Advective Controls on the North Atlantic Anthropogenic Carbon Sink

Though it is clear that the North Atlantic is the site of the highest storage of anthropogenic carbon (Cant) per area, it is uncertain whether the air‐sea Cant fluxes contributing to North Atlantic Cant storage occur in the subpolar gyre or upstream in the subtropical gyre. Using data and models, we show that air‐sea Cant uptake capacity is advected into the subpolar gyre along the same subsurface pathway as nutrients. This pathway is known as the nutrient stream. On the A22 section between Woods Hole and Bermuda, nutrient stream waters are the oldest in the upper 2,000 m and contain low Cant. These northward moving waters are sufficiently depleted in Cant such that they could sustain a subpolar air‐sea Cant flux of −0.19 Pg Cant year−1. The ocean hindcast model used here indicates that despite some subtropical re‐circulation, uptake capacity is transported into the subpolar gyre where it sustains subpolar air‐sea Cant uptake. With this model, we show that high‐ and low‐end estimates of subpolar air‐sea Cant flux are reconciled by accounting for a factor of two difference in their respective study areas. If half of the observed air‐sea Cant uptake capacity transport at A22 is ventilated in the subpolar region, it can fully support high‐end estimates of the subpolar air‐sea Cant sink (−0.09±0.01 Pg Cant year−1).


Introduction
The ocean is a sink for approximately 40% of anthropogenic carbon (C ant ) emissions resulting from fossil fuel combustion and cement production (Ciais et al., 2013;Khatiwala et al., 2009). Before the Industrial Revolution, it is believed that there was a steady state natural carbon (C nat ) cycle. C nat is brought to the surface by ocean circulation and is consumed by phytoplankton, whose remains and waste products ultimately sink and are remineralized, returning the carbon to depth. This process contributes to more C nat being stored at depth. Regional surface heating and cooling tendencies also play an important role in the C nat cycle. The cooling of northward moving subtropical waters favors absorption of atmospheric CO 2 and eventually results in the transformation of subtropical waters into deep waters, thus also enhancing deep water C nat storage. C ant has been added to the atmosphere and is absorbed because the surface ocean is out of equilibrium with the rapidly growing C ant perturbation. To date, there is no evidence of significant direct interaction between the C nat and C ant components of the carbon cycle; therefore we assume that the C ant behaves as a passive tracer, unaffected by biology. Contemporary carbon (C con ) is the sum of the C nat and C ant components, and C con is what is directly measured. Estimates of the relatively small C ant signal contained in C con must be inferred from other biogeochemical tracers.

