First results of Antarctic sea ice type retrieval from  active and passive microwave remote sensing data

Melsheimer, Christian; Spreen, Gunnar; Ye, Yufang; Shokr, Mohammed

doi:https://doi.org/10.5194/tc-17-105-2023

Articles | Volume 17, issue 1

https://doi.org/10.5194/tc-17-105-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/tc-17-105-2023

© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 17, issue 1

Research article

|

12 Jan 2023

Research article |

| 12 Jan 2023

First results of Antarctic sea ice type retrieval from active and passive microwave remote sensing data

Christian Melsheimer, Gunnar Spreen, Yufang Ye, and Mohammed Shokr

Download

Final revised paper (published on 12 Jan 2023)
Preprint (discussion started on 04 Jan 2022)

Interactive discussion

Status: closed

RC1:
'Comment on tc-2021-381', Anonymous Referee #1, 14 Feb 2022

Review of

Antarctic sea ice types from active and passive microwave remote sensing

by

Melsheimer, C., et al.

Summary: Little is known about the distribution of different sea-ice types of the Southern Ocean, while of the Arctic Ocean sea ice types several satellite data products exist, comprising binary products, partial concentration products, or sea-ice age. This study aims to contribute filling this knowledge gap by means of an empirical approach combining active and passive microwave satellite observations of the years 2013 through 2019. This approach, called ECICE, has been developed for Arctic sea-ice conditions. In this study it is applied with minimum modification to Antarctic sea-ice conditions. Multiyear ice partial concentration corrections developed for the Arctic Ocean are as well applied to Antarctic sea-ice conditions with almost no modification. A few selected examples of obtained ice-type concentration maps are shown and discussed in comparison to other satellite remote sensing products in an exclusively qualitative manner.

General Comments:

I find the goal of (improved) understanding and quantification of the different Antarctic sea-ice types timely, highly relevant, and important

However, I am not convinced that this manuscript can be seen as a substantial contribution towards accomplishing this goal. It does not make a significant contribution in this research field.

I provide arguments for my decision in the general comments that follow which the authors' please see as umbrellas for the numerous specific and detailed comments provided further below.

GC1: Physics

Arctic and Antarctic sea ice differ fundamentally in the physical properties and hence microwave signatures that are required to provide a reliable, credible, and accurate estimate of the distribution of Antarctic sea-ice types. The paper (and possibly the work behind) fails to adequately describe and discuss these differences. Examples are: 1) Arctic multiyear ice differs from Antarctic multiyear ice in its development history - aka how summer melt changes the sea ice characteristics, its vertical structure and its microwave signature. 2) Antarctic sea ice is experiencing an enhanced atmospheric influence year-round by a considerably larger amount of low pressure systems that regularly cross the entire sea-ice cover, thanks to the geographic setting of the Antarctic sea ice. This influence creates a substantially more dynamic environment for sea-ice growth and different forms of deformation. In particular is sea-ice formation along the ice edge governed by the so-called pancake ice cycle. 3) The snow cover on Antarctic sea ice is known to be much more variable in terms of its depth, composition, (density) layering, and interaction with the underlying sea ice in response to issues mentioned in 2).

Each of these three explicitly mentioned differences in the sea-ice physical properties have an influence on the microwave signature of the sea ice that needs to be understood and adequately quantified before a sea-ice type retrieval algorithm based on satellite microwave observations can be developed or an existing approach such as ECICE can be modified so that it is fit for purpose - aka Antarctic sea ice type concentration retrieval.

GC2: Previous work / References

The approach and the results are potentially of high relevance for the scientific community. So far, this manuscript does not really convince with arguments that the modelling community would benefit from knowing where Antarctic sea ice is of which type. More importantly, however, since this manuscript (and the work behind) arguably is one of the first attempts to derive Antarctic sea-ice type concentrations from combined active / passive microwave satellite data, it has to be embedded substantially better into the overall context. This refers to 1) Discussion of existing (binary) sea-ice type retrieval approaches (for the Arctic) and why (possibly) none of these has been applied to Antarctic sea ice. 2) Discussion of our current specific knowledge about the Antarctic sea ice type distribution and microwave signatures - aka: typical emissivities / brightness temperatures / radar backscatter values for the surface types encountered - including snow. Here the paper, to my opinion, fails to provide the required background to further develop / modify ECICE and to adequately interprete and discuss the results obtained.

GC3: Description of the methodology

The description of the methodology fails to adequately motivate the steps carried out - especially in light of the differences between Arctic and Antarctic sea ice (see GC1). It is not clear how the distributions of brightness temperatures and radar backscatter values were actually obtained. These values are not backed up with values from the literature / other studies (see GC2). The description of input data preparation is not complete.

GC4: Description and interpretation of the results

The description and interpretation of the results mimics the lack of background information about the physical properties and the related microwave signatures - as well as the general differences between Arctic and Antarctic sea ice in terms of its temporal behavior and the role snow has.

Clearly, the paper would benefit from including an Antarctic expert and an expert for snow on Antarctic sea ice. This appears to be mandatory to appropriately interprete and discuss the results qualitatively and quantitatively - be it the multiyear ice area time series, the spurious multiyear ice in the northwestern Ross Sea, the usefulness of the two Arctic multiyear ice concentration correction approaches under Antarctic conditions, or, in general, the likelihood that there is ample opportunity for Antarctic young ice and first-year ice microwave signatures to leak over into the one of multiyear ice - plus an understanding about where icebergs' microwave signatures belong to. Another issue that is completely missing in these descriptions is an investigation about the total concentration the partial ice-type concentrations sum up to. There is multiple evidence in the results shown, that this sum is not 100% but the actual total sea-ice concentration which accuracy is not further discussed and/or shown in comparison to independent data either.

Finally, the authors only aimed for a qualitative evaluation of their results, so this is all they included in this paper. Fine. However, the description of the data used for the evaluation is not complete to understand the scope and relevance of the authors' efforts. The results of the evaluation are very vague, leaving a user in doubt about regions and time periods the data are credible. It seems questionable whether presenting such results has the effect the authors potentially hope for - aka: an encouragement to continue their work and plenty of users of their data set. How can it be ensured that the results taken up by users are not misleading them?

