More efficient North Atlantic carbon pump during the Last Glacial Maximum

During the Last Glacial Maximum (LGM; ~20,000 years ago), the global ocean sequestered a large amount of carbon lost from the atmosphere and terrestrial biosphere. Suppressed CO2 outgassing from the Southern Ocean is the prevailing explanation for this carbon sequestration. By contrast, the North Atlantic Ocean—a major conduit for atmospheric CO2 transport to the ocean interior via the overturning circulation—has received much less attention. Here we demonstrate that North Atlantic carbon pump efficiency during the LGM was almost doubled relative to the Holocene. This is based on a novel proxy approach to estimate air–sea CO2 exchange signals using combined carbonate ion and nutrient reconstructions for multiple sediment cores from the North Atlantic. Our data indicate that in tandem with Southern Ocean processes, enhanced North Atlantic CO2 absorption contributed to lowering ice-age atmospheric CO2.

and multiple cores from ~intermediate depth in the North Atlantic, and supplement these with nutrient reconstructions of the same North Atlantic cores. I think this approach is very novel, although several major generalizations are made with regards to the interpretations of the various proxies involved, something I am sure the authors are aware off.
One generalization involves the mode of deep water formation in the North Atlantic, which was supposed to be considerable different and in a different location during the LGM compared with today. Potential effects of sea-ice also are ignored.
Furthermore, can ODP 999 from the Caribbean really be considered representative for the Gulf Stream over the entire interval? For example during early deglaciation, Leduc et al. (2007) suggest that orogenic blocking of the Andes during southward ITCZ movement caused this fresh water to be returned to the Atlantic Ocean, via the Amazon basin drainage, which led to lowered salinity of low-latitude currents in the Atlantic Ocean, such as the North Brazil current, the Guyana current and the Caribbean current feeding the Gulf Stream. Barker and Elderfield (2000) propose carbonate ion in the North Atlantic surface waters was up from Holocene values of about 200 umol/kg to glacial values of 260 umol/kg. This is a lot smaller compared with the increase at the Caribbean site.
Cd/Ca relationships discussed in the paper are based on very old papers of Ed Boyle. Marchitto and Broecker (2006, G-cubed) show that there are issues with using Cd/Ca in the intermediate North Atlantic for reconstructing phosphate. Real time measurements put phosphate concentrations at these locations below 1.2 umol/kg. Reconstructed values by the authors are all equal to or in excess of 1.3 umol/kg, which are found much further South today. Marchitto and Broecker (2006) suggest that Cd should be close to 0.2 nmol/kg. At the core locations in this study these seem to be values achieved during the glacial, not during the Holocene.
Actual phosphate values at the various locations should be shown in Figure 3, and the authors should discuss in the manuscript why they think their values can be considered robust given this discrepancy. Was Cd/Ca measured at ODP 999, as there is a value of 0 in Figure 5... In addition, it would be good to discuss about why Cd/Ca in intermediate bottom waters were even lower during the LGM compared with today where they are already one of the most nutrient depleted intermediate water masses.
Today the Caribbean may be considered as a slight source of atmospheric CO2. Is there any way the data can tell you whether it became less or more of a source of CO2 during the glacial, and what would the implications be for the carbonate ion gradient interpretation?
Finally, surface water d18O of the Caribbean became heavier during the LGM (+0.4 per mil after ice volume correction); but in the North Atlantic they decrease (Thornalley et al., 2010). If this is in relation to sea-ice/brine rejection, can nutrient versus d18Osw relationships in areas with seaice really be regarded as strong as ice-free open ocean? I find this paper fascinating. After reading through the main text and the supplement, I don't see critical flaws in this study. Therefore, I recommend its publication in Nature Communications after minor revisions. As I see it, this study should be of broad interest to the paleoclimate communities as well as the general readers.
It has long been a grand challenge for quantifying how much additional carbon was captured by the ocean during glacial periods. Using Last Glacial Maximum (LGM) as a classic example, the authors developed a novel and applicable way to tackle this problem. They utilized multiple proxies (e.g. boron isotopes, Mg/Ca, B/Ca, Cr/Ca ratios) to generate a powerful tracer, called [CO32-]as, which could be used to distinguish the air-sea exchange effect from the biological and other physical effects on changing the seawater carbonate system. By comparing the surface ocean carbonate chemistry in the Caribbean with deep ocean carbonate chemistry in the North Atlantic both in Holocene and LGM, the change of carbon sink of North Atlantic Ocean in the LGM relative to the Holocene is determined and the previously underappreciated contribution of North Atlantic Ocean is indicated. As the title suggested, the conclusion is that we have more efficient North Atlantic carbon pump during the LGM.
