Review of "An Evaluation of Antarctic Sea Ice Thickness from the Global Ice-Ocean Modeling and Assimilation System based on In-situ and Satellite Observations" by S. Liao et al.

The authors compare Antarctic sea-ice thickness (SIT) in the global sea-ice reanalysis GIOMAS with observational estimates from satellite radar altimetry and various in-situ observations. They find that anomalies and trend of Antarctic sea-ice volume (SIV) match quite well between model and observations. However, compared to satellite observations, the model simulates much lower mean SIV and SIT. Variability in the model also seems to be lower than observed, at least in some regions. Some hypothesis on the reasons for the discrepancies between model and observations are stated but not tested.

The study addresses an important gap in our knowledge of the Earth system: the thickness of Antarctic sea ice. It makes good use of novel satellite observations and the available in-situ observations and documents several intriguing facts such as the link between variability and mean state. It thus shows good potential. However, the manuscript is lacking maturity and substance. It stays purely descriptive and misses the chance to better understand discrepancies, which would really help the community to make progress with developing better observational and modelling products of Antarctic sea-ice thickness. There are also some methodological shortcomings when it comes to comparing model data with in-situ observations. I recommend publication of the manuscript after a major revision that addresses my concerns stated below.

Major comments:

1) p 5 to p6 l 165 and elsewhere: I appreciate the basic description of discrepancies between observed and modelled SIV and SIT, but more analysis on the reasons for the discrepancies is needed in order for the paper to make a substantial contribution to the field. As it stands, only hypotheses and speculations are offered (e.g. p 4 l 131f "These differences may lead to overestimation...", p 4 l 142f "... which may be due to model bias ...", p 4 l 149f "... may be partially attributed to the uncertainties of SIT retrieved...").

I would appreciate if the authors could provide at least some attempts to disentangle the respective roles of observation errors, model errors and insufficient data assimilation methods. To this end, I would suggest to

1a) discuss sea-ice thickness in a model run /without/ assimilation of sea-ice concentration. A complementary approach would be to discuss the assimilation increments (here: nudging tendencies) for sea-ice concentration. Assuming that GIOMAS still uses the original PIOMAS nudging methods (Lindsay and Zhang, 2006), there will also be implicit increments on SIT and SIV. It would be good to also discuss the implicit increments on SIT and SIV that arise from SIC the nudging, as their impact on the mean state might be substantial.

1b) better quantify observational uncertainty. From the figures presented, it is evident that there are major discrepancies between ES and CS2-derived SIT. The same is likely true when comparing the various in-situ observations with the satellite records - can we have some scatter plots of these please? To clarify, I am not asking for an intercomparison of observational products - this would clearly be outside the scope of what the authors wanted to present. However, it is necessary to present the model-observation discrepancies in the context of the observational uncertainties, so there needs to be some quantitative discussion of these.

1c) use at least some of the in-situ observations (the ones who are judged most reliable) as ground truth and verify both satellite observations and GIOMAS against these. This would allow to draw some much-needed conclusions on whether it is the observational data sets or the model (or both) that need to be improved.

2) The discussion of trends (p 6 ll 166 - 175) is very interesting but disappointingly rudimentary. I consider Figure 5 a success for GIOMAS and a central result of the manuscript: it convincingly shows that year-to-year variability as well as trends in GIOMAS and the satellite record correspond quite well. This is an important message, but it needs to be backed up with more analysis (see minor comments). I would also suggest to dedicate a separate subsection to the trends.

3) Comparing monthly-mean model fields to in-situ observations at a specific point in space and time might introduce substantial errors, because you are not comparing "like with like". Is it true that the authors use monthly-mean fields from GIOMAS? If daily-mean fields are available, these really ought to be used to compare to in-situ data. The mean and variability of sparsely sampled instantaneous in-situ observations might not meaningfully represent the monthly mean and variability over a large-scale region like a model grid cell. Please specify and discuss how exactly you match in-situ observations to model fields and give some justification why you think the direct comparison is not misleading (see e.g. Janjic (2018) for an introduction/overview of the problem).

Minor comments:

- The manuscript would benefit from language editing, as there are various places with small grammatical errors and slightly inappropriate word choices. Nothing serious though. To give two examples: p 1 l 9f: "not very clear" is too colloquial and imprecise. p 2 l 34: "limited by the short of..." needs to be replaced by "limited by the lack of..."

- p 3 l 81: nudging of SIC is not state of the art anymore, it can introduce grave problems with SIT (see Tietsche et al. 2012). Please add some discussion on implicit changes to SIT when nudging SIC in your system.

