Carbon isotope minima in the South Atlantic during the last deglaciation: evaluating the influence of air-sea gas exchange

Carbon isotope minima were a ubiquitous feature in the mid-depth (1.5–2.5 km) Atlantic during Heinrich Stadial 1 (HS1, 14.5–17.5 kyr BP) and the Younger Dryas (YD, 11.6–12.9 kyr BP), with the most likely driver being collapse of the Atlantic Meridional Overturning Circulation (AMOC). Negative carbon isotope anomalies also occurred throughout the surface ocean and atmosphere, but their timing relative to AMOC collapse and the underlying drivers have remained unclear. Here we evaluate the lead-lag relationship between AMOC variability and surface ocean δ13C signals using high resolution benthic and planktonic stable isotope records from two Brazil Margin cores (located at 1.8 km and 2.1 km water depth). In each case, the decrease in benthic δ13C during HS1 leads planktonic δ13C by 800 ± 200 years. Because the records are based on the same samples, the relative timing is constrained by the core stratigraphy. Our results imply that AMOC collapse initiates a chain of events that propagates through the oceanic carbon cycle in less than 1 kyr. Direct comparison of planktonic foraminiferal and atmospheric records implies a portion of the surface ocean δ13C signal can be explained by temperature-dependent equilibration with a 13C-depleted atmosphere, with the remainder due to biological productivity, input of carbon from the abyss, or reduced air-sea equilibration.


Introduction
The rise in atmospheric CO 2 during the last deglaciation was first documented over 30 years ago (Neftel et al 1982) yet the underlying mechanisms responsible for the signal remain unclear (Broecker 1982, Sigman et al 2010. The initial rise in CO 2 during Heinrich Stadial 1 (HS1) coincided with a decrease in the δ 13 C of CO 2 (Schmitt et al 2012), implying the carbon originated from a 13 C-depleted reservoir. The atmospheric δ 13 C record is remarkably similar to δ 13 C time series from the mid-depth Atlantic, implying both reflect the input of isotopically light carbon from a common source (figure 1). While the mid-depth signal is likely due to collapse of the AMOC , Oppo et al 2015, the atmospheric signal may be due to upwelling of light carbon in the Southern Ocean Lea 2002, Menviel et al 2018), weakening of the biological pump (Menviel et al 2015, Schmittner and, or some combination of these effects (Bauska et al 2016). In this paper, we review evidence for collapse of the AMOC during HS1, focusing on results from the Brazil Margin, a location with a welldeveloped depth transect of cores that can be used to monitor the relative timing of circulation and carbon cycle changes in the surface, mid-depth (1.5-2.5 km) and abyssal Atlantic. We focus on two Brazil Margin cores, including 46.3°W, 1800 m) and KNR159-5-33GGC (27.6°S, 46.2°W, 2100 m). Using benthic and planktonic δ 13 C records from these cores we evaluate the relative timing of AMOC collapse and changes in the oceanic carbon cycle. We also estimate the predicted changes in surface ocean δ 13 C at the Brazil Margin using available atmospheric δ 13 C records and a new planktonic foraminiferal Mg/Ca time series from core KNR159-5-78GGC. We then compare the predicted and observed δ 13 C records to assess the drivers of surface ocean δ 13 C variability during millennial-scale events of the last deglaciation.

