General comments:

I have two general comments.

One is that I found it difficult to figure out what are the contributions and findings of the author team and what is the information that was taken from other sources / previous work. Don't get me wrong here: You cite and give credits to this other work very well but this information is so much mixed with the description of your results that I found it hard sometimes to disentangle these.

Reply: Thank you for the constructive review. We will attempt to clarify the results and discussion section to separate better the results from the original datasets and our compilations based on clusters. See more detailed responses to your comments below.

The second general comment exactly hooks up on this last impression. I was wondering whether the overall results of your manuscript could be worked out more clearly by finding a way to better separate your results and their discription / interpretation by you on the one hand from the critical discussion of the results and their limitations plus hypotheses developed based on combining your results with those of previous work on the other hand.

Reply: See previous reply. We think it is important to discuss the original data and methodological discrepancies (see also comments from Reviewer 2), and thus we discuss the original data when interpreting sea-ice trends in the clusters produced here. Presenting the results from the local data, without directly interpreting it regarding its local informational value, would not be very meaningful.

Specific comments:

L67/68++: I can agree that there is truly very limited information about the past sea ice cover in the Arctic; there are however some useful extensions of information about the sea ice cover from the satellite era back into the past - especially in the European Arctic - based on observations. For instance: Divine and Dick, 2006, Historical variability of sea ice edge position in the Nordic Seas, J. Geophys. Res.-Oceans, 111; England et al., 2008, A millennial-scale record of Arctic Ocean sea ice variability and the demise of the Ellesmere Island ice shelves, Geophys. Res. Lett., 35; I am sure there are more.

There are also reconstructions of, e.g. the Arctic sea ice volume, based on numerical modeling carefully tuned to present day conditions (see Schweiger et al.,  2019, Arctic sea-ice volume variability over 1901-2010: A model-based reconstruction, J. Climate, 32).

Not sure whether you'd consider it worthwhile to harvest archives for more published work of others in this respect. It might be quite interesting.

Reply: The lines you mention list the very few compilations of proxy-based sea-ice reconstructions. It is true historical observations exist and we will mention this in the Introduction. However, because the focus of this study was on sea ice reconstructions for the whole Common Era, we did not include any data that didn’t cover at least 80% of it. Historical reconstructions don’t fit this criterion and therefore were not included in the statistical analysis. Unfortunately, the ice-shelve reconstruction focused more on the ice-shelve establishment than on sea-ice variability, and on longer time-scale trends. The Arctic sea-ice volume reconstructions from Schweiger et al (2019), on the other hand, dealt with sea-ice changes over the last century, over an annual to decadal timescale. Therefore, the timescales were not compatible with the criteria set (see methods). However, we will include the study of the historical sea-ice edge position in the Nordic Seas in the discussion.

L184-186: "Sea-ice extents ... gives the annual extent". In the sea ice community, sea ice extent is defined as the sum of the area of all grid cells covered by at least 15% sea ice. In contrast, sea ice area is the sum of the area of all grid cells covered by any sea ice weighed with the actual sea ice area fraction. From what you describe you seem to have computed the sea ice area and you should name it like this henceforth - aka use "sea ice area" instead of "sea ice extent". You might need to correct this in the entire manuscript.

Reply: Indeed, we used the sum of the area of all grid cells covered by any sea ice weighted with the actual sea ice area fraction. Thus, the terminology will be corrected to “sea ice area”. Thank you for pointing it out.

Figure 1: Please note the source of the bathymetry used and also note where the arrows denoting the currents stem from. If you plotted the latter by yourself you might want to at least state which source you used as blueprint. What is the source of the sea-ice median extent?

Reply: All the sources will be included in the revised figure caption.

Figure 2:

- I was wondering whether it would make sense to explain in a bit more detail what a dendrogram in general and in this case tells us.

Reply: We will change “Dendrogram” to “Dendrogram from cluster analysis” in the caption. A more detailed explanation will be added in the “2.3. Statistical analyses” and “3.1. Proxy-based sea-ice reconstructions” items.

