General comments

In this study, the authors describe a new operational sea ice forecasting system for the Southern Ocean using a regional MITgcm ocean/sea ice/ice shelf general circulation model along with an ensemble based localized error Kalman Filter data assimilation system that assimilates sea ice concentration on a daily basis. Results from forecasts ranging from 24 to 168 hours are compared against different observational products to show the model performance in terms of RMSE of sea ice concentration, integrated ice-edge error, mean absolute error of ice thickness, and mean absolute error of sea ice drift.  There was also a comparison of model sea ice convergence forecasts versus changes in MODIS imagery for one particular event (a sea ice opening in November 2021 that would be relevant for navigation to a particular coastal station).

I thought the manuscript was mostly (see below) clear and easy to understand.  A regional sea ice forecast system for the Southern Ocean would certainly be useful, not only for the scientific/resupply missions for different nations, but also for the many private operations (i.e. fishing and tourism) that are becoming more numerous in Antarctic waters.  MITgcm is a great tool for the ocean and ice shelf modeling.  While the embedded sea ice model may not be the most "up to date", I think it is fine for these purposes, especially in the Antarctic where I do not think the lack of different ice thickness categories is such an issue where there is not much multi-year ice.  I am not an expert on data assimilation and cannot comment on the appropriateness of the method used here (hopefully there will be a reviewer who can).  All in all, this seems like a solid numerical setup for a sea ice forecast system (although I do have some questions below).

My two concerns for a manuscript that is describing a forecast system is that I think more needs to be added to the model description and that it is hard to tell how well this setup is performing compared to either a simple forward model with no data assimilation or other existing global sea ice forecast models.

1)  I think there are important aspects of the forward model that are relevant to dynamically moving ice around that are not described.  Is there any parameterization of landfast ice processes?  I know one exists in MITgcm, but that is meant more for the Arctic and it is not mentioned here, or in the Zhao et al. 2023 paper describing the forward model, whether it is used (and/or modified).  Are icebergs (especially grounded ones that can limit ice transport) represented?  Is tidal forcing included?  All these processes would be relevant to ice motion and divergence.

2)  The manuscript has several descriptions of the performance of the data assimilative forecasts, but it is difficult to tell how much the data assimilation adds to the forecast skill.  Was there a control run (like in Liang et al., 2019, JGR, which has some of the same authors) without data assimilation over the same dates as the forecast runs?  If so, what does that look like compared to observations?  What do the ice concentration and integrated ice-edge errors look like over time with no forward modeling and just the initial analysis (i.e. persistence of the initial sea ice state throughout the forecast period)?  I do not expect the authors to go through all the different existing global sea ice forecasting systems, but I think it might be helpful to a reader if some information were given on how this system compares to others.  For example, the 168-hour sea ice concentration RMSE for this model (Figure 3) looks considerably better for most months than the GIOPS for the Southern hemisphere (Figure 3b in Smith et al., 2016).

There are also several minor awkward usages or subject/verb agreement mistakes.  I am not going to explicitly comment on most of them, and they generally do not make the manuscript more difficult to understand (I still feel the manuscript is pretty well organized and understandable), but I think they should be cleared up in the next version.

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 SOIPS may be a very good forecast system for sea ice, but this manuscript still needs some work before it can help an interested reader judge that for themselves.

Specific comments

Line 26:  By "with thin first-year ice dominating the majority" do the authors mean that the majority of the ice is thin first-year ice?

Line 28:  Since katabatic winds can blow sea ice away from the coast, as well as away from the front of ice shelves, suggest changing "off the ice-shelf" to "off the ice-shelf and coast".

Lines 61-63:  I agree with the authors that regional models "with higher resolution" still offer significant advantages, but isn't the resolution of this model (line 93: ~ 18 km) lower than the resolution at these latitudes of most of the global models (1/4 degree or better) listed in this paragraph?

Line 92:  I think it is also worth mentioning that the open boundaries are farther north than any likely northern extent of the sea ice.

Line 95:  Large and Pond (1981) is just the bulk formula for momentum flux (I think).  Is there a bulk formulation used for heat and salt/freshwater fluxes?

