What goes in must come out: the oceanic outgassing of anthropogenic carbon

About 25% of the emitted anthropogenic CO2 is absorbed by the ocean and transported to the interior through key gateways, such as the Southern Ocean or the North Atlantic. Over the next few centuries, anthropogenic CO2 is then redistributed by ocean circulation and stored mostly in the upper layers of the subtropical gyres. Because of the combined effects of (i) weakening buffering capacity, (ii) warming-induced lower solubility, (iii) changes in wind stress and (iv) changes in ocean circulation, there is a high confidence that the ocean sink will weaken in the future. Here, we use IPCC-class Earth System Model (ESM) simulations following the SSP1-2.6 and SSP5-8.5 climate change scenarios extended to the year 2300 to reveal that anthropogenic CO2 begins to outgas in the subtropical gyres of both hemispheres during the summer months of the 21st century. In 2100, about 53% of the surface ocean experience outgassing at least one month in a year in SSP1-2.6, against 37% in SSP5-8.5. After 2100, this fraction keeps increasing, reaching 63% by 2300 in SSP5-8.5 while stabilizing at 55% in SSP1-2.6. This outgassing pattern is driven by the rapid increase in oceanic pCO2, faster than the atmospheric pCO2, due to the combined effect of both rapid warming and long-term accumulation of anthropogenic carbon in these regions. These findings call for increased observation efforts in these areas, particularly in the subtropical gyres of the Southern Hemisphere, in order to detect future release of anthropogenic carbon and accurately constrain the future carbon budget.


Introduction
The ocean responds to the anthropogenic perturbation by uptaking roughly 25% of the human induced carbon emissions annually (Friedlingstein et al 2022), altering in the meantime its physical and biogeochemical state.Under a steady increase in atmospheric CO 2 , the ocean acts as a negative feedback by absorbing the excess heat and CO 2 from the atmosphere (Friedlingstein et al 2006, Roy et al 2011).Nevertheless, the feedback has been shown to be non-linear over long-term stabilizing or overshoot scenarios (Schwinger andTjiputra 2018, Asaadi et al 2023).Therefore, accurately predicting the future evolution of climate requires a better understanding of the ocean's long-term response and capacity to mitigate or reinforce human-induced climate change.
Presently, the CO 2 emitted by human activities enters the ocean interior through ventilation regions and is redistributed and stored predominantly in the North Atlantic, the subtropical gyres and the Southern Ocean (Sabine et al 2004, Gruber et al 2009, 2019a, Sabine and Tanhua 2010, Khatiwala et al 2013).Since anthropogenic carbon (C ant ) in the ocean cannot be directly observed, models have been adopted to determine its long-term air-sea fluxes and transport pathways variability (Wetzel et al 2005, Gorgues et al 2010, Tjiputra et al 2010).Ocean inversion methods, combining observational data with dynamical models, have also been applied to estimate of the C ant flux (Gloor et al 2003, Mikaloff Fletcher et al 2006, Gruber et al 2009, DeVries 2014).The Southern Ocean is one of the main gateways for C ant , representing about 40% of the contemporary oceanic C ant uptake (DeVries 2014).C ant is then redistributed in the interior by the ocean mixing and circulation.The net lateral transport of C ant is mostly northward, (Gruber et al 2009, DeVries 2014, Bourgeois et al 2022).However, this vertically integrated view obstructs the more complex patterns (Nakano et al 2015).C ant is subducted out of the mixed layer mostly in the mid and high-latitude (between 20 • and 60 • latitudes in both hemispheres, Bopp et al 2015, Davila et al 2022)).Below the mixed layer, C ant is transported following the mean large scale overturning circulation (Gruber et al 2009, Nakano et al 2015).It is transported equatorward in the intermediate water layers, by the subtropical or Subantarctic Mode Waters or the Antarctic Intermediate Water.In the deeper layers, it is transported northward in the Pacific, while it is transported southward in the Atlantic.
The anthropogenic CO 2 also re-emerges from the deep ocean, notably through the subtropical or shallow overturning circulation (Nakano et al 2015, Iudicone et al 2016).The upwelling at the equator transports carbon across the base of the mixed layer (Bopp et al 2015, Zhai et al 2017).Moreover, the subduction (downward flux) in the rest of the ocean results from a compensation between strong upward and downward fluxes (Toyama et al 2017).Such reemergence can strongly inhibit the ability of the ocean to further uptake C ant (Rodgers et al 2020).On top of that, due to the combined effects of (i) the slowing down of the CO 2 emissions rate, (ii) weakening of the buffering capacity (Jiang et al 2019, 2023, Arora et al 2020), (iii) decrease in warming-induced solubility (Tjiputra et al 2010, Katavouta and Williams 2021), (iv) changes in wind stress (Bronselaer et al 2016(Bronselaer et al , 2018) ) and ocean circulation (Tjiputra et al 2010, Gruber et al 2019a, Arora et al 2020, Bronselaer and Zanna 2020, Katavouta and Williams 2021), there is a high confidence that the ocean sink will weaken in the future even if the extent of the weakening is scenario-dependent Arora et al 2020, IPCC 2021).Considering the decline in oceanic carbon sink and the re-emergence of the C ant into the surface ocean, our study addresses the question of when and where the previously absorbed C ant by the ocean will be released back into the atmosphere, and whether these outgassing signals will be strong enough to emerge from the background inter-annual variability and be detectable.
The concept of time of emergence or departure has been used extensively in recent works to assess the emergence of anthropogenically induced signals such as warming, acidification, air-sea CO 2 flux trends, and primary production decline (Henson et al 2010, Mora et al 2013, Keller et al 2014, Rodgers et al 2015, Bertini and Tjiputra 2022, Tjiputra et al 2023).Time of emergence is defined as the time when the considered signal of a variable exceeds its respective background noise.The background noise is commonly defined as twice the standard deviation of either an ensemble of simulations (McKinley et al 2016, Schlunegger et al 2019) or time series from preindustrial simulations (Henson et al 2010, 2017, Christian 2014, Keller et al 2014, Tjiputra et al 2023), representing internal or inter-annual climate variability.Statistically, using twice the standard deviation ensures that the signal exceed 95% of the values in the background noise.Some alternative choices for the noise can be one standard deviation (confidence interval of 67%, Rodgers et al 2015) or minimum and maximum values (Mora et al 2013).The time of emergence is a useful concept for getting information about the detectability of a signal as well as the pace of climate change impacts.
The main goal of this work is to identify the time of emergence of C ant outgassing and its drivers in the different ocean regions.The next section describes the model and global warming scenarios used, the definition of the time of emergence used for C ant outgassing as well as the diagnostic applied to identify the drivers of the emergence.Following the method section, the results are presented and discussed.