Specific Comments:

Line 33-34 "... but it is ... satellite data" is perhaps an a bit too general statement which i) could be specified better by telling the approaches used of doing so (aka: using instantaneous microwave observations) [note: using multi-annual time series of satellite data would work as well], perhaps by including the work of Comiso et al. (2011?) who figured out the differences in the signature of Arctic SYI vs. MYI, and which ii) could be amended by the fact that sea-ice age data retrieved for the Arctic (but not the Antarctic) are based on ice motion data which are in fact derived using satellite data. Hence it IS possible but nobody looked into it yet.

L43-50: This paragraph is meant to provide the fundament for why ECICE needs some form of adaptation when applied to Antarctic sea ice. In that respect and given that this paper is the first attempting to derive partial ice-type concentrations for Antarctic sea ice, it would make a lot of sense to provide an adequate review of the difference in the sea ice AND snow properties year-round between the Antarctic and the Arctic that is back-up very well by a convincing set of references. This paragraph does not fulfil that role and should be re-written. --> GC1 / GC2

L47-49: "The ice cover ... The turbulent ..." --> I encourage you to provide 1-2 references each that underline these statements - particularly the notion that Antarctic sea ice is rougher - but also the evidence that the sea-ice structure is often different in the Antarctic compared to the Arctic.

L53/54: "Beside MYI ..." --> It would be very important to underline that in fact a substantial amount of the MYI along the East Antarctic coast is actually fast ice. This is often true (older than 2 years old) multi-year ice and is of even larger importance for the ecosystem and has effects on buttressing the ice shelves.

L55-56: "pancake ice can form" --> Isn't this underestimating the fact that a lot of the Antarctic seasonal sea ice is actually formed via the so-called pancake ice cycle first published (in the 1990ties or late 1980ties?) by Lange et al. ?

L60/61: Please check whether it is really the sea ice type that is required or whether these models wouldn't primarily be happy with using improved data of the sea-ice thickness (distribution), the degree of deformation and the snow load. Also, when it comes to validate a climate model I suspect that there are very few that already provide "sea-ice type" as a variable. They might provide ice age though.

L64++: "Recently, ..." --> While it is ok to already mention ECICE here, I ask you to provide a bit more background about algorithms that have been developed in the Arctic to separate FYI from MYI and to provide MYI concentration - first and foremost the NASA Team algorithm.

In addition to that, in order to put the value of your work into a wider context, I also ask the authors to provide more background about other attempts to discriminate between Arctic ice types. It is important that the reader understands that there is almost a full zoo of methods focusing on discriminating between different ice types in the Arctic - in addition to the NASA Team algorithm. To mention in addition to your and Ye's work is the work at met.no, at IFREMER, at BYU (David Long and his group) (and possibly others) that use coarse resolution satellite observations followed by the uncountable attempts to discriminate ice types using SAR. In contrast, activities in the Antarctic are very sparse.

You might argue that you are looking for ice type CONCENTRATION and not a simple discrimination. That is true, but even here your work is more upfront than any other work and this needs to be (implicitly) stressed.

You might also argue that ice type CONCENTRATION is the more important parameter, but if I understood your introduction so far correctly, then we are in need of ANY information about the ice-type distribution of Antarctic sea ice (other than land-fast sea ice), no matter whether this is a binary classification result or whether it is (already) an ice type concentration.

Because of this I ask you to one more time dig into the literature and try to find out what others did in this sector. If we omit polynyas / fast ice - for which a lot of studies exist - then there is not too many, perhaps add: Lythe et al., Classification of sea ice types in the Ross Sea, Antarctica from SAR and AVHRR imagery, International J. Remote Sensing, 20(15), 3073-3085, 1999, http://dx.doi.org/10.1080/014311699211624

and Ozsoy-Cicek et al., Intercomparisons of Antarctic sea ice types from visual ship, RADARSAT-1 SAR, Envisat ASAR, QuikSCAT, and AMSR-E satellite observations in the Bellingshausen Sea , Deep Sea Res. II, 58(9-10), 1092-1111, 2011, https://doi.org/10.1016/j.dsr2.2010.10.031 --> GC2

L84/85: "It takes input ... any given ice types" --> So, I can input 6.9 GHz AMSR2 TB H-pol and 5 GHz ASCAT observations and can obtain the partial concentration of pancake ice? Or I can input 91.6 GHz SSMIS TB at H- and V-Pol and get the partial concentration of MYI ice and FYI ice? If this is not the case then I suggest to re-write this sentence according to the actual capabilities of ECICE which seems to be oversold a bit here.

L87/88: "to account for anomalies ... One anomaly causes ..." --> I suggest to re-phrase these statements. It is not clear what you mean by "observations". To me observations are the data you obtain from the satellites, i.e. brightness temperatures or backscatter coefficients. Hence, I ask myself what anomalies are in this regard? You possibly refer to those cases where ECICE fails to interprete the input satellite data into the correct total and/or partial ice concentrations, creating anomalous high or low concentrations and/or an anomalous misclassification of MYI as FYI and vice versa. Therefore it might be more correct to state that one set of satellite observations can be the result of several different combinations of physical parameters, causing ambiguous retrieval of total and/or partial concentrations when input into ECICE.

L100-102: Please provide 2-4 references for publications that could underline your statement for sea-ice concentration and sea-ice type concentration - for both passive and active microwave observations.

L111-113: What happens, during the retrieval, if fractions do not add up to 1 and/or for fractions below 0 or above 1? Are these set to 1 (or 0) before the median of all realizations is computed?

L116: How is the spread around the median computed? How many valid values are required for a median and its spread to be computed (assuming that not all 1000 realizations provide a valid result)?

Lines 119-124: I am missing the physics and references in this paragraph. What are the physical properties of the ice types that cause the different radiometric and backscattering properties that allow us to discriminate between the three ice types? Which of these are influenced by which snow physical properties that make MYI to look like FYI? How about the ambiguities between YI and FYI?