10.1029/2019GB006457
C ant is stored primarily in the upper ocean, mostly in mode and intermediate waters (Gruber et al., 2019;Sabine et al., 2004). While approximately 35-50% of ocean C ant is stored in subpolar mode waters and intermediate waters ( = 26.5-27.5 kg m −3 ), these water masses account for only 20% of ocean volume (Iudicone et al., 2016). Mode and intermediate waters dominate air-sea C ant uptake, a result of winds and buoyancy fluxes shaping a water mass with high outcrop area to volume ratio (Iudicone et al., 2016). The Southern Ocean and North Atlantic are primary ventilation regions for mode and intermediate waters, and accordingly, air-sea C ant uptake occurs primarily in those regions (Mikaloff Fletcher et al., 2006). These regions sustain high C ant fluxes due to the upwelling of waters with a partial pressure of CO 2 (pCO 2 ) that is close to preindustrial values, and therefore these waters are low in C ant .
The North Atlantic basin is the most intense air-sea C ant sink (Mikaloff Fletcher et al., 2006) and the site of rapid injection of C ant into the deep ocean facilitated by Labrador Sea deep convection (Yashayaev & Loder, 2016). However, the spatial distribution of air-sea C ant uptake between the subtropical and subpolar North Atlantic is under debate (Mikaloff Fletcher et al., 2006;Pérez et al., 2013). Ocean inversions suggest subpolar air-sea C ant uptake is approximately half (−0.09 ± 0.01 Pg C ant year −1 ) of North Atlantic uptake (−0.22 ± 0.02 Pg C ant year −1 for 36-76 • N) (Mikaloff Fletcher et al., 2006). However, a recent observational analysis suggests that most air-sea C ant uptake occurs in the subtropics, and the subpolar air-sea C ant sink is minimal (Pérez et al., 2013). Differing estimates of subpolar air-sea C ant flux may be attributable simply to substantially different regional boundaries used to define the subpolar North Atlantic. Mikaloff Fletcher et al. (2006) utilize latitudinal boundaries of 49 • N to 76 • N to define the subpolar North Atlantic, whereas Pérez et al. (2013) set the southern boundary of the subpolar North Atlantic along the Portugal-Greenland A25 (∼40-60 • N) section and northern boundary at 78 • N. The latter definition leaves out the Labrador Sea and a significant portion of the Irminger Sea from the subpolar region. Here we explore the effects of various study regions on the magnitude of subpolar air-sea C ant flux estimates.
Mechanisms controlling the interannual variability of the North Atlantic C con and C ant sink have been investigated in the context of the changes in the NAO state from the early 1990s to the early 2000s (Pérez et al., 2013;Thomas et al., 2008;Ullman et al., 2009). We will expand upon this work in our effort to determine what controls the magnitude of the subpolar C ant sink. Pérez et al. (2013) investigate the role of the North Atlantic Oscillation (NAO) and the Atlantic Meridional Overturning Circulation (AMOC) in subpolar North Atlantic C ant sink variability from 1997 to 2004. Using hydrographic data, Pérez et al. (2013) find that lateral transport and storage rates of C ant were higher in 1997 than in 2004. From 1989 to 1995, there was an extended period of high wintertime NAO (Hurrell & National Center for Atmospheric Research Staff (eds), 2018). Under the assumption that circulation in 1997 is representative of the high NAO period (1989)(1990)(1991)(1992)(1993)(1994)(1995) and that 2004 is representative of the ensuing low NAO period (2002)(2003)(2004)(2005)(2006), Pérez et al. (2013) conclude that subpolar North Atlantic air-sea C ant flux was more intense in 1997 than in 2004 due to higher NAO and AMOC in 1997. However, the changes in the air-sea C ant flux are not statistically significant given uncertainty bounds. Using two different hindcast models, Thomas et al. (2008) and Ullman et al. (2009) find that as the NAO index declines from the 1990s to the 2000s, there is a reduction of air-sea C con uptake in the eastern subpolar gyre and increases in air-sea C con uptake in the western subpolar region. Because the available hydrographic cruise data require the southern boundary for the subpolar gyre to be at A25, the analysis of Pérez et al. (2013) likely misses a compensating region of increased air-sea C con uptake in the west that was identified these two modeling studies.
To improve mechanistic understanding of subpolar air-sea C ant uptake, we investigate the lateral supply of intermediate waters into the subpolar North Atlantic. Lagrangian drifters indicate very limited net flow of surface subtropical waters into the subpolar gyre (Burkholder & Lozier, 2011. Instead, parcels follow subsurface pathways to the surface of the subpolar gyre, meaning that mode and intermediate waters in the subtropics eventually reach the surface in the subpolar gyre. The pathways of mode and intermediate waters have also been investigated in the context of the nutrient stream (Palter & Lozier, 2008;Pelegri et al., 1996;Sarmiento et al., 2004;Williams et al., 2006). We expand upon these nutrient stream studies by considering how tracer distributions in nutrient stream waters can influence air-sea C ant uptake. The nutrient stream is a subsurface region of high nutrients embedded within the western boundary current that supplies the subpolar North Atlantic with nitrate (NO 3 ) and supports its intense subpolar primary production (Pelegri et al., 1996;Williams et al., 2011). NO 3 increases with depth due to biological removal at the surface and remineralization at depth. In contrast, the surface C ant source Global Biogeochemical Cycles 10.1029/2019GB006457 means that C ant decreases with depth. Steeply sloping isopycnals result in isopyncal transport bringing high NO 3 concentrations, and possibly low C ant concentrations, toward the surface where they then intersect with high velocity currents. Deep winter mixing at higher latitudes allows for these isopycnals to outcrop and for the nutrients to ultimately surface in the subpolar North Atlantic (Williams et al., 2011). Nutrient stream transport occurs primarily in intermediate waters ( = 27.0-27.5 kg m −3 ) that dominate net volume transport into the subpolar gyre (Burkholder & Lozier, 2011). This interpretation is supported by model experiments (Williams et al., 2006) where idealized tracer released in dense waters of the subtropical North Atlantic outcrops in the subpolar gyre, while tracer released in lighter subtropical surface waters tends to recirculate in the subtropical gyre. Lateral supply of nutrients in the nutrient stream is much greater than supply by Ekman suction within the subpolar gyre (Whitt, 2019;Williams et al., 2006). Thus, studies of both simulated tracer release and observed drifter trajectories indicate that the surface of the subpolar gyre is irrigated from the south with nutrients originating in mode and intermediate waters. Here we investigate whether the nutrient stream also irrigates the subpolar gyre with low C ant waters.
Nutrient stream waters are mostly composed of the remnants of Antarctic Intermediate Water (AAIW) (Palter & Lozier, 2008;Sarmiento et al., 2004;Sen Gupta & England, 2007), and their slow northward transit beneath the euphotic layer (Sen Gupta & England, 2007) allows for high remineralized nutrient concentrations (Williams et al., 2011). Respiration of organic matter also drives down O 2 concentration and results in a large difference between measured and saturation O 2 (Apparent Oxygen Utilization, AOU). The high AOU found in nutrient stream waters (Williams et al., 2011) indicates that they are relatively old, that is, have not been in contact with the atmosphere for many decades. The above evidence suggesting high nutrient stream water mass age is consistent with the nutrient stream containing low C ant waters.
The supply of waters with the potential to take up anthropogenic CO 2 , low in C ant , is necessary to sustain a subpolar C ant sink. In the Southern Ocean, the supply of low C ant comes from Circumpolar Deep Water that is upwelled to the surface, accumulates anthropogenic carbon, and is then subducted in mode and intermediate waters (Sabine et al., 2004). Relative to the rate of increase in atmospheric CO 2 , circulation timescales (Sen Gupta & England, 2007) are long enough that these waters have the potential to absorb more anthropogenic carbon when they eventually resurface in subpolar North Atlantic. If nutrient stream waters contain sufficiently low C ant , their ventilation in the subpolar gyre may contribute significantly to air-sea C ant uptake in the subpolar North Atlantic.
Our objective is to quantify the northward transport of nutrient stream waters with a substantially lower C ant content. We will determine the degree to which transport in the nutrient stream density band, which has been shown to outcrop in the subpolar North Atlantic, is sufficient to sustain subpolar North Atlantic air-sea C ant uptake. With our analysis of C ant transport, we improve our understanding of the mechanisms driving subpolar-subtropical partitioning of North Atlantic air-sea C ant uptake. We will test the following hypothesis: Sufficient C ant uptake capacity to sustain the subpolar North Atlantic C ant sink is advected northward in the nutrient stream.
In order to determine the pathways of low C ant waters in the North Atlantic, we use a gridded biogeochemical data product  and ocean circulation from the best estimate of the physical state (Forget et al., 2015). Though there are interpolation issues with the gridded data product (section 2.7), we can qualitatively understand these effects by comparing the gridded data product to a hindcast simulation (Yeager et al., 2018) with internally consistent circulation and biogeochemistry. We estimate transport of air-sea C ant uptake capacity along these pathways using biogeochemical and hydrographic measurements from a quality controlled bottle data product . Using the hindcast simulation, we can also look at how subpolar air-sea C ant flux changes with time. After validating the hindcast simulation against transport estimates from this study and others, we investigate variability in subpolar ventilation and relate it to variability in subpolar air-sea C ant flux.