Model configurations
We use the second version of the Norwegian Earth System Model (NorESM2-LM, Seland et al 2020a, 2020b) built with atmospheric, ocean, sea-ice, and land modules.The ocean component of NorESM2-LM is the Bergen Layered Ocean Model (BLOM) coupled with the isopycnic coordinate Hamburg Ocean Carbon Cycle (iHAMOCC) model for ocean biogeochemistry (Tjiputra et al 2020).iHAMOCC represents the lower trophic biological productivity in the upper ocean by including one phytoplankton, one zooplankton compartment, multiple limiting nutrients (nitrate, phosphate, and dissolved iron), dissolved organic carbon and particulate matters.A fixed stoichiometry redfield ratio is used to govern the fluxes of nutrients and carbon among the different ecosystem compartments.
The inorganic carbon chemistry in iHAMOCC is based on the Ocean Carbon cycle Model Intercomparison Project protocols.It computes the partial pressure of CO 2 gas in the surface layer (pCO 2 ) based on the temperature, salinity, dissolved inorganic carbon (DIC) and alkalinity concentrations.The air-sea CO 2 fluxes is computed according to Wanninkhof (2014) taking into account surface wind speed, Schmidt number, gas solubility, atmospheric pCO 2 , and surface ocean pCO 2 .In order to explicitly estimate the C ant fluxes and inventory in the ocean, NorESM2-LM simulates a set of 'natural' carbon tracers (include DIC, alkalinity, and CaCO 3 ) that are constrained by a fixed preindustrial control atmospheric CO 2 concentration of 284.7 ppm during the air-sea gas exchange.Thus, C ant is calculated as the difference between the total and natural DIC tracers (Tjiputra et al 2020).It should be mentioned that NorESM2-LM does not consider natural carbon tracers in the sediment, as we assume that the long timescale of sediment dynamic does not play a major role in our relatively short transient timescale.Indeed, C ant interaction with the sediment occur at multi-millennial time scales (Archer et al 1997), which is considerably longer than the periods (1850-2300) analyzed here.Finally, it is noteworthy that even though our methodological choice can provide an estimates of the partition between anthropogenic and natural carbon, other approaches also exist, e.g. using labeled carbon tracer (Holzer and DeVries 2022), which could give different results.

Scenarios and simulations
Three model experiments were performed following the CMIP6 framework and as described in O'Neill et al (2016): the historical (1850-2014) and the SSP1-2.6 and SSP5-8.5 extended future scenarios .All the simulations are forced with prescribed atmospheric greenhouse gas concentrations, aerosol emissions and land-use change forcings (Hurtt et al 2020, Ma et al 2020, Meinshausen et al 2020).In such concentration-forced simulations, the ocean and land carbon cycle react to atmospheric CO 2 concentrations, which are prescribed via global mean time series, but do not feed back on the atmospheric CO 2 .The historical scenario follows forcings largely based on available observations.The SSP1-2.6 scenario is on the low end of the radiative forcing.After a brief increase at the beginning of the 21st century, the CO 2 emissions decrease and atmospheric CO 2 concentrations stabilize.The SSP5-8.5 is on the high end of the radiative forcing with the carbon emissions increasing until 2080 after which they start to slowly decline to reach zero in 2250.The simulations outputs are available online (Tjiputra et al 2023).
Following the rising atmospheric CO 2 , the global surface air temperature increases, reaching +1.3In addition, to isolate the effect of atmospheric CO 2 increase from the impact of global warming, we also used the so-called biogeochemical coupled (BGC) simulation (as opposed to the fully coupled simulations described before).In this simulation, the atmospheric CO 2 follows that of SSP5-8.5, but the CO 2 added to the atmosphere is not radiatively active, and does not cause global warming.