You use snow metamorphism only in the context of "return of cold temperatures" albeit snow metamorphism encloses a wide variety of changes of the snows' crystal structure and composition under the action of temperature, humidity and wind. This should be re-phrased. In addition "warm spells" only cause "snow wetness" to develop if the temperatures are high enough; still, even with considerable below freezing (-5 degC) temperatures snow metamorphism (rounding of grains, etc.) is present. --> GC1

L126-134: This paragraph describes the temperature correction as developed for Arctic conditions. You appear to adopt it 1-to-1 to Antarctic conditions as is indicated by the last sentence in this paragraph. Without an adequate introduction and review of the physical, radiometric and backscattering properties of Antarctic sea ice and its snow cover compared to the Arctic, this raises my concerns. On the one hand differ Antarctic MYI and partly also FYI physical and microwave properties from Arctic ones. On the other hand differ Antarctic snow properties often fundamentally from those in the Arctic - not to speak of the frequency with which the weather influences the microwave signature of Antarctic sea ice compared to the Arctic. I am wondering whether a close collaboration with specialists in this field would not substantially improve both, set up of the algorithm and interpretation of the results.

L150-152: "this correction scheme ... to MYI" --> I suggest to separate this correction from the drift correction because it has nothing in common with it. I further suggest that you make clear which form of snow metamorphism you are refering to here. In L154 you introduce "HR" as being related to the "onset of snow melt" which, at first glimpse, would suggest an increase in snow wetness and hence elevated brightness temperatures, making MYI to look like FYI rather than the other way round as is stated here. You are possibly refering to melt-refreeze cycles or the like and need to specify this here to avoid confusion.

L150: To me "Ex-MYI" implies that this sea ice once was MYI and now is a different ice type. How about you name it "artificial MYI"?

L166-169: I suggest to add the actual resolutions and sampling interval of the AMSR2 channels used.

Please provide information about the native spatial resolution of the ASCAT data and how you gridded these into the NSIDC grid of 12.5 km grid resolution. It appears to me that the statistics is different for these data than for the AMSR2 data because of the different viewing geometry and swath width.

What would also be important to know is whether the sigma_nought values were corrected towards a certain common incidence angle (e.g. 40 degrees)? If this is not the case, please provide a comment why you deemed that as not being necessary.

L170-180: Your description about the choice of sample areas and time periods is not specific enough to my opinion. I have the following questions:

1) Apparently you used ASI SIC maps to define your sample areas. What is the requirement regarding the SIC to have a grid cell contributing to the sample?

2) How did you define "beginning of the cold season"?

3) For which time period (just 1 day?) did you select grid cells from the Weddell Sea defining the MYI distribution?

4) For which time period(s) and region(s) did you select grid cells defining the FYI distribution?

5) From which time period(s) and region(s) did you define the YI distributions based on the PSSM data set?

6) How did you take into account the YI that develops during the ubiquitous pancake ice cycle in the MIZ that might cover several hundreds of kilometers? Isn't this, not the one growing in the polynyas, the far more relevant YI type in the Antarctic?

7) Where exactly, with respect to the ice edge, are your open water sample areas located?

8) How representative are the open water sample areas in the regions and months (March, Ross Sea; August, everywhere?) chosen for the weather influence?

One way to answer at least some of these questions would be to create a map in which you show the locations of the sample areas and, via color coding, the time-periods and/or frequency with which you used selected the data.

Table 2:

- I have concerns with two values in this table. Why do you define the END of the warm episode with a positive (2degC) air temperature? Is this a typo? If not it is absolutely not understandable and needs some justification.

- What is the motivation for the very long maximum duration of the warm episode? This does not sound overly reasonable to me - neither for the Arctic nor for the Antarctic actually. I can guess that the length of this period is chosen this way because the melt and melt-refreeze processes change the physical and therefore microwave signature of the snow / sea ice system for a considerable number of days; even after freezing conditions have returned the modified microwave signature might still last (e.g. Voss et al., 2003, in Polar Research and his work related to that).

- Apart from that I am wondering how such a long maximum duration does match the comparably high frequency of warm events caused by cyclones passing over the sea ice. I'd say that such events can be quite short-lived. Therefore, depending on whether you aim for a monthly or a daily ice type product one could recommend to use a considerably shorter maximum duration of such events of just 5 or even 3 days.

- In case you comment on the choice of these parameters later in the paper, i.e. in the context of the discussion, please point the reader that already here to increase the credibility of your choices.

Table 3:

- I note that the choice of the values for these parameters specifically for Antarctic conditions has not been discussed and/or movitated so far. You might want to do that, please.

- In case you comment on the choice of these parameters later in the paper, i.e. in the context of the discussion, please point the reader that already here to increase the credibility of your choices.

L187-191: Three comments here:

1) What kept you from using ERA5 data? Is there are credible argument to stick to ERA-Interim data for surface temperature data in the Antarctic?

2) Tschudi et al. (2016) appears to be a bit outdated given the fact that there is a version 4.1 of the NSIDC sea-ice motion data set, referenced as Tschudi et al. (2019 or even 2020).

3) What is the motivation to use this rather low resolution OSI SAF sea-ice drift product? Doesn't it harmonize with the overall 12.5 km grid resolution you aim for much less than the NSIDC sea-ice drift product?

L192: In Table 3 you mention the TB at 37 GHz, not 19 GHz; please check.

Figure 1: I have a number of comments here; comment #1 and #2 are related directly to the figure content while comments #3 to #5 are related to the omission of relating the results shown to previous work.

1) What does the "Distr. set: AQ2" in the title of each panel refer to? Could it be removed?

2) What is the statistics behind the data? What is the time period? At how many data per surface type do we look?

3) What explains, to your opinion the fact, that GR3719 is a bit lower than is classically observed for open water and, particularly, for FYI (compare the tie point triangle used in the NASA-Team algorithm).

4) How do your values compare in general to tie points used by ordinary sea-ice concentration retrieval algorithms?

5) How do your backscatter values compare to values for C-Band radar backscatter of Antarctic sea ice cited in the literature?

L201-203: While I am fine with using AMSR-E instead of AMSR2, I have concerns to simply replace ASCAT (C-Band) with QuikSCAT (Ku-Band) as signal penetration into and interaction with the snow / sea ice system differ - in addition to incidence angle and resolution.