Hydrographic context
The Brazil Margin depth transect is located on the western edge of the South Atlantic subtropical gyre (figure 2). At the latitude of the core sites (∼27°S), near surface waters are influenced by the Brazil Current (BC), which is composed of warm, saline Tropical Water in the upper 150 m (Tsuchiya et al 1994) and cooler, fresher South Atlantic Central Water (SACW) from ∼150 to 800 m (Tomczak and Godfrey 1994). The BC originates from the southern South Equatorial Current, which approaches South America from the east and bifurcates at ∼15°S (Peterson and Stramma 1991), with approximately 8 Sv flowing northward, eventually becoming the North Brazil Current, and 4 Sv flowing southward as the weaker BC (Stramma et al 1990). Recirculating flow within the gyre causes BC transport to increase to ∼10 Sv Figure 1. Mid-depth Brazil Margin δ 13 C time series compared to atmospheric δ 13 C of CO 2 from 0-25 kyr BP. The benthic foraminiferal records are simple three-point running mean values of C. wuellerstorfi δ 13 C for 1800 m (black line) and 2100 m water depth (green line) (Tessin and Lund 2013). Note that the results at 2100 m are shifted by +0.2%. Triangles denote calendar ages for each Brazil Margin record, assuming a regional reservoir age (ΔR) of 0±200 years (1σ). Both time series indicate δ 13 C decreased abruptly at 17.8 kyr BP. Two records of atmospheric δ 13 C are shown, including the spline smooth of Epica Dome C (EDC) data (gray area; Schmitt et al 2012) and the discrete estimates from the Taylor Glacier record (red circles; Bauska et al 2016). The gray zone for EDC represents the ±1σ uncertainty around the mean value (see Schmitt et al 2012 for details). The shift in atmospheric δ 13 C during HS1 occurs 300-400 years earlier when either the AlCC2012 age model (Veres et al 2012) or (Parrenin et al 2013) age model is applied to the EDC record. Symbols represent cores with radiocarbon-constrained time series summarized in Lund et al (2015). Red symbols mark the depths of the two cores that are the focus of this study (KNR159-5-78GGC at 1800 m and KNR159-5-33GGC at 2100 m). Also noted are the approximate depths of Antarctic Intermediate Water (AAIW), North Atlantic Deep Water (NADW), and Antarctic Bottom Water (AABW). by 27°S (Peterson and Stramma 1991). Surface-mixed layer records based on planktonic foraminifera such as G. sacculifer and G. ruber should therefore reflect hydrographic conditions in the BC and the broader South Atlantic subtropical gyre.
Near 40°S, the approximate latitude of the Subtropical Front, the BC encounters the northward flowing Malvinas Current before combining and flowing offshore as the South Atlantic Current, which bounds the southern edge of the gyre (Peterson and Stramma 1991). Wintertime convection in the Brazil-Malvinas confluence region creates a range of Subtropical Mode Waters that are found from ∼150 to 400 m water depth at the Brazil Margin (i.e. the shallow portion of SACW) (Gordon 1981, Provost et al 1999. At 27°S, the hydrographic properties of the upper thermocline are set by the lightest of these mode waters, which originate in the confluence region from ∼32°S to 40°S in the Southwest Atlantic (Provost et al 1999). Further south, between the Subtropical Front and Sub-Antarctic Front, the formation of Sub-Antarctic Mode Water contributes to the deeper portion of SACW (∼400 to 800 m water depth) (Stramma and England 1999). Thus, records based on thermocline dwelling foraminifera such as N. dutertrei should reflect the influence of not only the BC, but also the Malvinas Current, which originates in the Southern Ocean.
Deeper in the water column, the Brazil Margin sites are influenced by North Atlantic Deep Water (NADW) and Antarctic Bottom Water (AABW) (figure 2). NADW is clearly outlined by the maximum in salinity centered at ∼2500 m, which is apparent in vertical profiles from the Holocene as a maximum in δ 13 C and minimum in δ 18 O (Hoffman and Lund 2012) (figure S1 is available online at stacks.iop.org/ERL/ 14/055004/mmedia). The Brazil Margin transect also monitors low salinity AABW as it enters the abyssal South Atlantic (figure 2), which is reflected in the δ 13 C minimum and δ 18 O maximum in the deepest portion of the Holocene profiles (figure S1). Core-top δ 13 C results for the deepest sites average 0.4% (Hoffman and Lund 2012), indistinguishable from the value for AABW (Kroopnick 1985). Core-top δ 18 O results for the deepest sites (3.2%±0.2%) are also similar to estimates of AABW δ 18 O based on modern hydrographic data (∼3.1%) (Hoffman and Lund 2012).

Background
As described in detail in the supplemental materials, the mid-depth Brazil Margin sites were strongly influenced by Glacial North Atlantic Intermediate Water (GNAIW) during the LGM. At 1800 m, the watermass mixture was approximately 75% GNAIW and 25% GAABW (figure 3(a)), while at 2100 m, the proportions were ∼60% and 40%, respectively (Tessin and Lund 2013). During HS1, the δ 18 O and δ 13 C at each site decreased by approximately 0.5% ( figure 3(b)). In contrast, deeper sites showed little or no change . The overall HS1 pattern at the Brazil Margin is therefore one of large signals at mid-depth and little or no variability below 2500 m. The trajectory of the mid-depth Brazil Margin anomalies parallels the trend in northern component water (NCW), indicating the sites responded to the shift in NCW composition. Modeling simulations and carbonate ion reconstructions suggest the Brazil Margin δ 13 C anomalies were due to collapse of the AMOC and accumulation of respired carbon at middepth Lund 2015, Lacerra et al 2017), while the δ 18 O anomalies were likely due to input of isotopically light melt water and subsurface warming (Zhang et al 2017). Modeled collapse of the AMOC also yields negative δ 13 C anomalies in the surface ocean due to reduced biological export of light carbon and air-sea gas exchange with a 13 C-depleted atmosphere . The Brazil Margin and atmospheric δ 13 C time series in figure 1 may therefore reflect distinct biogeochemical consequences of AMOC collapse, with the former reflecting accumulation of respired carbon and the latter responding to exchange with a 13 C-depleted surface ocean.