Carbon fluxes in the different scenarios
Globally, the total carbon uptake simulated by NorESM2-LM increases during the historical period (black line in figure 1(a)) and captures well the observed long-term trend, though slightly overestimated.In both global warming scenarios, the uptake keeps increasing, reaching a maximum rate then decreases.In SSP5-8.5 (figure 1(a), red line), the maximum global uptake occurs around 2080 then steadily decreases until the end of the simulation period.In SSP1-2.6 (figure 1(a), blue line), the maximum is reached around 2030 and the global uptake is very close to zero from 2100 onward.The long term evolution of ocean carbon uptake along the simulations is mostly due to the uptake of C ant (figure 1(b)).The global mean natural carbon fluxes are close to zero from 1850 to 2300.
The NorESM2-LM model reproduces the spatial patterns of the C ant air-sea flux as understood from the literature on the historical period (figure 1(c)).The uptake is particularly strong in the Southern Ocean, the North Atlantic and the tropics (Mikaloff Fletcher et al 2006, Corbière et al 2007, Gruber et al 2009, 2019b, Ridge and McKinley 2020, Brown et al 2021).In SSP1-2.6, in the 2090s (figure 1(d)), the close to zero global mean C ant air-sea flux results from the compensation between regions of uptake and outgassing C ant .The Southern Ocean outgases C ant between 40 • S and 60 • S while still uptakes C ant south of 60 • S. The subpolar North Atlantic and North Pacific, as well as the equatorial band, also outgas C ant .In the 2290s (figure 1(e)), the spatial distributions of C ant fluxes are similar, though attenuated.In SSP5-8.5, in the 2090s (figure 1(f)), C ant uptake patterns are similar to the 1990s but stronger, with particularly strong uptake in the Southern Ocean, the North Atlantic and the tropics.In the 2290s (figure 1(g)), the global mean C ant uptake results from a compensation between regions of C ant uptake and outgas.As in SSP1-2.6, the strongest fluxes are simulated in the Southern Ocean: outgassing between 40 • S and 60 • S and uptaking south of 60 • S. In addition, the subpolar North Atlantic and North Pacific, as well as parts of the equatorial band also outgas C ant .Positive values stand for ocean carbon uptake.

Time of anthropogenic carbon outgassing
The main goal of this work is to determine when and where C ant is released back into the atmosphere, i.e F tot co2 (t, i) − F nat co2 (t, i) ⩽ 0 where F xxx (t, i) is the total or the natural air-sea carbon flux for time t and location i.We analyze monthly average rather than annual average because there is a strong seasonality in the C ant flux (figure 2).For instance, net annual C ant uptakes are simulated in some regions even though they release C ant during summer, because of the stronger uptake in winter.Moreover, for detecting when the C ant outgassing will emerge from the background inter-annual variability, we use a 50 years running mean of the carbon fluxes ( as well as a 50 years running standard deviation of the total carbon flux, σ tot (t, i).Thus, for any location i, we search for the time t out (i) (month and year) such as: The time of C ant outgassing defined above depends on the time window used to calculate the mean and on the threshold used for the emergence (here the standard deviation).Additional analysis show that the outgassing spatial patterns are only weakly sensitive to the time window and threshold., i.e. the regions with the earliest C ant outgassing stay the same (e.g.figure S1).However, it may change the absolute year of outgassing but not the month of outgassing.

