncertainty in the Southern Ocean CO2 sink projections8,9, nevertheless, changes in future mechanisms remain poorly understood. Understanding how climate change alters the Southern Ocean’s ability to regulate CO2 and heat exchanges, and their governing mechanisms is crucial to strengthening our confidence in the simulated future changes. Further, improved process understanding of the behavior of the Southern Ocean under extreme conditions like the high-emission scenario is essential to anticipate related ecosystems impacts and climate feedbacks in the future.

The intergovernmental panel on climate (IPCC) high emission-scenario projects over 1000 ppm atmospheric CO2 ( ~ 2.5-factor increase) and associated warming of up to 4.5 °C by 2100 in the global ocean surface relative to the contemporary climate10,11. Simulated ocean related changes include, but are not limited to, sea-ice loss, ocean acidification, deoxygenation, and a decrease in CO2 buffer capacity12,13,14,15,16,17. These predicted changes are key drivers to biogeochemical, ecosystems and physical ocean related gradual and abrupt regime shifts in projections. The decrease of the CO2 buffering capacity for example is projected to enhance biological-induced CO2 uptake in summer, and amplify the seasonal cycle of air-sea CO2 fluxes (FCO2) as well as acidification both of which may have a significant implications for the efficiency of the ocean CO2 uptake and marine calcifying organisms in the Southern Ocean18.

In this study, we investigate how these projected anthropogenically-driven changes will impact the magnitude and mechanisms of the Southern Ocean CO2 uptake between the contemporary and the last decades of the 21st Century. We use nine state-of-art Earth System Models (ESMs) focusing on the historical and future high emission (SSP5-8.5) scenarios. The Southern Ocean is defined here as the region south of 30 °S, divided into three subdomains using dynamical frontal boundaries to reflect the three regimes of the Southern Ocean (Fig. 1). Namely, (i) the Subtropical region, the warmest and most stratified part of the Southern Ocean, north of the Subtropical Front, (ii) the Sub-Antarctic region, a highly productive, moderately warm region that hosts the subduction of Subantarctic Mode and Antarctic Intermediate Waters, located between the Subtropical Front and the Polar Front; and (iii) the Antarctic region, south of the Polar Front, is characterized by the seasonal sea-ice cover, which regulates seasonal upwelling and primary production.

Fig. 1: Contemporary air-sea CO2 flux (FCO2) in the Southern Ocean.
figure 1

Obtained from a the ensemble mean of six pCO2 products and c nine CMIP6 ESMs for the historical period (1995–2014), and b, d their variability computed as one standard deviation. FCO2 is given gC m-2 yr-1 in all panels. Negative values indicate a flux into the ocean. The fronts are defined according to Orsi et al.77 as black lines with the Subtropical Front to the North and the Polar Front to the South. The zones are defined as the subtropical region north of the Subtropical Front (outer line) to 30oS, the Sub-Antarctic region between the two fronts, and the Antarctic region to the south of the Polar Front (inner line). Individual models are shown in Fig. Supplementary Fig. 7.

Full size image

Results

Air-sea CO2 flux mean state change between present and future climate

Observation-based, pCO2-products annual mean air-sea CO2 fluxes in the Southern Ocean are characterized by a strong CO2 uptake within and north of the Subtropical Front (Subtropical region), a weak ingassing flux between the Subtropical and Polar Front (Sub-Antarctic region) and a weak outgassing flux south of the Polar Front (Antarctic region), (Fig. 1a, see Methods for description and data available at Supplementary Data 1). ESMs broadly show good spatial agreement with the pCO2-products ensemble north of the Subtropical Front (present climate 1995–2014, Fig. 1c). Differences between ESMs and pCO2-products in the Subtropics mainly occur in the eastern Pacific, where most of the ESMs show a CO2 outgassing feature. ESMs annual means are also comparable, overlapping within the envelope of the model spread with the pCO2-products ensemble in the Subtropics; -15.5 ± 2.4 gC m-2 yr-1 (ESMs) and -16.4 ± 1.1 gC m-2 yr-1 (pCO2-products), (Fig. 2a). In the Sub-Antarctic region, ESMs display a relatively larger model spread in comparison to the Subtropics, showing the largest disagreement in Malvina’s sea (> 10 gC m-2 yr-1) and south of Australia (~ 9-10 gC m-2 yr-1). The ESMs ensemble mean (-9.0 ± 4.5 gC m-2 yr-1) for the Sub-Antarctic is nonetheless comparable to the observationally-derived pCO2-products ensemble mean (-11 ± 0.8 gC m-2 yr-1) (Fig. 2b). The Antarctic region shows a comparable model spread with the Sub-Antarctic but does not overlap with the pCO2-products. The Antarctic region is the weakest CO2 sink of the three subdomains in the contemporary Southern Ocean, showing an annual mean air-sea CO2 flux (FCO2) of -8.4 ± 5.4 gC m-2 yr-1 in ESMs and -1.81 ± 1.5 gC m-2 yr-1 in pCO2-products (Fig. 2c).

Fig. 2: Present climate vs. end of the 21st century air-sea CO2 fluxes in the Southern Ocean.
figure 2

a The Subtropical region on the right, b the Sub-Antarctic in the middle, and c Antarctic region on the left panel for the ensemble mean from nine CMIP6 ESMs, and the ensemble mean of six pCO2 products for the historical period (1995–2015, blue and gray), and end of the 21st century (2080–2099, orange). The error bars reflect one standard deviation of the inter-model (product) spread. Individual models are shown in Supplementary Fig. 8.

