Arctic sea ice simulation in the PlioMIP ensemble

Howell, Fergus W.; Haywood, Alan M.; Otto-Bliesner, Bette L.; Bragg, Fran; Chan, Wing-Le; Chandler, Mark A.; Contoux, Camille; Kamae, Youichi; Abe-Ouchi, Ayako; Rosenbloom, Nan A.; Stepanek, Christian; Zhang, Zhongshi

doi:https://doi.org/10.5194/cp-12-749-2016

Articles | Volume 12, issue 3

https://doi.org/10.5194/cp-12-749-2016

© Author(s) 2016. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/cp-12-749-2016

© Author(s) 2016. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 12, issue 3

Research article

|

23 Mar 2016

Research article |

| 23 Mar 2016

Arctic sea ice simulation in the PlioMIP ensemble

Fergus W. Howell, Alan M. Haywood, Bette L. Otto-Bliesner, Fran Bragg, Wing-Le Chan, Mark A. Chandler, Camille Contoux, Youichi Kamae, Ayako Abe-Ouchi, Nan A. Rosenbloom, Christian Stepanek, and Zhongshi Zhang

Download

Final revised paper (published on 23 Mar 2016)
Supplement to the final revised paper
Preprint (discussion started on 07 Apr 2015)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC C308: 'Review', Anonymous Referee #1, 06 May 2015
RC C360: 'Review of article cp-2015-29', Anonymous Referee #2, 12 May 2015
AC C937: 'Response to reviewer 1 for paper: Arctic sea ice in the PlioMIP ensemble: is model performance for modern climates a reliable guide to performance for the past or the future?', Fergus Howell, 03 Jul 2015
AC C944: 'Response to reviewer 2 for paper: Arctic sea ice in the PlioMIP ensemble: is model performance for modern climates a reliable guide to performance for the past or the future?', Fergus Howell, 03 Jul 2015

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

ED: Reconsider after major revisions (06 Jul 2015) by Uwe Mikolajewicz

AR by Fergus Howell on behalf of the Authors (17 Aug 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (19 Aug 2015) by Uwe Mikolajewicz

RR by Anonymous Referee #2 (14 Sep 2015)

Suggestions for revision or reasons for rejection

The article “Assessment of simulations of Arctic sea ice in the PioMIP models” by F.W. Howell et al. analyzes sea ice representation in pre-industrial and mid-Pliocene climates in an ensemble of 8 different global coupled models. The inter-model spread is discussed and possible relations between pre-industrial and Pliocene simulations are investigated.
Since data are very sparse for the Pliocene and basically no robust information about sea ice distribution in the Pliocene is available, we do not gain much information about model performance by simultaneously comparing pre-industrial and mid-Pliocene model simulations. There are several studies existing, which either evaluate a larger number of global models for present day climate (better suited for model evaluation than pre-industrial) or relate behavior of ice properties in present day and future studies in a much more comprehensive way than this study does. Thus, I appreciate that this new version of the article focus more on sea ice conditions under pre-industrial and mid-Pliocene conditions and less on a potential ranking of model’s ability to simulate sea ice.
Unfortunately, large parts of the manuscript are very descriptive and mentioning trivialities; it is not necessary to describe in detail the ice thickness pattern of every single participating model. Instead more analysis should focus on the causes for the model spread and the causes for different pre-industrial – Pliocene differences. Why is the seasonal amplitude of ice extent different? How does the atmospheric circulation look like in the different models both in pre-industrial and pliocene? How does this affect the sea ice thickness distribution in the different models? How does the atmospheric and oceanic heat transport into the Arctic look like and how might it affect sea ice representation?
The argumentation that the representation of NAO or AMO could not be investigated at all due to too short time series is not entirely convincing. I agree 30 years are short but NAO and AMO are strongly affecting sea ice and if we can not say anything about NAO and AMO, then it might also be difficult to make robust conclusions on the sea ice conditions themselves. Here, the question arises why only 30 years from the 500-1000 year simulations are used for the analyses? To make results more robust, a longer period should be used.