Attribution of the anthropogenic carbon outgassing
The year of C ant outgassing emergence defined in equation ( 1 ⟩ .In areas where outgassing emerge earlier (see the result section), wind speed and solubility are weaker and even decrease along the 21st century (figure S3).Thus they do not strengthen the C ant flux and cause earlier emergence.Therefore C ant outgassing is primarily driven by evolution of the oceanic and atmospheric pCO 2 .
In response to the increase in atmospheric pCO 2 , the ocean takes up CO 2 , thereby changing its pCO 2 .In the long-term, the oceanic pCO 2 follows closely the atmospheric counterpart (Fay andMcKinley 2013, Tjiputra et al 2014).For the ocean to release carbon, its partial pressure has to be larger than the atmospheric one.We infer that the regions where carbon outgassing emerges are also the ones where the oceanic pCO 2 increase faster than the atmospheric pCO 2 .Changes in oceanic pCO 2 in turns depend on changes in the DIC content but also on the temperature, the salinity (or freshwater inputs) and the alkalinity.Following a first order Taylor expansion (Takahashi et al 1993, Lovenduski et al 2007, Goris et al 2015, Gallego et al 2018, 2020), the oceanic pCO 2 trend (dpCO 2 /dt) can be recosntructed the sum of three components: (i) the carbon system trend gathering the trends in DIC and alkalinity, (ii) the thermal trend and (iii) the freshwater/salinity trend (accounting for the dilution/concentration impact of freshwater on DIC and alkalinity) following equation (2).
Freshwater and salinity. (2) In this equation, T and S are the sea surface temperature and salinity.C s and A s are the salinitynormalized DIC and alkalinity concentrations: X s = S 0 /S × X. S 0 is the temporal average of S over the entire simulation.γ X are related to the sensitivity of pCO 2 to DIC, alkalinity, salinity and temperature (∂pCO 2 /∂X = pCO 2 /γ X ).The overbar, X, stands for seasonal average of X calculated over the 50 year window.The trends, ∂X/∂t, are computed as the difference in X between two consecutive 50 years period.γ C and γ A are computed with the pyCO2SYS module (Humphreys et al 2022) using the 50 years mean of DIC, alkalinity, salinity and temperature.γ T is 0.0423 • C −1 and γ S is 1 (Takahashi et al 1993, Sarmiento andGruber 2006).

Outgassing of anthropogenic carbon
Throughout the historical and the future warming scenarios, the simulations show that the ocean releases C ant at an increasing rate.During the historical period, between 5% and 20% of the ocean area is releasing C ant for at least one month annually (figure 3(a), thin lines).Nevertheless, these fluxes are too weak to emerge from their background inter-annual variability signals.Only by 2040 does outgassing signal become stronger than the interannual variability (figure 3(a), thick lines).The SSP1-2.6 simulation shows a larger proportion of the ocean outgassing C ant than SSP5-8.5 in the 21st century.By 2100, about 53% of the surface ocean release C ant for at least one month in a year in SSP1-2.6 versus only about 37% in SSP5-8.5.Nonetheless, the outgassing emerges only in 19% of the ocean in both simulations by the end of the 21st century.After 2100, the fraction of the ocean outgassing C ant continues to increase in SSP5-8.5, reaching 63% by 2300, while it stabilizes at around 55% in SSP1-2.6.The ocean fraction where outgassing emerge from inter-annual variability keeps increasing in SSP5-8.5, reaching 49% by 2300, while it decreases in SSP1-2.6,stabilizing at 12%.
The spatial pattern of the season when C ant outgassing emerges coincides essentially with wellknown oceanic regions (figures 3(b) and (d)).The subtropical gyres outgas C ant in summer: JAS for the Northern Hemisphere and JFM for the Southern Hemisphere (see also figure S4 for exact month) .These patterns are very similar in both scenarios.In the Southern Ocean (40-60 • S), the majority of C ant outgassing occurs in winter (JAS), although there are some regional differences (such as in spring or fall), but remain relatively consistent across both scenarios.In the eastern part of the the North Atlantic (40 • N-60 • N), Cant outgassing emergence occurs in summer (JAS) in SSP5-8.5, while it occurs mostly in winter (JFM) in SSP1-2.6.In the Arctic (<70 • N), C ant outgassing emergence occurs only in SSP5-8.5 in summer.
The spatial pattern of the year when C ant outgassing emerges from inter-annual variability also coincides with distinct oceanic regions (figures 3(c) and (e)).The earliest emergence occurs in the subtropical gyres in both scenarios.In some regions, C ant outgassing starts before 2030 (as shown by the black contour in figures 3(c) and (e)).The farther away from the centre of the gyre, the later the outgassing emerges.In SSP1-2.6,most of the C ant outgassing emerge before 2100.After 2100, outgassing emergence starts to become detectable in parts of the North Atlantic.In SSP5-8.5, C ant emergence continues to occur throughout the simulation (as expected from figure 3(a)).The model simulates no detectable C ant outgassing in the southernmost part of the Southern Ocean or in most of the equatorial band.