Full size image

At the end of the 21st century (2080–2099) a remarkably consistent pattern emerges in the ocean carbon uptake across ESMs; the region of the weakest CO2 sink in the contemporary climate becomes the most intense sink. Namely, the region of the strongest CO2 sink shift poleward from the Subtropics to the Antarctic (Fig. 2a–c). By the end of the 21st century, Subtropics contributes only about 23% (-22.0 ± 2.5 gC m-2 yr-1) of total Southern Ocean CO2 uptake under the high-warming scenario in comparison to 47% in the contemporary climate. The CO2 sink in the Subtropics increase by the smallest margin (6.6 ± 1.1 gC m-2 yr-1) in comparison of other Southern Ocean subdomains. The Sub-Antarctic region contributes 29% (-27.7 ± 5.4 gC m-2 yr-1) to the projected future, which is comparable to 27% in the present climate. The Antarctic region on the other hand displays the most extensive CO2 sink increase of about 450% (-37.9 ± 7.3 gC m-2 yr-1), becoming the largest CO2 sink region (48%) at the end of the 21st century (Figs. 1–2). The Antarctic region also has the largest annual mean FCO2 model spread in the projected climate (Fig. 2c).

Mechanisms of air-sea CO2 fluxes in the present climate

Air-sea CO2 fluxes (FCO2) are regulated by the thermodynamic and kinematic forcings19. The thermodynamic forcing, the air-sea pCO2 gradient (ΔpCO2), is considered the primary driver of FCO2; it determines the direction of the flux20. The kinematic forcing, on the other hand, controls the efficiency of gas transfer, and it is principally regulated by near-surface wind speeds. We note that kinematic forcing can induce indirect effects on the surface pCO2, e.g., through changing the ocean circulation or water mass ventilation patterns21. On short timescales (hourly to weekly), kinematic forcing can also determine the magnitude and direction of FCO222,23,24,25. However, ΔpCO2 plays a leading role in seasonal-scale FCO2 variability 26,27. Therefore, mechanisms regulating FCO2 variability can be estimated from processes regulating ΔpCO2. Further, considering that atmospheric pCO2 is almost uniform in the Southern Ocean, ΔpCO2 is ultimately controlled by the ocean pCO2. Indeed, observed and ESMs ΔpCO2 properties broadly delineate a similar latitudinal structure with FCO2 (Figs. 2 and 3), showing a strong annual mean CO2 ingassing flux in the Subtropics consistent with negative ΔpCO2 (~-40 µatm) and decreasing poleward; > -30 µatm in the Sub-Antarctic, and near-zero value in the Antarctic region (Fig. 3a–c). Given that ESMs mean state magnitudes differ for some variables (e.g. pCO2 and dissolved inorganic carbon (DIC) among others), comparing a multi-model seasonality is often impractical. Henceforth, we instead use monthly rates of change (first-order temporal derivatives, see Methods) for selected variables to highlight the changes in model and observed features at the seasonal scale.

Fig. 3: Seasonal and annual mean zonally averaged delta air-sea pCO2 (ΔpCO2) in the Southern Ocean.
figure 3

a–c from the ensemble mean of six pCO2 products and nine CMIP6 ESMs for the present climate (blue line) and the end of the century (2080–2099, orange). d–f Net changes in the projected future climate (note the change of scale). The first, second, and third columns depict the austral summer (DJF), austral winter (JJA) averages, and the annual mean, respectively. The vertical gray lines denote the frontal positions, the solid line the Polar Front, and dotted the Subtropical Front. The shading in all figures shows the uncertainty; inter-model one standard deviation. Individual models are shown in Supplementary Fig. 9.

Full size image

Surface ocean pCO2 and ΔpCO2 variability is controlled by the relative contribution of thermal and nonthermal components28,29. The thermal component can empirically be estimated using the Takahashi et al.30 formulation30, and we estimate the nonthermal component by subtracting the thermal component from the total (Fig. 4, see Methods). The thermal component is driven by temperature variations through changes in gas solubility (Henry’s law). The nonthermal component is mainly controlled by mixing and biology 31, it also includes the role of total alkalinity and salinity changes. However, total alkalinity and salinity have been shown to play a minor role in the seasonal cycle of ocean pCO2 in the contemporary Southern Ocean29,31, thus, here we focus on processes responsible for sources and sinks of DIC, i.e., primary production, respiration, and changes in seasonal mixing. The thermal and nonthermal components oppose each other on a seasonal scale32,33 (Fig. 4d–f), and hence the larger of the two determines the observed seasonal cycle phasing of pCO2, and ultimately FCO229. The relative contributions of thermal and nonthermal components are assessed here through a metrics that calculates the absolute difference in their monthly rates (MT-nonT, Eqs. 1–3, Methods). MT-nonT > 0 µatm mon-1 (Fig. 4j–l) denotes periods where (and when) the thermal component has a stronger influence on monthly pCO2 changes than nonthermal processes (mixing and biology), and vice versa for MT-nonT < 0 atm mon-1. For the contemporary climate, the observed ΔpCO2 seasonality in the Subtropics is broadly controlled by thermal processes, has a near-equal contribution of the two components in the Sub-Antarctic, and is predominantly nonthermally controlled in the Antarctic region (Fig. 4a–c & j–l). ESMs broadly display good inter-model agreement in MT-nonT with respect to the observed estimates in the Subtropics but show a degrading comparability poleward (Fig. 4d–l). Thermal constraints on CO2 solubility that regulate the seasonal cycle of pCO2 in the Subtropics are as follows: negative ΔpCO2 magnitudes in the autumn-winter seasons are due to higher gas solubility in cooler seasons, while the opposite is true in warmer seasons (Fig. 4a, d). The strong thermal dominance in Subtropics is partly because the largest seasonal temperature contrast (summer-winter difference) in the Southern Ocean occurs in the northern edge and decreasing poleward (Supplementary Fig. 2b). The Subtropics, therefore, has the highest seasonal warming and cooling rates in the contemporary Southern Ocean (Supplementary Fig. 3) and thus larger (dpCO2/dt)T magnitudes in comparison to (dpCO2 /dt)nonT (Fig. 4d). This thermally dominated seasonal cycle in the Subtropics is well captured by ESMs; they show a relatively small model spread and compares well with the observed estimates (Fig. 4a). In the Sub-Antarctic region, the seasonal scale contribution of thermal and nonthermal processes have a near equal contribution (MT-nonT ~ 0 atm mon-1) and hence the seasonal cycle of ΔpCO2 is generally less defined in comparison to the Subtropical and Antarctic regions (Fig. 4b, e & k). A full description of the thermal and nonthermal processes for the contemporary Sub-Antarctic region is provided in Supplementary Note 1.