If AMOC collapse triggered the surface ocean and atmosphere response, then proxies of the AMOC should lead proxies of surface ocean carbon cycle change during HS1. Model simulations suggest the time lag is approximately 1 kyr, due to propagation of the signal via the deep ocean circulation . While declining δ 13 C at the Brazil Margin appears to lead atmospheric δ 13 C (figure 1), uncertainty in the age models for both sediment and ice cores precludes a clear determination of the leadlag relationship (Tessin and Lund 2013). Here we address this problem by evaluating the relative timing of benthic and planktonic δ 13 C anomalies in the same sediment cores. We use mid-depth sites from the Brazil Margin, where benthic δ 13 C should reflect AMOC variability and planktonic δ 13 C should reflect equilibration with a 13 C-depleted atmosphere. Using atmospheric δ 13 C and planktonic foraminiferal Mg/Ca records, we assess whether temperature mediated airsea gas exchange on its own can explain the surface mixed layer and thermocline-depth δ 13 C records. We also address the recent suggestion by Lynch-Stieglitz et al (2019) that the mid-depth anomalies could be due to temperature-mediated air-sea gas exchange.

Methods
The δ 13 C of planktonic foraminifera reflects not only the δ 13 C of dissolved inorganic carbon (DIC) but also respiration, photosynthetic fractionation by algal symbionts, seawater carbonate ion concentration, and vertical migration (Curry and Crowley 1987, Spero and Lea 1993, 1996, Spero et al 1997, Spero et al 2003. Many of these vital effects influence the carbon isotopic composition in the surrounding microenvironment, which is the source of calcifying fluid for the shell (Zeebe et al 1999). One way to control for vital effects is to use multiple planktonic species, including those with and without symbionts and those that dwell at surface and thermocline depths. Given that many vital effects are size-dependent, it is also important to use shells from as narrow a size range as possible. To characterize the surface mixed-layer, we use G. sacculifer (300-355 μm size fraction), focusing on individuals without the sac-like terminal chamber Crowley 1987, Spero et al 2003). We also use the N. dutertrei data from Hertzberg et al (2016) to characterize the upper thermocline (Fairbanks et al 1982, Curry and Crowley 1987, Multiza et al 1999. Following Spero et al (2003), Hertzberg et al (2016) employed N. dutertrei individuals from the >350 μm size fraction that lacked evidence for secondary crusting. Note that the G. sacculifer record is from core 33GGC while the N. dutertrei record is from core 78GGC.
To minimize sampling uncertainty in the stable isotope time series, we crushed and homogenized 20-40 G. sacculifer tests from each sample and then ran four separate aliquots of powder to determine the mean δ 13 C for each stratigraphic level. This approach was necessary to extract the cleanest possible δ 13 C signal for comparison to the benthic δ 13 C time series in 33GGC. Given their larger size relative to G. sacculifer and therefore greater CO 2 yield, Hertzberg et al (2016) crushed and homogenized 4-8 N. dutertrei tests from each sample and then ran four separate aliquots of powder to determine the mean δ 13 C at each stratigraphic level. The analyses for both species were run on a Finnigan MAT 253 triple-collector gas source mass spectrometer coupled to a Finnigan Kiel automated carbonate device at the University of Michigan's Stable Isotope Laboratory. Isotope values were corrected to VPDB using National Bureau of Standards 19 (n=53, δ 13 C=1.92%±0.05%, δ 18 O=−2.18%± 0.05%). The benthic stable isotope records used to assess the relative timing of mid-depth and surface ocean δ 13 C anomalies were originally presented in Tessin and Lund (2013). We use the calendar-calibrated age models for 33GGC and 78GGC from Lund et al (2015), which are updated versions of those originally presented in Tessin and Lund (2013).