This paper also reads well. The experimental and sample details are given (mostly in the SI), the comprehensive numerical method descriptions are listed, and the conclusion looks robust to me. There are many definitions of new terms in this paper, such as [DIC]as, [CO32-]as, [CO3]EXP, [CO3]EXP−in-situ. These parameters are critical for understanding the essence of this research. I find the authors give a clear explanation for these conceptions and make them accessible to the readers. Their calculation procedures are provided in the method part and the supplementary materials, which I think is really crucial for the readers to understand how this study rigorously evaluates the North Atlantic carbon sequestration during the Last Glacial Maximum.
It is always a hassle when dealing with so many proxies because the error is difficult to quantify. To handle this problem, the authors adopt all three approaches -quadratic addition of individual errors, Monte Carlo resampling and creating several scenarios. In this way, errors are fully propagated and various situations are explored.
My major concern is what is the relative contribution of North Atlantic Ocean in sinking CO2 compared with the Southern Ocean. We know from this study that North Atlantic CO2 pump efficiency during the LGM was enhanced by a factor of ~2.7 relative to the Holocene. I am wondering if there is any way to quantitatively (or even qualitatively) constrain the absolute carbon sink fluxes from the NA. This is kind of important to know better the role NA played in LGM. If the carbon sink flux turns out to be small, then even if the efficiency of NA was greatly increased during LGM, it might still play a minor role. Some discussion of this aspect should be given in the text.
Below are several minor comments: Line 50 and Line 52: change n-dash to m-dash Line 74-76: I think strong cooling will decrease the Henry constant, and thus there will be more CO2 in the ocean relative to the atmosphere. It might be not proper to say "strong cooling causes surface-water pCO2 to be lower than atmospheric pCO2". The high nutrient utilization does causes surface-water pCO2 to be lower than atmospheric pCO2. Need to rephrase.
Line 94-96: Does this sentence indicate efficient biological pump is caused by "strong cooling of low-PO4 northward-flowing Gulf Stream waters and limited nutrient supply from the subsurface"? Need to rewrite the sentence.
Line 154: dissociated constant should be "dissociation constant". Similar to Line 110.
Line 245-246: change "for some of which" to "for which some"?

Dear Reviewers,
We have revised our manuscript substantially based on the valuable review comments received. The constructive comments have improved our manuscript significantly. In addition to changes to address the points raised, we emphasize the pragmatic recipe in the main text since it is easier to follow and apply. The second method, which involves frequent use of the CO 2 sys software, has now been moved to the SI. Some redundant supplementary figures (mainly related to the CO 2 sys sensitivity test) have also been removed from the SI. We hope you find the revision more streamlined.
Below, we list our point-by-point responses (blue) to your comments (black). The main text has been revised accordingly (blue). Arguably, the most obvious explanation would be a larger pool of regenerated organic matter, but this appears inconsistent with the Cd/Ca proxy for nutrients. Also after taking into account glacial-interglacial changes in temperature and salinity, the inferred larger LGM vertical [CO_3^{2-}] remains largely unexplained. This is a good summary of our study.
The authors attempt to explain the large vertical [CO_3^{2-}] gradient through a more efficient carbon pump during the LGM. This seems illogical to me, because the authors have already taken the soft-tissue and solubility carbon pumps into account in their calculations. If anything, the calculations by the authors tend to overestimate the impacts of the carbon pumps because of incomplete air-sea equilibration of carbon  or incomplete nutrient utilization (Ito & Follows, 2005). In other words, the main carbon pumps cannot be stronger than what the authors already accounted for, although they can be weaker.
There may be a misunderstanding here about our correction method to account for influences from the solubility and biological pumps.