Fig. 4: The seasonal cycle of delta air-sea pCO2 (ΔpCO2) and drivers of surface pCO2 in the Southern Ocean.
figure 4

a–c The seasonal cycle of ΔpCO2 for the ensemble model mean in blue and pCO2-products in black. d–i seasonal cycle of the thermal and nonthermal components of pCO2 changes for present climate d–f and the end of the 21st century g–i. j–l Seasonal cycle of thermal-nonthermal regimes from the metrics MT-nonT (Eq. 4, Methods). MT-nonT > 0 atm mon-1 denotes periods where the thermal component is leading monthly pCO2 changes, while MT-nonT < 0 atm mon-1, denotes dominance of nonthermal processes (biological and mixing). The shading in all figures shows the inter-modal variability, computed as the standard deviation. Individual models are shown in Supplementary Fig. 10.

Full size image

In the Antarctic region, the seasonal cycle of pCO2 is primarily nonthermally controlled for both pCO2-products and ESMs in the contemporary climate (Fig. 4 l), ESMs show a large spread for both ΔpCO2 and MT-nonT in this region. The nonthermal processes that can be isolated from model outputs are net primary production, respiration and remineralization, indicators of mixing and stratification, and changes in sea-ice cover. We will use the apparent oxygen utilization (AOU) to estimate respiration, and will decompose the total rate of change of DIC into preformed and regenerated components (Figs. 5 and 6). AOU is defined as the difference between oxygen at saturation and the in situ dissolved oxygen concentration and it is used here to estimate respiration within the MLD34. Physical mixing processes will be here diagnosed through the rate of change of SST, mixed layer depth (MLD) and a stratification index based on the vertical density gradient (Figs. 5, 7). In addition, sea-ice is an essential distinguishing feature of Antarctic FCO2 properties in comparison to the Sub-Antarctic and Subtropics. The seasonal presence of sea ice limits surface heat fluxes35,36 constraining the SST to near freezing (Supplementary Fig. 1). This keeps the Antarctic SST rate of change relatively low (Fig. 5j & Supplementary Fig. 1), and hence the observed thermal pCO2 component is lower than in the Subtropics and Sub-Antarctic region (Fig. 4f). While some ESMs shows this feature (e.g. CanESM5 and UKESM1-0LL, Supplementary fig. 3.), not all are consistent with the observed estimate, other models show a larger than observed rate of change of SST (Fig. 5j). ESMs also show a large model spread in the MLD (Fig. 5c, d) and stratification (Fig. 7a, b), which is overestimated in the Antarctic region. Model temperature bias in the Southern Ocean is a well-known feature of CMIP models since inception37. Some studies have linked this temperature bias to discrepancies in the simulated Atlantic Meridional Overturning Circulation (AMOC)38,39, and other studies suggest that cloud-related biases overestimate the incoming shortwave flux40. These compounded effects influence the vertical heat exchanges and stratification in the Antarctic region, and result in substantial model biases in the seasonal warming and cooling rates.

Fig. 5: The seasonal cycle of the physical and biological drivers of surface pCO2.
figure 5

Seasonal cycle of a, b net primary production in the surface ocean (NPP) and c, d mixed layer depth (MLD), monthly rates of change of e, f surface dissolved inorganic carbon (dDIC/dt), g, h apparent oxygen utilization rate (dAOU/dt, g, h), and i, j surface temperature (dSST/dt) in the Sub-Antarctic region (left) and the Antarctic region (right). The blue lines are the present climate (1995–2014) and the orange lines are the end of the 21st century (2080–2099). Black lines depict observational estimates. The shading in all figures shows the inter-model standard deviation.

Full size image
Fig. 6: The decomposition of surface dissolved inorganic carbon (DIC) in the surface ocean.
figure 6

Current and projected rates of change of preformed (orange line), regenerated (green line) and total DIC (dDIC/dt, black) for the ensemble mean of IPSL-CM6A-LR, NorESM2-LM, and CMCC-ESM2 in a the Sub-Antarctic region and b Antarctic region. The shading in all figures shows the inter-model one standard deviation.

Full size image
Fig. 7: Upper ocean dissolved inorganic carbon and density vertical gradients.
figure 7

a, b Vertical density gradient (\(d\rho /{dz},{top}\)) and c, d vertical dissolved inorganic carbon gradient (dDIC/dz, middle panel) for the Sub-Antarctic (left) and Antarctic (right) regions. Blue lines depict the present climate (1995–2014), orange lines at the end of the 21st century (2080–2099), and black lines are observations (GLODAP) version 2 (2019)). e, f Regression plots between the maximum annual primary production rate and the maximum subsurface DIC gradient for the contemporary climate (1995−2014). The shading in all figures shows the ensemble one standard deviation.