In order to assess the influence of sea surface temperature (SST) on δ 13 C, a new G. ruber Mg/Ca time series was generated for core 78GGC. We picked ∼40 shells from each sample (250-355 μm size fraction, average total weight ∼580 μg). The shells were gently crushed between glass plates, homogenized, and split into two aliquots for replicate analyses. Each sample underwent a cleaning procedure that included clay removal, metal oxide reduction, and organic matter oxidation Keigwin 1985, Rosenthal et al 1997). All samples were dissolved in 2% HNO 3 and run on a Thermo Element 2 ICP-MS at UCONN Avery Point. Samples were corrected for drift using a 2% HNO 3 blank and 100 ppm Ca standard (Mg/Ca=3.40 mmol mol −1 ). Analytical precision (1σ), determined using two external standards with Mg/Ca ratios of 3.20 and 3.60 mmol mol −1 , was 0.50% (n=12) and 0.38% (n=12), respectively. In addition, Al/Ca, Mn/Ca, and Fe/Ca ratios were measured to monitor for clays and metal oxides not removed during the cleaning procedure. Mg/Ca ratios were converted to SST using the species-specific G. ruber calibration equation of Anand et al (2003). This calibration yields an average late-Holocene SST of 23.2°C, similar to modern average annual SST at the core site of 23.6°C (Locarnini et al 2013).

Results and discussion
3.1. Planktonic foraminiferal stable isotope records Both G. sacculifer and N. dutertrei δ 18 O decreased by ∼1.5% during the deglaciation, with an initial decline during HS1, a pause during the B-A, and then a second decline during the YD (figure 4). There is a brief increase in δ 18 O at ∼5 kyr BP in the G. sacculifer record but it is unclear whether a similar feature exists in the N. dutertrei time series because of its lower resolution during the Holocene. In general, the two δ 18 O records maintain a 0.3%-0.5% offset from 0 to 20 kyr BP, with the exception of the B-A and early YD where the δ 18 O values converge (figure 4(c)). If the 0.3%-0.5% δ 18 O difference is entirely due to temperature, it would imply that G. sacculifer calcified in 1°C-2°C warmer water, consistent with its shallower habitat.
The δ 13 C time series also show broadly similar patterns during the deglaciation, including negative carbon isotope anomalies during HS1 and the YD (figure 4). The N. dutertrei HS1 anomaly, defined here as the δ 13 C difference between the 15-16 kyr BP and 18-20 kyr BP time intervals, is −0.66% ±0.07%. The uncertainty is based on the standard error of the mean values for time window. By comparison, the YD anomaly, defined as the difference between 11-12 kyr BP and 13-14 kyr BP, is −0.32%±0.17%. Thus, the δ 13 C signal is greater during HS1 than the YD. In the case of G. sacculifer, the HS1 anomaly is −0.49±0.08%, while the YD anomaly is −0.34%±0.10%. Here the HS1 signal also appears to be larger but the uncertainties preclude a clear statement on their relative magnitude. Overall the two records display remarkably similar millennial-scale variability ( figure 4(f)). Given that G. sacculifer dwells in the surface mixed layer, while N. dutertrei calcifies in the upper thermocline, our results suggest the δ 13 C of DIC declined during both HS1 and the YD. The G. sacculifer record also shows a negative excursion at ∼8 kyr BP (figure 4(e)) that is not resolved by the N. dutertrei time series ( figure 4(d)).
3.2. Relative timing of benthic, planktonic, and atmospheric signals The negative shifts in N. dutertrei and G. sacculifer δ 13 C during HS1 lag the respective benthic δ 13 C time series in each core (figure 5). The inset in each panel of figure 5 shows the correlation versus lag relationship for the planktonic and benthic δ 13 C records for the 20 to 15 kyr BP interval (i.e. the LGM to HS1). Positive values indicate planktonic δ 13 C lags benthic δ 13 C. In core 78GGC, the maximum correlation (r 2 =0.89) occurs at a lag of 800±100 years, which is nearly constant during the LGM to HS1 transition. In core 33GGC, the maximum correlation (r 2 =0.88) is similar but spans a wider range (800±300 years). The broader time span is due a smaller lag near the beginning of HS1 which steadily increases until ∼16 kyr BP. Overall, however, the planktonic records lag their benthic counterparts by approximately 800 years. While the absolute age of the δ 13 C anomalies may change with future age model updates, the relative timing of the benthic and planktonic records is set by the core stratigraphy. In each core, the decline in benthic δ 13 C occurs 10-20 cm prior the decline in planktonic δ 13 C, larger than can be explained by bioturbation. Furthermore, planktonic and benthic shells used in our analyses are from a similar size fraction (>250 μm), so preferential sorting based on shell size is unlikely.