Instead of working with piece-wise available sea-ice drift products it might be a very good idea to use one consistent data set, namely the NSIDC one - unless you find an alternative with year-round coverage (IFREMER?); yes, NSIDC is not an optimal choice but with that you avoid inconsistencies and jumps in your then much longer (by combining AMSR-E and AMSR2) time series.

You might want to consider to simply delete these three lines here.

Section 3.1:

- How many Sentinel-1 SAR data from which dates were used? Where were these located (provide a map with the frames)? What was the time difference between SAR image acquisition and ECICE product? What is the "time stamp" of the ECICE products [0 UTC, 12 UTC]?

- Where were the Sentinel-1 SAR images taken from. Which type of SAR images was used (Wide Swath, Extended Wide Swath, ...)? How were the SAR image (pre-)processed for the evaluation? Was any drift correction applied to the SAR images?

- You decided to provide a qualitative intercomparison without computing radar backscatter (sigma_nought) values. Why? Wouldn't your results be much more credible and useful if you would come up with fractions of MYI and/or FYI derived based on a rough (by means of sigma_nought value) classification from the SAR images and compare those to the ECICE MYI concentration maps?

- L214-215: "In SAR images, MYI ... sub-surface layer" --> This very qualitative and not overly scientifically formulated sentence applies to the Arctic. Melt processes during summer in the Antarctic differ considerably from the Arctic and I doubt that one can speak of a "bubbly sub-surface" layer here. Please revise your wording taking int account the specifics of seasonal changes in microwave signatures in the Antarctic compared to the Arctic.

- L221/222: "the iceberg ... as FYI" --> I don't agree. The SAR signatures inside that 70% FYI polygon encircling the iceberg are brighter than outside the polygon. The isoline does also not indicate at which side FYI concentrations are actually higher or lower. Given the fact that the area southwest of the iceberg is certainly dominated by FYI I suggest to re-phrase this statement along the lines that for that polygon both FYI and MYI concentrations are below 70% but that you don't know which is the dominant one. See also your Figure 5.

- For one grid cell, do partial concentrations sum up to 100%? I am asking because in the area indicated as > 50% YI fringing the Antarctic Peninsula there is evidence for MYI concentration > 50%. Did you actually check for maps like the one shown in Fig. 2 what the sum YI + FYI + MYI concentration is? It would interesting to see an example of this - perhaps in the appendix or in supplementary material.

- L222: This last sentence about the "quality" of this comparison I deem almost obsolete without information about how many SAR images of how many regions from which dates have actially been taken into account.

Section 3.2:

- How are the weekly charts derived with respect to temporal availability of the input data? Is always the latest highest quality data set used for a respective grid cell (or pixel)? Or what is the compositing method used?

- Does "microwave satellite imagers" include SAR? What is the dominant input data source for the charts you have used?

- What does "analysis ... by experienced specialists" mean? Is this a manual analysis? Is the analysis done by one specialist or a team of specialists and what are the quality measures?

- L228/229: What is the size of such a pixel? What is the grid that is used here? Is it the NSIDC polarstereographic one? Looking at Figures 3 and 4 I get the impression that in these ice charts the classification is not done pixel-by-pixel but rather in form of polygons that contain ice of similar characteristics and concentrations - such as done, e.g., by the Canadian and Danish Ice Services. Could you please check once again how the ice charts you show in your manuscript were generated, and if need be, re-phrase your description?

- Were the input data projected into a common grid prior to ice chart generation?

- L234/235: What is your "cold season"? What do you mean by "sporadic comparisons with data from other years"? How many, from where and which dates were these additional comparisons?

- L243-246: Your observations of the different labeling of ice as FYI or MYI between AARI and NIC charts could be the result of different definitions of when FYI is re-labelled MYI ice by the producing agencies? Did you check that? I note that a switch in March / April disagrees with the WMO recommendation you mentioned further up in your manuscript.

- I note in addition that there are more fundamental differences between the AARI and the NIC ice charts in the Eastern Weddell Sea regarding the location of YI and FYI.

Figure 3:

- I suggest to use a title for the MYI concentration that is consistent with the other two ECICE results.

- Putting the legend of the AARI ice chart into appendix is not a good solution. I suggest the following: You crop all maps to an area that excludes all the annotations in the AARI ice chart, put all ECICE results into the second row of panels and put the ice chart legend in the first row of panels next to the ice chart. In the second row of panels you could then also follow your approach from Fig. 5 and provide one legend with the title "Ice-type concentration", marking the ice type itself in the map (actually it is in the panels' titles but perhaps you consider to remove these anyways.).

- Finally, I note that you seem to use an old land mask to mask out Antarctica, still containing an overly long "Trolltunga" of the Fimbul Ice Shelf and the Mertz Ice Shelf. Given the fact that you focus here on AMSR2 data it might be a very good idea to use an more recent and hence more accurate land mask.

- L258 / Fig. 5: "with the summation of their fractions equal to 100%" --> When I look at Figure 5 I doubt that this statement holds. I guess it needs to be replaced by the actual total sea ice concentration that is obtained with the ECICE algorithm because there are quite some areas downstream of the Ronne-Filcher Ice Shelf polynya where YI conc. + FYI conc. + MYI conc. add up to something between 80 and 90%. I am sure you will get back to this in the discussion section. But it certainly does not hurt to either state that "theoretically" the partial concentration should add up to 100% but that this is not always the case, or correct your writing accordingly towards that the sum of the partial ice concentrations adds (of course) only up to the actually existing amount of sea ice. --> GC4

L261/262 / Figure 5: Your maps do also reveal that ECICE seems to have a problem discriminating between YI and MYI because the area just next to the Ronne-Filcher Ice Shelf appears to be characterised by some YI, no FYI and some MYI as one can observe a fringe of non-zero MYI concentration in that area.

Section 3.3:

- Please provide more information about the PSSM maps. What are the grey areas masking parts of the maps shown? Where did you get the data from? What is their temporal and spatial resolution? How many of these maps did you use for which regions? The scope of this part of your intercomparison remains vague.

