Pronounced centennial-scale Atlantic Ocean climate variability correlated with Western Hemisphere hydroclimate

Surface-ocean circulation in the northern Atlantic Ocean influences Northern Hemisphere climate. Century-scale circulation variability in the Atlantic Ocean, however, is poorly constrained due to insufficiently-resolved paleoceanographic records. Here we present a replicated reconstruction of sea-surface temperature and salinity from a site sensitive to North Atlantic circulation in the Gulf of Mexico which reveals pronounced centennial-scale variability over the late Holocene. We find significant correlations on these timescales between salinity changes in the Atlantic, a diagnostic parameter of circulation, and widespread precipitation anomalies using three approaches: multiproxy synthesis, observational datasets, and a transient simulation. Our results demonstrate links between centennial changes in northern Atlantic surface-circulation and hydroclimate changes in the adjacent continents over the late Holocene. Notably, our findings reveal that weakened surface-circulation in the Atlantic Ocean was concomitant with well-documented rainfall anomalies in the Western Hemisphere during the Little Ice Age.

if the authors have cherry picked which records they are showing to simply support their hypothesis. For example, given the requisite for the authors to show a North Atlantic wide coherent salinity response, there are additional subpolar North Atlantic salinity records resolving the LIA which could be used to complement the subtropical data (which seem sparse too) and that could be added to figure Thornalley et al., 2009 (Nature). The authors are to be commended for their replication of the Mg/Ca-d18O data in 3 multicore subcores, which show excellent agreement. A more challenging test would of course be to replicate at a slightly different site since all subcore would be susceptible to any errors introduced from any downslope transport, local alteration of ruber habitat depth or seasonality or compounding effects such as carbonate ion changes. To a large extent the similar trends seen in the regional compilation does this. I just urge the authors not to overplay the significance of replicating data (and stacking) from the same site, since it does not eliminate all errors, such as those mentioned above as well as systematic errors in the calibration. Following on from this, the errors presented for this study are unrealistically small since they rely too heavily on the idea that replication of Mg/Ca or d18Osw between replicates at the same site can counter the errors in temperature calibration; random errors contributing to this uncertainty will be reduced but not systematic proxy bias (seasonality, depth, other environmental factors controlling Mg/Ca common to the 3 subcores (eg change in growth rate, local carbonate ion). Lund and Curry 2006 had problems interpreting their d18Osw values and the range obtained, since the local d18Osw-S relation yielded unrealistically large changes, suggesting complications with its use as a SSS proxy. Further discussion of this would be useful. I find the final link to an AMOC cause for the changes a bit of a stretch. As is discussed, the model results can cause similar changes as observed not through an AMOC change but rather simply a change in the strength of the surface gyres. In this model the LIA is not linked to a weaker AMOC but is more strongly caused by a weakening subpolar gyre and its links to sea-ice and atmospheric circulation. Given the authors do not present a strong case for an AMOC link, and there is conflicting evidence for what the AMOC did during the LIA (stronger or weaker) I would urge a more speculative tone to the AMOC link. In my opinion it is not needed -this is an interesting paper that is drawing together lots of SSS records and trying to synthesize them with terrestrial data; the AMOC link weakens it.
Reviewer #2 (Remarks to the Author): I am very supportive of the kind of work that is reported on in this manuscript. The authors use a variety of paleoclimatic indicators to provide insights onto the role of centennial scale changes in the Atlantic ocean on the larger climate system. Specifically, they focus on records from the Gulf of Mexico that indicate sea surface temperature (SST) and sea surface salinity (SSS) conditions that may also be related to larger scale conditions in the Atlantic. From these records they infer centennial scale variations and their larger climatic associations.
I feel that this work may eventually be suitable for publication, but the primary problem I see now is that the writing is not at all sufficiently clear to be of utility for non-specialists. My background is as a climate modeler who is very familiar with issues related to Atlantic variability and climate. However, I found the discussions of the various processes and their variations to be very confusing. I strongly suggest that the authors fundamentally rethink exactly what they are trying to communicate, and that in doing so they very critically think how they structure their manuscript.
Specific comments 1. Please provide line numbers for ease of commenting on specific sections of the text.
2. Bottom of page 2 … I would really question the statement saying that the last millennium has "… well-known external forcing …" . Also, if the words "well-know" in that sentence also apply to "characteristics global oceanic and terrestrial fingerprints", then I challenge what is really meant by "well-known".
3. Page 3 I am not completely convinced by aspects of the discussion of the Loop Current, Gulf off Mexico reconstructions, and Atlantic conditions. There is an assumption that warmer and saltier conditions are associated with a stronger Loop current and transport through the Florida Straits. But the authors also suggest that it is really the frequency of eddy shedding and associated transport of warm, saline water into the Gulf of Mexico that influences SST and SSS at the reconstruction sites. Thus, what is the linkage between eddy shedding and the strength of the transport from the Caribbean through the Florida Straits and into the North Atlantic? This gets at the question of the real linkage between the Gulf of Mexico reconstructions and larger scale climate. This manuscript presents very interesting analysis on the linkage between centennial-scale Atlantic variability and western hemisphere hydroclimate over the last millennium using multiproxy synthesis. The Garrison Basin reconstructions of SST and SSS proxy are linked to the loop current strength and thus AMOC variations. These proxies are compared with other proxies for AMOC and rainfall. The results suggest that the AMOC was weakened during the Little Ice age and had important impact on western hemisphere precipitation at various locations. The results are further supported by a transient model simulation.
The topic on the linkage between AMOC and western hemisphere hydroclimate is important and will attract wide interests in the community and the wider field. The use of Garrison Basin reconstructions of SST and SSS proxy variations as a proxy for the loop current and AMOC strength are novel results. The multiproxy synthesis provide important information for the teleconnection between ocean dynamics and hydroclimate. The results will influence thinking in the field. Meanwhile, some important technical aspects need to be clarified in the manuscript. I recommend the manuscript be accepted for publication in Nature Communications after some minor revision outlined in the following specific review comments.