Data and Methods
We leverage a suite of observational products, model simulations, and direct observations to investigate the link between the nutrient stream and the North Atlantic C ant sink.

Gridded Observations
We use the GLODAPv2.2016b data product (GLODAPv2)  for climatological spatial distributions of biogeochemical variables. This product is an aggregation of high-quality observations of primary biogeochemical variables, collected from 1972 to 2013, and gridded to 1 • × 1 • . Most of these observations were collected as part of the World Ocean Circulation Experiment (WOCE) in the 1990s and the subsequent and ongoing CLIVAR repeat hydrography and GO-SHIP programs. For example, while GLO-DAPv2 incorporates observations over 42 years, half of the stations with DIC measurements used to make GLODAPv2 were sampled between 1997 and 2007. C ant and C nat are estimated using the Transit Time Distribution (TTD) method of Waugh et al. (2006) and are normalized to 2002. GLODAPv2 is available online at NOAA's Ocean Carbon Data System (OCADS, https://www.nodc.noaa.gov/ocads/).

Bottle Data
The GLODAPv2 bottle data product  is used to resolve tracer gradients within the Gulf Stream. Improved representation of tracer gradients, relative to the gridded GLODAPv2 product, is necessary for more accurate calculation of C ant transport because it depends critically on the intersection of velocity (section 2.5) and tracer gradients. The GLODAPv2 bottle data product features additional bias corrections and quality control on the original cruise data. Measurements of key biogeochemical variables (oxygen, DIC, and alkalinity) and transient tracers are from the October to November 2003 occupation of the WOCE A22 section (Bermuda-Woods Hole segment, Figure 1). These measurements and all derived variables are interpolated between stations along isopycnals, using Ocean Data View 4.7 (ODV), to be consistent with the physics of interior transport. The Bermuda-Woods Hole segment, centered at approximately 38 • N (Figure 1), is chosen because it is the northernmost WOCE section that is also south of the nutrient stream outcrop region at approximately 45 • N. These data are available online from NOAA's OCADS.

C ant Estimate
The C • T method (Pérez et al., 2008;Vázquez-Rodríguez et al., 2009) is used to estimate C ant along the Bermuda-Woods Hole segment. The C • T method is an upgraded version of the ΔC * back-calculation method (Gruber et al., 1996) that includes improved parameterization of preformed alkalinity and the time evolution of air-sea disequilibrium.
While the TTD method, used in GLODAPv2, and the C • T method provide similar estimates, they obtain their estimates in very different ways. The C • T method removes background biological remineralization and C nat from measured DIC to obtain a C ant estimate, while the TTD method derives C ant based on a distribution of transit times that is informed by transient tracers and assumptions about the importance of advection versus mixing. A detailed intercomparison of C ant estimation methods, including the C • T method, can be found in Vázquez-Rodríguez et al. (2009). Like the TrOCA method (Touratier & Goyet, 2004), local CFC measurements are not needed because relationships between the air-sea disequilibrium component and the transient tracers are predetermined for each basin. This allows the C • T method to be applied to sections lacking CFC observations. In the case of our analysis of A22, CFC data provide independent evidence that the waters of the nutrient stream are old. The uncertainty of C ant estimates made with the C • T method is ±5.2 μmol kg −1 (Vázquez-Rodríguez et al., 2009). Previous work has shown that estimates made with the C • T method compare well to estimates based on the TTD method and it is more accurate than the TrOCA method (Vázquez-Rodríguez et al., 2009;Yool et al., 2010).

C ant Deficit
We define a new quantity, the C ant deficit (C ant,def ), as the difference between the C ant concentration of a parcel and the C ant concentration at the surface of the subtropics. Parcels strongly negative in C ant,def , once transported to the surface, will act to lower surface ocean pCO 2 toward its preindustrial state. Thus, C ant,def serves as an estimate of air-sea C ant uptake capacity, having defined this uptake potential in reference to subtropical surface parcels that are equilibrated with the current atmosphere. This assumption is supported by observations (Pérez et al., 2013) taken along the A25 section that indicate subtropical waters in the surface layer off the coast of Portugal, and subpolar waters in the surface layer off the coast of Greenland, have similar C ant concentrations. The observational evidence is consistent with heat loss minimally affecting a parcel's C ant saturation state in the subpolar gyre (Pérez et al., 2013). Thus, transport of C ant,def is the transport of a downstream air-sea C ant uptake capacity. We estimate C ant,def in 2003 at A22 as follows: The left hand side, C ant,def is the difference between the saturation concentration of C ant the surface layer (⟨C ant ⟩ <26.5 ) and the local estimate of C ant . ⟨C ant ⟩ 2003 <26.5 is the mean concentration of C ant above the =26.5 kg m −3 surface along the Bermuda-Woods Hole A22 segment in 2003.
When this calculation is made for the hindcast model, it is made for a timeseries of years. Due the continued increase in atmospheric CO 2 , ⟨C ant ⟩ <26.5 increases over time. This means that in our model analysis C ant,def becomes increasingly negative at depth over time. This is simply due to the fact atmospheric CO 2 is continually increasing at a rate much faster than C ant can be transferred to depth by mixing and advection.

Geostrophic Transport Estimation
We use Bermuda-Woods Hole A22 hydrographic data from the GLODAPv2 bottle data product to estimate volume transport and combine this with C ant,def to estimate the transport of uptake capacity. Geostrophic velocity is calculated from baroclinic shear, assuming a level of no motion at 3,000 m. Geostrophic transport is calculated from the product of geostrophic velocity normal to the section and cross-sectional area. We choose 3,000 m because there is transport below the 2,000 m level of no motion (Pelegri & Csanady, 1991). A deeper reference level decreases carbon transport by less than 10% but increases our Gulf Stream transport estimate to 74 Sv, bringing it closer to the 97 Sv estimate of Rossby et al. (2014). We neglect any barotropic tranport; therefore our transport estimate is a low-end estimate for 2003.

