Increased horizontal resolution provides limited benefits for simulated Arctic and Antarctic sea ice climatology and recent extremes in a NEMO4-SI3 hierarchy

Richaud, Benjamin; Topál, Dániel; Massonnet, François; Fichefet, Thierry; Goosse, Hugues; Barthélemy, Antoine

doi:10.5194/egusphere-2026-3552

Preprints

https://doi.org/10.5194/egusphere-2026-3552

Preprints

24 Jul 2026

| 24 Jul 2026

Increased horizontal resolution provides limited benefits for simulated Arctic and Antarctic sea ice climatology and recent extremes in a NEMO4-SI3 hierarchy

Benjamin Richaud, Dániel Topál, François Massonnet, Thierry Fichefet, Hugues Goosse, and Antoine Barthélemy

Abstract. We assess the sensitivity of Arctic and Antarctic sea ice simulations to horizontal resolution using an ocean-sea ice NEMO4-SI3 hierarchy at 1°, 1/4° and 1/12° nominal resolutions, but with identical vertical resolution, forced by the ERA5 atmospheric reanalysis. Increased resolution improves selected regional and seasonal features, particularly where ocean circulation strongly controls the sea ice edge. In the Arctic, improvements are modest and mainly occur in the Greenland Sea, where the higher-resolution configurations (1/4° and 1/12°) better represent Atlantic inflow west of Svalbard and reduce the excessive winter ice extent simulated at 1°. In the Antarctic, the response is more widespread. Higher horizontal resolution improves the winter ice edge downstream of major bathymetric features, while the 1/4° configuration also improves austral summer sea ice extent and thickness in the Weddell and Ross Seas. Additionally, the 1/12° configuration enhances vertical mixing with warm subsurface waters, promoting thinner ice and a more poleward ice margin. Overall, across both poles, resolution does not systematically improve sea ice extremes or hemispheric biases and benefits depend strongly on the diagnostic considered.

Received: 17 Jun 2026 – Discussion started: 24 Jul 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Benjamin Richaud, Dániel Topál, François Massonnet, Thierry Fichefet, Hugues Goosse, and Antoine Barthélemy

Status: final response (author comments only)

RC1: 'Comment on egusphere-2026-3552', Anonymous Referee #1, 29 Sep 2026

This manuscript evaluates polar climate in forced ocean-sea ice simulations, isolating the role of horizontal resolution. The authors recommend the 1/4° configuration as the best cost–benefit compromise. In the Arctic, the impact of higher resolution is largest in the Greenland sea, with improved Atlantic inflow at both 1/4° and 1/12°. The Antarctic results are more mixed, although the winter sea ice extent appears to be improved at 1/12°.
This study is a welcome addition to the literature, particularly as high-resolution approaches become more popular. Inclusion of the computational expense (Sec 2.4) helps weigh up the costs of the high-resolution simulations. I found the manuscript interesting and generally well-written, and think it is a good fit for GMD once the following issues are addressed. I recommend major revisions because of one major comment below.

Major comment
This seems to be a fairly clean experiment design, where the only differences in the simulations besides the ocean/sea ice horizontal resolution are some changes to ocean parameters and a shorter spin-up for the 1/12° simulation. However, it appears from the text that a further difference between the resolutions is that cavities are not included at 1°, but are for higher resolution. Do the cavities have an influence on the simulations? How do the meltwater amounts compare, and does the 1° simulation without cavities prescribe meltwater? Please clarify exactly the differences among the simulations. If meltwater is treated differently among the simulations, please demonstrate that this does not impact the results, or discuss the implications for attributing differences to resolution alone.