- Given the fact that you look at years 2017 and 2018 I assume it is SSMIS data and not SSM/I anymore, am I correct? Please correct your writing accordingly.

- I would appreciate if you could comment on the quality and limitations of the PSSM based ice type maps. What the approximate thickness limit between thin ice and "other ice"? Does "thin ice" mean that there is 100% thin ice or could this potentially also be 50% thicker ice interspersed with open water?

Figures 6 through 9:

- I suggest to reduce the size of the panels considerably. In particular I recommend to make the PSSM map the same size as the white box shown in the left panel denoting its location. Even better would be if you'd crop the maps in the left panels to the size of the PSSM map. That way would would be able to reduce the number of these figures from 4 to 2 or perhaps even 1. Did you try, in this context, to combine the information from both panels into one? Perhaps by extracting isolines from the PSSM maps and superpose these onto the YI concentration maps?

- Did you chose the dates shown in the manuscript arbitrarily? If so, make a note.

- Were these the only PSSM maps you considered in your comparison? If not how did the comparison go for all the other maps? Did you derive any quantitative information?

- Looking at these YI concentration maps reminds me one more time the issue of how the ECICE ice-type concentration maps deal with cases of considerably less than 100% total sea-ice concentration because I note that all the YI concentration maps shown in Fig. 6-9 reveal lower YI concentrations in the core of the polynyas.

L280: "given the limitation that a rigorous validation ..." --> Given the fact that you kind of advertise the data set obtained with this paper and given the fact that this is first attempt to provide such a data set, I don't take it as a positive sign of credibility of the data set produced, when this paper only deals with a very general, little quantitative evaluation. The results presented are partly very vague and the description of the physical background being the foundation for the approach used and the data set is not overly exhaustive and - at least for me - not convincing.

L281: "Large icebergs are often erroneously retrieved as FYI" --> Is this your result? You could state this more clearly. But, when doing so, please take into account my comment made to Fig. 2 with respect to this issue.

L288-300: I was kind of expecting that you would run into problems with weather-induced variations in the snow physical properties and resulting microwave signatures. Since in your manuscript the physical foundation and description of the processes and properties resultung in specific microwave signatures is not overly detailed and mature, it is of course difficult to discuss these observations. I find that your attempt to explain your observations go into the correct direction but are far from being conclusive and is too vague. I'd say you could delineate the reasons that caused the MYI concentration over-estimation much better and much more specifically by means of checking the input data values and compare these with what is known from literature. It might make sense to take into account ERA-Interim and/or ERA5 data (you use them anyways) to discuss you observations also in the context of melt-refreeze, ice-snow interface flooding, slush refreezing, snow-ice formation and the like. I again recommend to take a look at the work of Voss et al. (2003) and the related doctoral thesis.

Figure 11: Looking at that figure again makes me to think whether you ever tried to look at maps of YI conc + FYI conc + MYI conc? It appears to me that there are patches of spuriously large MYI concentration that coincide with a total sum of partial concentrations above 100%.

L299: None of the references listed in this line deal with pancake ice and its backscatter. These are all references dealing with the snow cover and should be put into L298 behind"... MYI in that respect."

L292: How credible are - to your opinion - these MYI occurrences "far offshore in the outer Ross Sea"? Which process can cause these?

Final question to L288-300: How did you compute the total MYI area shown in Fig. 10? Did you apply a threshold MYI concentration or did you count from 1 % onwards? What did you use as gridcell area to compute the total area?

L302-303: You can look yourself into the likelihood of (1) by checking the drift data you used. How did you cope with data gaps in the drift product? Did you include the quality flags?

L306-307: "such seeding points" --> please explain this in more detail or delete it. Questions I would have is how this happens and why this should have an influence on the MYI concentration in particular and not on the other partial concentrations.

L309/310: I don't understand why you refer to an observation of Ted Makysm when you yourself used the data for the drift correction. Didn't you yourself take a look at the data once you suspected that these could include spurious drift estimates? This is inconclusive.

Line 311-314: I suggest to not look into the data used but first try to understand which sea ice and snow physical properties you encounter during the course of one cold season and to further understand how the microwave signature looks like. This might require to look into 1-dimensional numerical modelling of microwave emissivities and of microwave backscatter as a function of sea ice and snow properties. There is a paper by Willmes et al. (2014) in the Cryosphere and there is work by Tonboe et al. that might help here.

L319-322: "The most likely reason ..." --> While your observation from Fig. 10 seems to be credible, I am wondering whether this isn't an over-simplification of the situation. I agree, wettening of the snow cover can mask MYI so that it looks like FYI. But at the same time the re-freezing of the slush at the ice-snow interface, ice lenses, whatsoever causing larger grain sizes can have the adverse effect and making FYI looking like MYI. A deep snow pack and/or substantial deformation of FYI has the same effect as demonstrated by one of the co-authors for the Arctic ocean. In addition, and here the authors were right earlier of course, pancake ice is a nasty fellow and could possibly also likely to be misclassified as either of the two thicker ice types - adding to their partial concentration.

L323-326: Two more thoughts on this: Beginning in October the expansion of the Antarctic ice cover stops and the lateral movements switches to a retreat / compaction type. In addition, due to the dispersion the fraction of MYI per grid cell has decreased to a value that is likely not large enough anymore to be adequately detected by ECICE. I am sure this is something you can check in your data. One could hypothesize that computing the total ECICE MYI area is reasonable as long as ECICE is capable to derive the MYI concentration with high accuracy ... which I doubt is the case when the partial concentration has fallen below 30% and when the MYI coverage has dispersed in many small floes embedded in a mixture of YI and FYI.

Also any MYI that has arrived in the MIZ (in the Weddell Sea) is now likely to melt as air temperatures are not cold enough anymore out there to keep it alive. From that point of view I find a rather decay of the MYI area in the September / October time frame not overly surprizing.

L337: "and melt" --> Where did I find examples of these in your manuscript?

L343-344: "The new time series ... outweighs the shortcomings that still exist." --> I do not agree to this statement because of 1) the unmature physical foundation, 2) the vague interpretation of spurious ice type concentrations and 3) the very qualitative evaluation.