State Estimate
A gridded estimate of the physical state is provided by ECCOv4, (Forget et al., 2015), a nominal 1 • global ocean state estimate. ECCOv4 is based on MITgcm and is used as a best estimate of the ocean's physical state. ECCOv4 uses adjoint data assimilation to bias correct the model with observations. This creates a state estimate that can be used to calculate internally consistent global integrals and trends. These are best-fit estimates that are consistent both with physical laws and sparse data (Wunsch, 2016). These observations include profiling floats (ARGO), satellite altimetry, and mass distribution derived from satellite (GRACE). Monthly output, provided by NASA's Jet Propulsion Laboratory, is available online for the period 1992-2015. We focus on a 1997-2007 period, centered on 2002, because carbonate system measurements in the gridded GLODAPv2 product are normalized to 2002.

Hindcast Model
We also employ a hindcast model from which C ant can be precisely determined and where biogeochemistry and circulation are internally consistent. For GLODAPv2 to provide a climatological three-dimensional fields of C ant and nutrients, significant smoothing in space and time are required. A natural consequence of this processing is to damp spatial gradients in the resulting fields. The hindcast model avoids these limitations but has significant biases (Tagklis et al., 2017). We therefore validate that the model suitably represents the large-scale circulation (section 3.4). Hindcast model output comes from the CESM Decadal Prediction Large Ensemble (CESM-DPLE) (Yeager et al., 2018). The model, NCAR's CESM1.1, is run in a forced ocean/sea-ice configuration. In this model configuration, the surface ocean is forced with historical surface forcing over the simulation period 1955-2015 (Yeager et al., 2018). In this modeling framework, there is no nudging of properties in the ocean interior. The ocean model is the nominal-1 • resolution POP version 2 with 60 vertical levels. Ocean biogeochemistry, including lower-trophic ecosystem dynamics, is simulated in the hindcast. The model includes two tracers, a C nat tracer that experiences time evolving physics but constant preindustrial atmospheric CO 2 , and a contemporary DIC tracer, C con . These can be differenced for an exact calculation of simulated C ant (Long et al., 2013). Model output is available on NCAR's GLADE file system on the Cheyenne supercomputer and online at NCAR's Earth System Grid.

Results
Here we evaluate the distribution of C ant in the nutrient stream and calculate transport of air-sea C ant uptake capacity (C ant,def ) into the subpolar North Atlantic. We also examine interannual variability of ventilation of nutrient stream C ant .

Observations and Simulation of Low C ant Pathways
The Gulf Stream flows northward along the east coast of North America, separating and then flowing into the subpolar North Atlantic (Figure 2). Currents, which are visualized as streamlines (Figure 2), are averaged over the densities of nutrient stream core ( = 27.0-27.5 kg m −3 ). The Gulf Stream separates from the coast of North America and tracks zonally at Flemish Cap off of Newfoundland (approximately 45 • N, 60 • W) at which point it is known as the North Atlantic Current. The western boundary current in the hindcast simulation ( Figure 2, bottom row) is broader than in the more-realistic state estimate. In the hindcast, the North Atlantic Current segment is biased southward and more zonal (Figure 2). These biases are common in coarse resolution models (Tagklis et al., 2017).
In the observations/state estimate, North of the Grand Banks, elevated NO 3 is embedded in the western boundary current (Figure 2c). The narrow streak of low NO 3 , and also AOU, extending from Cape Hatteras to the Grand Banks is likely an interpolation artifact; other interpolations (Palter & Lozier, 2008) lack this feature. The elevated region of nutrients in the western boundary current is bounded by =27.0-27.5 kg m −3 , is known as the nutrient stream (Palter & Lozier, 2008;Pelegri et al., 1996;Williams et al., 2006). In the hindcast model, the southward bias in the North Atlantic Current shunts the simulated NO 3 maximum southward (Figure 2d). The fact that biases in the circulation of the hindcast simulation yields biases in NO 3 is consistent with circulation being the primary determinant of nutrient gradients. In both the direct observations of nutrients combined with the best estimate of the observed circulation, and in the hindcast model, the western boundary current transports high concentrations of NO 3 and AOU into the subpolar gyre.
Correspondence between high AOU, high NO 3 , low C ant , and high velocities are present in both observations and the hindcast simulation (Figure 2). AOU and NO 3 concentration decreases, and C ant concentration increases, along the streamlines leading into the subpolar gyre. South of the subpolar gyre, the GLODAPv2 C ant estimate is higher than the CESM hindcast. Ventilation timescales in the CESM hindcast simulation that are longer than observed (Long et al., 2013) contribute to differences between the C ant field in the CESM hindcast simulation and the C ant field from the GLODAPv2 gridded data product. Also, the GLO-DAPv2 TTD based C ant estimate is biased high by approximately 30% in intermediate waters in the subtropics due to assuming a uniform ratio of advective/diffusive transport (Steinfeldt et al., 2009). The assumption of a constant air-sea disequilibrium in TTD calculations can also contribute to a high bias in observations. The discrepancy between C ant concentrations in the hindcast simulation and data-based estimates is due to biases in both products.
In the observations/state estimate, high nutrients and AOU (Figures 2c and 2e) are found in layers that have not recently encountered the surface, suggesting nutrient stream waters have not had significant contact with the anthropogenic atmospheric CO 2 transient. The simulation also features high AOU and nutrients in the nutrient stream (Figures 2d and 2f). Consistent with an older water mass, there is a C ant minimum embedded within the nutrient stream in both observations (Figure 2a) and the hindcast simulation (Figure 2b).