Minor comments
Title – On the basis of the analysis in the manuscript, which shows some improvements (and the striking differences in ocean currents) with increased resolution, ‘limited’ doesn’t seem quite appropriate. ‘Modest’ might suit better.
Introduction – a very recent study to include is Bushuk et al. (2026, https://doi.org/10.1029/2026GL124043)
L108: In addition to the major comment, are the cavities included for both Greenland and Antarctica? Are they static? Please specify.
L118: Why use ERA5 and not JRA55 as per the OMIP protocol? Do you have a sense of how the choice of reanalysis affects the results? Is the difference between forcing datasets possibly greater than the difference between the resolutions? Please comment.
Sec 2.5 – Why not also include OSI SAF for sea ice, if both OSI SAF and NSIDC sea ice drift are included? What is the resolution of PIOMAS itself compared to the NEMO configurations? Please comment in the text.
L122 – I think it’s worth emphasizing elsewhere (Introduction/Conclusions) that the reanalysis is 1/4°, so the 1/4 and 1/12° simulations see the same forcing.
L185 – Please comment on the uncertainty in the EN4 product.
L265: ‘known characteristics of atmosphere-forced NEMO-SI3’ – is this also true for JRA55 forcing? Please comment.
Figure 1: Are the differences among resolutions larger for sea ice area rather than sea ice extent? Please check and comment.
L260: This feature (thinner sea ice north of Chukchi) is quite hard to see in the figure. Is it significant?
Figure 2 (and similar figures): These panels are quite small. The figure could be made larger and span the full text width. I find red=more sea ice and blue=less sea ice somewhat counter-intuitive - consider inverting the colorbar.
Figure 6: Why combine Arctic and Antarctic here? I would suggest separating them so the Antarctic one appears where it is discussed.
Sec. 3.1.4, Fig. 8, Sec. 3.3, Fig. 16 – The authors find ‘no clear differences in either melt or freeze onset exist across the three configurations’ in the Arctic or Antarctic, so I would suggest removing these subsections and the corresponding two figures. This will help with readability. A sentence or two could be added elsewhere to state that no differences were found.
Fig. 10: The winter SIE looks to be substantially improved in the 1/12° - please comment on this.
L404: This shift in peak is a bit unexpected. To consider shifts in seasonality, it is important to look at daily values. Daily sea ice extent is shown in the anomaly plots, so it is available. Switching both Arctic and Antarctic seasonal cycle analysis to daily SIE might affect the interpretation here.
L442: The sea ice thickness pattern in March in the Weddell Sea in the 1° simulation is also quite different to the other resolutions.
Fig. 14: I find this figure difficult to read. Why not separate temperature and salinity in different panels?
Fig. 17: The 1° simulation also performs quite well for the total extent anomaly for May-Jul 2023. Please comment. Does the comparison look different for sea ice area?
L598: I would expect to see more discussion of how the findings might translate to coupled configurations. Would coupled interactions lead to a greater or lesser impact of resolution? Please comment in the Discussion.
L600: Zapponini et al. (2026) find ‘consistent and substantial improvements’ relating to resolution, rather than ‘modest.’ Please discuss why your interpretations differ, or align the sentence towards their findings.
Data Availability: The two PSC links are missing a hypen in sea ice. E.g. the first one should be ‘https://psc.apl.uw.edu/research/projects/arctic-sea-ice-volume-anomaly/’. As per the GMD code and data policy, the namelists should also be made publicly available.

Typographical comments
L6 and elsewhere: I would suggest ‘varied’ rather than ‘widespread’
L10: add a comma after ‘biases’
L15: perhaps ‘hypothesis’ rather than ‘promise’
L18: perhaps ‘conform to this assumption’
L22: perhaps ‘premature’ rather than ‘too hasteful’
L25: perhaps ‘climate-scale’ rather than ‘regular-scale’?
L96: Would be helpful to add the resolutions in km also for the Southern Ocean, say at 50S.
L98: suggest ‘allowing proper resolution of the ..’
L208: Remove ‘time-step’
L256: Please break this long sentence into two, and correct the ‘-,’
L264: Rephrase ‘where they span less than 1’ to ‘where they span close to 1’
L284: add a comma after ‘resolutions’
L290: replace ‘well reflected in’ with ‘consistent with’
L312: add a comma after ‘configuration’
L322: consider reminding the reader here what the ‘feature’ is (thinner sea ice)
L369: perhaps ‘preceding’ or ‘previous’ rather than ‘foregoing’
L376: million km2 should have a superscript ‘2’
L377: ‘still leads to’ replace with something less causal, like ‘is associated with’
L486: The discussion of ‘fast Antarctic sea ice’ is a little confusing because ‘fast ice’ is an ice type. The Arctic section mostly used ‘higher speeds’ – suggest doing the same here.
L553: This sentence is repeated, and both paragraphs should be checked for repetition.
L595: Perhaps avoid repetition of ‘progress’ in the paragraph openings (also at L587)

