“Impacts of CICE sea ice model and ERA atmosphere on an Antarctic MetROMS ocean model, MetROMS-UHel-v1.0” by Ӓijälä and colleagues evaluates a circumpolar Southern Ocean model including sea ice and ice shelf cavities. Compared to a previously published version of the model, both the atmospheric forcing and sea ice component have been updated. The impacts of each change in isolation are evaluated, as well as both changes together. Evaluation focuses on sea ice properties, with ocean hydrography and ice shelf melting given a secondary focus. In general the updates to the model are beneficial for the representation of sea ice, but problematic for ocean hydrography, particularly a freshening of the continental shelves.

General comments:

In general this is a solid and thorough evaluation paper including some clever analysis of sea ice model performance. I believe it is suitable for GMD after moderate revisions.

In particular, there are a few missing pieces which would help to fill in the oceanographic gaps in the story:

1. Some quantification of ACC strength, ideally a figure showing transport through Drake Passage. The text mentions it slows down, which is concerning, but does not quantify either the initial bias or the degree of drift.
2. A figure of the overturning streamfunction in each model configuration. I suspect that the coastal freshening induced by the update to CICE6 is shutting down dense water export from the shelf. Quantifying the strength of the lower cell would confirm this.
3. Some analysis of the sea ice formation rates in each experiment, as this process is crucial for setting the shelf water masses.
4. A figure (bar chart?) showing ice shelf mass loss integrated over each main ice shelf or region, compared to observations. This would make it much easier to gauge the regional dependence in model bias. From the existing figures it is very difficult to parse the magnitude of the bias or model sensitivity in the small, high-melt cavities in the Amundsen Sea, as well as the large, cold cavities which experience both melting and refreezing.

The manuscript also errs too much on the side of simple description, with some missed opportunities for attribution (i.e. which physical processes are behind the model bias visualised?) Of course it would be excessive to fully analyse the causes of all model biases in one manuscript, but at least some discussion of the likely possibilities is warranted. In my specific comments below I have highlighted examples where this is particularly needed.

Specific comments:

Line 5: It’s worth mentioning more explicitly in the abstract that ice shelf cavities are included, as this is not generally true of Southern Ocean models.

Line 37: You could make the argument that modelling Antarctic sea ice is also challenging because historically, most development and tuning has been focused on Arctic conditions, which are not entirely transferable to Antarctica.

Line 68: What specific changes have been made to the CICE physics parameterisations? If we knew which processes in particular had been updated or added, it could make attribution of the model behaviour easier later. Were any of the bug fixes critical?

Line 71: Naughten et al. 2018, which the baseline MetROMS-Iceshelf simulation seems to follow, used elastic-anisotropic-plastic rheology rather than EVP. What was the rationale for changing this and what was the impact? Are there any other major changes from the original model? I note that the total ice shelf basal mass loss was higher in Naughten et al. than any of the four simulations presented here, and I’m curious as to why.

Figure 1b: The two different blue shadings are hard to distinguish, could you use two more different colours?

Line 128: As with CICE6, the improvements made in ERA5 are summarised as “development in model physics”, which could mean any number of things. Which specific processes have been improved?

Line 134: Note that this salinity restoration is only at the surface. 

Line 139: Why is the salinity restoring necessary? It’s an unfortunate limitation on the model and the sort of experiments for which it is suitable. Given the updates to ERA and CICE, have you tried switching it off? 

Line 139: ABW should be AABW. 

Line 141: Why is there zero zonal velocity at the northern boundary? This seems equivalent to a no-slip condition, i.e. treating 30S as a solid wall. I wonder if this is contributing to slowing down the ACC, as noted later. Or, is it sufficiently clear of the northern front of the ACC?

Line 185: Are the biases in spring (too much ice offshore and too little close to the coast) linked to the strong sea ice drift shown later (i.e. excessive export of ice)? Or are they the beginnings of the summertime low bias, driven by thermodynamic melting?

Line 198: Why might the Pacific and Indian Ocean sectors show the largest underestimation? They are quite a different regime to the other sectors, oceanographically.

Figure 4: It would be interesting to fit linear trends to the observations and models, perhaps piecewise breaking around 2014.

Line 310: Do the CICE6 simulations generally have younger sea ice (there should be an age tracer to analyse)? This would make sense, together with less ridging, thinner coastal ice, stronger export, and possibly stronger sea ice formation (see my general comment).

Figure 7: It’s really hard to see the differences between simulations. Could you plot anomalies of the ice speed without the vectors, perhaps in supplementary?

Line 338: I don’t understand what is meant by “simple ocean boundary”. Do you mean the sea ice-ocean interface? What is simple about it?

Line 358: How appropriate is a comparison to EN4 on the continental shelf (let alone the missing ice shelf cavities)? Does it include enough reliable observations on the shelf?

Figure 9: This figure is uncomfortably similar to Figure 4 of Naughten et al. 2018, down to the placement of the labels and the nonlinear colour bar. Was the same code used? If so, no attribution is given. I am not sure of the journal’s policies on this.