- I can see that panel b) shows a standardized quantity that somehow seems to be related to sea ice (area?) reconstructions. But the y-axis annotation says "Standard deviation" and the text in the caption speaks of "average composite ... solid ... mean values of all records .... dashed values represent the amplitude". These descriptions seem not to match well with each other and some explanation might be helpful.

Reply: The y-axis label will be fixed to “standardized sea-ice reconstructions” and the individual reconstructions will also be plotted in the figure to improve the understanding. The same will be done for Figure 3.

L225: "Thus, an increase ..." --> I agree that this interpretation can be made. I was wondering however, to what degree a change in the dominant ice type, i.e. from multiyear ice to seasonal ice, could also have resulted in an increase in the seasonal sea ice - being interpreted as an overall increase in sea ice - which would be contrary to what we have been observing over the past decades in the Arctic.

Reply: Most of the cores are located in regions that have not been covered, during the Common Era, by multiyear sea ice. The only cores located in areas that might have been influenced by multi-year sea ice are cores D and M. And, for these two cores, the change in the dominant ice type will be included in the discussion, which was especially relevant for site M (composite core 03TC-41GC-03PC).

L282: "The similarity ..." I am not so sure I go with this similarity here because G1 seems to have an accelerated increase from -1 to +2 until 1700-1800 CE and a rather moderate decrease afterwards while G2 kind of ramps up from -1 to near 0 around 300 CE, followed by a rather linear and comparably weak increase to values around +1 in1500 CE followed by a quite remarkable decrease to values below 0 until present day. The time of the maximum value appears to coincide with a remarkable ramp up in G1 values. This is what I see there and I am not sure this could be explained by large-scale climate forcing that easily.

Reply: We agree that the sentence is a bit vague, and we will describe the details more clearly in the revised version. There is a common long-term trend from around -1 to +1 until around 1500 (so ¾ of the Common Era) in both clusters. G2 is less smooth compared to G1 including your mentioned jump at 300 CE and hence includes temporal deviations from the overall similar trend. A clear divergence dominates the last 500 years. We will edit the text accordingly. Partly the reason can be methodological, i.e., the sea-ice maximum was not captured by the IP₂₅ proxy. This will be explained in the text, which we tried to express better.

L289: "It was probably caused by ... of the coast." --> I take this sentence as the example to express the impression that a lot of what is written in this paragraph is less the presentation of results but rather a lot of discussion and hypothetical statements.

Reply: We intentionally chose to merge results and discussion. Representing only the results for such local variations would be very descriptive and we would need to repeat all details in the discussion again, if separated. In our view, these local variations are only interesting or relevant if they are directly explained. Hence the direct inclusion of original references and/or own explanations. We will, however, try to clarify the reasoning behind the “hypothetical statements”.

I note that section 3 is indeed entitlled "Results and Discussion" but I was wondering whether a more distinct separation of what are your results (i.e. what is new) and what are the points of discussion and hypotheses where you mix in a lot of information from other authors. I find it difficult to focus and lose traction on the results of this paper.

Reply: See previous reply. Our result is the synthesis of local data to identify common vs. local changes which still requires to account for the local information and/or specifics of the sea-ice proxy from original studies. We are wording out “the original authors” when we look at their interpretations of the data; we think it is important to carefully consider the data origins beyond the cluster behaviour. But we will change the wording in some places to highlight better our results/interpretations and the interpretations by the original authors.

If you would try to separate discussion issues more clearly from the results issues it might also become more clear what the different influencing factors as well as the various limitations of the approach used actually are. Among these would be the first-year ice - multiyear ice issue repeatedly mentioned, or issues like changes in the location of the land/ice - sea ice transition zone and thereby in the strength of katabatic winds / polynya existence by changes in ice sheet extent.

Note: This comment applies also to the previous and the following paragraphs

Reply: See our response above.

Table 2: I was wondering whether you at all thought about comparing the output of these two models also against observations of the sea-ice concentration from satellites. There you could only use data from the late 1970ies onwards but it might provide you with an idea whether any of the models is potentially biased in its representation of the sea ice extent (or area?) See my previous comment about your description of how you compute sea ice extent as well.