Citation: https://doi.org/10.5194/egusphere-2026-3552-RC1
RC2: 'Comment on egusphere-2026-3552', Anonymous Referee #2, 04 Oct 2026

Main comments and recommendations

This manuscript aims at a systematic analysis of the impact of spatial resolution on the simulation of sea ice essential climate variables, namely sea ice extent, thickness (and volume), and drift. This is presented in the framework of the most used configurations of the NEMO-SI3 model, which makes it of relevance to the readers of GMD due to the permanent special issue that represents a reference for developers and users. There is a substantial merit in this manuscript, and I am convinced that it should be published to become a reference guide for the improvement of sea ice models that are lagging with respect to other components of earth system models. However, I would recommend to improve on a few main concerns expressed in the points below and in the specific comments.

1.     In its present form, the manuscript presents an extensive but often speculative and in parts disorganized analysis of the impacts. As a developer, I would struggle to find clear avenues of development for improving the model based on the presented findings. As a climate scientist, I would probably conclude that low-resolution sea ice simulations are sufficient to capture the climatic role of sea ice. The message and the role of the paper should be made evident earlier, and the discussion and conclusion need more focus and possibly boldness. The presentation is dry and agnostic, as if the authors expected any resolution to perform equally. I understand that this is due to their findings, but also, there appears to be a limited improvement relative to what is already found in the literature when moving from coarse-resolution CMIP5 to HiResMIP in CMIP6. Selivanova et al. (2024a,b) came to very similar conclusions, although they only looked at coarse and intermediate resolutions from HiResMIP. There is a fundamental problem in sea-ice model development that does not match other improvements in ocean modelling: the hypothesis that high resolution should perform better does not hold. The authors have a great opportunity here given the extensive set of simulations they performed, and I encourage them to take a stronger stance. They make very clear hypotheses at the beginning, in the form of bullet points. These points at page 3 lines ~80 are not followed up directly in the discussion and conclusion section. The presentation of the results would benefit from streamlining and closer alignment to these hypotheses instead of following a scattered narrative like “the 1/12 is better here, the ¼ is better there”. For instance, it is immediately clear that the ORCA25 resolution plays a pivotal role and separates from the lower and higher resolutions. Hence, relative comparisons using the ¼ degree model as reference in Figure 6 would be more useful. Instead that figure assumes that there is monotonic response to increased resolution (ORCA1-ORCA025, ORCA025-ORCA12), which makes the interpretation more complicated. This statement is only done very late in the manuscript (L505).

2.     The authors have provided an extensive report-like presentation, which is in line with the GMD objectives. The introduction is excellent and contains all the relevant details, but descriptive fatigue becomes evident along the length of the paper, while going through the symmetry of the Arctic and Antarctic sections. Repetitions and attempts to offer similar interpretations in both hemispheres leads to redundancy and lack of focus. This affects the reader's attention, despite the importance of this work. I see the value of this type of exposition in a PhD thesis and in a technical report, but less in a scientific paper that should offer more depth and recommendations on the model development. The authors indicate that most of the biases that persist have been studied in literature, therefore, there is limited value in describing the origin of these biases, if not when there is an expectation that resolution would be the main cause or that resolution works in favour of reducing biases. Finally, the authors state that SIT and volume are only used as references in assessing the relative role of each simulation; however, I have found a few instances in which GIOMAS is used as if it was a reliable reference for the mean state.