Observed Vertical Structure of C ant at A22 (38 • N)
The lower isopycnals of the nutrient stream are on density surfaces that outcrop in the subpolar North Atlantic (Pelegri et al., 1996;Sarmiento et al., 2004;Williams et al., 2006). The outcrop location is where these waters can restore the surface ocean pCO 2 toward its preindustrial value and support significant air-sea C ant fluxes. Using A22 observations, we investigate the structure and transport of C ant,def in the nutrient stream.
There are three water masses in the Gulf Stream region at 38 • N. Distribution of all water mass properties generally follows isopycnals (Figures 3a-3d); thus we use isopycnal boundaries in our analysis. Waters down to the = 26.5 kg m −3 surface are subtropical waters (Palter et al., 2005;Williams et al., 2011). These waters are low in nutrients and AOU and high in C ant (Figures 3a, 3b, and 3d), consistent with relatively recent ventilation. At =26.5-27.0 kg m −3 are thermocline waters that form the upper portion of the nutrient stream (Williams et al., 2011). Below this layer, between =27.0-27.7 kg m −3 , lie the intermediate waters, that contain the highest nutrient concentrations. Waters in this density range outcrop in the subpolar gyre (Williams et al., 2011). Intermediate waters have a high nutrient concentration and high AOU, consistent with a large contribution from an AAIW end member (Palter & Lozier, 2008). Our analysis illustrates that these nutrient stream waters ( =27.0-27.7 kg m −3 ) also contain very little C ant (Figure 3d).
As noted in section 2.3, local CFC-11 data are not used in our C ant estimate and thus can be used for qualitative comparison to the C ant distribution. The minimum of CFC-11 (Figure 3c) in the nutrient stream is in accordance with older waters with low C ant . Waters below the nutrient stream ( > 27.7 kg m −3 ) are elevated in C ant and CFC-11. These waters are primarily composed of North Atlantic Deep Water, formed in the Labrador Sea, and have been ventilated more recently than the nutrient stream source waters (Hinrichsen & Tomczak, 1993;Smethie et al., 2000). Both AOU and CFC-11, independent tracers with age-dependent concentrations, indicate nutrient stream waters at this location are the oldest waters in the upper 2,000 m (Figurea 3b and 3c).

Observed Lateral Transport of C ant at A22 (38 • N)
Here we investigate lateral transport of C ant through the A22 (Bermuda-Woods Hole segment) section at approximately 38 • N. We also estimate the degree to which lateral transport of C ant,def can sustain the subpolar North Atlantic air-sea C ant sink. Values given here are synoptic estimates for 2003. The highest estimate of C ant concentration is in subtropical surface waters ( <26.5 kg m −3 , Table 1). This layer intersects with the Gulf Stream (Figures 3d and 3e), where we estimate velocities normal to the Bermuda-Woods Hole A22 section of over 1 m s −1 . Our observational estimate of volume transport normal to the Bermuda-Woods Hole section is 46 Sv. Associated with the total volume transport, 0.80 Pg C year −1 of anthropogenic carbon are transported northward across the Bermuda-Woods Hole section. Due to the strong vertical gradient of C ant , lateral C ant transport is maximized in the upper layers (Table 1).  Observational estimates of air-sea C ant uptake capacity (C ant,def , see section 2.4) are most negative between the =26.5 kg m −3 and =27.7 kg m −3 isopycnal surfaces. Waters in the nutrient stream are depleted in C ant relative to surface waters and have an enhanced C ant,def . Our sign convention is such that a negative C ant,def transport corresponds with potential downstream air-sea C ant uptake. C ant,def transport through this section within the nutrient stream ( =26.5-27.7 kg m −3 ) is −0.19 Pg C ant year −1 (Table 1), indicating a northward transport of potential air-sea C ant uptake. The most intense negative C ant,def transport occurs on the =27.0 kg m −3 surface between 275 and 400 km (Figure 3f). There is anomalously high C ant above the nutrient stream resulting in northward transport of positive C ant,def , indicating transport of waters that have little capacity to absorb additional C ant . Adjacent vertical streaks of less intense positive (440 km) and negative (500 km) C ant,def transport are indicative of eddies along the section and minimally affect the large-scale transport.

Simulated Transport at A22 and the Subpolar Gyre Boundary
Here we validate simulated transport of C ant,def against observations in order to determine if we can use the hindcast simulation to investigate basin-wide air-sea C ant fluxes and interannual variability. We also estimate the fraction of C ant,def transport at A22 that reaches the subpolar gyre. The subpolar gyre is defined as the area between 49 • N and 65 • N, consistent with the winter outcrop region of the nutrient stream core (Figure 1). At A22 and AR19, overturning and volume transport are well simulated in the CESM hindcast (Table 2). Simulated C ant,def transport is higher than observed (Table 3). This is due to a higher surface C ant concentration (Equation 1) in the hindcast simulation (observed: 45.0 μmol kg −1 , simulated: 51.7 μmol kg −1 ).
In observations, there is a cross stream surface C ant gradient with slightly lower C ant slope waters along the coast, high C ant waters in the Gulf Stream, and slightly lower subtropical waters further offshore (Figure 3d). This gradient is greatly reduced in the hindcast simulation at A22, because there is no slope water simulated  (Pelegri et al., 1996;Williams et al., 2006). b This section is comparable to the downstream distance of A22, ∼800 km from where the simulated Gulf stream separates from the coast. at this location. This gradient does exist, however, further downstream. This downstream displacement is due to a simulated Gulf Stream that separates further downstream than is observed (Figure 2).
The A22 section is approximately 800 km northeast of Cape Hatteras, where the Gulf Stream is observed to separate from the coast. A north-south section along 60 • W in the hindcast is at the observed downstream distance of A22 and is potentially more consistent with observed conditions at A22. If we sample the model at this longitude, with the same section length as A22, then volume transport is the same as at 60 • W (43 Sv), but the C ant,def transport is closer to observations (Table 3), consistent with a Gulf Stream surface concentration (48.7 μmol kg −1 ) that is closer to observed.
The observed fraction of C ant,def reaching the subpolar gyre, relative to the portion that is recirculated in the subtropical gyre, was calculated by others from observations of nutrient transport (Pelegri et al., 1996;Williams et al., 2006). The observed fraction is approximately 50% for layers between = 26.5-27.7 kg m −3 .
In the hindcast simulation, we can calculate this fraction directly from C ant,def transport. To calculate the fraction reaching the subpolar region in the hindcast, we take the C ant,def transport at A22 and 60 • W and divide by the C ant,def transport at subpolar boundary (49 • N). In our hindcast simulation this fraction is lower than this previous observation of 50%. We find 33% of the transport at A22 is transferred to the subpolar gyre but closer to observed if transport at 60 • W is used (41%) ( Table 3).

