Comment

Comment

Author’s Response

RC1-1

Methods, section 2.1 – it’s not entirely clear which cores are recalculated for SIC and which for SST as part of this study; I think all the cores should be mentioned in the methods and clarified which are analysed from the scratch and which have had their results recalculated; and which are jut cited. Caption to Figure 1 is confusing in this matter.

Agreed – we will add all cores to methods and provide more clarity around which cores provide what information, and where this information was collected from.

For reference:

TAN96 => new data both for SSST and WSIC

SO136 => WSIC recalculated through augmented modern database (249 analogs vs 195 analogs in Crosta 2004)

E27-23 => Published data (Ferry et al., 2015)

RC1-2

Results – what is missing here is the figure and description for the results of the recalculated core SO136-111; it is a part of this study and needs to be described.

Agreed – we will include a description of the recalculated SO136-111 results.

RC1-3

Discussion, section 4.1 – this part belongs to Results, not in Discussion and the sentences that do belong to Discussion should just briefly describe the past conditions and trends for SIC and SST, e.g., line 276-278 and 282-283 so I suggest restructuring. And please provide time intervals for the periods you describe in text. Also, there is first mention of the core E27-23, which was not mentioned in Methods or Results and if it is a part of recalculation then it should be properly described. Otherwise, please provide a citation for this core.

Agreed – we will restructure this section to fit into Results and will provide time intervals described in text.

E27-23 is not recalculated within this study, but a citation (Ferry et al., 2015) will be provided. 

RC1-4

l. 26 – what quantitative technique was used to reconstruct SIC and SST? Transfer function? Please clarify

We used a diatom-based transfer function. We will update the language and clarify accordingly.

RC1-5

l. 30 – please provide percentage info for the SIC (consolidated)

Generally, 0-15% open ocean; 15-40% unconsolidated sea ice; >40% consolidated sea ice (Armand et al., 2005 and references therein).

We will clarify the text to something like:

“Following the modern concept (Armand et al., 2005 and references therein), we find that winter sea ice was consolidated (wSIC = >40%) over the core site …”

RC1-6

Abstract overall – seems like there might be too much detail regarding the past conditions, could be simplified and generalized a bit, e.g. SST values could be mentioned only for the minimum and maximum values and otherwise just refer to trends

Agreed – we will rework to remove excess details (e.g., lat/long coordinates and water depth) and align with other comments regarding the Abstract (e.g., RC1-23, 24).

RC1-7

l. 51 – what does it mean ‘dynamically linked’?

We use the term ‘dynamic’ in the convention sense to describe a “force that controls or influences a process of growth, change, interaction or activity” (from Merriam-Webster).

We therefore describe the link between sea ice and carbon sequestration as being ‘dynamically’ linked because each factor exerts some force or influence over the other.

RC1-8

l. 66 – either 23 to 19 ka or 23.000 to 19.000, please use consistent time scale; also, is it BP?

Yes – ages are presented in BP.

We will update and standardize to ‘ka BP’ throughout.

RC1-9

l. 64-84 – are there any other proxies providing information on reconstructed oceanic variability in the region? Such as foraminifera etc? Would be nice to mention

Yes  - there are other proxy reconstructions from the region that provide information on oceanic variability (dust, nitrogen, temperature, etc.), but not many foraminifera reconstructions. However, these proxies don’t necessary look at sea ice variability or capture it in the same capacity as do the use of diatoms and transfer functions.

We’ve kept this paragraph and paper primarily focused on sea ice, and while we acknowledge and appreciate the reviewer’s suggestion to include other key proxy reconstructions from the region, we feel as though discussing additional proxies from the region may detract from the tightly focused narrative.

It is also worth mentioning that forthcoming submissions from Chadwick et al. and Kohfeld et al., which will be submitted to this special issue, will address some of the larger topics concerning regional oceanic variability and reconstructions from the region and will supplement this manuscript.

RC1-10

l. 88-92 – this belongs to methods; introduction should mainly state general information on the materials studied

After some consideration, the co-authors have agreed that the text provided on lines 88-92 would likely be useful to readers who skip the methods section and only quickly read the paper.

We will remove unnecessary information (e.g., latitude & longitude, water depth) to streamline the reading, but believe the additional references to SO136-111 and E27-23 should remain; however, we are open to further discussion on this topic.

RC1-11

l. 196-206 – this part belongs to Results section

The text provided in the manuscript on line 196 may be slightly misleading – our analysis did not establish these taxonomic groups, as these have been used in other publications and are established methods (e.g., Crosta et al., 2004, Ghadi et al., 2020).