Full size image

The seasonal presence of sea-ice also regulates biological and physical-driven variations of upper ocean DIC. The NPP seasonal cycle timing is linked to sea-ice variation in the Antarctic36; NPP increase initializes only after the sea-ice maximum (September), this leads to one to two months offset in comparison to the ice-free Sub-Antarctic (Fig. 5b, d). The NPP-related surface DIC consumption is evident in the negative DIC rate of change (Fig. 5f) during the high production season, and it coincides with the (dpCO2/dt)nonT minima (Fig. 4f). Further, the timing of the NPP seasonal maxima also aligns with a minimum in the apparent oxygen utilization rate (Fig. 5h). Negative AOU rate of change reflects oxygen production during photosynthesis, whereas positive AOU rates of change is indicative of respiration or the oxidation of organic matter back to DIC in the near surface. Indeed, ESMs show positive AOU rate of change after the NPP maxima indicating subsequent remineralization. The alignment of AOU and DIC rates of change highlights that the role of biology in setting the DIC levels (Fig. 5a, b & e–h). Further, the decomposition of DIC into preformed and regeneration components (see methods for description, Eqs. 5–10) further substantiate that AOU is indeed a reliable indicator of biological-driven DIC variations (Fig. 6). Namely, dDICregenerated/dt, and dAOU/dt display a similar seasonal cycle phasing, and both follows the NPP seasonality.

In addition to the NPP links to biological DIC variation in the upper ocean, NPP plays an essential role in regulating the physical DIC component41,42. According to the models, NPP levels sets vertical DIC gradients between the epipelagic and mesopelagic layers, which in turn determines the DIC entrainment rates during seasons (or events) of upper ocean mixing (Fig. 7). ESMs with high NPP tend to have stronger vertical DIC gradients (Fig. 7e, f, Supplementary Figs. 4d, f & 5a, b). The relationship between (dDIC/dz)max and seasonal NPPmax is robust in the Sub-Antarctic (p < 0.01) but is non-significant in the Antarctic region (Fig. 7e, f). We note that the low number of models may affect the significance in this region where models have a large spread in the seasonal sea ice.

In summary, the processes characterizing the present-climate Southern Ocean DIC seasonality is primary production, respiration and entrainment mixing. In the Sub-Antarctic, entrainment fluxes are responsible for the DIC seasonal maximum; the dDIC/dt seasonal maximum occurs in early winter consistent with deep MLDs when maximum entrainment mixing is expected43 (Fig. 5a). On the other hand, the dDIC/dt maximum occurs earlier than MLD maximum in the Antarctic, while MLDs are relatively shallow (~ 80 m) (Fig. 5b, f, h). In this region, the dDIC/dt maximum coincides with respiration and remineralization, as demonstrated by AOU and regenerated DIC (Figs. 5f, h and 6). The Antarctic DIC seasonal maximum is therefore first-order driven by near-surface respiration which peaks in autumn, but is sustained by entrainment mixing through winter when vertical mixing onsets in the models. This outcome suggests that the near surface respiration may be playing a more essential role than previously thought in the seasonal variations of surface DIC in the Antarctic region.

Poleward migration of the dominant CO2 uptake zone in a warmer climate

At the end of the high emission scenario (2080–2099), the ocean is warmer; the Southern Ocean warms the most in the Subtropics (>3 °C), and the warming signal decreases poleward, reaching a maximum of 1 °C in the Antarctic by the end of the 21st century (Fig. 8d). Upwelling of the cool deep circumpolar water and sea-ice minimizes warming in the Antarctic region, keeping the surface waters relatively cool44. In the projected future, the surface ocean will be saltier in the Subtropics and fresher in the Antarctic region relative to the present climate (Supplementary Fig. 5e, f). Further, the increase of atmospheric CO2 lowers the ocean CO2 buffering capacity as the ocean take more CO2, diagnosed by the increased Revelle Factor11,12,13,17. The combination of these factors leads to a poleward migration of the dominant region of CO2 sink from the Subtropics to the Antarctic region (Fig. 8a, b). The mechanistic insight related to this poleward shift and why CO2 uptake in the Subtropical region, although being the largest sink region in the present climate, do not increase as atmospheric CO2 increase in the projected future is described below.

Fig. 8: Summary figure for the drivers of the projected poleward shift in the dominant CO2 sink region under high emissions.
figure 8

Annual and zonal mean air-sea CO2 fluxes for the ensemble mean (FCO2) south of 30 °S a, change in annual mean sea-ice fraction b, Air-sea ΔpCO2 c, sea surface temperature seasonal amplitude d and surface ocean temperature with respect to the historical period (1995–2014) e.

Full size image

At the end of the 21st century, ΔpCO2 annual averages change by nearly -150 µatm in the Antarctic region (>100 µatm model spread) and less than 50 µatm north of the Polar Front (Fig. 3f). North of the Polar Front, ΔpCO2 seasonal averages (winter-summer) change has opposite signs and hence the annual mean is relatively small, whereas south of the Polar Front, ΔpCO2 winter-summer averages have the same sign (Figs. 3 and 8c). Differences in the net ΔpCO2 change north and south of the Polar Front at the end of century can be explained by differences in the relative contribution of the thermal and nonthermal pCO2 components. In the Subtropics, the projected thermal component remains larger than the nonthermal contribution as in the present climate, and hence surface temperature remains the primary driver (Fig. 4 g, j). Therefore, higher atmospheric CO2 enhances CO2 uptake during cooler seasons when CO2 solubility is high (autumn-winter) (Fig. 3e), but warming weakens CO2 solubility in summer. This creates a positive feedback on the Subtropics CO2 sink; the increase in CO2 uptake during winter as atmospheric CO2 increase is nearly all compensated by the weakening of CO2 solubility in summer (Fig. 3d, e), and thus the net annual mean ΔpCO2 is almost unchanged in the projected climate (Fig. 3f). The reduction of gas solubility in summer decreases the ΔpCO2 which weakens CO2 ingassing in the projected climate (Fig. 8c). The extension of the high emission scenario up to 2300 available for CanESM5 shows that the Subtropical CO2 sink weakens post-2100 even though atmospheric CO2 continues to rise (Supplementary Fig. 6), suggesting that the warming-driven weakening of CO2 solubility becomes increasingly important as the ocean warms in the Southern Ocean except for the Antarctic region. ESMs general show a good agreement in simulating this mechanism in the Subtropics (Figs. 3–4), suggesting it is a robust feature under the high emission scenario.