Air-Sea C ant Fluxes and the Subpolar C ant,def Budget
Here we compare the air-sea C ant flux of the CESM hindcast to observational estimates. We also calculate C ant,def transports into and out of the subpolar North Atlantic, between 49 • N and 65 • N. We analyze air-sea Previous studies have estimated air-sea C ant fluxes over quite different areas (Figure 4). Using simulated air-sea C ant flux from the CESM hindcast we estimate the effect varying areas has on subpolar air-sea C ant flux estimates. The simulated air-sea C ant flux is approximately twice as high in our study region as compared to the study region of Pérez et al. (2013) that is substantially smaller (Figure 4). The difference between the two prior observational estimates of the air-sea C ant flux (Mikaloff Fletcher et al., 2006;Pérez et al., 2013) is consistent with the difference in the simulated air-sea C ant flux integrated over the respective study regions. The smallest study region excludes the Labrador Sea and much of the nutrient stream outcrop region (A25-65 • N; Figure 1). The inclusion of the Labrador Sea is necessary to encompass the outcrop region of the nutrient stream (Figure 5d). For our subpolar study region, we select a northern boundary of 65 • N. This marks the boundary between the subpolar gyre and the Nordic Seas, which are separated physically by the Iceland-Scotland Ridge.
Using the hindcast, we estimate transport of C ant,def into the study region, to determine how the air-sea C ant flux is sustained. Northward simulated C ant,def transport is maximized above 830 m; therefore we select this model vertical level as the bottom boundary of our subpolar box. Deep mixing in the subpolar gyre reaches these depths every year and ventilates these waters. Supply of low C ant waters, as indicated by negative C ant,def , supports the air-sea C ant flux by lowering the carbon concentration of surface waters. The largest negative source of C ant,def to the upper 830 m of the subpolar box is from lateral advection in the nutrient stream ( Figure 5). The mean nutrient stream lateral advective transport of −0.12 Pg C ant,def year −1 is sufficient to support the mean air-sea C ant flux, −0.11 Pg C ant year −1 ( Figure 5). These waters enter the region in the upper (northward moving) limb of the overturning circulation, moving along isopycnals (Burkholder & Lozier, 2011) toward the surface. Because the circulation is rapid in the upper limb of the overturning circulation, there is little growth in atmospheric CO 2 during the transit, and thus there is little time for C ant,def concentration to change ( Figure 6). The C ant,def that is supplied to the mixed layer in the nutrient stream is eliminated as C ant is added to the mixed layer by the air-sea C ant flux. By the time a parcel leaves the surface, C ant,def is ∼0 (Figures 2 and 6). In the lower (southward moving) limb of the overturning, below =27.7 kg m −3 , waters move southward and C ant,def is removed from the subpolar region. The transport is positive below =27.7 kg m −3 because C ant,def is negative and the velocity is southward, also negative, therefore advective transport, which is the product of velocity and C ant,def , is positive. Waters below =27.7 kg m −3 have not been in contact with the atmosphere for several years, and over time atmospheric CO 2 has increased. Thus the deeper waters, which have maintained relatively constant C ant after leaving the mixed layer in the subpolar gyre, have more negative C ant,def . Diapycnal diffusion across the bottom boundary acts to make C ant,def more negative by removing carbon from the upper 830 m, while vertical advection acts to make it more positive via the downwelling of low carbon waters. The downwelling occurs primarily in the southeast corner of the study region, where the subtropical gyre extends slightly into the study region. The diapycnal diffusion C ant,def source is canceled by a vertical advection C ant,def sink that is twice as large as the diapycnal diffusion source ( Figure 5). Fluxes at the northern boundary are small. Integrating the fluxes over the region, there is a net positive growth of C ant,def ( Figure 5, Storage), that is, a loss of uptake capacity within the subpolar box as C ant is absorbed from the atmosphere. Despite the loss of C ant uptake capacity, air-sea C ant uptake continues, because the surface C ant concentration is following the increase in atmospheric CO 2 . As long as atmospheric CO 2 continues to increase, subsequent years will see more C ant,def supplied to the upper 830 m. In summary, the northward flux of C ant,def is satisfied by the air-sea C ant flux at the surface and the export of C ant,def below =27.7 kg m −3 can be attributed to increasing atmospheric CO 2 ( Figure 6).