We will therefore change the wording to something like:

“Based on previously established taxonomic groups, diatoms were grouped into one of three  categories based on temperature preference and sea ice tolerance: …”

RC1-12

l. 201 – what is the sea ice concentration range for this group?

The highest abundances of the diatom species composing this group in the modern sediments are found at WSI greater than 60-70% (Zielinski and Gersonde, 1997; Armand et al., 2005; Esper et al., 2010). They are therefore all suited to record past changes in WSI (Esper et al., 2014).

RC1-13

Table 1 – just curious, did you identify any Thalassiosira antarctica var. antarctica? Its northern equivalent is pretty common in the Arctic and sub-Arctic region

Only a few specimens of Thalassiosira antarctica var antarctica form 2 (warm variety; Taylor et al., 2002) were found, and they were generally identified during glacial periods.

TAN => up to 1% of the total diatom assemblages

SO136 => up to ~2% of the total diatom assemblages

E27-23 => up to 1.5% of the total diatom assemblages

RC1-14

l. 212 – why did you choose this period only? Is the present-day diatom succession limited to January-March? Please clarify

January-March is mentioned only for the SST. In the Southern Ocean, diatom production is restricted to the sunlit period (spring to fall). Production starts earlier in the SAZ-POOZ than in the Sea Ice Zone, which is especially late in the coastal zone due to high sea-ice cover (Nelson et al., 2001; Arrigo et al., 2004; Grigorov et al., 2014).

Although there is a succession in diatom production from spring to fall (Grigorov et al., 2014) and that spring production may exceed summer production in some regions (Fiala et al., 2002), most of the export occurs during the summer months (Fiala et al., 1998; Kopczynska et al., 1998; Fischer et al., 2002; Armand et al., 2008; Grigorov et al., 2014; Rigual-Hernandez et al., 2015).

For these reasons, summer SST is generally a better explanatory variable than spring or annual one (Esper et al., 2014).

We will add additional clarity (not to this degree) to the manuscript to resolve any confusion.

RC1-15

l. 227-228 – it would be nice to consider other quantitative, transfer function tests at some point, such as ML (MLRC) and WA-PLS to show that MAT is indeed the best choice.

Other transfer functions have been tested using the modern diatom database used here (Ferry et al., 2015) and using another modern diatom database (Esper et al., 2014).

In Esper et al. (2014), MAT performed better in term of R2 and RMSEP. Though the G-IG patterns were reconstructed with both IKM and MAT, the latter reconstructed more variable sea ice at the multi-millennial timescale as IKM is known to smooth down records due to its approach (regression and paleo-environmental equation; Esper et al., 2014). Conversely, GAM and MAT provided similar results in core SO136-111 (Ferry et al., 2015).

Finally, it is worth noting that MAT provides SST and WSI reconstructions that are in agreement with other type of SST and WSI reconstructions (Gersonde et al., 2005; Civel et al., 2021), other downcore proxies and, more globally, Southern Ocean paleoclimate at any timescales (Crosta et al., 2004; Nair et al., 2019; Ghadi et al., 2020; Orme et al., 2020; Crosta et al., 2021; Shukla et al., 2021).

This topic will also be discussed in the forthcoming Kohfeld et al. manuscript, which will be submitted to the same special issue.

RC1-16

l. 242 – which periods specifically? Looks like MIS 1, 4 and 5

We will update wording to something like:

“The Sub-Antarctic Zone (SAZ) group had relatively low abundances, with higher values occurring generally during the warmer interstadial periods MIS 1 and 5, and briefly during MIS 4 at 67 ka.”’

RC1-17

l. 289 – I can’t find the description of cores MD06 in Discussion

We will rework the Methods & Results section to include a description of all cores that were used in the manuscript that were not already introduced (in line with comment RC1-21).

The cores that will be introduced for the %AAIW calculation include:

[1] MD06-2990;

[2] MD06-2989; and

[3] MD97-2120

From Pahanke & Zahn (2005) & Ronge et al. (2015).

In addition, the following cores are used in Discussion 4.3 for the SST gradient:

[1] SO136-GC3;

[2] FR1/94-GC3;

[3] ODP1119;

[4] Q200; and

[5] DSDP594.

We will add a sentence in the manuscript that points to these cores (and references) so that all cores used in this analysis are included within the text and cited.

RC1-18

l. 308 – please clarify that you mean explanation no. 3

Noted - we will rearrange the numbering as follows and update lines 305-307 accordingly:

[1] Different statistical applications;

[2] lateral sediment redistribution;

[3] differences in laboratory protocols;

[4] differences in diatom identification/counting methodology; and

[5] selective diatom dissolution;

We will then correct the numbering in the appendices but leave the wording from lines 308-327 as is.