Reply: We are not aware of any notable link between biases and differences in long-term variations and trends. However, we agree that it is a relevant point to add a verification of the simulations. We hence did such a comparison now and suggest adding the comparison between the models’ results and the satellite data, for the period between 1979 and 2000, and will include the table as Supplementary Material (added as supplement here as well - Table S2). 

Typos / editorial comments:

Line 47: "preventing heat and moisture transfers" --> perhaps better: "reducing or even almost preventing heat and moisture transfers"

Reply: The text will be changed as suggested.

Line 95: Suggest to replace "high-resolution" by the actual resolution in years.

Reply: We cannot inform one specific resolution value because each of the 14 sea-ice reconstructions have their own resolution, which also varied through the records. But the average resolution will be added to Figures S8 to S14 in the Supplementary Material.

L104 / L141: "original authors." --> Not clear what this means. The co-authors of this manuscript? Are the data associated with this part of the archived data you downloaded or is this additional data. If the latter I recommend that you emphasize this more.

Reply: In the first case (L104), it was additional data that were not found in the searched databanks. In the second case (L141), the original age-depth models were used. In both cases, the sentences will be rephrased to improve readability.

L195/196: "offset of 0.5 to avoid zeroes" --> So you are actually working with sea ice concentration data sets that have values ranging between 0.5 and 1.5, is this correct?

Reply: The offset is an arbitrary number to allow for a log-transformation of fractional data for statistical analysis. So, the used numbers are the log-values derived from the fractional data across the range 0.5 to 1.5. It is a common procedure when using fractional data like sea ice or similar.

L251/252: "common non-linear trend" --> Two questions here: 1) why non-linear? and 2) would you mind to share the values of these trends?

Reply: The “non-linear trend” is the average composite observed in Figure 2. The values of all the composites will be included in the Supplementary Material of the revised manuscript.

L277/278: "since most of the results ... period" --> You could put more emphasis on this issue by providing information and/or referring to the actual time coverage.

Reply: We will emphasize the information regarding the cores that do not cover the last two centuries of the Common Era.

Figure 5: I strongly recommend to have identical ranges of the extent displayed at the respective axis, i.e. N. Hemisphere Summer sea ice extent has the same axis range, N. Hemisphere, Winter sea ice extent has the same axis range.

Reply: The idea behind the variable Y-axis ranges was to emphasize the similarity between sea-ice evolution from the whole Northern Hemisphere and from the area around Greenland. However, we agree and will set the same Y-axis ranges as suggested as it is still possible to observe the similarity between the two regions.

 Comments:

- Generally, the manuscript is well written.

-The literature referenced is current.

- The database of proxy-based sea-ice reconstructions is solid and the generation of the data is consistent with established methods of organic-geochemical analysis.

Reply: Thank you for the constructive review.

-The authors have used nearly all relevant proxy-based sea-ice reconstruction datasets currently available for the Common Era in this area. The only data set, I’m missing, is from Core MD99-2275 (Iceland) published by Massé in 2008 (Guillaume Massé, Steven J. Rowland, Marie-Alexandrine Sicre, Jeremy Jacob, Eystein Jansen, Simon T. Belt; Abrupt climate changes for Iceland during the last millennium: Evidence from high resolution sea ice reconstructions. Earth and Planetary Science Letters 269, 564-568).

Reply: We looked at the Core MD99-2275 data. Unfortunately, it does not fit our time coverage requirements (80% of the Common Era), since the data ranges between 800 and 1950 CE. Adding shorter timeseries would deteriorate the consistency of long-term trends as these would then partly depend on data availability rather than only climate. But we will reference the publication in the discussion.

- Personally, I would have liked to see some additional explanations of the figures so that also non-specialists could somewhat better evaluate the inferences made based on the results. Thus, at its present stage, it limits the readership, even though the topic is certainly of interest to a broader community due to its topicality.

Reply: See our responses to Reviewer 2. We will change axis title names and add explanations to the captions.

- In the text, the authors mention the Little Ice Age several times. It would certainly be helpful if such events would be highlighted in the figures (e.g., fig. 2d). As an example, see Kolling et al. (2017) figs. 2 and 6. The same applies to any recognizable warming events such as the Medieval Climate Anomaly (especially since Wang et al. 2022 is cited).