Our approach is to isolate the air-sea exchange component CO 2 in the ocean. Corrections are necessary for each of the Holocene and LGM. To be simple, let us consider the Holocene. If our method tends to "overestimate" carbon pump impacts as described by the reviewer, it would be impossible to explain the vertical [CO 3 2-] gradient even for the Holocene where our reconstructions are consistent with estimates using the GLODAP dataset (Figs. 2, 5).
We think that the misunderstanding surrounds what is corrected by our method. Below, we explain details about our correction method.
Regarding the solubility pump, it affects [CO 3 2-] in two ways: any T/S change would change CO 2 system dissociation constants and thus [CO 3 2-], and (ii) when at the surface, T/S variations would change air-sea pCO 2 gradients, alter air-sea exchange component CO 2 in seawater, and thereby affect [CO 3 2-].
Regarding the biological pump, its effects on [CO 3 2-] include two parts: (i) changes in nutrient utilization and respiration redistribute DIC and ALK within the ocean and hence [CO 3 2-] (see changes from Fig. 1a to Fig. 1b), and (ii) biological processes would change air-sea pCO 2 gradients, alter air-sea CO 2 exchanges (see changes from Fig. 1b to Fig. 1c), and thereby affect [CO 3 2-].
For both pumps, our corrections account for influences only from (i), assuming zero net air-sea CO 2 change from (ii).
Therefore, our calculations represent the minimum impact (i.e., not to overestimate any effect) of the solubility and biological pumps on [CO 3 2-], because it assumes no air-sea CO 2 exchange. This is critical to interpret the [CO 3 2-] gradients presented in our manuscript.
After the above corrections, any [CO 3 2-] gradient would reflect air-sea CO 2 exchange. The more airsea CO 2 uptake, the larger the [CO 3 2-] gradient. In other words, after our corrections, more CO 2 invasion would cause greater vertical [CO 3 2-] gradients.
Given the Reviewer's comment above, we have further clarified our correction method in the main text wherever appropriate (lines 108-109, 169-178, 188, 720). In particular, we make use of welldefined sensitivities ( Fig. 4; Methods) and a plot of [CO 3 2-] Norm versus PO 4 ( Fig. 5) to demonstrate the corrections associated with the solubility and biological pumps, because they are more straightforward to understand.
We hope that our clarification resolves the reviewer's comment.
How to resolve this conundrum? The authors make use of many different proxies involving many different assumptions. Could one of those proxies perhaps be less accurate than the authors seem to think? There has been much debate about the range of applicability of d11B and B/Ca as pH proxies ( The reliability of our [CO 3 2-] reconstructions is supported by the following evidence: Regarding the benthic Cd/Ca proxy, there is some debate. However, all published studies unanimously suggest lower PO 4 for the Glacial North Intermediate Water (GNAIW), which is consistent with its much elevated benthic  13 C (~1.5‰) [4][5][6][7][8][9][10] . To have no increase in the North Atlantic CO 2 uptake efficiency would require a higher PO 4 for GNAIW ( Fig. R1; next page), which contradicts proxy data.
Instead of the proxies, we suspect that the real issue with the comments from Reviewer 1 surrounds a misunderstanding of our correction method. With the clarifications provided here, we hope that the reviewer can now agree with our conclusions.
Additionally, temperature, salinity, and pressure are estimated using established approaches. Their effects on seawater [CO 3 2-] are based on well-defined sensitivities (Fig. 4). All uncertainties are fully propagated as described in the SI (see appraisal of our error estimates from Reviewer 3).
Finally, we thank Reviewer 1 for the valuable comments, which we have used to significantly improve the description of our correction method. This is a very interesting manuscript, a first to try and tackle changes in the Atlantic carbon pump efficiency over glacial-interglacial timescales, with implications for atmospheric CO2. The authors propose that the carbon pump of the North Atlantic trippled during the LGM compared with the Holocene. To come to this conclusion they use estimates of estimate air-sea CO2 exchange signals using carbonate ion reconstructions for surface waters in a core from the Caribbean (not Gulf stream), and multiple cores from ~intermediate depth in the North Atlantic, and supplement these with nutrient reconstructions of the same North Atlantic cores.
I think this approach is very novel, although several major generalizations are made with regards to the interpretations of the various proxies involved, something I am sure the authors are aware off. One generalization involves the mode of deep water formation in the North Atlantic, which was supposed to be considerable different and in a different location during the LGM compared with today. Potential effects of sea-ice also are ignored.