RC1-19

l. 308-330 – I suggest to put this text in a separate sub-chapter as it stands out of the description of past conditions

We initially had this section broken out as its own sub-chapter (as suggested), but after reading and having discussions around the chapter’s flow, we decided to embed part of the discussion within the text and append the non-essential part of the discussion.

We are open to reworking this section and separate it into a sub-chapter if the reviewer feels this is important; however, in our own writing/re-writing exercises we have found the current state of the manuscript to have the best reading flow.

RC1-20

l. 462 – reference for the core is needed here

Noted - we will add the Pahnke & Zahn (2005) reference for core MD97-2120.

RC1-21

l. 528 – again, if these two cores are a key element of Discussion and overall conclusions, then we need more info in Methods and Results

Noted - we will include a description of these cores in previous sections. See response to RC1-17 for more information.

RC1-22

l. 549-556 – perhaps this part fits better to Introduction

We agree that the current reading of this paragraph would fit better within the Introduction. We would like to keep these ideas at the end of the paper, so we will rework the paragraph to read more as a conclusion.

We will change the text to read something like:

“In conclusion, this paper has focused exclusively on sea ice as a driver of physical change…”

and

“We recognize that these processes may not act independently, and as such, have contributed new data to help advance our collective understanding…”

RC1-23

l. 29 - coordinates etc should be removed from Abstract, too much detail

Agreed - coordinates and water depth have been removed from the Abstract.

RC1-24

l. 36 – ‘…coolest values, respectively…’

Agreed – Line 36 now reads:

“WSIC and SSSTs reached their maximum concentrations and coolest values, respectively, by 24.5 ka…”

RC1-25

l. 38 – SSST – too many S or sSSTs

This was a typing error and should have been “sSST”; however, based on other comments, we are updating all “sSST” to “SSST” throughout the manuscript.

RC1-26

l. 87 – SSSTs – should it be singular?

Yes – SSST should be singular. The text currently reads:

“SSSTs and wSIC are estimated by applying the Modern Analogue Technique …”

We will change the wording to:

“SSST and WSIC estimates are produced by applying the Modern Analog Technique …”

RC1-27

l. 114 – ‘published cores providing recalculated sea ice extent data’?

Only SO136-111 has been recalculated for this study. We will update the language to:

“… and additional published cores providing sea-ice extent data”.

RC1-28

Figure 1 – please add abbreviations SSI and WSI in legend

Agreed – we will update Figure 1 accordingly.

Comment #

Comment

Author’s Response

RC2-1

sSST is sometimes written as SSST throughout the manuscript

Noted - we will standardize and update throughout the manuscript to SSST.

RC2-2

From this paper alone, it is not clear what the percentage changes in SIC (%wSIC) represents. Does a value of 40% indicate that the amount of sea-ice is 40% of modern sea-ice concentrations or some other reference point? Or does it indicate that only 40% of the region around the core site is covered by sea-ice at this time? Furthermore, does %wSIC give any indication about what thickness of sea-ice is present? A couple of sentences in the methods section clarifying what “%wSIC” is would address this issue.

Sea-ice concentration (SIC) is a pixel/grid-scale observation defined as the fraction of ocean area that is covered by sea ice. Sea-ice concentration thresholds are generally: 0-15% open ocean; 15-40% unconsolidated sea ice; >40% consolidated sea ice (Armand et al., 2005 and references therein; Hobbs et al., 216). Therefore, a value of 40% indicates that 40% of the region over the core site was covered by sea ice during the winter at the considered time slice. These values represent a mean state integrated over the time period covered by the sample.

As requested, we will provide additional clarity on what wSIC is measuring more specifically.

RC2-3

Line 117-119: make it clear that these are modern(?) positions of sea ice extent and the subtropical/polar front.

Agreed – we will clarify that these are the modern positions of the sea-ice edge and fronts.

RC2-4

Line 235: what proportion of the overall numbers of frustules counted in each sample are in the transfer function training set? If the number (percentage) is low (<60%?) in any sample, are the sSST and %wSIC values compromised?

For TAN96, the downcore proportion of diatoms included in the TF is >82% (mean = 92%). The Sea Ice group accounts for <1% during interglacials, and up to ~7% during glacials.

For SO136-111, the downcore proportion of diatoms included in the TF is >79% (mean = 91%). The Sea Ice group accounts for <1% during interglacials, and up to ~4% during glacials.