Reply: We will add the indication of the Little Ice Age and the Medieval Climate Anomaly in Figures 2 and 3.

- The authors have critically discussed the differences between the matches (or non-matches) of the results of the proxy-based datasets and the modeling results of the long-term trends and short-term variability and substantiated them with recent publications.

Reply: Thank you for your kind comment. We tried our best.

-The authors have also not failed to point out the problems of interpreting different proxies, which are, for example, due to different habitats and different seasons of reproduction of the producing/synthesizing organisms and are additionally influenced by sea-ice independent parameters. In this context, I would additionally recommend Spielhagen et al. 2011 (MSM5/5-712; Enhanced Modern Heat Transfer to the Arctic by Warm Atlantic Water, 28 January 2011, vol. 331, Science) for the discussion, even though foraminifera are not used as a proxy in this manuscript.

Reply: We will add the suggested reference in our discussion.

- From the proxy point of view, it would be desirable if at least one proxy record as an example of each of the three groups would be transferred from the supplement to the main body of the manuscript.

Reply: To avoid bias when choosing the only one proxy record per group, we added all the standardized proxy records in Figure 3. For the individual records with their identification, the readers are then referred to the Supplementary Material.

One last comment to Chapter 2.1. (line 131-133): ........"When available"...... In case of “not available”, does this mean, that you have used different units in some cases! This is not the usual procedure.

Reply: In those cases where the data was not normalized by TOC, we used the concentration data (µg/g dw or ng/g dw). But we will change the sentence to make the information clearer.

 Comments:

Line 60: “Most human observations of sea ice are derived from satellites and date back only to the 1970s”

The authors are advised to check the older publications of Vinje et al., 2001 (J Climate , V14, p 255-), and  Divine&Dick2006 (doi:10.1029/2004JC002851) where sea ice retreat after 1850 based on historical sea ice observations in the Nordic seas is discussed.

Reply: We will mention historical sea ice observations and the references in the Introduction, and also in the Discussion (Section 3.3).

Line 85: “…an integration of existing sea-ice reconstructions in a systematic way…”  consider adding “for the entire northern North Atlantic ”, since one can find earlier publications where regional scale compilations of sea ice proxies were made.

Reply: The sentence will be completed as suggested.

Line 91: “… large model biases exist in Arctic regions, attributable to complex feedback framework…” consider adding  “… and remaining deficiencies in the implementation of sea ice in GCMs… “

Reply: The sentence will be completed as suggested. We will now also include a verification of the two used models vs. observations as suggested by the other reviewer.

Line 108: F. cylindrus was demonstrated to be rather a cold water species than the one associated with sea ice (see Oksman et al., 2019, von Quillfeldt (2004)). You mention it later in the text, anyways, so what was the point to list it here as one of the indicator species?.

Reply: This is a good point, in some locations F. cylindrus could be a relevant sea-ice proxy and that is why we added it in the initial data search. But since we did not find these data, we will edit the text as suggested.

Line 164: “These represent the percentage of ice cover in each grid square and both variables are on atmospheric grids”

“percentage of ice cover in each grid square” is associated with sea ice area.

Why atmospheric grid (lower 2 degree resolution) is used for the analysis of sea ice data, while you state earlier that the model has a ~1° resolution in the ocean and sea?

Reply: In the CESM1, the sea-ice model is coupled more “tightly” to the atmosphere and land models than to the ocean model, in order to better resolve the diurnal cycles (Craig et al., 2012). As we focus on long-term variability and trends, the spatial resolution for the analysis is not that important as detailed local variability is not the main interest here.

Craig, A. P., Vertenstein, M., and Jacob, R.: A new flexible coupler for earth system modeling developed for CCSM4 and CESM1, Int. J. High Perform. Comput. Appl., 26, 31–42, https://doi.org/10.1177/1094342011428141, 2012.

Line 191: Some details on wavelet analysis are missing. Have you detrended the records prior to the wavelet analysis? Were the series resampled (I assume so), and to which common time increment? How the autoregression in the series was estimated prior to significance testing?