We thank the reviewer for commenting on the novelty of our work. Below we address specific points. Thank you for this insightful comment on the use of ODP 999 for Gulf Stream reconstructions.
We use ODP 999 to constrain Gulf Stream chemistry because (i) this core has published  11 B ready for use 1 , and (ii) most North Atlantic subtropical gyre water circulates through the Caribbean Sea before being transported to the subpolar North Atlantic via the Gulf Stream 11 .
As a first attempt with our approach, we think it is reasonable to use ODP 999 to constrain the firstorder carbonate chemistry of warm surface source water changes. We hope the reviewer agrees.
We agree with the reviewer's point about possible complications during the last deglacial. Therefore, we have removed the deglacial calculations. Fig. 3 has been revised accordingly.
In the revision, we now focus on two time intervals: the Holocene and LGM.
Following the reviewer's advice, we have further investigated whether ODP 999 can be used to represent the first-order carbonate chemistry changes for the Holocene and LGM. To do so, we have tried to reconstruct surface water [CO 3 2-] for four additional cores from the wider tropical western Atlantic. For these cores, we assume quasi-equilibrium in pCO 2 between surface waters and the atmosphere. We believe that this is a reasonable assumption because subtropical surface waters cycle multiple times through the North Atlantic gyre (refs 1,2 ). As can be seen from Fig. S6, ODP 999 and the other four sites show very similar [CO 3 2-] Norm during the Holocene and LGM. We therefore believe that it is reasonable to use ODP 999 to represent the first-order Gulf Stream carbonate chemistry for the two time intervals of interest here.
We have explained our reasoning above in the main text (lines 131-140) and SI (Section 3.3). Actual phosphate values at the various locations should be shown in Figure 3, and the authors should discuss in the manuscript why they think their values can be considered robust given this discrepancy.

Fig. S6. Similar [CO 3 2-] Norm at ODP 999 and other sites from the broader subtropical western North
Thank you. We have updated Fig. 3 to show PO 4 values.
To justify the robustness of our down-core nutrient estimates, we have compared our reconstructions with those based on H. elegans. Compared to Cibicidoides (used in this study), D Cd into H. elegans is far less variable 15 . As can be seen from Fig. S8 (next page), for cores bathed in water masses with similar benthic  13 C, our Cd reconstructions match favourably with those based on H. elegans 10 . This supports the reliability of our reconstructions.
Using updated PO 4 reconstructions, we find glacial North Atlantic carbon uptake efficiency was doubled compared to the Holocene. Note: our conclusion of a more efficient LGM carbon pump remains unchanged, as long as LGM PO 4 was lower than in the Holocene even without knowing the exact value of LGM PO 4 . See Fig. R1 above for details. Was Cd/Ca measured at ODP 999, as there is a value of 0 in Figure 5... This is assumed due to its oligotrophic setting (lines 147-148). Any PO 4 increase at ODP 999 during the LGM would strengthen our conclusion (see sensitivity test in Fig. S17b). Despite the well accepted view of decreased GNAIW nutrient levels, the mechanisms causing this change are not fully clear. Possible reasons at least include: (i) lower preformed nutrient (and hence a stronger biological pump), (ii) reduced respiration related to faster ventilation, and (iii) reduced nutrient supply from glacial AAIW. Further work is needed to distinguish their relative roles in decreasing PO 4 of GNAIW.
Based on the reviewer's advice, we have now briefly discussed reasons for the lower Cd in GNAIW (lines 266-268).
Today the Caribbean may be considered as a slight source of atmospheric CO2. Is there any way the data can tell you whether it became less or more of a source of CO2 during the glacial, and what would the implications be for the carbonate ion gradient interpretation?
The reviewer is correct in that Caribbean today (preindustrial) is indeed a weak source of CO 2 to the atmosphere. This may also have been true for the Holocene and LGM when surface water pCO 2 was higher than the atmosphere by ~20 ppm (Fig. R2; next page). Everything else being equal, comparable pCO 2 gradients would suggest similar CO 2 outgassing from ~10°N (Caribbean) to ~35°N (latitudes north of which we define as the North Atlantic) for the Holocene and LGM. Thus, our conclusion remains unaffected.