Author Response to major comments by Reviewer #1:
Comments: Reviewer #1 writes that our work is commendable, valuable, and worthy of publication with the following four major points that could stand to benefit our manuscript: 1. Reconcile our study with a new, long-term SSS dataset: Our analyses and interpretation need to be reconciled with and discussed alongside a newly published, long-term (1896-2013) surface salinity dataset over the northern Atlantic Ocean from Friedman et al. (2017).
2. Bioturbation and sample resolution versus replication: Reviewer #1 cautions against over-interpretation of our replicated results and asks us to clarify our interpretation with regard to the sample resolution available from the Garrison Basin cores, reconstructed climate variability, and potential effects of bioturbation.
3. Inclusion of previously published records: There are previous studies reconstructing salinity changes that we have neglected to include in our multiproxy synthesis and as such, by pointing this out, Reviewer #1 feels that we might have "cherry-picked" our analysis to support our hypothesis.
4. Causal link to AMOC during the LIA: Reviewer #1 suggests using a more speculative tone concerning our linkage of the Garrison Basin reconstructed δ¹⁸O-seawater record to potentially altered AMOC during the LIA. Moreover, the timing of the LIA hydroclimate changes and potentially coeval changes in ocean circulation requires further discussion and explanation.
Firstly, we thank Reviewer #1 for a constructive, in-depth, and positive review of our manuscript and also for supporting publication of our study. Below, we address the major points put forth by Reviewer #1 and detail how we have incorporated her/his suggestions.

Author Reply to Point 1 (New SSS Dataset):
We thank the reviewer for pointing out this new study by Friedman et al. (2017). At the time of submission of our manuscript, this paper was not yet published and as such, we did not include it in our study. We have rectified this and included the study in our discussion and references. See e.g. Lines 116-122. In the study, they show that there is a century-long salinification trend in the TATL PC time series with an EOF pattern indicating basinwide salinification. Although their SSS dataset does not include any datapoints from the Gulf of Mexico ( Fig. 1 Friedman et al., 2017), such a basinwide salinification of the northern tropical-to-subtropical Atlantic Ocean is very much in line with results presented in our study. Our newly reconstructed, replicated δ¹⁸O sw record from the Garrison Basin, displays a secular trend from lesser-to-higher values, also indicating a centurylong trend of salinification. However, as Reviewer #1 suggests later, it is important to note that our temporal resolution and associated uncertainty (in both age and δ¹⁸O sw values) precludes a quantitative comparison with the new SSS dataset. We stress upon this caveat in our revised text.
Qualitatively, the Garrison Basin record and other proxy records of δ¹⁸O sw shown in Figure 4 seem to indicate that this century-long salinification trend is part of a multi-centennial salinification trend following anomalous freshening during the onset of the Little Ice Age (LIA), a major finding in our study. We have updated our discussion to reflect this.
On the other hand, Friedman et al. (2017) find that the NATL PC time series seems to display a long-term freshening trend whereas it switches into a salinification trend from 1970-2013 (Fig. 3,  Friedman et al., 2017). Reviewer #1 feels that this 1896-2013 trend is at odds with our inference, however, we feel that this is not the case for two primary reasons: 1. The δ¹⁸O sw proxy records that we utilize in our synthesis, except for three records, are situated south of 45°N, the southernmost latitude for the NATL region as delineated in Friedman et al. (2017). The three proxy records that are located in this region are from the Labrador Sea 1 , the South Iceland rise 2 , and the Feni Drift 3 (newly included based on Reviewer #1's suggestion). Unfortunately, the temporal resolution, age and analytical uncertainty in these records preclude a quantitative comparison with the newly compiled SSS dataset from Friedman et al. (2017). Furthermore, even if there was a freshening trend from 1896-2013 (i.e. one century-long) in the region as purported by Friedman and others, the three paleo-records indicate a multi-centennial salinification trend from 1400CE-present, and thus, a century-long, higher-resolution, freshening trend would have to be superimposed on top of this longer trend. Thus, these records support our inference that there was a basin-wide freshening event during the LIA and a multicentennial salinification trend into the instrumental era. 2. The standard deviation of the TATL salinity timeseries is almost an order of magnitude larger than the NATL timeseries (Fig. 1b, Friedman et al., 2017) and hence there is a lower probability of proxies being able to capture and resolve potential trends in the NATL domain compared to the TATL domain.
Despite not being able to perform quantitative comparisons with these new SSS data, the Our study does not preclude the influences of the NAO or AMO on the AMOC or vice versa. We refrain from mentioning those modes of variability as they operate on timescales shorter than available temporal resolution of our reconstruction and synthesis. However, we have included a line to highlight the importance of subcentennial processes and the role they might play in explaining SSS variability. Lines 230-233.