The lag between benthic and planktonic δ 13 C is similar to the observed lag between benthic and atmospheric δ 13 C in figure 1. Here, however, atmospheric δ 13 C appears to lag benthic δ 13 C not only during HS1 but also later in the deglaciation (8-12 kyr BP). To quantify this difference, we linearly detrended each record and determined the correlation coefficient between the resulting time series at a range of lags spanning 0 to 3000 years ( figure 6). Positive values indicate atmospheric δ 13 C lags benthic δ 13 C. To assess the sensitivity of our results to the choice of atmospheric record, we used EDC (Schmitt et al 2012), EDC on the timescale of Veres et al (2012), and the Taylor Glacier δ 13 C results (Bauska et al 2016). For core 78GGC, we find that the maximum correlation with atmospheric δ 13 C occurs at lags of 900-1300 years. Similarly, the maximum correlation for core 33GGC occurs at lags of 800-1200 years. For both cores, the smallest lag (800-900 years) occurs using the Veres et al (2012) timescale for EDC. The largest lags occur using the Schmitt et al (2012) timescale for EDC (1200-1300 years), while the Taylor Glacier results have intermediate values (1000-1100 years). On average, atmospheric δ 13 C lags benthic δ 13 C by ∼1000 years, similar to the 800 year offset between benthic and planktonic δ 13 C (figure 5).
The generally greater lag for the atmospheric records is likely due to uncertainty in the ice core and sediment core age models. One possibility is that reservoir ages at the Brazil Margin were higher during the deglaciation (e.g. 600 years versus 400 years), which would bring the lag estimates in figure 6 into agreement with the benthic-planktonic offset. Alternatively, the Veres et al (2012) timescale for EDC, which is similar to that of Parrenin et al (2013), may be the most accurate age model for atmospheric δ 13 C. If this is the case, then the benthic-atmosphere lag (800-900 years) would overlap the benthic-planktonic offset. Regardless of the details, it is clear that atmospheric and surface ocean δ 13 C lagged the mid-depth signal, which is inconsistent with the idea that air-sea equilibration drove δ 13 C variability throughout the upper Atlantic (Lynch-Stieglitz et al 2019). Instead, our results are consistent with AMOC collapse causing the accumulation of respired carbon at mid-depths (Lacerra et al 2017), with a subsequent carbon cycle response in the surface ocean.

Inferred sea-surface temperatures
The SST time series for core 78GGC shows the anticipated glacial to interglacial pattern, with a total LGM to Holocene SST increase of ∼2°C ( figure 7(a)). The SST rise occurs late in the deglaciation, however, with nearly all of the signal occurring from 15 to 13 kyr BP (i.e. the Bølling-Allerød or B-A). After 12 kyr BP, SSTs remained persistently high, with values ranging from 23.5 to 24.5°C. SSTs peaked at ∼6 kyr BP, followed by a cooling of 0.5°C-1.0°C during the late Holocene. The SST pattern in 78GGC is similar to that reconstructed by Carlson et al (2008) using G. ruber Mg/Ca from core KNR159-5-36GGC ( figure 7(b)). Thus, two Brazil Margin Mg/Ca time series show that deglacial warming in the surface mixed layer occurred during the B-A, indicating that the G. sacculifer δ 18 O signal during HS1 was likely due to changes in δ 18 O sw . The Mg/Ca results also imply that the negative anomaly in G. sacculifer δ 13 C during HS1 was due to factors other than SST, with the caveat that the long equilibration time for δ 13 C means that it likely reflects the integrated temperature history of subtropical gyre surface waters at the Brazil Margin. The relevance of the G. ruber Mg/Ca results for N. dutertrei δ 13 C are less clear given their different depth habitats. However, the small δ 18 O offset between G. sacculifer and N. dutertrei δ 18 O during the deglaciation (<0.5%) implies they experienced a similar temperature history. During the first half of HS1, when N. dutertrei δ 13 C decreased by 0.6%, the corresponding decrease in δ 18 O was relatively modest (∼0.2%; figure 4(c)), equivalent to a maximum warming of 1°C.