Drivers of the carbon outgassing in SSP5-8.5
Here, applying equation (2), we seek to unravel the drivers of the C ant outgassing.In particular, we aim to explain why the outgassing signal emerges earlier in the subtropical gyres than in the higher latitude or at the equator.Since the patterns of outgassing emergence are relatively similar between the SSP1-2.6 and SSP5-8.5, suggesting similar processes are at play, we focus our analysis on the SSP5-8.5 scenario, which has the strongest signal.
The season of the C ant outgassing emergence is controlled by the drivers of pCO 2 seasonality (see figure S5 and Sarmiento and Gruber 2006, Landschützer et al 2018).C ant outgassing signal emerges when pCO 2 is maximum.In the midlatitudes (15-40 • N/S), pCO 2 reaches its highest level in summer when the temperature is maximum.In the Southern Ocean (40-60 • S), the highest pCO 2 occurs in winter as a result of non-thermal processes.With the deepening of the mixed layer, water rich in natural carbon but poor in C ant are entrained to the surface, leading to an increase in natural carbon but a decrease in C ant at the surface.However, the total carbon content is higher than in summer and higher than it would be without C ant .In addition, because of the higher carbon content (and Revelle Factor), pCO 2 is more sensitive to the increase in DIC.Altogether, it leads to the stronger outgassing of carbon (in our approach, this excess outgassing is considered as C ant ).In the North Atlantic, >40 • N, and in the Arctic, the outgassing signals emerge in the summer (in SSP5-8.5)when pCO 2 is maximum because of temperature and when the sea-ice extent is minimum (Notz and Community 2020).
For the global average, the rate of change of oceanic pCO 2 closely follows the time evolution of atmospheric pCO 2 rate of change (solid and dashed lines in figure 4(a)), primarily driven by the ocean carbon uptake that increases the DIC content (blue patch in figure 4(a); carbon system term in equation ( 2)).Until 2080, the pCO 2 growth rate is higher in the  2)).(a) Global mean trends of surface ocean (solid black line) and atmospheric (dashed grey line) pCO2 as well the contributions from trends in surface (i) dissolved inorganic carbon and alkalinity (blue, Carbon syst.), (ii) temperature (orange, Thermal) and (iii) fresh water and salinity (green, FW & sal.).The grey patch shows the residual from the difference between the trend in pCO2 and the sum of all component trends (i)-(iii).(b to (m) Maps of the spatial anomalies of (i) the pCO2 trends (1st column), (ii) the Carbon syst.trends (2nd column) and (iii) the thermal trends (3rd column).The spatial anomalies are relative to the globally average trend (panel (a) for the time period considered.Each row shows the spatial anomalies of the trends for a different time period: January-March of the 21st and 22nd centuries (1st and 3rd rows) and July-September of the 21st and 22nd centuries (2nd and 4th rows).Dotted patches show regions where the C ant outgassing emerge from inter-annual variability for each time period.The grey contour shows where ocean pCO2 trend is stronger than pCO atm 2 .
atmosphere than in the ocean.From 2080 onward, pCO 2 growth rate is higher in the ocean than in the atmosphere, reducing the difference and thus the carbon uptake (figure 1(a)).During the 22nd and 23rd centuries, the reductions in atmospheric growth rate and oceanic carbon uptake lead to a slow-down in oceanic pCO 2 increase.At the same time, the ocean keeps warming, and thus the thermal contribution becomes growingly more important (figure 4(a), orange patch).The freshwater input and salinity term only plays a marginal role (figure 4(a), green patch).We note that the sum of the carbon system trend, thermal trend and freshwater and salinity trend (right-hand side in equation ( 2)) does not exactly match the trend in oceanic pCO 2 (left-hand side in equation ( 2)).However, the residual is small (figure 4 As inferred above, C ant outgassing emerge earlier in regions where the increase in oceanic pCO 2 is faster than its atmospheric counterpart (grey contours in figures 4(b) and (e)).In this areas the increase in pCO 2 is also faster than the global average.For the 21st century, the faster increase in oceanic pCO 2 occurs in the subtropical gyres where outgassing signals emerge before 2100 (figures 4(b) and (e)) and this expands at the edges of the subtropical gyres during the 22nd century (figures 4(h) and (k)).
In the subtropical gyre of both hemispheres (20-40 • N/S), the fast increase in oceanic pCO 2 is driven by the accumulation of C ant at the ocean surface with a significant contribution from warming in the Northern Hemisphere (figures 4 and 5).This is particularly evident during the 21st century.In the Southern Hemisphere, the fast C ant accumulation (figure 5(a)) combine with rather high values of mean oceanic pCO 2 (figure S7) lead to a faster increase of the carbon system-driven oceanic pCO 2 (figure 4(d)) while warming plays a secondary role (figures 4(c) and 5(b)).In the Northern Hemisphere, the fast C ant accumulation dominates in the western part of the North Pacific while fast warming prevails in the eastern North Pacific and in the North Atlantic (figures 4(f), (g) and 5(c), (d)).In the 22nd century, the expansion of the emergence along the edges of the subtropical gyres is also driven by the fast C ant accumulation (figures 4(j), (m) and 5(e) and (g)).In the subpolar gyre of the North Atlantic (40-60 • N), the rapid summer warming causes the anomalously strong pCO 2 growth (figures 4(l) and 5(h)).
The analysis of the BGC simulation (i.e.without C ant -induced warming; figures S8 and S9) shows that, in the gyres core, the fast C ant accumulation is still responsible for the early outgassing even without warming.Overall, warming makes the C ant outgassing to emerge earlier and in a larger portion of the ocean.In regions where warming has a major contribution in C ant outgassing, the emergence signals seen in the fully coupled simulation (e.g.Arctic, North Atlantic, Equatorial Atlantic) are delayed or do not happen in the BGC simulation.