I am still not convinced by using CV as measure for the variability even it has been used by Stroeve et al. 2014 – your CV is also not exactly what Stroeve et al. (2014) used and they used it in a somewhat different context. CV assumes that sea ice concentration and thickness variability should linearly increase with ice concentration and thickness, respectively. To my knowledge, no one ever showed this.
In any case, CV can not be a reliable measure for ice variations if ice thickness/ concentration are near 0 as it is in four of your Pliocene-summer simulations. Thus, the conclusions that especially summer ice variability increases in Pliocene compared to the pre-industrial simulations should not be taken based only on CV-values. I would even call this statement wrong.
CV could be used if a) the “normal” standard deviation/ variance of the models is shown as comparison and b) evidence is shown that we can assume a linear growth of ice concentration and ice thickness variability with increasing ice concentration and thickness, respectively (e.g. cite an article that shows this).

Specific comments:
1. Line 6: I would say that the statement that the model spread is 3 times larger in summer than in the rest of the year is just wrong: Figure 13 shows that ice varies between 0.7 and 2.7m thickness in winter (quite evenly distributed) and between 0 and 2 m in summer (5 out of 8 models between 0 and 0.3 m, the other three about 1, 1.7 and 2.2 m). See discussion above.
2. Line 40 and following, line 306-310: I would suggest making a single section discussing model-setup and the experiments a bit more in detail. What is the difference to the CMIP5-model versions? How have sea ice and SST been prescribed in Pliocene-Experiment 1 (AGCM-simulations) if observations are so uncertain?
3. Line 54: additional to what?
4. Line 54: “reduces”
5. Lines 59-61: Are you sure all models provide ice thickness as grid-box mean and not as mean over the ice-covered part only (which is the case in most CMIP5 models)?
6. Line 83: Figure 1: Extend the plotted area to the south – if this figure does not show the position of the ice edges, it is useless. It does not matter that the other figures show a different area, you could indicate this in the figure caption.
7. Lines 92-98: You should also indicate the observed amplitude for comparison.
8. Line 106: I do not see any ice thickness anomaly at 0E, 80-90N, to be clear it is maybe better to say 180E, 80-90N instead of “Greenwich meridian”.
9. Line 110: mention that ice thickness is likely overestimated at the Siberian coast in MIROC, COSMOS.
10. Lines 111-113: This is not needed. However, if you want to mention that sea ice is thinner in Nordic Seas compared to Siberian coast you should also shortly mention why this is the case.
11. Lines 120-143: This discussion is too detailed and difficult to read. The section should be shortened and the most important points mentioned. The reader can find details in the figure.
12. Line 164: In contrast to this, in the discussion it seems to be stated that CCSM shows a larger ice extent in the mid-Pliocene – please check.
13. Line 187: delete one “amplitude”
14. Line 205: “Many” sounds a bit strange with a total of 8 models whereas 4 do not show any ice in summer.
15. Figure 7: As figure 1: please show a larger area in the plot.
16. Line 218/219: difficult to judge from the Figures 7 and 8, there are very large differences across models in winter as well and Figure 9 shows about the same max-min difference in winter compared to summer. As mentioned earlier I do not think CV should be used and comparing Figures 9 and 12 clearly indicates why. Just from looking at Figure 9: do you really want to suggest that model spread is 3 times larger in summer than in winter?
17. Line 284-296: Please make clearer if this is a summary from Shu et al. 2015. Please specify which time period Shu et al. (2015) analyzed. I would suggest adding these numbers, at least annual mean ice extent to table 2.
18. Lines 299-302: I do not understand this sentence: Please clarify.
19. Lines 303-305: This sentence is not clear.
20. Lines 326-327: Of course they are not the same: maybe better: “…vary strongly: The summer sea ice …”
21. Line 324-325: maybe better: “almost ice free” or “ice free in late summer”
22. Line 326-327: sounds like HadCM3 simulates summer sea ice in the entire Arctic Basin. Ok, not really wrong but actually ice concentration is very low along all ice edges.
23. Line 405: Please specify what you mean with “CMIP5”. Here and elsewhere CMIP5 is compared to pre-industrial simulations: Pre-industrial simulations are also part of CMIP5 as historical and future and many more simulations are. It seems you mean a certain time-period with CMIP5 (historical, satellite period …)?
24. Line 435-440: SST and SAT are not necessarily drivers of sea ice variations but could also be driven by ice variations: One reason for better correlation in Pliocene could be that larger parts of the ocean north of 60N are ice-free for longer periods in the year and could thus warm up much stronger than in the pre-industrial period. The longer, the ice-free period, the more the ocean can warm. In the pre-industrial period instead, summer SST and SAT in the Arctic will almost be very near melting temperature of ice, it does not matter, if sea ice concentration is 100% or 50% in a certain gridbox or smaller region: as long as some ice is left, the ocean can hardly warm up.
25. Lines 453ff: The albedo discussion would fit better into section 4.3.1 “Influence of sea ice models”
26. Lines 521-526: According to the introduction, it is debated if sea ice was seasonal or perennial? Yes, HadCM3 agrees with the findings of perennial Arctic sea ice but the other models agree with findings of Cronin et al., Moran et al and Poyak et al.. Is there any particular reason to believe more in the perennial assumption? Furthermore, HadCM3 shows a very unrealistic sea ice concentration distribution in both pre-industrial and mid-Pliocene summer, thus even if HadCM3, probably by chance, keeps the points of Knees (2014) at 80N ice-free year around, we can be quite sure that Arctic sea ice distribution will not look like HadCM3. All the years with very low observed ice concentrations (e.g. 2007, 2012) still showed the thickest ice with highest concentration north of Greenland and the Canadian Archipelago.