3.4. The effect of air-sea gas exchange on planktonic δ 13 C Given the similar lag between: (1) benthic and planktonic δ 13 C, and (2) benthic δ 13 C and atmospheric δ 13 C, the relative timing between the planktonic and atmospheric δ 13 C records is likely correct to within several hundred years. The records can therefore be directly compared to assess the influence of air-sea gas exchange on planktonic  ) and benthic (C. wuellerstorfi) δ 13 C records from KNR159-5-78GGC (1800 m water depth). Inset depicts correlation versus lag relationship for the two records; positive lag values indicate planktonic δ 13 C lags benthic δ 13 C. (Bottom) Planktonic (G. sacculifer) and benthic (C. wuellerstorfi) δ 13 C records from KNR159-5-33GGC (2100 m water depth). Inset depicts correlation versus lag relationship for the two records. In each case, the decline in benthic δ 13 C precedes planktonic δ 13 C by ∼800 years. δ 13 C. Overall, the atmospheric and oceanic data display similar millennial-scale variability; in each case, δ 13 C declined during HS1, increased during the B-A, decreased again during the YD, and finally rebounded to LGM-like values during the early Holocene (figure 8). While the timing is similar across records, the magnitude of the planktonic anomalies during HS1 and the YD (0.4%-0.6%) is approximately twice the atmospheric signal (0.2%-0.3%). (Note the y-axes in each panel of figure 8 are scaled so that the relative magnitude of the δ 13 C signals is preserved.) Surface ocean δ 13 C DIC can be influenced by a range of factors, including air-sea gas exchange, biological export of carbon from the surface ocean, and mixing with watermasses of different δ 13 C composition. For the LGM, we can estimate how changes in atmospheric δ 13 C and equilibration temperatures might impact surface water δ 13 C DIC at the Brazil Margin using ice core-derived atmospheric δ 13 C data and SSTs inferred from planktonic foraminiferal Mg/Ca. This approach assumes that the SST signal at our site is broadly representative of locations in the subtropical gyre where surface water exchanges carbon with the atmosphere. At isotopic equilibrium, the fractionation between seawater DIC and gaseous CO 2 at 22°C is approximately 8.2% (Zhang et al 1995). Assuming an atmospheric δ 13 C of −6.4%, the expected δ 13 C of DIC (δ 13 C DIC ) would be 1.8%, which falls within ±0.1% of the mean G. sacculifer and N. dutertrei values for the 18-20 kyr BP interval ( figure 8(b)). This result is surprising considering the observed δ 13 C offsets between foraminifera and DIC; the δ 13 C of N. dutertrei is enriched relative to δ 13 C DIC by ∼0.5% (Mulitza et al 1999) while the δ 13 C of G. sacculifer (250-355 μm) has been shown to be depleted relative to δ 13 C DIC by ∼0.1% (Spero et al 2003). Thus, the inferred δ 13 C DIC for the upper thermocline would be 0.5% less than the N. dutertrei δ 13 C record, making it more depleted than the surface mixed layer δ 13 C DIC (based on G. sacculifer). Relatively depleted δ 13 C values for the upper thermocline compared to the surface mixed layer are consistent with the observed upper ocean gradient in δ 13 C DIC driven by photosynthesis and respiration of organic matter (Kroopnick 1985, Mulitza et al 1999. To determine the drivers of millennial-scale δ 13 C variability at the Brazil Margin, we separately consider the influence of SST and atmospheric δ 13 C. In the first case, we assume atmospheric δ 13 C was constant during last deglaciation and equal to the LGM value (−6.4%) but take into account temperature-dependent fractionation between DIC in seawater and CO 2 in air (Zhang et al 1995). The resulting δ 13 C DIC time series (dashed black line in figure 8(b)) shows no signal until the B-A, when δ 13 C declines by approximately 0.2% due to the 2°C increase in SST ( figure 7). While the temperature-only δ 13 C DIC estimate shows lower values during the YD, similar to the planktonic δ 13 C data, the lack of signal during HS1 indicates additional factors were responsible for the HS1 signal.