Typos / Editoral Comments:

L27: I suggest to look for a more recent paper making this statement, e.g. Kwok 2018 in Environmental Research Letters.

L30: There should be another reference from Parkinson and DiGirolamo from 2022 in Remote Sensing of Environment.

L49: Typo: "...sea ice For ..." --> "... sea ice. For ..."

L79: "existing a ice chart" --> "existing ice chart"

L114: Typo: "coast" --> "cost"

L152/153: Please explain all the mathematical expressions that are used here for the first time.

L159-164: You have introduced the sensors' acronyms further up and can omit that here.

L207: I am sure a reader would appreciate to see 1-2 references here.

L210/211: It might make sense to mention already here that the polynya maps used in the comparison are from a different year than those used for algorithm tuning.

L222: "Sentinel-1 scenes" --> "Sentinel-1 SAR scenes"

L256: Typo: "forth" --> "fourth"; see also L262

L266: Typo: "where" --> "were"

L286: "often" --> since you deal with a limited number of years in this paper you could perhaps mention all years during which you observe this increase.

L332: "outside the melt season" --> "during the freezing season"

"spatial" --> "grid"

L334/335: "... is well captured" --> You could add "by our ECICE results" to make clear that this is your result.

L334: "... in the Antarctic" --> add: "in addition to ship-based observations of the ice conditions."

Citation: https://doi.org/10.5194/tc-2021-381-RC1
- AC1: 'Reply on RC1', Christian Melsheimer, 01 Apr 2022
  
  (see attached PDF file)
  
  Citation: https://doi.org/10.5194/tc-2021-381-AC1
RC2:
'Comment on tc-2021-381', Anonymous Referee #2, 17 Feb 2022

Please see my comments in the attached pdf.

Citation: https://doi.org/10.5194/tc-2021-381-RC2
- AC2: 'Reply on RC2', Christian Melsheimer, 01 Apr 2022
  
  (see attached PDF file)
  
  Citation: https://doi.org/10.5194/tc-2021-381-AC2
RC3:
'Comment on tc-2021-381', Anonymous Referee #3, 01 Mar 2022

Review of “Antarctic sea ice types from active and passive microwave remote sensing” by Melsheimer et al.

Summary

This manuscript describes a method to retrieve three Antarctic sea ice types from passive microwave and scatterometer data: young ice, first-year ice, and multiyear ice. The approach adapts the ECICE method originally developed for the Arctic. Corrections are made for temperature and ice motion to improve performance. Comparisons are made with SAR data and ice chart stage of development fields. The ECICE fields generally compare well, but there are some discrepancies, notably the “formation” of MYI in July in the outer Ross Sea. The authors hypothesize that this is due to snow metamorphosis on FYI that results in a MYI-like signal. Other errors occur due to motion tracking errors.

General Comment

An Antarctic MYI data product is a great contribution. As the authors note, it is challenging to create such a product. The general approach here is a good one. The authors adapt the ECICE method that has been successful in the Arctic. The method appears to work well based on initial comparison with SAR and stage of development ice charts. The comparisons are generally qualitative in nature. I think that at this point, introducing a new method, this is okay. It is hoped that the authors will follow up this work with more quantitative comparisons to develop statistical error estimates. And it is hoped that the data set is expanded to a longer time series in future work. One thing that would help with the qualitative comparisons is if the SoD images could be custom made with a color scale indicating only the three ECICE ice types. This is discussed further in the comments below. But I’m not sure how feasible or easy this would be with the SoD fields. So, I don’t think this is essential, though it would make the results easier to interpret. The other issues are all quite minor. The writing is good and the figures are otherwise well done. I recommend acceptance after minor revisions.

Specific Comments (by line number):

51: One reason why most MYI is in the Weddell Sea is that the gyre that transports MYI away from the coast to the north and northwest also transports in ice from the north and northeast. This is seasonal ice that gets transport into the Weddell, where it compacts along the ice shelve and Antarctic Peninsula and, along with less solar insolation and colder temperatures, allows that FYI to survive into MYI. This seems a salient point to make here as it is the mechanism to form MYI.

90-92: No SSMIS sensor data are used?

174: How is the “beginning of the cold season” defined? Is it the minimum total extent? But at the minimum, there may be regional gains and regional ice losses occurring (the minimum marks when the gains start to outpace the losses). Ideally, you would use the minimum at given grid cell or at least regionally.

187: How accurate are the ECMWF 2 m temperatures over the sea ice? There are several coastal stations that I assume provide observations, but over the sea ice, the observations are quite sparse, with few buoys (compared to the Arctic). It is reasonable to use ECMWF as that is what is available and better than nothing. But I think a mention on potential uncertainty is worthwhile here.

188-189: And likewise for the ice motions. Antarctic motions typically have higher errors because of the variability of the ice (flooding ice, etc.) and lack of buoy validation. Again, don’t need to go into great detail, but a comment on the uncertainty would be helpful.

234-235: In what format are the SoD charts provided? It seems they are used here merely qualitatively. If they are just images, that makes sense. But if they are in some sort of data format (e.g., GeoTIFF), they could be used to do some quantitative comparison with the ECICE. And also, as noted below, they could be manipulated to consolidate the different ice classes into the main three with a clear color scale to more easily visually compare with ECICE.

236, Figure 2: This figure seems a bit odd and confusing to me. It seems like there are two SAR images overlaid on the ECICE image. But they overlay, so block the ECICE. Once can see some continuity, so the performance looks reasonable, but it seems odd to show only one figure with one or the other (SAR or ECICE). The ECICE color scale seems to have several more gradations than the 5 indicated in the legend. The legend color scale should match the colors plotted. It seems like creating a two-panel image – one with the ECICE and one with the SAR images and then overlay the contours on both – would be clearer?

239, Figure 3: I guess it is okay to have the SoD color scale in the Appendix – at least the authors acknowledge that it isn’t legible in the figure. But ideally, a better color scale would be included/added to the figure. And it’s clear that the SoD figure has more categories than the ECICE, so it is a bit hard to directly compare, though the overall patterns are clear. It would be more work, but if it were possible to actually take the SoD and create a custom plot with the SoD categories combined into the three ECICE categories, that would be quite helpful.