Author Reply to Point 2 (Bioturbation and resolution):
We include a section in the supplementary discussion about sample resolution and bioturbation. As stated there, we used X-Ray computed tomography and visual examination to investigate the effect of bioturbation on our cores. These inspections seemed to indicate a minimal influence of bioturbation (e.g. negligible burrow sizes etc.) Regardless, as the Reviewer suggests, since the Garrison Basin is not anoxic, it does not contain laminated sediments. Therefore, we used statistical modeling to investigate the effect of bioturbation and how it would smooth climatic signals in a foraminiferal record based on our reconstructed results. We used the TURBO2 code, a MATLAB algorithm 7,8 that simulates bioturbation, based on input mixing depths. We used a range of published mixing depths based on radionuclide measurements in the Gulf of Mexico 9 as input. Essentially, we can summarize our results as follows: despite the range of mixing depths applied, simulated signals of bioturbation did not result in the removal of centennial-scale variability and more importantly, each simulation resulted in the positive identification of a changepoint in δ 18 O sw records during the onset of the LIA, a central point of our study.
We have reworded our manuscript and supplementary materials to echo the Reviewer's concern regarding over-interpretation of our replicated results and associated errors. We ensure to caution the readers that though we have accounted for many forms of uncertainty, structural uncertainty in interpretation and reconstruction is still prevalent.  15 ) or seasonally-sensitive proxies (such as N. pachyderma [12][13][14] to facilitate a mean-annual, surface-ocean-only comparison."

Author Reply to Point 4 (Link to AMOC):
We agree with the Reviewer that there is conflicting evidence regarding whether AMOC variability was higher/lower during the Little Ice Age. Records of sea-surface salinity alone, such as the new Garrison Basin reconstructions and those compiled in our synthesis, cannot comprehensively constrain changes in the AMOC system. Regardless, they can provide a perspective on changes in surface-ocean circulation variability [16][17][18] , that may or may not be tied to changes in the AMOC system. Moreover, as we demonstrate, such a synthesis of SSS can be correlated to proxy reconstructions of rainfall/atmospheric variability. Although the development, refinement, and production of future records that can directly track processes related to deepwater formation, for example sortable silt 19 or radiogenic tracers 20 , will provide greater constraints on changes in AMOC, we feel that δ 18 O sw records are currently underutilized in advancing knowledge on surface-ocean circulation and advection.
More pertinently perhaps, we also agree with the reviewer that sensitivity experiments performed on the MPI-ESM model indicate that anomalous changes in subpolar gyre strength (and resultant changes in Atlantic surface-ocean circulation) need not be tied to direct changes in the AMOC system as a whole. We incorporate the reviewer's suggestions by revising our Abstract, Introduction, and Discussion to take a more speculative role for the causal link with AMOC changes and deepwater formation. We have now more explicitly stated that our SSS synthesis alone cannot confirm the hypothesis that AMOC was reduced during the LIA, but rather, is consistent with this hypothesis. Furthermore, changes in the AMOC system are not needed to cause such shifts in ocean-atmosphere process that bring with it resultant spatial pattern in SSS and western Hemisphere hydroclimate during the LIA, as evinced by the model-data comparison. Current transport and more frequent eddy shedding into the northern Gulf of Mexico and we then tie this framework to greater Atlantic Ocean variability using correlation analyses ( Fig. 1 and   5). On the reviewer's suggestion, we have clarified our text to emphasize this control of eddy shedding on SST/SSS anomalies and its link to the strength of transport (See Lines 142-154; 212-220).