No samples report using <60% of total identified specimens and therefore our SSST and WSIC estimates are not believed to have been compromised by low proportion of the TF diatom assemblages.

RC2-5

Line 308/309: is the Ferry et al (2015) data available for you to run through your transfer function?

The MAT has been applied on Ferry’s data (core E27-23; Figure 1 included below). Results appear very similar to the published ones, especially in the timing of sea-ice changes. This was observed and published for core SO136-111 in Ferry et al., 2015.

Comment #

Comment

Author’s Response

RC3-1

Sedimentation rate in core TAN1302-96 is much higher during interglacial/warmer period than during glacial period and MIS 2, 3 and 4 are represented by less than 30cm in that core from ~90 to 120cm. The period that the authors discuss as MIS 3 is part of MIS 5. The evidence comes first from the d¹⁸O stratigraphy measured on N. pachyderma (senestre? should be indicated by the authors). The values measured between 120 and 170cm are clearly too low to represent MIS 3. They indicate that from 120 to 300cm the sediments were deposited during MIS 5. This is also indicated by the 14C data: measurements at both 130 and 170cm indicated dates undistinguishable from background because both are older than 70kyr. Further evidence of the “extended” MIS 5 and shrinked MIS 2, 3, 4 could come from the carbon isotopic record of N. pachyderma but they are not presented in the paper. The authors could/should compare their isotopic record to the isotopic record of core SO136-111 that they are also using in this paper if they want further evidence. Correcting the chronology for the studied core TAN1302-96 will make it possible to reconcile the sea ice record of this core with those of the 2 cores from the same area: core E27-23 and core SO136-111. It is not clear if MIS 6 is represented in the core. There is no corresponding isotopic value but it might be due to the low resolution of the isotopic data

We appreciate the reviewer’s deep engagement with the data provided in this manuscript. This comment has provided valuable discussion surrounding the robustness of the age model as currently outlined in the manuscript, and we welcome discussions to improve the reliability of our data and interpretations.

Age model construction

To test the reliability of our age model, we have constructed 4 additional age models (5 total) and have set up a series of tests to determine their reliability. All age models use the youngest 5 radiocarbon samples outlined in the manuscript but use different tie points for the older portions of the core (incorporating comparisons with the EDC (dD as was done for SO136-111). The new age model versions also make different use of the 2 older, NDFB radiocarbon samples at 130 and 170cm. The 4 additional age models that we compare are as follows:

[1] EDC 1: includes both the 130cm and 170cm radiocarbon samples and is tied to the EDT temperature record (dD data from Stenni et al. (2010) on the AICC2021 timescale from Veres et al. (2013)). The NDFB radiocarbon ages used were 57.5 ka and 57.7 ka, respectively, as these were the lower bracket of the NDFB results supplied by the CAS laboratory.

[2] EDC 2: includes only the 130cm radiocarbon sample (using an age of 57.5 ka) and excludes the 170cm sample. This model uses the same tie points to the EDC SST record as were used for EDC 1.

[3] EDC 3: excludes both the 130cm and 170cm radiocarbon samples and uses the same tie points to the EDC SST record as were used for EDC 1.

[4] R3: based on the reviewer’s comments, we attribute an age of 80 ka to the 130cm radiocarbon sample (reviewer suggests >70 ka) and a date of 104 ka to the 170cm sample. These date attributions are based on tying the TAN96 d¹⁸O record to the LR04 stack at the location of the radiocarbon samples. We then tie the TAN96 d¹⁸O record to the LR04 stack and assume that all sediment between 120 and 300cm accumulated during MIS 5, and that sediments between ~90 and 120 cm correspond to MIS 2, 3, and 4.

[5] d18O 1: original age model used in the manuscript, which uses 5 youngest radiocarbon dates, excludes the NDFB dates, and tie points between the TAN96 d¹⁸O record and the LR04 stack.

Age model comparison:

Following their selective tuning to EDC and d18O records, we compare these 5 age models based on:

[1] match of the d¹³C records between TAN 96 and nearby core SO136-111

[2] sensible behavior and magnitude of sediment accumulation rates relative to nearby SO136-111

[3] overall fit to d18O and EDC SST

Observations

Our comparisons suggest:

[1] Our current age model (d18O 1) fits very well with the d13C record from SO136-111 (Figure RC3-1), suggesting that it is consistent with the age model provided for SO136-111.

[2] R3, EDC 3, and d18O 1 all provide reasonable fits to the LR04 stack

[3] d18O-1 and EDC 3 provide the most sensible sedimentation rates when compared with SO136-111 rates.