Discussion and conclusions
Our work shows that the accumulation of C ant in the subtropical gyres of both hemispheres combined with the continuous surface warming leads to an increase in oceanic pCO 2 faster than elsewhere and faster than the atmospheric pCO 2 growth in these areas.In addition, the seasonal cycle of temperature pushes the pCO 2 toward even higher values during the summer months of the respective hemispheres.These factors lead to C ant outgassing signals, which emerge beyond the inter-annual variability in most of the subtropical gyres by the end of the 21st century in two contrasting future scenarios (i.e. with and without strong climate mitigation of SSP1-2.6 and SSP5-8.5),as simulated by state-of-the-art ESM.Finally, similar results are obtained when we applied our analysis on another CMIP6 ESM, ACCESS-ESM1-5 (figures S10 and S11).This ESM is "independent" and has a climate sensitivity stronger than NorESM2-LM.Thus, despite using only one model, we consider our results to be relatively robust.
The accumulation of C ant in the subtropical gyres has been previously identified in the literature with observations (Khatiwala et al 2013, Gruber et al 2019a) and models (Tjiputra et al 2010, Bopp et al 2015).This C ant accumulation pattern is driven by the large-scale circulation, e.g. the convergence of Ekman transport at the surface of the subtropical gyres and the formation and equatorward transport of mode and intermediate waters at sub-surface.The convergence of subsurface C ant into the subtropical gyres is also consistent with the early emergence of subsurface ocean acidification signals in these regions projected across an ensemble of CMIP6 ESMs (Tjiputra et al 2023).The surface warming pattern simulated in NorESM2-LM model over the 21st century is consistent with that identified in other ESMs (IPCC 2021).This future warming pattern is is also consistent with the simulated warming over the historical period in models (IPCC 2021), consistent with the observed long-term temperature change (Olonscheck et al 2020). Finally, Tjiputra et al (2014) show that ESMs simulate the strongest long-term annual trends in pCO 2 in subtropical gyres, in good agreement with observational data.
Since it is not possible to clearly identify the CO 2 molecule originating from human activities, the distinction between natural and anthropogenic air-sea CO 2 flux is often a methodological choice.In this work and as usually in the Earth system modeling world, the C ant air-sea flux is the difference between the total and natural carbon fluxes.With this choice, changes in the natural CO 2 flux driven by global warming are correctly not accounted for as anthropogenic.However, with this approach, we cannot make the distinction between C ant and natural carbon outgassing due to an increase in the Revelle Factor.This can notably be the case in the Southern Ocean during winter, where pCO 2 is more sensitive to the entrainment of deep water rich in natural DIC.This caveat has been explored in recent work suggesting that the invasion of C ant into the ocean would lead to the release of natural carbon because of changes in carbon chemistry (Holzer and DeVries 2022).In other words, the net uptake of the C ant that the model simulates in our work would actually result from an even stronger uptake of C ant balanced by the release of natural carbon.Thus, the actual C ant outgassing may occur later than estimated here.
In addition to the definition of C ant , the time of emergence identified here is not without its uncertainties.The inter-annual variability represented in ESMs is generally underestimated compared to observation, even though the variability in observation is potentially overestimated due to the sparse spatial and temporal coverage of surface ocean pCO 2 , especially in the Southern Hemisphere (Gloege et al 2021, Hauck et al 2023).These uncertainties are likely to modify the year of emergence, even if the spatial patterns should remain the same (figure S1).Our model simulations are configured as CO 2 concentration driven and not emission driven.Thus, the fluxes of carbon by the ocean does not feed back to the atmospheric concentration, which may affect the determination of C ant outgassing timescale.Finally, though a useful indication, the time of emergence usually defines a lower bound for the detection horizon of a signal (Schlunegger et al 2019) because measurement uncertainty is not included.
Our findings show that key regions of future C ant outgassing are the subtropical gyres in the summer months and the Southern Ocean in the winter months, implying that a long-term monitoring system in these domains would be highly valuable.Our analysis focuses on the monthly timescale, allowing us to consider different mechanisms acting at seasonal time scales (e.g.stronger thermal effect during summer).Indeed, previous works looking at the emergence of the trends in annual ocean CO 2 uptake give opposite results, with the subtropics having the latest time of emergence (McKinley et al 2016, Schlunegger et al 2019).This late emergence is due to the weak annual trends resulting from the opposing trends toward outgassing in summer and uptake in winter (Schlunegger et al 2019).This is a consequence of the strong seasonality in pCO 2 , which is further enhanced in the future climate scenarios (Gallego et al 2018, Landschützer et al 2018).Though surface pCO 2 in the Northern Hemisphere subtropical gyres are relatively well monitored, subtropical gyres in the Southern Hemisphere are considerably under observed (Tjiputra et al 2014, Gloege et al 2022).Notably, the South Pacific subtropical gyre has vast areas without any observational data.Given that these regions will likely release anthropogenic carbon early, our work calls for largely reinforcing the observational effort in the subtropical gyres of the Southern Hemisphere.
Finally, our study underlines the importance of understanding the long-term response of previously absorbed anthropogenic carbon by the ocean in future climate change scenarios when considering the allowable future carbon emissions toward certain climate targets (e.g. the Paris Agreement).For instance, in a future scenario with net zero emissions, one should take into account the ocean reversing role from a net carbon sink to a net carbon source, as the previously absorbed carbon is released back to the atmosphere.CE2COAST (318477), and the EU-funded project Ocean-ICU (101083922).The authors thank the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6.We also thank the climate modelling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF.The authors acknowledge KeyCLIM project (295046) for coordinating access to the CMIP6 data and Alok Kumar Gupta for his assistance in generating the NorESM2 simulations.