Referee Report: PDF

Hide

RR by Anonymous Referee #1 (17 Sep 2015)

Suggestions for revision or reasons for rejection

The new version of this paper is in my opinion still not suitable for publication and still requires some major work to become publishable. This is not, as before, too much related to issues related to inconsistent model-data comparison, but instead to too superficial an analysis of the results presented here. It does not become clear in as how much this paper really provides new insights regarding the evolution of sea ice in a warmer climate relative to existing studies based on, say, CMIP simulations. This paper currently remains very descriptive, rather than providing the reader with any robust results. This fact, however, is not stated in the paper, but instead is hidden behind a language that is very speculative throughout much of the text. I still believe that these simulations can provide new, broadly relevant and interesting insights regarding the evolution of sea ice, but more in-depth analysis is required to extract those from the available data.

Alternatively, this paper should be shortened significantly and only provide a description of model results. This would then be a helpful reference for these simulations, and it would be made clear to the reader that this is not meant to provide an "assessment" or the like. In that case, the title should be changed to "Description of simulations of Arctic sea ice in the PlioMIP models"

If the authors decided to keep the current scope of this paper I suggest that they consider the following remarks for a possible revised version:

l.5 and section 3.3: I am still not convinced that much can be learned from using CV in the current context. What is the geophysical relevance of CV that makes this measure preferable over simply using ensemble spread? If in a warmer climate all simulations are ice free, but one simulation still has a tiny ice floe of 2 m² lying around somewhere, then CV will be more than 10. But this high value would be totally irrelevant, as is expressed by the geophysically more relevant ensemble spread given by ensemble standard deviation. I disagree in particular with the statement tat standard deviation does not allow one to compare data sets with different mean values (l.64). Why not?
If the authors decide to keep the analysis of CV, it'd be helpful to give geophysical reasons for its relevance - rather than simply stating that others have used this metric before. Please also note that "ensemble spread" is very different from "variability", but currently these terms are used as if they were to describe the same thing.

l.11: "suggesting that the dominant atmospheric and oceanic influences may be different in the [two] simulations": This is one example of the speculative language. All data is there to test this suggestion, so why not do it? In particular since I doubt that this is true.

l.24: The Arctic is only "widely predicted to become seasonally ice free before the end of the 21st century" for a specific evolution of CO2

Introduction in general: This should include some short discussion of what we do know from previous studies on sea-ice ensemble spread, correlations between individual sea-ice metrics and drivers, temporal correlation of sea-ice evolution, generally evolution of sea ice in a warmer climate, etc., which is necessary to allow the reader to identify the open questions that are addressed by the present study.

l.56 leading to Figure 14: I was wondering if some of the results of this study are simply related to the fact that sea-ice extent is used to describe the areal coverage of sea ice, rather than sea-ice area. If in a cold climate sea-ice concentration reduces because of some warming from, say, 90 % to 45 %, sea-ice extent would remain the same, even though the area decreases by 50 %. This then renders the correlation of extent and temperature very weak. Sea-ice extent is only a useful metric when comparing data to observations, since it allows one to account for some observational uncertainty. In the present context, where most of the analysis is only carried out in the model realm, sea-ice area would give much more robust results, in particular given the very low sea-ice concentration that is obtained in the warm climate runs.