3.     From a more scientific perspective, the authors do not seem to acknowledge that the physical scales may be different in both hemispheres (although this is made explicit later when presenting and discussing the effects of improved mesoscale circulation, but not in the introduction). Why do we expect high resolution to influence the mean ice mass balance given the dominating role of seasonal thermodynamic processes? Nonlinear terms are not resolved in sea-ice models therefore the only nonlinearities are most likely to operate through the stress terms, and, maybe, eddy advection. Is it really expected that these nonlinearities would propagate into the sea-ice mean state? Higher spatial resolution is essential to resolve topographic features in ocean circulation, which play a role in most of the Arctic due to the geographic land constraints, but only along the Antarctic Circumpolar Current and along the eastern Antarctic rim current in the Southern Hemisphere. From the results of the eddy-resolving simulations, it appear clear that a mesoscale resolving ocean does not enhance corresponding nonlinear responses, which reminds me of similar issues emerging in the field of ocean biogeochemistry. It is therefore not surprising that the higher resolution enhances vertical mixing in the Southern Ocean which ultimately warms the upper ocean because sea ice is melted, in turn leading to a bias in the upper ocean heat content itself. This is a typical case of unresolved feedback loop, because sea-ice resiliency to melting should probably contrast this warming leading to a dynamical equilibrium; there is a missing process in sea ice that is not just dependent on the resolution.

4.     Another main point is related to the lack of display of variances and of a quantitative estimation of model biases or improvements thereof. The inclusion of measures of dispersion in the time series of the climatology by adding shadings (e.g. Fig. 1) would strengthen the comparisons. The mean state is relevant, but it may be achieved with different uncertainties by different models and products. The statement at Page 27 L542 indicates that much of the benefit is achieved in the transition between 1 degree to ¼ degree. This statement would be much stronger, especially for publication in GMD, if backed up by quantification during the presentation of the results, instead of basic face-value comparisons. Several metrics have been proposed (Lin et al introduced a few of them, which may be also biased by the CMIP6 focus, but I believe the field of biogeochemical modelling presented the more thorough interdisciplinary effort: Stow, C.A., Jolliff, J., McGillicuddy Jr., D.J., Doney, S.C., Allen, J.I., Friedrichs, M.A.M., Rose, K.A., Wallhead, P., 2009. Skill assessment for coupled biological/physical models of marine systems. Journal of Marine Systems 76, 4–15.). For instance, the improvement in the modelling efficiency (Nash and Sutcliffe, 1970) could be weighted by a factor that considers computational costs. This would help to corroborate quantitatively the conclusions.

5.     I would like the authors to reconsider the role of the analysis of extreme events in the economy of the manuscript. As it stands, it is the weakest part of the manuscript because the selection of the extremes is not demonstrated. The chosen extremes are known in the observational records, but are they also extremes in a statistical sense in all three simulations? Anomalies are extremely dependent on the mean state, and interannual variability may not be captured as expected. Given the emphasis on the mean state as the target diagnostic, I would find more convincing a comparison of how the different resolutions capture the distribution of the seasonal extremes over the 2000-2024 period.