Figure 1 .
Figure 1.Air-sea carbon fluxes between 1850 and 2300, for the historical and two extended scenarios with the NorESM2-LM.(a) Time series of the total carbon flux averaged globally in the historical (black), SSP1-2.6 (blue) and SSP5-8.5 (red) simulations.The yellow line shows the observed carbon flux from Landschützer et al (2020).The thick lines show the yearly average, while the shadings show the monthly average.(b) Same as (a) but separating the natural (dashed line) from the anthropogenic (plain line) carbon fluxes.Maps of the C ant flux average on (c) 1990-1999, (d), (f) 2090-2099 and (e), (g) 2290-2299 for the historical, SSP1-2.6 and SSP5-8.5 simulations.Vertical grey shading in (b) shows the 3 different time periods depicted in panels (c)-(g).Positive values stand for ocean carbon uptake.

Figure 2 .
Figure 2. Illustration of the time of C ant outgassing, tout(i) as defined in equation (1).On the top, maps of the air-sea C ant flux in (a) August and (b) February 2099 in the SSP5-8.5 simulation.Below, monthly time series of the total (F tot co2 , thin grey lines) and natural (F nat co2 , thin blue lines) carbon fluxes in grid points (c) without and (d) with C ant outgassing.In (d) 50 years running mean time series of the total ( ⟨ F tot co2 ⟩ , thick orange line) and natural ( ⟨ F nat co2 ⟩ , thick blue line) carbon fluxes in February are shown.Orange shading shows the 50 years standard deviation of the total carbon flux.In February 2040 (year indicated by the vertical dashed line), C ant outgassing emerges from inter-annual variability for the first time in this grid point.Positive values stand for ocean carbon uptake.The chosen grid points for (c) and (d) are shown on (a) and (b).

⟩
and σ tot .Since, (1) the emergence patterns are weakly sensitive to σ tot (figure S1), (2) ⟨ F nat co2 ⟩ does not vary much along the simulations (figures 2(c) and (d)), (3) the emergence patterns are similar when using the preindustrial flux instead of ⟨ F nat co2 ⟩ (figure S2), then the outgassing emergence patterns are essentially determined by ⟨ F tot co2

Figure 3 .
Figure 3. (a) Time series of the surface ocean fraction where C ant is released to the atmosphere (thin lines) and where C ant outgassing signal emerges from inter-annual variability (thick lines).Earliest season (b), (d) and year (c), (e) of emerging C ant outgassing in two future scenarios: (b), (c) the SSP1-2.6 and (d), (e) the SSP5-8.5 scenarios.OND stands for October-November-December, JFM for January-February-March, AMJ for April-May-June and JAS for July-August-September.Exact month of emergence is shown in figure S4.Black contour indicate C ant outgassing occurring before 2030.

Figure 4 .
Figure 4. First order Taylor expansion of the oceanic pCO2 trends in SSP5-8.5 (equation (2)).(a) Global mean trends of surface ocean (solid black line) and atmospheric (dashed grey line) pCO2 as well the contributions from trends in surface (i) dissolved inorganic carbon and alkalinity (blue, Carbon syst.), (ii) temperature (orange, Thermal) and (iii) fresh water and salinity (green, FW & sal.).The grey patch shows the residual from the difference between the trend in pCO2 and the sum of all component trends (i)-(iii).(b to (m) Maps of the spatial anomalies of (i) the pCO2 trends (1st column), (ii) the Carbon syst.trends (2nd column) and (iii) the thermal trends (3rd column).The spatial anomalies are relative to the globally average trend (panel (a) for the time period considered.Each row shows the spatial anomalies of the trends for a different time period: January-March of the 21st and 22nd centuries (1st and 3rd rows) and July-September of the 21st and 22nd centuries (2nd and 4th rows).Dotted patches show regions where the C ant outgassing emerge from inter-annual variability for each time period.The grey contour shows where ocean pCO2 trend is stronger than pCO atm 2 .
(a), grey patch).From now we focus on the spatial anomalies of the trends (figures 4(b)-(m)) to identify regions where the trends are steeper/shallower than their global average.The absolute values of the trends are shown in figure S6.