In the Sub-Antarctic region, the sign of the projected MT-nonT (Fig. 4k) also remains unchanged with respect to the present climate; the majority of the ESMs still show a weak thermally driven seasonal cycle of ocean pCO2 as in the present climate. The thermal and nonthermal components are increased by nearly equivalent amounts (Fig. 4h). Since the seasonal warming and cooling rates show little to no change in the projected climate (Fig. 5i), the increase in the thermal pCO2 component is primarily due to the ocean pCO2 increase (Eq. 1 in Methods). pCO2 is more sensitivity to temperature in a high pCO2 environment because of the Revelle factor increase12. The Revelle Factor increase enhances the sensitivity of pCO2 to DIC and temperature changes, this effect increases poleward and is strongest in the Antarctic consistent with Revelle factor pattens12,13,17 (Supplementary Fig. 11). In particular, the Revelle Factor increase enhances the sensitivity of pCO2 to primary production and respiration (and mixing) driven DIC changes on the nonthermal pCO2 components with nearly equivalent magnitudes but opposing directions (Fig. 4h, i). Although DIC rate of change declines in the projected future (Fig. 5e, f), its impact on the nonthermal pCO2 contribution is larger than the present climate. This decline in DIC rates is driven by two factors. First, the upper ocean is more stratified; stronger density vertical gradients (Fig. 7a, b) due to warming (Fig. 8e) and freshening of the upper ocean (Supplementary Fig. 5f, Fig. 5c, d). Secondly, the anthropogenic ocean DIC increase from rising atmospheric CO2 propagates from the surface: DIC increases more at the surface than at depth (Supplementary Fig. 4), consequent weakening of the vertical DIC gradients (Fig. 7c, d), leading to a weaker DIC entrainment potential during vertical mixing. In summary, the impact of the Revelle factor increase partly self-compensates in Sub-Antarctic; it enhances the impact of mixing and respiration-driven DIC changes on surface pCO2 in one direction, and primary production in the opposing direction, Fig. 4i.

In the Antarctic region, the magnitudes of the thermal and nonthermal components are also larger at the end of the 21st century compared to the present climate consistent with the expected Revelle factor increase in a high pCO2 environment (Fig. 4i). Nevertheless, the warming-driven melting of sea-ice adds unique characteristics to how the Antarctic ocean pCO2 and FCO2 are impacted by the high atmospheric CO2 and warming. We illustrate this by examining long-term changes on the thermal and nonthermal components. In addition to a lower CO2 buffer capacity, dpCO2T/dt magnitudes in the Antarctic increases by a larger margin in comparison to the ice-free regions because of the enhancements of seasonal cooling and warming rates related to sea-ice melt (Fig. 5j). The lengthening of open water seasons, and shallower MLDs due to warming and sea-ice melt (Fig. 5d) increases the summer-winter surface temperature contrast (Fig. 8d), leading to increases in the SST rate of change. The Antarctic surface waters require less energy to warm in the projected climate because of less sea-ice and shallower MLDs, this amplifies the seasonal warming and cooling rates, ultimately summer temperatures are higher than the present climate (Supplementary Fig. 1a, Fig. 8d). On the other hand, winter temperatures remain similar to the present climate because seasonal sea-ice growth is still present at the end of the century (Supplementary fig. 1a). Nevertheless, since not all Antarctic sea-ice is lost by the end of the century, surface waters remain cool even in late summer (below 5oC), and hence ocean CO2 solubility increases (Supplementary Fig. 1a); this explains the ΔpCO2 increase in the Antarctic (Fig. 8c). In summary, the Revelle factor increase, larger seasonal warming and cooling rates (Fig. 5j) together with higher ocean pCO2 in the Antarctic leads to a larger thermal pCO2 rates increase in comparison to the ice-free regions of the Southern Ocean (Fig. 4h, i).

Ultimately, the melting of sea-ice, stratification increase, MLD shallowing, DIC rates decline, and SST rates increase leads to a regime change in the primary driver of ocean pCO2 from the nonthermal to thermal drivers in the winter to mid-spring seasons (JJASO) in the Antarctic region (Fig. 4l). Ocean pCO2 shifts from the seasonal mixing-driven CO2 outgassing in the present climate (Fig. 4l) to a solubility-driven CO2 uptake during winter in the projected climate. The shift to gas solubility as the primary driver of ocean pCO2 changes in winter allows ESMs to take up CO2 in both the winter and summer seasons by a combination of solubility and biological CO2 uptake. A higher Revelle factor also enhances the effect of biological driven DIC changes and hence CO2 uptake in summer. Thus, ΔpCO2 magnitudes in the Antarctic region have the same sign in winter and summer in the Antarctic, this leads to an extensive net annual mean ΔpCO2 change in comparison to the Sub-Antarctic and Subtropical regions where ΔpCO2 seasonal averages have opposing signs (Figs. 3 and 8c). The same-sign change in seasonal of ΔpCO2 averages (Fig. 3d, f) in the Antarctic region also applies to ESMs with a year-round thermally-driven ocean pCO2 (e.g. CanESM5 and NorESM2, Supplementary Fig. 10). Antarctic surface waters are constrained by seasonal sea-ice presence and upwelling of the circumpolar deep water from warming significantly above the freezing temperature (Fig. 8e). In contrast, sea-ice-free regions (Sub-Antarctic and Subtropical) are already significantly above freezing temperature (8 °C minimum, Supplementary Fig. 1), and hence further warming reduces gas solubility. In the Antarctic region, warming-driven sea-ice melt increase the volume of near freezing surface waters which has a lower molecular kinetic energy of CO2 and therefore strengthening the solubility of atmospheric CO2. Subsequently, warming reduces the ΔpCO2 in the Subtropics and enhances it in the Antarctic (Fig. 8c). Thus, in principle, the Antarctic surface oceans can still take up CO2 through gas solubility even at the end of the 21st century in the high-emission scenario. Therefore, in the projected climate, the Antarctic operates in a hybrid mode between biologically-driven summertime and solubility-driven wintertime uptake. While the analyzed ESMs show a large model spread in the Antarctic, the emergence of the Antarctic region as dominant CO2 sink region in the projected climate is evident in all analyzed ESMs. This outcome suggests that although ESMs still show large differences the representation of biological and physical characteristics in the Antarctic region, the high-emission forcing projects a > 450% enhancement of the Antarctic CO2 sink, suggesting this feature a robust.