We consider the combined influence of atmospheric δ 13 C and SST variability using the EDC record and assuming temperature-dependent 13 C equilibration. The resulting δ 13 C DIC time series, depicted as the solid red line in figure 8(b), shows a ∼0.3% decline during HS1, followed by an additional ∼0.2% drop during the YD. Here the predicted HS1 signal is due to exchange with an isotopically lighter atmosphere while the YD signal is due to both the atmospheric influence and warmer SSTs. A similar pattern occurs if we use the Taylor Glacier δ 13 C record rather than EDC (dashed red line in figure 8(b)). Both of the predicted time series are similar to the G. sacculifer record; δ 13 C declines during HS1, levels out during the B-A, and then declines again during the YD, reaching values lower than at any point during the deglaciation. The overall agreement suggests that temperature-dependent gas exchange with a 13 C-depleted atmosphere can account for the broad features of the G. sacculifer record. However, G. sacculifer shows a larger than predicted signal during HS1, which is most easily observed in the difference between   figure 4 for detailed records). The thin dashed black line depicts the anticipated surface ocean δ 13 C assuming atmospheric δ 13 C=−6.4% and the temperature-dependent 13 C fractionation between DIC in seawater and CO 2 in air (Zhang et al 1995), using the Mg/Ca-based SST record in figure 7. The red line with gray error envelope shows the anticipated surface ocean δ 13 C based on the combined effect of variable atmospheric δ 13 C (EDC record in top panel) and SST. The dashed red line is the same but for the Taylor Glacier δ 13 C record. Note that no corrections have been applied to the planktonic δ 13 C records. (C) The difference between atmospheric and planktonic δ 13 C (Δδ 13 C), including G. sacculifer and EDC (blue line), G. sacculifer and Taylor Glacier (blue circles), N. dutertrei and EDC (green line), and N. dutertrei and Taylor Glacier (green circles).
atmospheric δ 13 C and G. sacculifer δ 13 C (Δδ 13 C) (blue lines in figure 8(c)). The G. sacculifer Δδ 13 C values based on both EDC and Taylor Glacier decreased by 0.2%-0.3% during HS1. The G. sacculifer δ 13 C anomaly is therefore 0.2%-0.3% greater than can be accounted for by equilibrium exchange with the atmosphere. During the YD, Δδ 13 C also decreased 0.2%-0.3%, but about half of this signal can be attributed to warmer SSTs. The largest mismatch between the predicted and observed G. sacculifer δ 13 C signals occurs during HS1.
In the case of N. dutertrei, the HS1 δ 13 C anomaly is ∼0.4% larger than can be accounted for by simple ocean-atmosphere equilibration ( figure 8(c)). While it possible that temperatures warmed in the upper thermocline, the δ 18 O data for N. dutertrei show only a modest decrease during HS1 (0.2%), equivalent to a maximum warming of ∼1°C. During the YD, the N. dutertrei δ 13 C signal (0.3%) is closer the predicted anomaly due to SST warming (0.2%), suggesting that the combined effect of variable atmospheric δ 13 C and SST can account for the majority of the N. dutertrei response at this time. Therefore, most of the YD signal for G. sacculifer and N. dutertrei can be explained by air-sea gas exchange but both species display a larger than expected δ 13 C signal during HS1, particularly N. dutertrei. Given that atmospheric CO 2 increased during HS1, this would likely yield lower surface ocean [ -CO 3 2 ] values. As a result, we would expect the carbonate ion effect on planktonic δ 13 C, which has a slope of ∼−0.01% per μmol kg −1 (Spero et al 1997), would yield positive δ 13 C anomalies. Thus, other factors need to be invoked to explain the HS1 δ 13 C signals.
3.5. Additional drivers of surface ocean δ 13 C What caused the planktonic δ 13 C anomalies during HS1? One possibility is that the surface ocean was less equilibrated with the atmosphere, perhaps due to weaker winds or a reduced residence time for surface waters in the South Atlantic. This seems unlikely, however, as G. sacculifer reflects conditions in the Brazil Current and broader subtropical gyre while N. dutertrei should track ventilated thermocline conditions further south. Alternatively, isotopically light carbon upwelled in the Southern Ocean during HS1 Charles 1997, Spero andLea 2002) could have been transmitted northward through mixing and ventilation in the Brazil-Malvinas Current confluence region and then carried to the Brazil Margin via upper thermocline waters. This scenario could explain the larger than expected N. dutertrei δ 13 C signal, but it is unclear how it would yield an anomalous response in G. sacculifer. Recently published model results suggest that increasing Southern Hemisphere westerlies by 50% relative to the LGM and decreasing freshwater input into the Southern Ocean by 0.2 Sv can yield substantial surface ocean and atmospheric δ 13 C anomalies at the Brazil Margin it unlikely that regional freshwater fluxes decreased early in the deglaciation. Further data and model inter-comparison studies are necessary to assess the role of freshwater forcing in the Southern Ocean and its role in driving release of isotopically light carbon from the abyss.