Line 370: The freshening of HSSW in CICE6 is concerning. Hopefully some further tuning of the new sea ice model parameters could alleviate this. Why do you think it occurred? Presumably there’s been a decrease in ice formation and/or a local increase in melting - either way, what could cause this? I don’t understand how this agrees with the other sea ice variables suggesting higher formation (see my above comment on line 310) or deeper mixed layers in the key formation regions (analysed later).

Line 384: The deep waters offshore will basically just be initial conditions so early in the simulation; this should be made more explicit in the note of caution.

Line 398: How is the model surface more saline than EN4 when surface salinity is nudged to observations? Is the nudging really weak, or do the two datasets (WOA and EN4) disagree?

Line 407: A better way to mitigate the freshening problem would be to tune up the CICE6 sea ice formation rates. Salinity restoration should be a last resort, especially on the continental shelves.

Figure 13: Presumably the colour scale is saturated, as observed melt rates in the Amundsen and Bellingshausen sectors are much higher than 4 m/y. The colour scale should indicate this with triangle caps.

Line 464: The wording implies that the biases in the BellAm sector are secondary, but I suspect they are the main driver of the low total mass loss. The cavities are small, so hard to see in Figure 13, but their mass loss is huge. This would be easier to see in the bar chart I suggested.

Line 367: Does the Amundsen Sea have the highest melt increase in percentage terms, or only absolute terms? An extra 2 m/y is not a big change for warm cavities like these, compared to cold cavities with much lower initial melt rates.

Line 471: It looks like there is also a loss of refreezing beneath the Ross and Filchner-Ronne Ice Shelves. This would make sense if fresher HSSW is dampening the magnitude of the ice pump.

Line 473: I disagree that tides and spatial resolution are the main drivers of low simulated melt rates. I strongly suspect the main driver is the cold bias in the Amundsen Sea (Figure A4), and it’s arguable whether or not spatial resolution contributes to this. It’s definitely possible to simulate enough CDW transport onshore in a quarter-degree C-grid model; see for example Mathiot et al. 2017 (doi:10.5194/gmd-10-2849-2017). The Nakayama study about topographic flow used a model equivalent to an Arakawa A-grid, whereas a C-grid as used for ROMS is a lot more forgiving. But even if the topography is adequately resolved, the heat from CDW could be wiped out if there is convection on the Amundsen Sea continental shelf, as sometimes happens with ERA forcing even at much higher resolution (eg Bett et al. 2020, doi:10.1029/2020JC016305). Tides could matter for the Filchner-Ronne Ice Shelf, but this is not a big contributor to the Antarctic total. 

Line 523: It’s a bit of a red flag that ChatGPT was used to write some of the code. Was this just for things like figure layout, or did it actually handle the data analysis? If so, was there sufficient human oversight to make sure it didn’t introduce any bugs? I am not sure of the journal’s policies on this matter.

General comments

In this study, the authors describe an updated version of the MetROMS ocean/sea ice/ice shelf regional Southern Ocean circulation model.  Four simulations were run: a base case with the original setup, a simulation where just the atmospheric forcing is updated from the ERA-Interim reanalysis to ERA5, a simulation where just the sea ice model is updated from CICE5 to CICE6, and a simulation where both the atmosphere and sea ice model are updated.  The hope is that these changes improved the model simulation of Antarctic sea ice and upper ocean hydrography.  Results from the different simulations were compared against each other and observations for sea ice concentration, sea ice area, sea ice edge, sea ice volume (although this is not compared to observations), sea ice drift, ocean temperature and salinity, surface mixed layer depth, and ice shelf basal melt.

I thought the manuscript was clear and easy to understand and generally met its objective of explaining how the upgrades impact the model simulation of different aspects of the Southern Ocean.  This setup (ROMS ocean model and CICE6 sea ice model) is a good tool and I’m glad the authors have updated it and made it available to everyone and are using GMD to tell the community what they have done.  I felt the specific quantities being compared did a good job of showing the general impact of the upgrades for simulating the physics of the Southern Ocean (although I do have specific suggested additions below).

However, there is not very much on what specific changes in either the forcing or the ice model code led to these differences in the model simulation (there is a little in lines 235-240 about biases in cloud cover and temperatures impacting summer sea ice).  I certainly understand how a thorough examination of the causes is well beyond the scope of what the authors are trying to do here.  However, are there any large changes between the forcing or ice code versions that are likely to lead to any of the changes seen in the results?  For example, does the significant increase in the resolution from ERAI to ERA5 (thus better resolving steep coastal orography) lead to generally stronger winds along the coast, thus impacting ice motion near the coast (e.g. lines 340-341) and perhaps ice production and mixed layer depths over the continental shelf?