Discussion

Examining multi-ESM projections of physical and biogeochemical processes, a robust pattern emerged whereby anthropogenic impacts manifested themselves through altering the seasonal carbon cycle with a large-scale shift in the CO2 uptake region from the Subtropics to the Antarctic region. This major finding was robust even when accounting for uncertainties in multi-model projections.

Mechanistically, it is caused by a change in the mechanisms of seasonal CO2 uptake in the Antarctic (Figs. 1–2, 8a) causing a poleward migration of the dominant CO2 uptake region. This poleward shift in the region of the largest CO2 sink was noticed by earlier studies16,45,46 without mechanistic links to seasonal carbon dynamics at polar latitudes. A recent study26 based on a single ESM showed that warming enhances the outgassing of natural CO2 under the high-emission scenario, which explains the decline of subtropical uptake. The Antarctic is a challenging region to model ocean carbon cycling. ESMs show a large spread and extensive biases with respect to observations for ΔpCO2, MLD, DIC, and thermal & nonthermal pCO2 components (Figs. 4–5). Relatively low model skill in the Antarctic region is partly due to known biases in sea ice36,47,48,49, cloud-related shortwave biases impact on heat fluxes38,40, the AMOC37 and their related stratification biases, which reflect on the simulated biological properties, e.g., phytoplankton phenology 36. This is because sea-ice extent and seasonality are central in setting the timing of the biologically-driven DIC variability, and the respiration-entrainment mixing driven CO2 outgassing in the contemporary Antarctic region; seasonal FCO2 variability, and uptake (Figs. 5–8). With these uncertainties in mind, this multi-model study provides two further insights in the poleward migration of the largest CO2 sink in the Southern Ocean under the high-emission scenario.

First, the melting of sea-ice plays a major role in setting the conditions for the major shift in the Antarctic region. The melting of sea-ice will freshen and stratify the upper ocean, which on the one hand shoals and decreases the differences in MLD seasonal characteristics among ESMs, reducing the model spread significantly by the end of the 21st century (Fig. 5d). It remains unclear why ESM’s MLD differences are reduced under the high-emission scenario. The role of sea-ice melt is well illustrated by a stronger corelation between sea-ice loss and FCO2 post-2060 (Fig. 8a, b) when sea-ice loss surpasses 5-10%. It is at this stage (post-2060) that the domain of the largest CO2 uptake transitions to the Antarctic region (Fig. 8a). The strong correlation between CO2 uptake and sea-ice melt post-2060 reinforces the central role of sea-ice in driving the CO2 dynamics of the Antarctic, and further highlights the importance of improving the representation of sea-ice in ESMs among other biases (Fig. 8a, b). Moreover, links between sea-ice changes and CO2 uptake through solubility elevate the role of solubility as an important mechanism of CO2 uptake in the Southern Ocean. The largest CO2 sink domain in the present climate and projected climate resides where solubility plays a primary role. Surface warming weakens the solubility in CO2 uptake in the Subtropics and enhances through sea-ice melt in the Antarctic. Beyond 2100, the high warming scenario in CanESM5 shows that the Subtropics cease to take up CO2 although atmospheric CO2 is still growing while the Antarctic region becomes even a stronger sink of CO2 (Supplementary Fig. 6). This suggests that climate change will have regionally variable effect on the Southern Ocean FCO2. Understanding the mechanisms behind this regional difference will be crucial to improve model projections, and elucidate the broader climate and ecosystems impacts.

Secondly, the accumulation of anthropogenic carbon in the upper ocean in the high warming scenario diminishes (or even reverses) the vertical gradient of DIC (Figs. 6–7). This mechanism works through the shift in the seasonal cycle of upper ocean DIC balance. The wintertime entrainment of DIC is reduced due to the weakened vertical gradients of DIC, leading to the dominance of thermally-driven pCO2 changes (Figs. 4l and 7). This sets up the seasonal regime change to the thermally-driven CO2 uptake in the Antarctic region.