267, Figure 5: As for Figure 3, it would be nice to have SoD in a simplified form with all types consolidated into the three ECICE types and with a color scale legend provided with the figure.

Minor Comments (by line number):

45: I’ve seen “snow-ice” with a dash to connect the two nouns and denote a unique type. But this is perhaps simply more of an editorial/style decision.

114: Typo, “cost” not “coast”

174: Not sure why the ASI reference is given as a footnote? If that is The Cryosphere style guideline, I guess that’s okay, but in my view, datasets should generally be cited as regular references.

346: It seems like the chart color legend (Table A1) should be after the beginning of the Appendix text? But as noted, it would be helpful to create a new legend that combines the relevant classes into the three main types for the figures in the main text of the manuscript.

Citation: https://doi.org/10.5194/tc-2021-381-RC3
- AC3: 'Reply on RC3', Christian Melsheimer, 01 Apr 2022
  
  (see attached PDF file)
  
  Citation: https://doi.org/10.5194/tc-2021-381-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Reconsider after major revisions (further review by editor and referees) (05 Apr 2022) by Yevgeny Aksenov

AR by Christian Melsheimer on behalf of the Authors (25 May 2022) Author's response Author's tracked changes Manuscript

ED: Reconsider after major revisions (further review by editor and referees) (14 Jun 2022) by Yevgeny Aksenov

ED: Referee Nomination & Report Request started (12 Jul 2022) by Yevgeny Aksenov

RR by Anonymous Referee #3 (29 Jul 2022)

RR by Anonymous Referee #1 (08 Aug 2022)

Suggestions for revision or reasons for rejection

Review of

Antarctic sea ice types from active and passive microwave remote sensing - Revision 1

by Melsheimer, C., et al.

Summary: I do not provide a summary of this manuscript because I reviewed the previous version and refer therefore to my summary given there.

Some aspects of the manuscript have improved compared to the version I reviewed. I still have a number of concerns, though, that make me to suggest another round of major reviews. The reasons for that I lay out in my general comments, further detailed by my specific comments.

General Comments:
GC1: The authors stated in their response to my 1st review that this is a proof-of-concept study and used this as an argument to keep a lot of the content of the manuscript at a level of depth where is was before. I therefore ask the authors to reflect the nature of this study in their title, in the abstract and also in the conclusions.

GC2: There are certain elements of the methodology that are still not described sufficiently well. To these belong how the initial MYI distribution is found, or how ice drift information is employed. In addition, the intercomparison of TBs and NRCS values from the sampling areas and periods with independent values of these parameters needs to be improved. At least the obvious discrepancies between the values used and those published in the literature should be mentioned properly and taken into account when discussing the limitations of the approach and the potential way ahead.

GC3: The interpretation and discussion of the results still requires a more thorough incorporation of the physical properties of sea ice and its snow cover and their influence on the microwave properties. The discussion appears still to be biased to much towards improving the correction scheme rather than improving ECICE and the selection of adequate sampling areas and periods for the adoption of ECICE from Arctic to Antarctic conditions.

GC4: There is a number of other smaller issues where the information provided is either not correct or needs to be complemented or where a more critical discussion of the own results appears to be useful for a more complete understanding of the main message of the manuscript.

Specific Comments:
Title and abstract: You have decided to not build your paper on a solid review of our understanding of the different physical properties of the Antarctic sea ice compared to the Arctic sea ice and their implications on active and passive microwave satellite data. Fine. You have also decided to keep out expertise about snow on Antarctic sea ice. In that case, however, I cannot accept this paper without the following suggestions for amendments in the title and the abstract.

1) The title is far too general and points into the wrong direction. As you stated in your comments to my 1st review this is a "proof-of-concept" study. This should be mentioned in the title.

2) You use ECICE and you discriminate young ice from first-year ice from multiyear ice - similar to what ECICE does in the Arctic - without modifying ECICE for the Antarctic for the reasons laid out in your response to my 1st review. Please mention ECICE in the title (there are Arctic ECICE papers, so it does not harm to mention it here). Hence a title such as: About a proof-of-concept study applying the ECICE algorithm to discriminate Antarctic multiyear ice from younger ice types" would be much more to the point.

3) In the abstract I suggest you write that ECICE uses probability distributions that decouple its application from the need of a deeper understanding and consideration of the emissive and backscattering microwave property differences between the two polar regions. This would lend much more credibility to the abstract and would fit much better to the content of the paper. Note that in L8-10 you explicitly write "Due to differences in physical and crystalline structural properties of sea ice and snow between the two polar regions, it has become difficult to identify ice types in the Antarctic". Since you do not deal with an exhaustive investigation of the different properties on the AMW and PMW signals but still come up with a solution - which is only possible thanks to the nature of the ECICE algorithm using probability distributions - I feel it is mandatory to tell the reader this piece of information, i.e. that you propose to, as a first proof-of-concept, apply a method based on probability distributions of the involved AMW and PMW data. See also my next comment.

4) The sentence "Until recently, no method ... time scales" in L10/11 should be deleted to my opinion. You did not develop a new method here. You applied a method that is known to work for Arctic conditions to Antarctic conditions to see whether it works.

5) I also strongly recommend to re-formulate the sentence "Although there are ... sea ice types" in L16/17. This is not the first time we learn about the evolution and dynamics of Antarctic sea ice types. I would rather say: "Our results of applying ECICE to Antarctic sea ice conditions for the first time demonstrate its potential for Antarctic sea ice type discrimination. They also confirm existing knowledge about typical spatial pattern in the sea ice type distribution and their movement."
Thank you for your consideration.

L39/40: "Unlike the Arctic ... still unclear" --> I object to this statement. It is well known where sea ice typically survives summer melt. It is well known where it usually drifts. It is well known that a lot of the sea ice is either formed along the marginal ice zone in the so-called pancake ice cycle or in the ubiquituous coastal polynyas from where it drifts north or northeastward. Also the location and age of landfast sea ice is well known.