Variability of Lateral C ant,def Transport
We also use the CESM hindcast to examine variability of the C ant,def transport and air-sea C ant flux. We focus on the southern boundary at 49 • N because we have shown above that the dominant source of C ant,def is provided through the southern boundary. We compare the high NAO period of the 1990s to the low NAO period of the 2000s, focusing on sustained periods of high (1992)(1993)(1994)(1995) and low NAO (2003NAO ( -2006 (Yashayaev & Loder, 2016).
For the two periods of interest, the simulated air-sea flux is within ∼0.01 Pg C ant,def year −1 of the C ant,def transport at 49 • N and 0.04 Pg C ant,def year −1 at 60 • W (Figure 7b, Table 3). Lateral C ant,def transport in the upper limb of the overturning circulation exceeds the air-sea flux for both periods. In the lower limb (southward moving), C ant,def is exported southward (Figures 5 and 6).
Air-sea C ant uptake increased (more negative) from the high NAO period (1992)(1993)(1994)(1995) to the low NAO period (2003)(2004)(2005)(2006) (Figure 4). An increase in ocean uptake is the expected response to the observed increase in atmospheric CO 2 . Assuming a transient steady state, we estimate this expected increase as in Mikaloff Fletcher et al. (2006) and Gruber et al. (2019). Figure 7a are the difference between expected and simulated variables and are the result of interannual climate variability. The air-sea C ant,def flux shown in Figures 5 and 6 is the same as the air-sea C ant flux, except that we have assigned a negative sign to the air-sea C ant flux, indicating atmospheric C ant removal. Air-sea C ant uptake increased less than the increase expected due to increased atmospheric CO 2 . Thus air-sea C ant uptake was anomalously high (more negative) in the 1992-1995 period and anomalously low (more positive) for the 2003-2006 period. At the same time, C ant,def transport was more negative in the early period compared to the late period, consistent with the simulated enhancement of air-sea C ant uptake anomalies. However, the simulated variability in C ant,def transport is greater than that for the simulated air-sea C ant flux.

Anomalies shown in
Changes C ant,def transport are due either to changes in C ant,def or changes in volume transport. Anomalously negative C ant,def transport during the high NAO period is due to increased volume transport in the nutrient stream core (Figures 7a and 7d). The ocean response to high NAO periods is increased formation of dense Labrador Sea Water (Yashayaev & Loder, 2016). In the high NAO period of 1992-1995, simulated northward volume flux was higher in dense waters ( =27.0-27.7 kg m −3 ) that form the core of the nutrient stream ( Figure 7d). During positive NAO events, formation of denser waters in the subpolar region acts to enhance the east-west density gradient and increase northward transport at depth (Lozier et al., 2010). In the low NAO period, there was reduced deep water formation and reduced volume transport in dense waters. Reduced northward transport in the nutrient stream core results in anomalously positive C ant,def transport (Figures 7a  and 7d). This is consistent with the transition seen in air-sea C ant flux. During periods of positive NAO, more uptake capacity is advected northward relative to what would be expected due to atmospheric CO 2 increase, which allows for anomalously negative air-sea C ant uptake.
Denser surface waters are not necessarily limited to regions of deep convection during high NAO events. Wherever the surface is denser relative NAO neutral periods, the deeper mixed layers there will penetrate further into the nutrient stream and therefore access more negative C ant,def . Likewise, when the state of the NAO is lower, mixed layers are shallower, and less C ant,def is accessed. Enhanced ventilation of the nutrient stream layer that occurs during positive NAO periods also contributes to anomalously negative air-sea C ant uptake.
Although there is decreased northward volume flux in the dense waters, total C ant,def transport becomes more negative as the NAO index transitions from high to low. The more negative northward C ant,def transport 10.1029/2019GB006457 is consistent with the increase in atmospheric CO 2 (Figure 6b). Uptake capacity grows along with atmospheric CO 2 increase, permitting the continued growth of the sink over time.

Discussion
The nutrient stream has been demonstrated to be an important source of nutrients to the subpolar North Atlantic Ocean (Palter & Lozier, 2008;Pelegri et al., 1996;Williams et al., 2011). We show that low C ant waters are also transported through the subtropics and into the subpolar gyre. Our results indicate the transfer of strongly negative C ant,def (air-sea C ant uptake capacity) in intermediate waters from the subtropical to subpolar gyre sustains subpolar air-sea C ant uptake. The magnitude of subpolar air-sea C ant fluxes are only ∼10% relative to the advective fluxes of C ant at A22 0.8 Pg C ant year −1 . This is consistent with the C ant in subpolar mode water being primarily sourced from upstream air-sea C ant uptake in subtropical waters (Iudicone et al., 2016;Levine et al., 2011). In the absence of a supply of low C ant waters, surface cooling would result in air-sea C nat uptake and lower the DIC to alkalinity ratio, which would progressively limit C ant absorption (Vlker et al., 2002). Here we show that the subsurface supply of low C ant waters and the subpolar exposure of these waters supports significant air-sea C ant uptake. While most uptake occurs upstream, downstream air-sea exchange of C ant in denser waters establishes the subpolar region as the site of approximately half (−0.09±0.01 Pg C ant year −1 ) of North Atlantic air-sea C ant flux (−0.22±0.02 Pg C ant year −1 ) (Mikaloff Fletcher et al., 2006), and as one of the most intense regions of air-sea C ant flux per area.
The magnitude of the subpolar North Atlantic air-sea C ant sink is still an important question. Forward models and inverse estimates agree that this region is an air-sea C ant sink (Long et al., 2013;Mikaloff Fletcher et al., 2006). Pérez et al. (2013) suggest the region is a minor air-sea sink for C ant , because northward flowing intermediate waters attain C ant saturation via air-sea exchange in the subtropics; however, a southern boundary at A25 (Pérez et al., 2013) excludes much of the western outcrop region of the nutrient stream, including the Labrador Sea, and reduces subpolar area by 50% (7.3 × 10 6 km 2 to 3.8 × 10 6 km 2 ) relative to the subpolar area enclosed by the southern boundary of Mikaloff Fletcher et al. (2006). A larger subpolar region is supported by the results of Ullman et al (2009) and global biome mapping efforts (Fay & McKinley, 2014). Our results illustrate that a 49 • N southern boundary ( Figure 6, Mikaloff Fletcher et al. 2006) for the subpolar North Atlantic is more fully representative of the region because it includes the important outcrop area for nutrient stream waters that carry a large air-sea C ant uptake capacity.
Low C ant intermediate waters ultimately outcrop in the subpolar region (49 • N-65 • N). Northward transport of C ant,def at A22 provides a potential air-sea C ant uptake capacity of −0.19 Pg C year −1 , but only some of this reaches the surface of the subpolar region. If half of the C ant,def transport reaches the subpolar region (Pelegri et al., 1996;Williams et al., 2006), we estimate a subpolar North Atlantic air-sea C ant sink of −0.10 Pg C ant year −1 . Our air-sea C ant sink estimate is comparable to the −0.09 ± 0.01 Pg C ant year −1 ocean inverse estimate of Mikaloff Fletcher et al. (2006) and the CESM hindcast model used here.
North Atlantic carbon cycle variability occurs in response to the NAO (Corbière et al., 2007;Pérez et al., 2013;Schuster et al., 2013;Thomas et al., 2008;Ullman et al., 2009;Watson et al., 2009). We find enhanced nutrient stream C ant,def transport during high NAO periods, consistent with an enhanced subpolar air-sea C ant sink during these periods (Fröb et al., 2016;Pérez et al., 2013). A recent observational analysis indicates that after the high NAO period of the early 1990s, subpolar C ant accumulation was anomalously low, with the greatest North Atlantic reductions occurring in the outcrop region of the nutrient stream (Gruber et al., 2019). The results of Gruber et al. (2019) are compatible with nutrient stream ventilation variability, illustrated here, as a driver of long-term subpolar air-sea C ant flux variability. Decreased transport in dense waters in the CESM hindcast is consistent with reduced deep water formation and reduced nutrient stream ventilation. Reductions to nutrient stream ventilation are consistent with reduced vertical supply of C con (Ullman et al., 2009) and reduced air-sea fluxes during neutral NAO periods in hindcast simulations of the carbon cycle (Thomas et al., 2008;Ullman et al., 2009). Over a similar period, observations indicate that the air-sea pCO 2 gradient decreased in the vicinity of the North Atlantic Current (McKinley et al., 2011;Schuster et al., 2009). In summary, our proposed mechanism is consistent with historical subpolar carbon cycle variability and enhanced subpolar air-sea C ant uptake when the wintertime NAO index is high.
Here we approach this problem with a traditional Eulerian framework in a coarse resolution model. The CESM hindcast suggests more C ant,def transport than observed at A22, and a greater fraction recirculates in the subtropical gyre than suggested by observations (Pelegri et al., 1996;Williams et al., 2006). Lagrangian Global Biogeochemical Cycles 10.1029/2019GB006457 methods could be used to resolve where the nutrient stream waters eventually outcrop (Burkholder & Lozier, 2014). These methods could yield more accurate estimates of the fraction of nutrient stream waters reaching the subpolar gyre.