This outcome suggest that solubility-induced carbon uptake may be as important or even possibly more important than biological CO2 uptake during major climate transitions50,51. Although biological CO2 uptake is an essential mechanism for CO2 uptake in the Southern Ocean34,46,52 and has been shown to play a key role in glacial time scale regime changes53,54. Biological CO2 uptake alone is partly offset by the remineralization of organic matter post-production, limiting its role in CO2 uptake in the Southern Ocean. Because of this respiration effect55, even if primary production were to increase in the projected climate, respiration will likely partly compensate for the CO2 uptake due to a larger Revelle factor. On the other hand, CO2 solubility is capable of becoming an important contributor to CO2 uptake when sea-ice melt is evident by subverting the role of near-surface respiration and/or seasonal mixing in winter CO2 outgassing through the shifting surface carbonate system to a thermally driven system. We postulate that this solubility feature is only active during relatively cool sea surface temperatures to allow solubility-driven CO2 ingassing. If the polar oceans were to warm significantly above the freezing temperature, the drivers of FCO2 may become similar to the Subtropical region or the Arctic Ocean, where warming reduces kinematic gas solubility56. Further warming and complete loss of sea ice in the Arctic for example has been shown to cause the reversal of the seasonal cycle of ocean pCO2 under the high-emission scenario in the projected climate56. This type of regime change is however unlikely to occur in the Southern Ocean because of dynamic differences. The Arctic Ocean relies on winter mixing to entrain nutrients and export carbon to the deep ocean and is thus more affected by global warming through stratification51. In contrast, westerly winds in the Southern Ocean will likely continue to provide the main forcing for the upwelling of cold and nutrient-rich water57. Instead, we postulate that the future of the polar Southern Ocean may operate in a hybrid mode of biologically-driven summertime uptake and solubility-driven wintertime uptake, taking a leading role in the projected Southern Ocean carbon sinks.

Sustaining carbon uptake in the polar Southern Ocean requires a mechanism to export the excess carbon into the interior ocean, away from the region of uptake. A recent study45 demonstrated that the increased CO2 sink south of the Polar Front under the high-emission scenario was linked to the Southern Ocean’s ability to advect subsurface low DIC waters equatorward allowing a continuous surface uptake of atmospheric CO2. The long-term perspective of this carbon sink may depend on the circulation changes that transfer carbon absorbed from the atmosphere to the water masses in the intermediate and deep-water reservoirs. On the other hand, anthropogenic ice sheet melt in Antarctica is projected to slow down the Southern Ocean overturning and enhance surface stratification58 which may weaken this northward DIC advection. Moreover, ice sheet melt is also projected to enhance Antarctic sea-ice and slow-down warming through the albedo feedback58,59. Stronger stratification may continue to constrain winter DIC surface-subsurface mixing and allowing the surface ocean to take up CO2 through solubility in winter although sea-ice is abundant, but this CO2 sink may eventually be weakened by poor overturning. It remains unclear how these processes will work together; the inclusion interactive ice sheet in the next generation of ESMs will be key to understanding this mechanism.

Methods

Earth system models

We use the output from nine CMIP6 ESMs (Supplementary Table 1), contrasting the contemporary period (1995–2014) and the end of the 21st century (2080–2099) under the socioeconomic pathway 5–8.5 (SSP5–8.5) climate scenario60. Model selection was based on the availability of essential variables at the time of writing; we used a single ensemble member for each model. All ESMs output was downloaded from the earth system grid federation (ESGF) hosted at the German climate computing center (DKRZ, https://esgf-node.llnl.gov/search/cmip6/) except for AWI-CM1. The downloaded model output was stored and analyzed on the Cheyenne National Center for Atmospheric Science (NCAR) server. The following fields were downloaded: monthly mean surface ocean partial pressure of carbon dioxide (spco2), air-sea delta pCO2 (dpco2), air-sea CO2 flux (fgco2), total dissolved inorganic carbon (dissic), total ocean alkalinity (talk), sea surface temperature (tos), sea surface salinity (sos), net primary production (intpp), mixed layer depth (mlotst), sea ice fraction (siconc), dissolved oxygen (o2), dissolved oxygen at saturation (o2sat), practical ocean salinity (so) and potential temperature (thetao). Since not all ESMs have the complete output variables available at the time of writing, some figures do not display all ESMs. Model outputs were analyzed on their native grids except for the unstructured mesh ocean model of AWI-CM1, which was regridded to a 1° × 1° regular grid. Stereographic projection figures (Fig. 1 was made using 1° × 1° regular grid data.

Observation-based pCO2-products

We use six observation-based surface pCO2 products available in the SeaFlux dataset61 downloaded from: https://zenodo.org/record/5482547. The individual products are: CMEMS-LSCE-FFNN62; CSIR-ML663; Jena-MLS64; JMA-MLR65; MPI-SOMFFN66; and NIES-FNN67—see Supplementary Fig. Table 2 for more details. The pCO2 products fill the gaps in sparse ship-based observations of pCO2 with satellite observations, reanalyzes outputs, and observational monthly climatologies. Statistical models (e.g., artificial neural networks) are used to estimate the relationship between proxy variables and pCO2, meaning that these approaches incorporate little to no mechanistic understanding of the drivers of pCO2. Proxy variables include, but are not limited to: sea surface temperature (remote sensing), salinity (reanalysis), mixed layer depth (climatology), and ocean color estimates of phytoplankton chlorophyll which is understood as a proxy for biomass (remote sensing). All methods use pCO2 from the Surface Ocean CO2 Atlas (SOCAT version 2020 or later68). All observation-based products are estimated on or re-gridded to a monthly 1° × 1° regular grid. Further, we use the ocean potential temperature and salinity dataset from the World Ocean Atlas69 to estimate observed vertical density. We use a monthly climatology of mixed layer depth by70 based on various temperature and salinity profiling instruments to compare with model output. Lastly, we use DIC from the Global Ocean Data Analysis Project (GLODAP) version 2 (2019) dissolved inorganic carbon dataset71 to compare with models.