Section 3: I found this section unnecessarily long. The reader can simply look at the figures, and doesn't need a detailed description of every single panel. In particular since much of the language remains very vague, repetitive and sometimes contradictive, such as "Most of the models display patterns that are broadly similar to ensemble mean - but there is appreciable variation with respect to the location of maximum ice thickness". Either the patterns are broadly similar (which includes their key characteristics), or they are not (as given by the location of maximum thickness). Or: "The thickest ice in COSMOS [...] is located in approximately the same region as in the ensemble mean." followed by "In COSMOS, the thickest ice is concentrated into a smaller area." I found this entire section very cumbersome to read.

l.143: What is "relatively" reduced ice?

l.144ff: Why should multi-year pre-industrial ice-thickness patterns match two months of observational record from 2009?

l.152: Another example for very vague language: "The ensemble mean thickness patterns appear to broadly match the observations."

l.187: I did not understand the logic (and meaning) behind: "The finding that sea-ice extent amplitude in the mid-pliocene is 64 % greater than the pre-industrial simulation amplitude holds for the ensemble mean at a lower amplitude extent amplitude."

l.210: Another example for vague and somewhat contradictive language: "A similar finding to the fact that MIROC has similar patterns in winter in both simulations holds for COSMOS, where the central Arctic sea ice thins by a greater amount in comparison to sea ice in other regions."

section 4.1: There is no assessment of pre-industrial simulations in this section, hence the title is misleading. Instead, this section primarily summarizes results from other studies on the historcial simulations from CMIP5.

l.289: Another example for very vague language: "The fact that historical extent simulated by MRI is almost 25 % greater than observations may suggest that its Arctic sea-ice cover is too extensive."

l.292: Why is it a contradiction that a model has a sea-ice extent closest to observations "although" it has the lowest sea-ice extent amplitude?

section 4.2: Again, this section does not really give an assessment of mid-Pliocene simulations, but instead comes to the conclusion that such assessment is not possible.

l.324: Unnecessary repetition, I find.

l.335: This is not very clearly spelled out: Why may a reasonable performance of a model relative to mid-Pliocene sea ice improve confidence into this model, while a the same time a match to present-day observations does not necessarily mean that the model is good?

l.359: Why does HadCM3 only appear to be in closest agreement with proxy-data indications? Either it is, or it isn't.

Section 4.3: Many of these results are known from earlier studies. This should be spelled out here, to allow the reader to see what really is new here.

l.365: Why is it a contradiction that CCSM and NorESM use the same sea-ice component "although" NorESM has a coarser atmosphere and a different ocean?

l.406: In section 4.1, there is no analysis of pre-industrial or mid-Pliocene performance, which would require some comparison against data to actually assess performance.

section 4.3.3: Much of this section remains unnecessarily vague. All data to support or reject the suggestions is in the data that the authors have available, so I find that the analysis should move beyond quoting existing studies by Hill et al., Zhang et al., etc.

l.424ff: I did not fully understand what is meant by "stronger correlation": A higher slope of the linear fit, or less spread around the fit?

l.473: I found this confusing: Models with lower sea-ice albedo have less ice-albedo feedback. Why would they have greater potential to amplify warming from greenhouse gas emissivity?

l.505: What is a "relatively consistent level of variabililty"?

l.518: Again, the data is there to examine this, rather than having to say that "If models see an enhanced ice-albedo feedback, than this is likely to affect those models predictions of future Arctic sea-ice change".

l.521: Why does the fact that HadCM3 produces the thinnest pre-industrial sea ice imply that this model generally has difficulty in simulating observed sea-ice thickness?

l.530: see l.359

Hide

ED: Reconsider after major revisions (23 Sep 2015) by Uwe Mikolajewicz

AR by Fergus Howell on behalf of the Authors (09 Dec 2015) Author's response Manuscript

ED: Referee Nomination & Report Request started (22 Dec 2015) by Uwe Mikolajewicz

RR by Anonymous Referee #2 (18 Jan 2016)

Suggestions for revision or reasons for rejection

The authors made a large effort and substantially improved the manuscript.
The main problem of this study is still, that as long as it is unclear how ice conditions in the mid-Pliocene really were no new insights on model performance and reliability, compared to existing studies, can be presented. However, despite this problem, I find that the study now provides a sufficiently large number of interesting results to be publishable after responding to a number of minor comments.