Additional thoughts:

Overall, we find it unlikely that the TAN96 sedimentation rates would be reduced to near-zero (~0.2 cm/ka) during glacial periods (i.e., ~13cm (117 to 130cm) deposited over >60 ka for MIS 2, 3, & 4), as was suggested by Reviewer 3.  While it makes sense that polar cores with >80% sea ice cover would experience greatly reduced sedimentation rates, the wSIC estimates for TAN 96 are 40-50% during glacial periods, suggesting that the study area would experience some productivity during glacial times. Furthermore, the proximal site SO136-111- which has comparable wSIC estimates during glacial periods - does not exhibit this behavior, as sedimentation rates are between 2-3 cm/ka during glacial periods.

Conclusion:

Our final response to reviewer document will provide the details of these comparisons, and our revised manuscript will provide a more comprehensive explanation of our age model determination including all supporting information.

RC3-2

Chronology: As the authors indicate , significant MRA variability occurs over a glacial cycle, specifically in the southern high latitudes. They should use as a minimum ±100 years for the uncertainty on the MRA as it is the variation indicated by Paterne et al., 2009, for the last century. The authors do not indicate the uncertainty they evaluate for the tie points used to correlate the planktic isotopic record to the LR04 benthic record. From figure 3 it seems that they also choose a too small uncertainty. Anyway the authors should give more details. Furthermore as they present a planktic isotopic record and a SST record and as their goal is to discuss the impact of sea ice extent on atmospheric CO₂, it would make more sense to establish the chronology of MIS 5 comparing their records to EDC deuterium record, following Govin et al., 2015, Capron et al., 2014. Anyway the record resolution is pretty low (partly due to the low sedimentation rate of the core) so the real uncertainties are large and this comment is not that important.

We have taken this comment into consideration and are in the process of comparing alternative age models that use the EDC deuterium record.

As noted above, we will also provide more information on the age model that is selected, including a minimum ±100 year uncertainty for the MRA and more information on tie point uncertainty.

RC3-3

SST and wSIC: the authors should give more details: how many analogues have been used for reconstructions? Is the error indicated on the figure the standard deviation between the different analogues? The tables should be available to reviewers.

In this version, the MAT has a ~1°C RMSEP for SSST and a 10% RMSEP for WSIC on the modern validation step. These errors are generally applied downcore as other TF (IKM, WA-PLS…) and geochemical proxies (TEX86, UK37, Mg/Ca) only provide a mean error on the calibration. However, MAT allows for a sample-only error, calculated as the standard deviation of the chosen analogs. In core TAN96, SSST standard deviation varies between ~0.2°C and ~2°C (mean of 1.28°C, in good agreement with the modern calibration) (Figure RC3-2). WSIC standard deviation varies between 0% during interglacials to ~30% in glacials (mean of 8%).

In SO136-111, SSST standard deviation varies between ~0.5°C and ~2°C (mean of 1.14°C) (Figure RC3-2). WSIC standard deviation varies between 0% in interglacials to ~25% during glacials (mean of 9.96%). All these data are provided in the appendix table. In TAN96, the dissimilarity of the fifth analog varies between ~0.15 and ~0.4 (mean of 0.23), far below the threshold of the first quartile (0.7). All five analogs are always preserved and estimates & SD are done on 5 analogs. In core SO136-111, the dissimilarity of the fifth analog varies between ~0.05 and ~0.3 (mean of 0.14), far below the threshold of the first quartile (0.7). All five analogs are always preserved and estimates & SD are done on 5 analogs. These data are not pivotal to the manuscript but a mention to the good dissimilarity and to the calculation on 5 analogs will be added to the revised manuscript.

RC3-4

Results: what is indicated in the text is not what is presented on the figures. Some examples: line 253, the SST increase seems to be ≤3°C on the figure. Taking into account uncertainties ~1 to 4°C would be precise enough. Line 254: on the figure the 2 methods indicate ~22 to ~33% wSIC for the oldest point. Where does the 48% comes from? Line 256: I do not see a rise in SST during MIS 5e, only variability.

This is a good catch – after looking through the datafile, an additional blank line of data was accidently added causing some of the data points to be shifted and/or not included in the figure.

An updated figure is provided below (Figure RC3-3) which shows the wSIC of 48% at 140 ka. We note that the SSST value at the MIS 5e/6 boundary is 4 °C in the updated figure, and the text will be updated to reflect the corrected value.

Finally, we do not disagree with your observations regarding SSTs during MIS 5e. We will change the wording to something like:

“Reconstructed SSSTs were variable throughout MIS 5e, reaching a maximum…”