Reply to Point 3 (Model Simulations):
Although it is beyond the scope of this manuscript to investigate the nature of forced versus unforced variability in the model, we have reworded our discussion to comment on this aspect and reference new papers that evaluate such forcings over the last millennium (Wang et al. 2017). We also note that there is a new paper that has been published (Moreno-Chamarro et al. 2017), which we now reference, that goes into detail regarding the model's ability to reproduce a Little Ice Age. We also reference the three papers from our group that have investigated the models' ability to simulate the Reviewer #2 writes that "I do not find that the model results add much ... what is the relevance of the fact that these are simulations of the last millennium rather than control simulations?" We note that there are problems with all models, although, having another source of information, and the ability to investigate mechanisms is immensely useful. We justify keeping the model simulations in the manuscript as there is remarkable similarity between the correlation analysis in the model on centennial timescales and the observations on decadal timescales. As the reviewer also points out, there are important and significant differences as well: we note that these differences can be due to model biases, observational shortcomings, and/or realistic differences in the climate signal between multidecadal and centennial timescales. We have included all of these caveats into our revised discussion and incorporate the Reviewer's suggestion to make more use of the model simulations.
Reviewer #3 writes that our manuscript and analyses makes for an interesting study that will attract wide interest in the community and larger field. We thank the reviewer for their constructive comments and suggestions. They recommend publication in Nature Communications with the following suggestions for minor revisions. We address these below: Comment: "It would be nice to show the low frequency time series of GOM SSS, loop current strength, and AMOC strength in this coupled simulation to verify this linkage and also compare the simulated time series of GOM SST/SSS and precipitation at various locations with the corresponding paleo records." Reply: The MPI-ESM model configuration that was used for the last millennium transient simulation, though it is of relatively high-resolution (1°x1°) compared to most state-of-the-art climate models, it is not sufficiently resolved to accurately simulate the Loop Current or associated eddy-resolving process that can be achieved with a higher-resolution, ocean-only model [1][2][3] . It does however accurately simulate several features of the climate system including the large-scale circulation and associated ocean-atmosphere processes [3][4][5][6][7] . Several modelpaleodata comparison studies demonstrate that it is not appropriate to compare gridpoint-togridpoint model output to proxy reconstructions 8,9 without interfacing either isotope-enabled outputs [10][11][12] , or using forward-modeling approaches 13,14 , or both 15 due to several factors 8,9,16,17 including model biases, non-standardized variance, stationarity issues, and spatiotemporal uncertainty in forcing and output. It is beyond the scope of this study to accurately compare climate output at various locations in the transient simulation to the several multiproxy records synthesized, although, there are many papers in the literature that investigate the robustness of the MPI-ESM transient output as well as its deficiencies in this regard 3,7,[18][19][20][21][22] . We note that climate models, and specifically transient simulations, are the best available tools to investigate mechanisms of the climate system in the past and large-scale analysis of model output such as our correlation map (Fig. 5) can delineate important modes of climate variability that might explain synthesized multiproxy records. For the purposes of our discussion, the MPI-ESM simulation reveals that there is a strong linkage between surface salinity changes in the Atlantic Ocean and precipitation variability in the continental Western Hemisphere on centennial timescales, similar to that observed in the multidecadal correlation map of reanalysis observations as well as the changes observed in the proxy records between the Little Ice Age and the modern era. We have updated our discussion section to incorporate several of these points. For illustrative purposes, based on the reviewer's suggestion, we include a figure that compares the reconstructed salinity signal in the Garrison Basin stacked record in the supplemental section.
Comment: "Does the coupled model also simulate a AMOC weakening during the Little Ice Age?" Reply: Based on the reviewer's suggestions, we have revised our manuscript discussion to include a paragraph that details the evolution of AMOC in the transient simulation. In short, although the model does simulate time periods of weakened AMOC over the last millennium, as detailed in Moreno-Chamarro et al. 2015, 2016, and 2016, changes in the strength of the subpolar gyre are primarily responsible for altered Atlantic Ocean surface-circulation changes during the Little Ice Age. We clarify in our revised text that our synthesis and model comparison provide strong evidence that surface circulation was altered in the Atlantic during the LIA and only additional proxy records that directly track changes in deepwater formation can confirm that such surface-circulation changes were not coeval with changes in the AMOC system as a whole.
Comment: "Page 10, last paragraph, the manuscript discussed the southward shift of the ITCZ in response to a weakening of the AMOC strength, and could cite previous coupled modeling studies on this topic using water hosing experiments, such as Zhang and Delworth 2005; Stouffer et al. 2006, etc." Reply: We thank the reviewer for suggesting these references and have incorporated these hosing studies into our revised manuscript's discussion and references.