1. Line 7-8: Why is this in contrast to proxy-data? If I understood right (see also introduction lines 27ff), it depends on the proxy data if we assume ice-free summer in Pliocene or not. In this way, the model uncertainties are in line with the uncertainties from the different proxies.
2. Line 24-25: I guess you mean RCP8.5?
3. Lines 34-36: I totally agree, these are key questions but unfortunately this article is not really investigating these questions.
4. Line 59: add “in” before Haywood et al. (2011)
5. Line 113: “in Figure 4” instead of “by Figure 4”
6. Figure 4: What do you show? Annual means, FMA, ASO? Please clarify in the figure caption. Do you calculate the RMSE only for north of 80N? Please clarify in the figure caption.
7. Figure 4/lines 112ff: It seems that figure 4 is showing the RMSE from one specific model with respect to another specific model (respect to the ensemble mean). Please clarify this in the figure caption. To avoid misunderstanding you might want to call it RMSD (root mean square deviation) and not RMSE because it is not really an error you are calculating but the deviation or difference between two models (if I understood correctly).
8. Line 132: Here a line or two on a possible relation between ice extent in pre-industrial and in Pliocene would be nice. Table 2 shows that those 3 models with highest ice extent in the pre-industrial period show also the highest extent in the mid-Pliocene period. However, the reductions of ice extent between mid-Pliocene and pre-industrial periods seem to be independent of the ice extent in the pre-industrial period. The ice extent in MIROC4m and MRI-CGCM is almost the same in the PI-period but MIROC4m simulates a twice as large reduction in the mid-Pliocene period.
9. Lines 138/139: This is only true for MIROC4m and NorESM. The other 6 models show almost the same annual cycle in pre-industrial and Pliocene climates.
10. Line 145: I do not really like this “overlap”. Maybe better: “This indicates the large spread/ uncertainty in the representation of sea ice extent in the models.”
11. Lines 138-139 and Lines 150-156: These discussions should be merged. And as stated earlier, only 2 models show a substantially reduced annual cycle, the others more or less the same. These two models are thus also responsible for the increase by 20% in the ensemble mean.
Why are these two models so outstanding in this respect?
12. Line 173: “overall no more or less similar” sounds awkward. Maybe better: “but the differences between thickness patterns are comparable”
13. Line 183: “have” instead of “had”
14. Line 191: delete “on the other hand”
15. Line 193: you could add that SD is much smaller in mid-Pliocene in summer because little ice is left in almost all models.
16. Line 194: delete one “of”
17. Lines 261-264: These sentences are not clear, please reformulate.
What is indeed interesting (and maybe this is meant by lines 261-264) is that CCSM4 and MRI-CGCM show a higher annual ice extent in mid-Pliocene than in the recent past – CMIP5 (1979-2005) simulations. Further, HadCM3 simulates a larger ice extent in the 1979-2005 period than in the pre-industrial period. Here, it might be useful to look at the ice area as well and not only at the ice extent.
18. Another interesting addition to the comparison with CMIP5 would be to compare the mid-Pliocene ice conditions to the ice conditions in a future period with similar greenhouse-gas concentrations in CMIP5-simulations.
19. Line 307ff: I absolutely agree. But given the fact that proxies obviously allow for a variety of conclusions on the sea ice conditions in mid-Pliocene, I wonder if it is possible to realistically reconstruct mid-Pliocene sea ice? This touches also the main problem of this study. As long as it is unclear how ice conditions in the mid-Pliocene really were, we do not get a lot of new insights on model performance or reliability of models.
20. Lines 368ff: Are you sure NorESM and CCSM really use exactly the same values for their albedo? Even if NorESM and CCSM use the same ice model, this does not necessarily mean that the albedo is exactly the same. Often, the albedo is used for model tuning.
21. Figure 12: I appreciate that you show the wind patterns and it is fine to only show selected models with quite some differences in the wind patterns (and the other models in the supplement) but I wonder why you chose different models for Pliocene and preindustrial. It would be more straightforward to show the same models to understand differences between the atmospheric circulations and their possible influence on sea ice distribution in Pliocene and pre-industrial.