- p 4 l 115 "grid of observations": in-situ data are not gridded. Please rephrase, e.g. "... take place in observation space, which means that GIOMAS was converted to the locations of the observations"

- p 4 l 119: You claim the difference ES - CS2 is "much less" than the variability, but 574 is almost half of 967! Can you please rephrase to acknowledge that there is considerable differences in CS2-ES SIV? I would like to see a figure with the time series of SIV anomalies as well.

- p 4 l 120: Please be more precise on your method to "splice together" the two data sets

- p 5 ll 137-150: The problem with the manuscript that I describe above in major comment (1) is most evident in this paragraph. There is lots of speculation of where discrepancies may come from, but no further insight offered at all. Look at these phrases: "... may be due to model bias...", "... maybe caused by smaller freeboard of CS2 than ES2...", "... still disputed whether radar altimeter signals originate from the snow/ice or snow/air interface...". Without at least an attempt to decide whether discrepancies are due to deficiencies in GIOMAS or the satellite observations, it is difficult to draw any conclusion on where improvements are needed. Maybe it makes sense to treat (at least some of) the in-situ observations as the ground truth, and verify both model and satellite observations against them?

- p 6 l 166ff: Can you please show a figure of Antarctic SIE, to clarify whether SIV anomalies and trends are mostly explained by SIE anomalies and trends, or there is some independence? A scatter plot with SIE/SIV anomalies would also be helpful.

- p 6 l 171 and Figure 5: The trend lines after 2013 are quite different. Can you please give the numbers for the trends in GIOMAS and the satellite record, and add to the discussion that GIOMAS seems to underestimate the trend seen in the satellite record?

- p 6 l 176f: The variability of SIT anomalies in satellite observations quite probably have a contribution from measurement noise, which the model should not try to imitate/simulate. Can you offer some comment or quantification on that point?

- p 6 l 180 ("... which is consistent with Fig. 4b."): This looks like a mistake. Figure 4b shows difference in mean not difference in variability. Please clarify and correct.

- p 6 l 181: I agree there appears to be a relationship to some extent, but the authors might be over-simplifying this relationship. Look at ULS 232: GIOMAS underestimates variability by a factor of 3, but has no bias. To clarify this, can we please have a scatter plot of standard deviation ratio GIOMAS/satellite versus mean bias?

- end of section 3: I am not clear what the reader should take away from the frequency comparison. Can you please add a paragraph with the conclusions?

References:

Janjić, T., Bormann, N., Bocquet, M., Carton, J. A., Cohn, S. E., Dance, S. L., Losa, S. N., Nichols, N. K., Potthast, R., Waller, J. A., & Weston, P. (2018). On the representation error in data assimilation. Quarterly Journal of the Royal Meteorological Society, 144 (713), 1257–1278. https://doi.org/10.1002/QJ.3130

Tietsche, S., Notz, D., Jungclaus, J. H., and Marotzke, J. (2012). Assimilation of sea-ice concentration in a global climate model - physical and statistical aspects. Ocean Sci. Discuss., 9 (4), 2403–2455. https://doi.org/10.5194/osd-9-2403-2012

The authors compare the sea-ice thickness climatologies from the Global Ice-Ocean Modeling and Assimilation System (GIOMAS) with satellite observations from the ESA CCI as well as other in-situ observations (ULS, Airborne measurements).

Measurements and analyses of Antarctic sea ice remain a tricky task as on the one hand  in-situ measurements are rare or very localized and satellite measurements on the other hand still feature a high uncertainty, especially due to the Antarctic sea-ice snow cover. A key outcome of the study is the good agreement between GIOMAS and the observed change in sea-ice volume anomalies before and after Summer 2013. However, compared to the satellite-based estimates, GIOMAS yields overall lower sea-ice thickness estimates.

The authors state several potential reasons for all discrepancies between GIOMAS and the observational data, however, these mainly remain suggestions. The study in general poses a good review on GIOMAS, satellite and in-situ observations and their respective limitations and some of their causes. However, the study stays in most cases at the surface of things and some methodological parts remain unclear – at least to this reviewer.

Therefore, I would recommend publication after some major revisions following my suggestions/comments outline below. These suggestions are meant to be challenged and do not necessarily have to be addressed in entirety. As I am no model expert, my comments are more related to the observational parts but model questions that arise for me, likely are also questions that would arise for potential future readers.