Thermal and nonthermal ocean pCO2 decomposition

Surface ocean pCO2 and ΔpCO2 variability is controlled by the relative contribution of thermal and nonthermal components28,29. We here estimate the thermal component using the Takahashi et al.30 formulation30 (Eqs. 1–2) and we estimate the nonthermal component by subtracting the thermal component from the total (Eq. 3). Thermal component is driven by temperature variations through changes in gas solubility (Henry’s law). The nonthermal pCO2 component on other hand is mainly controlled by mixing and biology31, which also includes the role of total alkalinity and salinity changes. However, total alkalinity and salinity have been shown to play a minor role in the seasonal cycle of ocean pCO2 in the contemporary Southern Ocean29,31, thus, here we focus on processes responsible for sources and sinks of DIC, i.e., primary production, respiration, and seasonal buoyancy change-driven mixing. The thermal and nonthermal components of pCO2 oppose each other on a seasonal scale32,33 (Fig. 4d, f), and hence the larger of the two determines the observed seasonal cycle phasing of pCO2, and ultimately FCO229. The relative contributions of thermal and nonthermal components are assessed here through the absolute difference in their monthly rates (MT-nonT, Eq. 4). The larger rate of change between |(dpCO2/dt) T| and |(dpCO2/dt) nonT| is therefore considered the dominant driver of ocean surface pCO2 change (monthly in our case). This is estimated by the absolute difference of the time derivative of the thermal and nonthermal components, and we term this diagnostic metric as MT-nonT (Eq. 3), consistent with Mongwe et al.29. MT-nonT > 0 indicates periods when temperature variance drives the pCO2, while MT-nonT < 0 is indicative of periods when nonthermal processes play a leading role in surface pCO2. While simple, MT-nonT provides a useful diagnostic for identifying the predominant mechanisms driving seasonal pCO2 variations, in particular, given that the thermal and nonthermal pCO2 oppose each other on a seasonal scale, thus isolating the leading driver provides key information.

$${pC}{O}_{2T}=0.0423\times \overline{{{pCO}}_{2}}\times {SST}$$
(1)
$${\left(\frac{\partial {pC}{O}_{2}}{\partial t}\right)}_{T}=0.0423\times \overline{{{pCO}}_{2}}\times \left(\frac{\partial {SST}}{\partial t}\right)$$
(2)
$${\left(\frac{\partial {pC}{O}_{2}}{\partial t}\right)}_{{nonT}}={\left(\frac{\partial {pC}{O}_{2}}{\partial t}\right)}_{{Tot}}-{\left(\frac{\partial {pC}{O}_{2}}{\partial t}\right)}_{T}$$
(3)
$${M}_{T-{nonT}}=\left|{\left(\frac{\partial {{pCO}}_{2}}{\partial t}\right)}_{T}\right|-\left|{\left(\frac{\partial {{pCO}}_{2}}{\partial t}\right)}_{{nonT}}\right|$$
(4)

DIC decomposition

The variability of ocean DIC outside of external drives is driven by ocean physical (circulation and mixing) and biological processes (i.e. primary production and respiration)72,73. DIC is consumed through photosynthesis to form organic matter in the upper ocean and is produced by the reverse process of respiration in the near surface to the deep ocean, where organic matter is remineralized, this component of DIC is referred to as regenerated DIC72,74. The non-biological component of DIC is mainly controlled by the subduction at mass transformation and mixing is referred to as preformed DIC. In this study, we decompose the surface DIC seasonal cycle into preformed and regenerated following72,74 (Eq. 5). Regenerated DIC comprises of the soft pump, which represents the biological production and remineralization of particulate organic carbon (Csoft, Eq. 7), and the carbonate part representing the consumption and remineralisation of molecularly charged nutrients (Eq. 8)73. Since our analysis is focused on the surface ocean, we assume that Talk = Talkpre, hence Eq. 8 excludes the total alkalinity contribution. Preformed DIC comprises of the saturated DIC concentration in equilibrium with the atmosphere (Csat), and the disequilibrium contribution; estimates deviations from Cast when mixing occurs before air-sea exchange is complete (Cdis)72,74,75. RC:O2 is the carbon to oxygen stoichiometric ratio (-106/170), RN:O2 is the stoichiometric ratio of nitrate to oxygen (-16/170), S is salinity, T is surface temperature, O2sat is oxygen at saturation, O2 is dissolved oxygen and AOU is the apparent utilization of oxygen.

$${DIC}={C}^{{pre}}+{C}^{{reg}}$$
(5)
$${C}^{{reg}}={C}^{{soft}}+{C}^{{carb}}$$
(6)
$${C}^{{soft}}={-R}_{{C:0}_{2}}\times {AOU}$$
(7)
$${C}^{{carb}}={\frac{1}{2}}\left(-R_{{N:0}_{2}}\times {AOU}\right)$$
(8)
$${AOU}={{O}_{2}^{{sat}}\left(T,S\right)-O}_{2}\left)\right.$$
(9)
$${C}^{{pre}}={C}^{{sat}}+{C}^{{dis}}$$
(10)

Study domain

In this study, we divide the Southern Ocean into three subdomains using dynamical frontal boundaries to reflect the three regimes of the Southern Ocean. Namely, (i) the Subtropical region, the warmest and most stratified part of the Southern Ocean, (ii) the Sub-Antarctic region, highly productive, moderately warm, and host the subduction of the Subantarctic Mode Water and Antarctic Intermediate Waters; and (iii) the Antarctic region, seasonally productive, cool, seasonally sea-ice covered and upwelling region. We use the76 dynamical boundary definition based on potential temperature, (i) the Subtropical region is between 30 °S and the Subtropical Front, (ii) the Sub-Antarctic region between the Subtropical Front and Polar Front; and (iii) the Antarctic region, south of the Polar Front to the sea-ice edge or continental margin. Orsi et al.76 boundaries are nearly similar to the commonly used typical Fay and McKinley et al.77 biomes but differ most in the Subtropical region, where the Fay and McKinley biomes exclude a large part of the subtropical Pacific. Furthermore, not all the analyzed ESMs have all the required output to estimate the77 biomes for both the present climate and the end of the century. Dynamic boundaries are calculated for each model and averaged for the present climate (1995–2014) and end of the 21st century (2080–2099). And for the observed estimate we use the same dynamic boundary based on the WOA dataset. Regional averages are estimate using each model’s native dynamic boundary and averaged for the ensemble mean afterward.