Specific comments
L215 Please give a quick summary of how the sea-ice mass budget is computed regionally (in Fig. 6). This diagnostic is as important as the advective heat flux in eq. (1). Are the authors using the average transport over a given region or the integral? In the latter case, circumpolar transport should be zero, although I guess only regional fluxes are sued in the manuscript. Also, it is not completely clear how to interpret the results in Fig. 6 since transport can be both negative and positive, and therefore the difference is unconstrained. More explanations are needed to clarify whether, for instance, a greater basal melt difference between configurations mean that the ice is melting less. Finally, it should be clarified if the units of the fluxes are per cell area
L265-270 As expressed in the points 1 and 2 above, there is less value in investigating biases. It would be more relevant to demonstrate whether changes in resolution alter them.
Sec. 3.1.1 and 3.1.2 (and 3.2.1/3.2.2) Having two separate sections for winter and summer tends to make the text and the description rather dispersive. The seasonal cycle is not just summer or winter and I suggest joining the sections describing Fig. 1 and Fig. 10.
L293-294 “modest”. This qualifier needs to be quantified. I would use the interannual dispersion of the models first, and of the products eventually (using the geometric mean of the standard deviations, so that the total variance is the sum of the variances).
L295-297 I think this message can be streamlined and simplified. All models overestimate the most extended MIZ region in the Arctic (actually, the only one because the Bering Sea is limited). The bias is reduced with increasing resolution.
L300 Are the changes in ocean velocity and drift related?
L301 The NAC is broad and unresolved at 1 degree, and the topographic circulation around Iceland and Svalbard is missing. However, the transport is the same, which indicates that the climate role of this region is included, and the coarse resolution could still be valued in climate simulations.
L318-319 Please explain what you mean exactly with “do not dominate”.
L322-323 It is unclear what “the second resolution-dependent winter Arctic sea ice feature” is.
L326-328 Can this be determined from a difference? We only see that the transport difference is seasonally much larger than the other terms.
L331-332 Please explain if this is related to the increased export in spring and summer.
L334-335 These are not really comparable, since the magnitudes are much smaller and there is no rebounding back in summer.
L359 Fig. 5. This is likely a wrong cross-reference.
L375 See main comment 4. This statement should be quantified.
L376-379 Please consider main point 5. The role of the initial mean state would be enhanced by an analysis of anomalies over the whole 2000-2024 period.
L308-382. Regionally, the results are difficult to interpret, as the Arctic SIE anomaly appears to be matched by compensation.
L395-399 Please check main comments 1 and 2. It is important to remark clearly that there is no coherent relation with the increase in resolution.
L404-406 The peak is still in the same month, but it is flatter. Also, SIV from GIOMAS should only be used as a reference for intercomparison.
L406 The reference to Barthelemy et al is unclear in this context.
L407-409 Please explain whether this is positive or negative. It is difficult for the reader to understand what is being referred to.
L412 The Amundsen Sea should have geographic coordinates ~130W
L410-419 This paragraph is quite difficult to follow. The emphasis seems to be only on the edge, which is indeed shaped by topography.
L420-422 This sentence is rather speculative. How all of this affects SIC is unclear.
L429-436 Please check the main comment 3. The role of the enhancement of the oceanic heat flux with the 1/12 deg resolution seems to be the main cause, but it is unconstrained and the model overestimates surface ocean warming with respect to the climatology. The high-res model does not seem to necessarily improve the overall representation of the upper ocean because there is an evident overestimation.
L438-441This part needs strengthening and less speculation. How much of the poleward edge is due to this surface warming in the 1/12° configuration? The meandering and the currents seem to lead to a mismatch in the ocean density structure with respect to the climatology.
L442 “systematically thicker” This statement requires a more refined qualification. Where? Figure 12 does not show a major feature. The Weddell, Ross and Bellingshausen regions (also at 1°) are more visible in the additional figures.
Figure 14 is quite busy and not much informative due to the many lines. Unless salinity plays a specific role in one of the regions, I would suggest to use density or to separate salinity and temperature.
L447-449 This reference supports what, exactly?
L457-458 This does not seem evident from the figure. Is the new ice formation smaller than the negative transport?
L474-475 “broader” The authors are being quite generous with the models and they should be more quantitative. The products also have major uncertainties. I suggest focusing on relative differences.
L478-479 The model shows little seasonality, and it happens that the data get closer to it.
L486-487 It is not known whether the drift is stronger in these regions. The statement is not supported by the presented data.
L494 Maybe the reference should be to Fig. 16. Please check.
L508 Unclear what is meant with atmospheric boundary forcing. This is resolution dependent. This comment is better understood in the discussion, but at this stage of the manuscript, it is unclear.
L513 A reduction is observed, which makes the reference to a “contribution” misleading.
L520-521 This is not surprising, since anomalies are linearly related to the mean state. Summer SIE is badly captured by all configurations, while winter is better. This is not an indication that resolution improves the capture of extremes, unless this is done systematically.
Section 4. The discussion appears to have been written hastily and deserves more attention. More specifically, it needs more structure: either go by region or by variable, as there is currently a mix of presentations.
L540-542 Please be more specific about the ocean dynamic control.
L547-548 The reference to model improvement is rather generic. Please be more specific.
L550-555 Please check main comment 3. How much is gained relative to the effort? We surely want to run the ocean at high resolution, but are we missing some feedback from sea ice that is unresponsive? The sentence “it can also expose or amplify biases in vertical mixing, stratification and ocean heat delivery to the ice” seems to imply that without being explicit. There are important questions to which the authors may contribute with their understanding of resolution-dependent responses.
L556-558 Repeated sentence.
L558-561 It is not clear what the real improvement is in the ice or in the ocean. This statement appears quite generic and I recommend strengthening.
L562 Thicker Than observed? The comparison needs to be clarified.
%66-560 What is the role of spatial resolution? The issue of configuration is unclear in this sentence.
L571-573 Improvements on drift are not evidently shown in the results.
L574-576 The study of extremes has been linked to the mean state, but this finding is not discussed in this section.
L578-579 I do not seem to have found any specific discussion on incomplete physical processes. Please be more specific.
L583 and L587 Please give an example of ocean or sea ice parameters that may require tuning, or of complex physical polar processes that should be represented.
L590-594 Is this also the case in the Antarctic?
L595-600 The question is how much of the apparent improvement in polar ocean circulation would be modulated or affected by a more responsive sea ice. Also, how sensitive is the need for coupling to the resolution? It appears that coarse resolution may still provide the proper heat transport and overall SIE.