Reply: During the analysis, the data was internally detrended. For that, a degree of time series smoothing of 0.75 was used over 100 simulations. For the surrogate time series, we used an autoregressive model (AR-1). The time resolution for the model data was 1 year and the time resolution for the proxy data varied among the records. We will add more details about all the settings used in the wavelet analysis. The time resolution for each proxy record will be added to Figures S8 to S14 in the Supplementary Material.

Line 195: What was the point in log-transforming the data prior to analysis?

Reply: It is common practice to log-transform fractional data for statistical analysis to avoid the sharp cut-off at 0 and 1 that does not exist for the other variables we compare to (both proxies and simulated temperature etc.). Log-transformation helps to normalize the data distribution and reduces skewness and heteroscedasticity to make it comparable with non-fractional data.

Line 208: Diatoms; as an option MD99-2322 series with a relatively good resolution: Miettinen, et al., doi:10.1002/2015PA002849.

Reply: We looked at MD99-2322 data, but only the “sea ice concentration” data is available, and not the abundance of diatom indicator species. To avoid the input of more bias sources, we decided not to use quantitative estimates of reconstructed sea-ice concentrations and sea-ice duration.

Line 223: “…group refer to increases in sea-ice cover.” As the authors notice later, for IP25 with its unimodal response to sea ice, the presence of perennial ice cover leads to the opposite tendency.

Reply: Indeed, that is an issue related to the interpretation of IP₂₅ data. However, because most of the sediment cores were not retrieved in regions with perennial ice cover, the unimodal response of IP₂₅ to sea ice was kept only in the discussion.

Line 284: it is not a "binary behavior", rather a manifestation of a unimodal response model quite typical for many climate proxies. See e.g. in Oksman et al., (2019) in Section 2, or elsewhere on quantitative methods of paleoreconstructions.

Reply: The term “binary behaviour” will be replaced by “two end-member scenarios for absent IP₂₅”, as used in Belt (2018).

While it is true that microfossil species typically have unimodal responses to environmental variables (optimum and tolerance), increases in sediment IP₂₅ concentration is generally associated with seasonal sea ice but not quantitatively. Moreover, the absence of IP₂₅ can indicate either absence of sea ice or perennial sea ice, described with "two end-member scenarios" or "binary" (Belt, 2018, 2019). Thus, we do not think unimodal is the correct response model here, but we will change "binary" to "two end-member scenarios for absent IP₂₅".

Belt, S. T.: Source-specific biomarkers as proxies for Arctic and Antarctic sea ice, Org. Geochem., 125, 277–298, https://doi.org/10.1016/j.orggeochem.2018.10.002, 2018.

Belt, S. T.: What do IP25 and related biomarkers really reveal about sea ice change?, Quat. Sci. Rev., 204, 216–219, https://doi.org/10.1016/j.quascirev.2018.11.025, 2019.

Line 287: “… switch from long seasonal sea-ice cover to a multiyear sea-ice scenario”  i.e. transition to a perennial sea ice cover in the area

Reply: The sentence will be fixed as suggested.

Line 304 “…Therefore, it does not have enough temporal coverage to register a potential IP25 decrease. “

Or this could actually be due to a more southerly location of the record. This is a very dynamic area close to the oceanic front, and spatial surface gradients are large.

Reply: Unfortunately, this line of discussion does not work because core C (MSM05/5_723-2) was retrieved from a slightly northern location than core B (MSM05/5_712-1). Figure 3 will be fixed to avoid this misinterpretation.

Line 309 see my earlier comment, this is not a "binary behaviour"

Reply: As mentioned earlier, the term “binary behaviour” will be replaced by “two end-member scenarios for absent IP₂₅”.

Line 324: “One possible explanation for this difference is the impact of warm waters on the melting rate of the drift ice”.  What is actually  implied here? Stronger oceanic fronts in the area? Higher sea ice export to the area for this particular time period?

Reply: We are referring to the presence of stronger oceanic fronts. More specifically, the presence of stronger currents carrying warm waters in the Denmark Strait between 1450 and 1650 CE. Because sediment cores JR51-GC35 (record F) and MD99-2269 (record G) were collected in regions relatively more protected than sediment core MD99-2263 (record H), they were not as influenced by this intrusion of warm waters as the area of the sediment core MD99-2263 (record H). This sentence will be rephrased to improve clarity.