Line 99:  Suggest adding the Losch 2008 reference that describes the implementation of ice shelves in MITgcm.

Line 107 (and line 376):  I do not think Zwally, 1990 is the best reference for the initial ice thickness data and the URL given on line 376 is the ICESat 500m DEM, not the sea ice thickness.  Is this the Kurtz and Markus (JGR, 2012) data?

Line 128:  Were any experiments done with more or less ensemble members?

Lines 129-132:  I assume the ensemble is generated in the method described in the PDAF wiki (https://pdaf.awi.de/trac/wiki/EnsembleGeneration), but there is no reference, and very few details, on how it is generated.

Lines 174-175 and Figure 3:  I agree that the bias between AMSR2 and OSISAF partially explains the ice concentration forecasting errors, but the shorter term forecasts (24 and 72 hours) look to generally be better than the assimilated AMSR2 data.  Do the authors have any explanation for this?  What would the forecast errors be with no data assimilation?

Figure 4:  I think it would be helpful to have a figure like this (monthly RMSE for ice concentration) for the AMSR2 vs. OSISAF for comparison.  I can certainly understand the authors not wanting to add any figures to the primary manuscript, but perhaps in a supplementary material section?

Lines 218-221:  As mentioned above, what about other non-simulated (I think, it is never explicitly stated one way or another) barriers to sea ice free drift such as fast ice or grounded icebergs?

Figure 6:  The sea ice edge forecasts look really good at 24-hours.  If the authors do create a supplementary material section, I would be curious what the ice edge looks like at 168-hour lead time.

Figure 8:  As for figure 6, what does the ice thickness look like at 168-hour lead time?

Lines 279-280 and Figure 9b:  I agree that it is mostly true that "the MAEs of both magnitude and direction of the sea ice drift forecasts do not exhibit significant amplification", but the MAE of the drift angle does increase significantly at 168 hours (compared to shorter lead times) in Oct-Nov and Jun-Sep.

Lines 288-289:  The mean absolute errors and mean magnitude of the NSIDC drift velocities are given earlier, but I did not see anything to indicate whether the mean model bias with respect to NSIDC was positive or negative until here.  Apologies if I missed it, but is the mean difference (not mean absolute error) or mean drift velocity from the model given anywhere?

Line 298:  Since landfast ice also floats, I suggest changing "Floating sea ice" to "Drifting sea ice" or "Moving sea ice".

Line 299:  Same comment as above about "floating sea ice zone".

Figure 10:  What is the lead time for the forecasts on the left?  Also, Figures c and d are really impressive!

Lines 337-338:  Same point as for Lines 218-221 above.

Lines 342-343:  Do the authors have any thoughts on if the forecasted convergence rates near the coast would be improved if the model included fastice processes?

Lines 358-363:  I still think more needs to be done to show if this model can do a better (or at least similar) job compared to those other forecast systems.

Lines 380-381:  The zenodo link to SOIPS (https://doi.org/10.5281/zenodo.10457661) did not work.

Technical corrections

Again, this is not a complete list and there are many minor grammatical errors that should be cleaned up in the next version.

Abstract lines 18, 19, and 20 and many other places:  Suggest changing "leading time" to "lead time".

Line 36:  "Grahams Land" should be "Graham Land".

Line 83:  Suggest changing "promise capacity" to "promise" or "capacity".

Line 124:  Should "5-order" be "5th-order"?

Line 195:  Suggest changing "just a number of sea ice extent" to "just a sea ice extent number".

Line 277:  Suggest changing "In contrary" to "In contrast".

In this study, Zhao et al. present an operational Southern Ocean Ice Prediction System and exhibit its ability for Antarctic sea ice prediction on synoptic time scales. They developed the prediction system based on MITgcm and assimilated satellite-derived sea ice concentration data, making predictions for the future 7 days. The prediction system shows promising skill in predicting the sea ice concentration, sea ice thickness, sea ice drift, and sea ice convergence.