Citation: https://doi.org/10.5194/egusphere-2026-3552-RC2

Benjamin Richaud, Dániel Topál, François Massonnet, Thierry Fichefet, Hugues Goosse, and Antoine Barthélemy

Data sets

Global Sea Ice and Ocean Fields Simulated with the Model NEMO 4.2.2 - SI3 in Three Configurations of Increasing Horizontal Resolution A. Barthélemy et al. https://doi.org/10.14428/DVN/JSDC8Q

Interactive computing environment

Bnjmnr/eORCA_comparison: NEMO-eORCA-comparison-paper, v1.2.0 D. Topál and B. Richaud https://doi.org/10.5281/zenodo.21508270

Benjamin Richaud, Dániel Topál, François Massonnet, Thierry Fichefet, Hugues Goosse, and Antoine Barthélemy

Viewed

Total article views: 391 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
237	116	38	391	33	36

HTML: 237
PDF: 116
XML: 38
Total: 391
BibTeX: 33
EndNote: 36

Views and downloads (calculated since 24 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	87	28	8	123
Aug 2026	29	26	11	66
Sep 2026	64	33	7	104
Oct 2026	57	29	12	98

Cumulative views and downloads (calculated since 24 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	87	28	8	123
Aug 2026	29	26	11	66
Sep 2026	64	33	7	104
Oct 2026	57	29	12	98

Viewed (geographical distribution)

Total article views: 354 (including HTML, PDF, and XML) Thereof 354 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 17 Oct 2026

Short summary

How does increased horizontal resolution impact the skills of an ice-ocean numerical model to simulate sea ice in the Arctic and Antarctic? We address this question using the NEMO-SI3 model at three different resolutions. While some regional improvements are noticeable in winter in both Arctic and Antarctic, particularly where the ocean is known to influence sea ice, the overall added benefits of using a higher resolution are not systematic.


Total:	0
HTML:	0
PDF:	0
XML:	0