Line 393: “mean chunks” consider changing to “data segments”..

Reply: The term “data segments” would be quite unspecific. “Mean chunks” are meant to highlight that non-overlapping mean segments are used mimicking the samples of proxy records. We will define the expression in more detail when first using it.

Line 394: Refs to figures in Supplementary. Remarkable that CESM1 EOF1 shows very little loading south of Greenland and in the Labrador sea compared to MPI. What could be the reason for this, lack of (sea ice) related variability in the region, or it just went into another, presumably second EOF?

Reply: Since it’s about the area south of Greenland, we assume this question is about the winter season (year maximum area). We checked, and the sea-ice variability was not “transferred”, or better captured by the second EOF (see figure attached – “EOF1_2_CESM_MPI_winter.pdf”). The apparent lack of sea-ice variability on southern Labrador Sea in CESM1 might have been caused by two reasons. One is an artifact of the plots’ scales: EOF1 ranges from -0.3 to +0.3 in CESM1, but from -0.2 to +0.2 in MPI. Another possible explanation is that the sea-ice variability in the Greenland Sea is simply stronger than in the southern Labrador Sea in the CESM1 than in MPI model. As we focus on the large-scale variability patterns, inter-model differences in local variability will have little impact on long-term variations.

Line 411: “close to Greenland east coastline” better change to the "Greenland sea", as "close to east coast" can be interpreted ambiguously

Reply: It will be corrected as suggested.

Line 416: "...which affects...that reaches Greenland sea. " Not clear what the authors meant here. I suggest a full stop after the "the Fram Strait" an leave out the rest of the sentence.

Reply: It will be corrected as suggested.

Line 419: “..PC1 in both models contains some cyclicity…”. Better use the term “quasi periodic variability” for this particular case, not "cyclicity". Same in line 421

Reply: It will be corrected as suggested.

Line 455   “…the variability was mostly concentrated in periods around 30 and 50 years.”

Since some information on wavelet analysis is missing, my thoughts below can be a bit speculative, but good to check it anyways. This "statistically significant" variability emerges only for some shorter periods when the data sampling is high enough to resolve any changes at these time scales. For significance testing, most likely an AR1 model was used based on autocorrelation coefficient estimated from the data directly. Due to data resampling/interpolation the autocorrelation will be overestimated, hence leading to lower CI 95% ranges at shorter timescales. With such background model for the analyzed time series, sporadically emerging variability at shorter time scales will be very much likely identified as significant anyways, yet being an artifact of testing procedure.

I therefore would suggest the authors to consider how meaningful is to make any numerical comparisons for the sub-centennial timescales of variability between the proxies and models since a temporal resolution of the proxy based reconstructions for most of the records is fairly low.

Reply: We agree that sub-centennial variability may not be very reliable in proxy data which we mention two sentences later and again in line 508 below. The question to which extent it is meaningful is part of why we included it in comparison to climate models, albeit only as a side topic with supplementary figures.

As part of explaining the details of the wavelet analysis in more detail in the revised version, we will also add an additional sentence here on the issue. We interpolated all proxy records onto a regular timescale using a linear interpolation before performing the wavelet analysis. So, the wavelet does not directly “know” about data sampling issues but indirectly suffers from the overestimated persistence from the interpolation. Obviously, as you correctly assume, the AR-1 test will also suffer from an overestimated autocorrelation due to the interpolation procedure. This is a common problem with all kinds of proxy data. The deteriorated signal-to-noise-ratio in combination with overestimated persistence would, however, rather underestimate significance and can be assumed to be on the conservative side. As we do not focus on sub-centennial variability in proxies in the main text, we think it is acceptable to keep the significance test as is in the supplementary figures, but we will expand the explanation of the issue in the main text.

Line 508.  “…variability and noise in the proxy data have a marked impact on short-them variability” Definitely in this case the lack of temporal resolution in the proxy series as well as dating uncertainties should be mentioned.

Reply: Indeed, and we will expand the explanation as written above. We will also mention in the conclusion the importance of poor temporal resolution and dating uncertainties on the interpretation of short-them variability which is why we focus mainly on low-frequent variations in this study.