Considering the limited effort for the operational Antarctic sea ice prediction when compared to its Arctic counterpart, this study is valuable by providing evidence of the model’s ability for skillful Southern Ocean and sea ice prediction. In addition, the manuscript is well-organized and easy to understand. However, I found some points to be further clarified, which are listed below. I suggest a major revision is needed.

Major comment:

1. Despite the main point of this work being to demonstrate the ability of the prediction system for the operational Antarctic sea ice prediction, the added scientific discussions will improve the manuscript a lot. The following are a few examples, but not limited to these.

（1）Why is the RMSE of prediction in Fig. 3 smaller than the RMSE of observation February and March? Why does the RMSE of prediction peak in April?

（2）L180-190: it’s interesting to know how many errors can be explained by the difference between OSISAF and AMSR2 and how many are caused by error growth during the model integration.

（3）What model deficiency in Fig. 5 leads to an increase in predicted IIEE in March-April and a decrease in April-May? Why is there little difference in IIEE for different lead times in January-June, but significant differences in other months?

（4）In Fig. 9, why the evolution of forecast errors in magnitude of sea ice drift is different from that in direction? Additionally, due to the complexity of the South Pacific Ocean current system, it is recommended to showcase the drift forecast capability in more ways, such as its spatial distribution.

2. Compared to other studies, an important feature of this research is the incorporation of ice-shelf model. Thus,

(1) Please provide more details on the ice-shelf model and coupling method. According to Line 98-99, it’s hard to realize the differences between the ice-shelf model used here and boundary conditions used in previous studies.

(2) More analyses should be conducted to highlight the advantages of this feature. For example, the Larsen-B ice shelf collapsed in January 2022 (doi: 10.5194/tc-2023-88), which occurred during the experimental period, so it is advisable to investigate the impact of this event on sea ice assimilation and prediction.

Minor comment:

Line 66-71: Because the preceding paragraph mentioned the advantages of regional models, it might be better to illustrate data assimilation studies based on regional models, such as SOSE.

Line 82: Considering the submission is in 2024 and an operational forecasting system is involved, the experiment should be extended to include 2023 when the Antarctic sea ice reaches its minimum extent.

Line 93: Considering that one important application of this system is for shipping services, the higher model resolution would indeed be preferable. Therefore, why not use a higher-resolution model such as MITgcm with 1/6° (doi: 10.1002/2016jc012650)?

Line 130-132: Please provide more details on the initial field perturbation process, such as which variables are perturbed? What is the explained variance of the first 11 EOF modes?

Line 135-136: Please provide more information about the observational errors used in the assimilation. For example, is 0.15 the representative error of observations? If so, how are instrument errors identified?

Line 138-140: The author's previous study used JRA55 as the atmospheric forcing, while this study uses GFS. Given the importance of atmospheric forcing for Antarctic sea ice simulation, did the author optimize the model parameters after changing the atmospheric forcing, as in doi: 10.1016/j.ocemod.2023.102183? If optimization has been conducted, are there significant changes in the model parameters? If not, could some of the subsequent results be attributed to the mismatch between the atmospheric forcing and the model, such as Line 213-214?

Line 155: is it OSI-401-d?

Line 163-165: I would argue that the RMSE increases to the end of March, followed by a decrease starting from April.

Line 208-209: It’s hard to follow and please rewrite this sentence.

Line 219-221: It’s very interesting and It would be more valuable if the author could present the correction method and the corrected IIEE.

Line 252-253: It's recommended to add the uncertainty of ICESat-2 to Fig. 8. From Fig. 7, the uncertainty appears to be around 0.5m, while in Fig. 8, the prediction error in the southern Weddell Sea and the western Ross Sea seem to reach up to 0.6m. Are these errors beyond the uncertainties of the observation? Why are the prediction errors of SIT larger in these areas?

Line 295: Please provide the specific definition of Sea ice convergence rate. What are the similarities and differences between the sea ice convergence rate and the divergence of sea ice drift?

There are quite a few typos. For instance, an extra hyphen of  “synoptic-scale” in Line 332 and an extra left parenthesis in Line 359.