Figure 5 .
Figure 5. Surface trends of C ant concentration (left column) and temperature (right column) for different time periods: January-March of the 21st and 22nd centuries (1st and 3rd rows) and July-September of the 21st and 22nd centuries (2nd and 4th rows).Dotted patches show regions where the C ant outgassing emerge from inter-annual variability for each time period.
Goris N, Schwinger J and Tjiputra J F 2022 Stratification constrains future heat and carbon uptake in the Southern Ocean between 30 • S and 55 • S Nat.Commun.13 340 Bronselaer B and Zanna L 2020 Heat and carbon coupling reveals ocean warming due to circulation changes Nature 584 227-33 Bronselaer B, Zanna L, Munday D R and Lowe J 2016 The influence of Southern Ocean winds on the North Atlantic carbon sink Glob.Biogeochem.Cycles 30 844-58 Bronselaer B, Zanna L, Munday D R and Lowe J 2018 Southern Ocean carbon-wind stress feedback Clim.Dyn.51 2743-57 Brown P J et al 2021 Circulation-driven variability of Atlantic anthropogenic carbon transports and uptake Nat.Geosci.14 571-7 Corbière A, Metzl N, Réverdin G, Brunet C and Takahashi T 2007 Interannual and decadal variability of the oceanic carbon sink in the North Atlantic subpolar gyre Tellus B 59 168-78 Davila X, Gebbie G, Brakstad A, Lauvset S K, McDonagh E L, Schwinger J and Olsen A 2022 How Is the Ocean Anthropogenic Carbon Reservoir Filled?Glob.Biogeochem.Cycles 36 e2021GB007055 DeVries T 2014 The oceanic anthropogenic CO2 sink: storage, air-sea fluxes and transports over the industrial era Glob.Biogeochem.Cycles 28 631-47 Fay A R and McKinley G A 2013 Global trends in surface ocean pCO2 from in situ data Glob.Biogeochem.Cycles 27 541-57 Friedlingstein P et al 2006 Climate-carbon cycle feedback analysis: results from the C4MIP Model intercomparison J. Clim.19 3337-53 Friedlingstein P et al 2022 Global carbon budget 2021 Earth Syst.Sci.Data 14 1917-2005 Gloege L et al 2021 Quantifying errors in observationally based estimates of ocean carbon sink variability Glob.Biogeochem.Cycles 35 e2020GB006788 Gloege L, Yan M, Zheng T and McKinley G A 2022 Improved quantification of ocean carbon uptake by using machine learning to merge global models and pCO 2 data J. Adv.Model.Earth Syst.14 e2021MS002620 Gloor M, Gruber N, Sarmiento J, Sabine C L, Feely R A and Rödenbeck C 2003 A first estimate of present and preindustrial air-sea CO2 flux patterns based on ocean interior carbon measurements and models Geophys.Res.Lett.30 10-1-10-4 Gorgues T, Aumont O and Rodgers K B 2010 A mechanistic account of increasing seasonal variations in the rate of ocean uptake of anthropogenic carbon Biogeosciences 7 2581-9 Goris N, Tjiputra J, Schwinger J and Heinze C 2015 Responses of carbon uptake and oceanic pCO2 to climate change in the North Atlantic: a model study with the Bergen Earth System Model Glob.Biogeochem.Cycles 29 1567-83 Gruber N et al 2009 Oceanic sources, sinks and transport of atmospheric CO2 Glob.Biogeochem.Cycles 23 n/a surface pCO2 trends from observations and models Tellus B 66 23083 Tjiputra J and Gupta A K 2023 What goes in must come out: the oceanic outgassing of anthropogenic carbon: simulations outputs [Dataset] Zenodo https://doi.org/10.5281/zenodo.10027151 Toyama K, Rodgers K B, Blanke B, Iudicone D, Ishii M, Aumont O and Sarmiento J L 2017 Large reemergence of anthropogenic carbon into the ocean's surface mixed layer sustained by the ocean's overturning circulation J. Clim.30 8615-31 Wanninkhof R 2014 Relationship between wind speed and gas exchange over the ocean revisited Limnol.Oceanogr.Methods 12 351-62 Wetzel P, Winguth A and Maier-Reimer E 2005 Sea-to-air CO2 flux from 1948 to 2003: a model study Glob.Biogeochem.Cycles 19 GB2005 Zhai P, Rodgers K B, Griffies S M, Slater R D, Iudicone D, Sarmiento J L and Resplandy L 2017 Mechanistic drivers of reemergence of anthropogenic carbon in the equatorial Pacific Geophys.Res.Lett.44 9433-9