Conclusion
The nutrient stream has been defined as the region of dense waters within the western boundary current that is elevated in nutrients. As indicated by high AOU and CFC-11, source waters of the nutrient stream ( =26.5-27.7 kg m −3 ) are some of the oldest waters in the upper 2,000 m. They have been sequestered from the atmosphere for long enough that they are highly depleted in C ant relative to subtropical waters. These waters are from the tropical Atlantic (Palter & Lozier, 2008) and ultimately may have an AAIW source (Sarmiento et al., 2004;Sen Gupta & England, 2007).
Global ocean C ant uptake occurs primarily in the North Atlantic and Southern Ocean (Khatiwala et al., 2013;Mikaloff Fletcher et al., 2006;Sabine et al., 2004). Here we focus on the North Atlantic which features high intensity air-sea fluxes (Landschützer et al., 2014;Takahashi et al., 2009) and rapid sequestration of C ant into the deep ocean (Fröb et al., 2016;Pérez et al., 2013). We find the that the low C ant waters of the nutrient stream provide significant air-sea C ant uptake capacity to the surface of the subpolar North Atlantic.
Upward tilt of isopycnals in the Gulf Stream region allows for the low C ant waters to be laterally transported into the Gulf Stream along isopycnals and thus preserving their properties. C ant uptake capacity of −0.19 Pg C ant year −1 is transported northward at A22. The hindcast simulation used here indicates that the simulated fraction of A22 C ant uptake capacity transport that reaches the subpolar North Atlantic is sufficient to support the simulated air-sea C ant flux.
Observations of nutrient transport (Pelegri et al., 1996;Williams et al., 2006) suggest that half of the transport at A22 reaches the subpolar North Atlantic, while the hindcast simulation indicates 33-41% of the air-sea C ant uptake capacity reaches the subpolar North Atlantic. If half of the C ant,def transport at A22 reaches the surface in the subpolar North Atlantic, it would suggest a subpolar North Atlantic uptake capacity of −0.10 Pg C ant year −1 in 2003. This is comparable to the total subpolar C ant uptake, −0.09±0.01 Pg C ant year −1 (Mikaloff Fletcher et al., 2006).
The hindcast simulation indicates that differences between the Mikaloff Fletcher et al. (2006) estimate and that of Pérez et al. (2013) is that the smaller study region of the latter excludes regions of the nutrient stream outcrop. Ventilation of low C ant in nutrient stream waters fully sustains a significant subpolar North Atlantic air-sea C ant sink.
In the future, circulation and biogeochemistry in the subpolar North Atlantic are projected to undergo significant changes (Goris et al., 2018;Halloran et al., 2015;Tagklis et al., 2017;Whitt, 2019). C con uptake is projected to decline mid-century after an initial period of growth (Halloran et al., 2015). CMIP5 models project that O 2 inventory is likely to decline, but patchy regions of O 2 increase may also occur, driven by declines in AOU transport in the nutrient stream (Tagklis et al., 2017). Analyzing the historical response to anthropogenic forcing of the subpolar air-sea C ant sink and nutrient stream, as well as the future response (e.g., Whitt, 2019) can improve our understanding of the mechanisms driving subpolar C ant uptake. Earth-system models and observations, utilized in complementary fashion, provide a potent framework for the study of regional air-sea C ant uptake in the present and future.