One other possible explanation of the larger than expected δ 13 C anomalies at Brazil Margin is weakening of the biological pump. Paired high resolution planktonic and benthic δ 13 C records from multiple intermediate depth sites indicate the upper ocean δ 13 C gradient decreased during HS1 (Hertzberg et al 2016). These data are consistent with the accumulation of isotopically light carbon in the surface ocean and less remineralization at intermediate depths. In this scenario, it is possible that N. dutertrei responded to reduced productivity in the Southern Ocean, with the associated surface ocean δ 13 C anomaly transmitted northward to the Brazil-Malvinas confluence region. Consistent with this idea, alkenone flux results from the Sub-Antarctic suggest export productivity declined early in the deglaciation, most likely due to reduced iron fertilization (Martínez-García et al 2014).
Weakening of the biological pump is unlikely to a have a direct impact on G. sacculifer δ 13 C given that nutrients are completely utilized in subtropical gyre surface waters. The δ 13 C DIC of these waters, and therefore the δ 13 C of G. sacculifer, is more likely to reflect the influence of atmospheric equilibration . This makes the larger than expected δ 13 C anomaly for G. sacculifer during HS1 more difficult to explain. One possibility is that the signal was transmitted by the Brazil Current from the equatorial Atlantic, where either reduced biological productivity or upwelling of 13 C-depleted mode waters (as implied by the N. dutertrei results) created negative carbon isotope anomalies. Consistent with this idea, modeling results suggest that weakening of the AMOC causes an overall reduction in export production in the South Atlantic and basin-wide depletion in surface ocean 13 C (Menviel et al 2015). Alternatively, warming in the equatorial Atlantic during HS1 (Arbuszewki et al 2013) may have contributed to lower surface ocean δ 13 C DIC in this region, which was then transmitted to the core sites by the Brazil Current. Testing of these hypotheses will require additional high-resolution planktonic δ 13 C, Mg/Ca, and productivity records from the tropical, subtropical, and sub-polar South Atlantic. For the time being, however, it appears that air-sea exchange on its own is unable to explain the full magnitude of planktonic δ 13 C anomalies at the Brazil Margin.

Conclusions
The goal of this paper has been to: (1) constrain the timescale on which AMOC perturbations propagate through the oceanic carbon cycle, and (2) evaluate the role of air-sea exchange in driving surface ocean δ 13 C variability during the last deglaciation. In the first case, we compare benthic and planktonic foraminiferal δ 13 C records from sediment cores that monitor NCW in the Southwest Atlantic. Our findings, which are based on two different cores and planktonic species, indicate that negative carbon isotope anomalies at mid-depth led those in the surface ocean by 800±200 years. This lag reflects the propagation time of AMOC collapse through the oceanic carbon cycle and its eventual expression in the surface ocean and atmosphere. This result is important because it implies that weakening of the AMOC plays a central role in surface ocean carbon isotope minima during glacial terminations. The timescale of signal propagation will be useful for testing simulations of the AMOC influence on the ocean carbon cycle. Our results also suggest that air-sea gas exchange is an unlikely explanation both mid-depth and surface ocean δ 13 C anomalies in the Atlantic (Lynch-Stieglitz et al 2018). Such a scenario would likely yield synchronous carbon isotope responses or benthic records that lag planktonic δ 13 C given the advection time of deep waters in the Atlantic.
Direct comparison of the planktonic and atmospheric δ 13 C records indicates that surface ocean δ 13 C anomalies during HS1 and the YD were approximately twice as large as those in the atmosphere. Assuming isotopic equilibration between atmospheric CO 2 and surface ocean DIC, we estimate the anticipated δ 13 C DIC for the Brazil Margin using available atmospheric δ 13 C records and local Mg/Ca-based SST estimates. The resulting curve is broadly similar to our G. sacculifer δ 13 C results, which suggests that temperature-mediated air-sea gas exchange had an important influence on surface ocean δ 13 C DIC in the South Atlantic subtropical gyre. Both G. sacculifer and N. dutertrei display a larger than predicted δ 13 C signal during HS1, however. In the case of N. dutertrei, this may reflect transport of isotopically light carbon from the sub-Antarctic where either weakening of the biological pump or upwelling of 13 C-depleted water yielded negative carbon isotope anomalies. Alternatively, less air-sea equilibration in the source region for upper thermocline waters could feasibly yield larger than expected N. dutertrei anomalies during HS1.