Given large pCO 2 reconstruction uncertainties (~±27 ppm) and other factors (e.g., wind, air-sea contact time) that affect air-sea CO 2 exchange fluxes, we think it is more meaningful to consider temporal magnitudes of changes in other carbonate chemistry variables (e.g., DIC as , [CO 3 2-] Norm ) as we have done here.
Preindustrial Caribbean hydrographic sites show similar DIC as values as sites located close to the Gulf Stream (Fig. S3). For the Holocene and LGM, ODP 999 also shows similar [CO 3 2-] Norm values to the other four sites from the western tropical Atlantic sites (Fig. S6). Because we have assumed air-sea CO 2 equilibrium and hence minimal net air-sea CO 2 flux at these four sites, comparable [CO 3 2-] Norm values suggest that air-sea exchange CO 2 signals do not change much in the tropical Atlantic region.
Therefore, any CO 2 outgassing in Caribbean would have an insignificant impact on our conclusions, because waters at low latitudes (<~35°N) are more or less at equilibrium with atmospheric CO 2 16,17 (lines 115-116).  It is always a hassle when dealing with so many proxies because the error is difficult to quantify. To handle this problem, the authors adopt all three approaches -quadratic addition of individual errors, Monte Carlo resampling and creating several scenarios. In this way, errors are fully propagated and various situations are explored.
We thank the reviewer very much for the appraisal of our work. The above is a fantastic summary of our work.
My major concern is what is the relative contribution of North Atlantic Ocean in sinking CO2 compared with the Southern Ocean. We know from this study that North Atlantic CO2 pump efficiency during the LGM was enhanced by a factor of ~2.7 relative to the Holocene. I am wondering if there is any way to quantitatively (or even qualitatively) constrain the absolute carbon sink fluxes from the NA. This is kind of important to know better the role NA played in LGM. If the carbon sink flux turns out to be small, then even if the efficiency of NA was greatly increased during LGM, it might still play a minor role. Some discussion of this aspect should be given in the text.
We thank the reviewer for this great suggestion. We have added a new section in the main text to quantify CO 2 uptake change between the Holocene and LGM. Calculation details are given in Methods. We also acknowledge uncertainties associated with our calculation in the main text (lines 230-254).
Our best estimate is an extra ~100 PgC carbon uptake by the LGM North Atlantic compared to the Holocene ( Fig. 6; next page). We have discussed the role of the North Atlantic in relation to the Southern Ocean in the context of the global carbon budget change (lines 274-285, 290-292).
Below are several minor comments: Line 50 and Line 52: change n-dash to m-dash This has been changed.
Line 74-76: I think strong cooling will decrease the Henry constant, and thus there will be more CO2 in the ocean relative to the atmosphere. It might be not proper to say "strong cooling causes surface-water pCO2 to be lower than atmospheric pCO2". The high nutrient utilization does causes surface-water pCO2 to be lower than atmospheric pCO2. Need to rephrase.
We have removed "along strong cooling" to avoid confusion.
Line 94-96: Does this sentence indicate efficient biological pump is caused by "strong cooling of low-PO4 northward-flowing Gulf Stream waters and limited nutrient supply from the subsurface"? Need to rewrite the sentence.
Good point. It has been reworded to read (lines 94-96): North Atlantic CO 2 absorption is driven by (i) an efficient solubility pump due to strong cooling of northward-flowing Gulf Stream waters and (ii) a strong biological pump associated with high nutrient utilization [25][26][27] .
Line 154: dissociated constant should be "dissociation constant". Similar to Line 110. LGM-HOL extra C uptake, PgC I think the authors did a great job in addressing the comments raised by the reviewers. The authors emphasized the pragmatic recipe in the main text, which helped clarify the misunderstanding of the reviewers. They also added a section to quantify CO2 uptake change between the Holocene and LGM following my suggestion, which is very critical to highlight the importance of North Atlantic in carbon pump during last glacial maximum. The calculation detail was also given in the Methods and it looks robust to me.
As I was generally satisfied with the initial draft and the authors made great progress following the suggestion of all three reviewers, I still recommend its publication in Nature communications and hope it will stimulate more discussions about the carbon pump in North Atlantic during glacialinterglacial cycles.