Major Comments:

L115-120: I have a few questions that I think need to be clarified in more detail to be understandable by potential readers.

1. i) How do the authors “convert the GIOMAS data to the observed grid”? Do the authors use a nearest neighbor approach? How do you handle multiple entries in case of large resolution differences? Mean? Median? Min? Max? Please elaborate!
2. ii) how exactly do the authors define the “climatological annual cycle”? From all years available in GIOMAS? Following some standard reference period like, e.g., 1970-2000? Please elaborate. Additionally, do the authors only use the averages or also daily GIOMAS data for their comparisons?

iii) What exactly do the authors mean by “spliced together”? This definitely needs clarification!

L130-L136 and L145-L150: While I enjoy reading these beginning analyses about the climatologies of Envisat and Cryosat-2 data which I haven’t seen before, I disagree with some of the authors suggestions.

1. i) For example, while the resolution, i.e. the spacing between footprints I assume the authors mean(?), and the footprint size and shape differ substantially between both sensors, the used retracker algorithms are intentionally the same. Following your mentioned reference of Paul et al. (2018), the effort of the SICCI project was to keep things a consistent as possible between sensors to also allow for an as consistent as possible time series.
2. ii) Furthermore, following the authors own illustrations between Envisat/Cryosat and GIOMAS, the made statements following Schwegmann et al. (2016) are just not applicable anymore. Both references, Paul et al. and Schwegmann et al. refer to different versions of the SICCI data and can to my understanding not be compared. I strongly urge the author to reiterate this paragraph. This also relates to L145/146

iii) In L146, I disagree with the fact that this is still disputed. I think there is a clear understanding in the community that the scattering surface is NOT the snow/ice interface but everything in the snow is too complex to be able to say WHERE it scatters.

And just in general, while I like to see these data sets being used, one clearly has to point out their limitations as also mentioned on the official ESA CCI data source: “Note, the southern hemisphere sea ice thickness dataset is an experimental climate data record, as the algorithm does not properly considers the impact of the complex snow morphology in the freeboard retrieval. Sea ice thickness is provided for all months but needs to be considered biased high in areas with high snow depth and during the southern summer months. Please consult the Product User Guide (PUG) for more information.” (see https://catalogue.ceda.ac.uk/uuid/b1f1ac03077b4aa784c5a413a2210bf5)

A last general point that I find clearly missing in the authors study with all the given data at hand is a proper comparison between all observations, i.e. how do satellite observations and GIOMAS perform directly compared to the observational data? Using not averages but the daily or monthly data. While it is nice to assume the satellite data is correct. As stated above, this is likely not true and an overestimation of reality. A proper analysis of this could really benefit the manuscript and would be of great use to the scientific community!

Minor Comments and Technical Remarks:

As a non-native English speaker (and writer), I still find some of the phrasing and choice of words unintuitive and I recommend some language editing. Example comprise the lack of several “the”’s, the potential use of correct hyphenation (e.g., sea-ice thickness), as well as multiple sentences starting with “And”.

e.g., L18: “of _the_ model”

L35: “lack” instead of “short”? or “shortage”?

L35: What do the authors mean with “freshwater flux of the SO?” by means of melting sea ice?

L38/39: This sounds like a pretty exhaustive list- what other data sets would there be? The way the sentence is phrased suggests this is very limited. While of course the _amount_ of data from these sources might be – its not by the number different available sources if the authors know what I mean.

L42: The sentence starting with “while” reads clunky

L49: I think the plural from reanalysis is reanalyses.

L50: Stop sentence after “alone.” and start the next one with “Reanalyses merge the information […]”

L53: Do not start the sentence with “And”.

L63: I think it should be algorithmS.

L67: see comment to Line 53.

L92: CCI is not properly introduced as abbreviation (only in the context of SICCI)

L93: The authors should probably introduce freeboard a bit earlier already.

L108/109: This means an EM Bird like instrument?

L120: THE trend!

L158: Maybe this is a result from the coarse resolution of the assimilated SSM/I SIC data? Clearly, this would be unable to resolve especially polynya, fast-ice and thin-ice signatures close to the coast.

Figure 1: As a figure enthusiast, I would urge the authors to use a better fitting land/iceshelf mask for their figures. While the Ronne/Filcher areas are shown as “land” the Ross ice shelf is not visible at all. The authors could consider using data from Natural Earth or other similar sources.

Figure 2: It might be my printer, but the two different shade colors for the uncertainty are hard to differentiate. I would assume the wider ones belongs to Envisat but I can not tell for sure.