Hide

RR by Anonymous Referee #1 (29 Jan 2016)

Suggestions for revision or reasons for rejection

This version of the paper is much improved over the previous versions in that it now has almost become a scientifically sound description of the PlioMIP ensemble. I'm hence happy to have this paper published subject to the authors addressing the following remaining issues:

1. More details are needed regarding the calculation of sea-ice thickness. It remains unclear whether the actual (floe) thickness is analyzed, or the so-called equivalent thickness that gives mean sea-ice thickness including the contribution of the open water part of the grid cell. Since the authors describe an impact of sea-ice concentration on thickness in the caption of figure 7, I assume that the equivalent thickness is used. However, this thickness is not related to observations of ice thickness (which give actual floe thickness), and is not related to any of the existing studies examining the ice-thickness patterns in CMIP simulations. Hence, either the authors change their analysis to an analysis of floe thickness (by dividing equivalent thickness by concentration), or the reference to existing studies becomes impossible to interpret since these are all based on an analysis of floe thickness. This important point was made already in my first review of this paper, I believe.

2. On a related note, how is spatial mean ice thickness calculated? How is open water considered in the averaging?

3. Why does RMSE of ice thickness not have any unit throughout the entire paper?

4. Around line 135: A greater relative change of summer sea ice does not imply an enhanced annual cycle. Only a greater absolute change does.

5. There is still quite a bit of unnecessarily vague/repetitive language throughout the paper, e.g. l. 151: "The ensemble mean extent amplitude of the mid-Pliocene simulations is by 20 % greater than the pre-industrial ensemble mean amplitude, further indication of the enhanced seasonal sea ice extent cycle in the mid-Pliocene simulations." I simply don't get why an increased amplitude is only an indication of an increased amplitude.
e.g., l.210 "Sea ice volume appears to show a stronger relationship with SAT, with a correlation coefficient of -0.76 versus -0.18" Why is this only apparently a stronger relationship?
e.g., l. 289: CCSM4 has a change of 4.1 % in the extent amplitude, so I don't understand why this model only "appears" to not have a substantial change in the annual cycle
Please also check the logical flow within parts of the paper. For example, why does the fact that MRI-CGCM historical simulations have a larger extent than observed show that that model "consistently simulates Arctic sea ice extent larger than the ensemble mean". To me, this only shows that it has a larger extent than observed.
Sorry for maybe being somewhat too frustrated about these kind of flaws, but it seems that these kind of things should have been caught through a careful reading of this paper by the co-authors rather than by the reviewers.

6. l.295: Why does the fact that pre-industrial ice is very thick in summer in MRI imply that there is a greater reduction in the maximum extent than in the minimum extent? I would expect the ice to then be even thicker during winter, hindering a loss of extent also then.

7. l.305: Why is there disagreement "particularly in summer"? From Figure 6 it seems that absolute model spread is even larger in winter, which is surprising given the geometry of the Arctic Ocean with its rather narrow passages at the edge of winter sea ice.

8. l.470: Please give more details as to which lengths of simulations are required for such analysis? From table 1 it seems that some of the models that were not analyzed have even longer runs than the three models listed here.

9. Figure 4:a "jet" colorbar is very hard to read for color-blind readers. In addition, for the lower panel a color scale that diverges from 0 seems useful, or, indeed, a color scale that has no negative values since these do not seem to exist in your analysis.

10. Related to Figure 4: Details should be given on how RMSE and spatial correlation were calculated.

11. Much of sections 4.1 and 4.2 up to line 305 do not read like a discussion, but instead like a continuation of the results section. It might be a good idea to indeed incorporate those subsections into section 3.

Hide

ED: Publish subject to minor revisions (review by Editor) (11 Feb 2016) by Uwe Mikolajewicz

AR by Fergus Howell on behalf of the Authors (26 Feb 2016) Author's response Manuscript

ED: Publish subject to minor revisions (review by Editor) (08 Mar 2016) by Uwe Mikolajewicz

AR by Fergus Howell on behalf of the Authors (09 Mar 2016) Author's response Manuscript

ED: Publish as is (10 Mar 2016) by Uwe Mikolajewicz

AR by Fergus Howell on behalf of the Authors (10 Mar 2016) Manuscript

Short summary

Simulations of pre-industrial and mid-Pliocene Arctic sea ice by eight GCMs are analysed. Ensemble variability in sea ice extent is greater in the mid-Pliocene summer, when half of the models simulate sea-ice-free conditions. Weaker correlations are seen between sea ice extent and temperatures in the pre-industrial era compared to the mid-Pliocene. The need for more comprehensive sea ice proxy data is highlighted, in order to better compare model performances.