I also felt that there could have been more to tie the presented individual changes to each other.  For example, for the basal melt there is a discussion of what may cause the systemic underestimation for all the models compared to observations, but there is no discussion of what differences in the modeled hydrography between the simulations would lead to the shown differences in basal melt between the different simulations.  Changing the ice model generally leads to greater ice shelf basal melt (especially over the Amundsen, Fig. 13d), which could be related to the temperatures at depth over the continental shelf generally being warmer (Fig. 10e), and the temperature difference between C6EI and C5EI might even be greater over the deep Amundsen shelf (hard to tell as there is no plan view of the temperature difference) as the winter mixed layer depths are significantly shallower there with the ice model change (Fig. 12f).  Updating the winds leads to generally lower ice shelf basal melt, which can be related to lower temperatures at depth (Fig. 10d) over the continental shelf.

I have some other specific comments and suggestions below, but most of these are minor and should be easily dealt with by the authors.  I think this upgrade is important, but I also think this description of the impacts needs a bit of work before it will be fully useful for the intended audience.

Specific comments

Lines 39-41:  Is it worth adding something here that while satellite based estimates of thickness are now becoming available in the Arctic, these efforts are not nearly as well advanced in the Antarctic since it is harder to estimate freeboard?  I know this is briefly mentioned at the beginning of section 4.1.3, but since this is another example of a difference between southern and northern sea ice, should it be mentioned here?

Lines 70-71:  I think it would be helpful to any potential users to explicitly list somewhere (maybe a supplemental table?) the instances in which CICE default values are not being used.

Lines 86-87:  Is the freeze-melt potential from ROMS integrated over the top 5 m of the ocean model for both versions of CICE as in Naughten et al. 2017?  If so, since there could be multiple ROMS layers in the top 5 m, is supercooling in the layers below the surface layer also handled as described in Naughten et al. 2017?

Line 123:  I really feel the ACC transport for the different experiments should be given.

Section 3.2:  I know it is mentioned in section 4.3, but I think it should be explicitly mentioned somewhere in this section that there is no tidal forcing.

Lines 150-152:  Is there also a sponge region near the northern boundary with increased viscosity and diffusivity as in Naughten et al. 2017?

Line 162:  Suggest changing “increases the concentration” to “generally increases the concentration” since there are areas where the ice concentration decreases with the updates.

Lines 221-222 and Figure 4:  I think it would be helpful to the reader to include a numerical comparison (correlation or skill score maybe?) between each of the different modeled sea ice area time series and the observations.

Lines 307-316:  One other metric that I think would help in determining why there are increases in concentration but decreases in volume near the coast when updating the ice code is the thermodynamic sea ice production, which could be compared to observations (e.g. Nihashi and Ohshima, 2015; Nakata et al., 2019) and help determine if volume decreases near the coast are due more to decreased production or changes in ice movement.

Figure 7 caption:  Is the “climatological sea ice concentration” the climatology during winter (JAS) or over the entire year?

Line 346:  “Robust” seems like a strong word for the level of increases shown in Fig. 8 (e.g. from 0.457 to 0.466 in the Indian Ocean sector or 0.483 to 0.493 in the BellAm sector) between C5EI and C5E5.

Figure 9:  I believe the ROMS model output is not absolute salinity, so my guess is that the authors converted the output to absolute salinity.  Apologies if I missed it, but should that be explicitly mentioned somewhere?

Lines 405-407:  This is more reason why I think it would be helpful to show the sea ice production from the different runs.

Lines 428-429:  It is hard to tell from the figures here that there is “a small improvement of the summer MLD”.  Can the authors make a table of overall RMSE between the MLD of each of the different runs and Sallée (or add a column to Figure 11 with C6E5 – Sallée)?

Line 441:  Suggest adding “excess” between “possible” and “deep water formation” since there should be deep water formation in these areas even if the MLDs there were perfect.

Line 457:  The 578 Gt/yr for C5EI is a fair bit lower than the 642 Gt/yr in Naughten et al. 2018b.  Do the authors have ideas on what is causing the difference other than the different (1996-2018 vs. 2002-2016) time periods?

Line 461:  Suggest adding “model” between “The” and “refreezing” to differentiate the melting in this sentence from the preceding one.

Lines 472-485:  This discussion on why all the models have too little melting is interesting, but most of this is already discussed in Naughten et al., 2018b (their sections 4.3 and 4.3.6), which should be referenced here.  Also, while it’s certainly important to mention this large difference, since this paper is mostly about the upgrades, shouldn’t there be something on why there is a difference in the melting with the atmospheric forcing and ice model updates?

Technical corrections

Lines 6-7:  I don’t think the sentence beginning “Both CICE sea ice models…” is necessary.

Line 8:  “increase” should be “increases”.

Line 10:  Should “ocean mix layer” be “ocean mixed layer”?

Line 96:  Should the Filchner-Ronne, Amery, and Ross ice shelves be labelled somewhere (Fig 1b)?

Line 208:  “being C6E5” should be “with C6E5 being”.

Figure 6 caption:  The units should be “([model volume]/[C5EI volume])”.

Line 312:  Typo, “A3)a-d” should be “A3a-d”.

Line 358:  Should the “;” be “ and”?

Line 370:  “them clearly fresher” should be “them are clearly fresher” or “them clearly are fresher”.

Line 475:  The “.” After “basal melting” is not necessary.