L41-42: "In the austral summer ... 2012a)." --> What do studies by Hobbs et al. (2015) and Parkinson and diGirolamo (2021 and before) state in this regard?

L53/54: "For these reasons ... two regions" --> I don't agree to put this sentence here because you have not yet made any link between the physical properties that you just described and how these influence radiometric and backscattering observations.

L62: Please provide at least one reference each for these two important implications of landfast sea ice in the Antarctic.

L76/77: "exclusively to the Arctic" --> Why is this the case? Because the radiometric signature between MYI and seasonal ice in the Antarctic was found to be not well suited to carry out a similar analysis in the Antarctic - at least when using the same channels as in the Arctic.

- "Besides, the retrieved ... see Section 3.2" --> This is a too global statement. First of all it needs to be stated whether you refer to MYI extent or area. It is the area which cannot increase during winter while the extent may increase during periods of MYI fracturing and dispersion during divergent conditions - depending on the threshold set to compute extent and depending on the capabilities of whatever algorithm used to reliably discriminate MYI from other ice types. Secondly, this tendency of MYI whatever to increase during the cold season is something that may occur but that not occurs regularly (I know this from own experience working with the NASA-Team MYI concentration data). Hence, I suggest to rephrase your statement.

L124/125: "the radiometric ... or even vanish" --> Since you are referring to Arctic conditions here it would be suitable to cite respective papers dealing with sea ice type retrieval in the Arctic, for instance the two Lindell and Long papers from 2016 in Transactions of Geoscience and Remote Sensing, 54, or Remote Sensing, 8.

L175/176: "but is FYI ... of the ice" --> I suggest to add that this can also be FYI with a thick snow cover known to resemble radiometric properties similar to MYI ice (as is one of the problems using the 37 GHz / 19 GHz vertical polarization gradient ratio for snow thickness retrieval).

L179: "non-MYI" --> You write that you keep this new pseudo-ice type which is either FYI or YI - so simply seasonal sea ice. But I cannot see it in any of the maps you show later. To which of the concentrations is the concentration of this non-MYI added? If it is not added anywhere ... where does it remain? Or is this the "magic" missing 20% in some of the maps of areas with 100% sea ice concentration?

L204/205: "This is the common grid used ..." --> I don't think that the information who uses this grid is required. Widely used as well is the EASE grid which has several advantages over the polarstereographic grid and is used, e.g., by OSI SAF for its sea ice concentration products. What is required, however, is to mention the latitude of the tangential plane of that polarstereographic grid.

L210/211: "using a simple linear approach ..." --> Did you develop this by yourself? In that case it would be appreciated to learn more about the method and its application frequency (daily? monthly? seasonally? How often are the linear regression coefficients updated? What is the data set used to define where there is sea ice and where there is open water? Is a sea ice concentration threshold used? What about summer conditions?). In case not please provide an appropriate reference.

L214-217: "For FYI and MYI ... per day" --> These lines and your comments to my 1st review do not lay out sufficiently well what you did why. Please put yourself into the position of a student who wants to redo your analysis ... The following points need clarification:

1) Why did you choose 2018 as the "master" year for your initial discrimination of sea ice types to find the probability distributions of the parameters laid out in Table 1?

2) While you attempt to describe better in your footnote 2 how you define what MYI is, it is not sufficient. First of all, the moment when the sea ice cover starts to grow again, resulting in an increase of the sea ice extent depends pretty much on the drift conditions. I can envision - and this is also an issue in publications where melt and freeze onset are derived - that on a day-to-day basis you cannot adequately define when sea ice indeed begins to advance again. Those publications / methods use - for good reason - a period of 3 or even 5 consecutive days of the respective geophysical process to happen (in your case sea-ice advance) to define that the conditions have switched from summer melt to freeze-up. Secondly, you do that "regionally" ... okay ... What are the regions? How did you define them? How large is the difference between freeze-up (or ice-advance and hence MYI definition) days between different regions? I can envision that MYI is defined as such as early as mid / end of February in the Weddell Sea but as late as May in parts of the Bellingshausen Sea. Here, your manuscript lacks essential information.

3) "Later in the season ..." --> I probably understand what you did but it is not described properly. When sea ice is defined as FYI becomes not clear. Do you apply a sea-ice motion data set to actually track the border of the MYI defined in the previous step? Or did you apply a ball-park estimate of 10 km drift per day? In the latter case the question would be: i) into which direction and ii) every day?

L223/224: Did you check the weather conditions in the regions and during the time periods chosen to take your open water samples? Are these from predominantly quiet conditions (low wind speed, low atmospheric water vapor load, low cloud cover) or do these cover a wide (an as wide as possible) range of different conditions and hence atmospheric influences on the AMW and PMW observations? It would be useful to provide this information in the context of showing and describing the respective distribution functions.

L224/225: "Details of all ..." --> Thank you for providing this information at least in form of a table. For the understanding and the credibility of your proof-of-concept study it would be much more useful, though, to have a map with the sea-ice concentration of, say, the first week of March 2018 (noting that several of your samples are dated around this period) superposed with the lat/lon boxes from where you obtained the data for the FYI and MYI samples. You could use such a map also to illustrate the regions that you apparently used to define freeze-up day and hence the date from which onwards a grid cell is defined MYI.

- I note that YI samples are chosen across the entire freezing season and also FYI samples originate from three different time periods, increasing the likelihood to cover the potential range of surface conditions. Only for MYI you restricted the samples to March which I believe is sub-optimal. Particularly in the Weddell Sea Sentinel-1 SAR imagery allows to clearly separate MYI from FYI also later in the season (just take a look at the SAR images provided regularly by the Polarview consortium). This is a lack of rigorosity in the selection process of the MYI samples. I note in addition that you have substantially fewer MYI samples (about 1000) than other samples (FYI about 6000, OW about 14 000, YI about 2000). Given the fact that in the interpretation of your results you focus quite a bit on the MYI cover distribution this seems to be a sub-optimal and not sufficiently well-thought through way to generate a data set of samples to be used. MYI signatures can vary a lot during winter and particularly in March they might not be representative for conditions met later at all.

L233/234: I am glad that you put your observations into the context of values published in the literature. I am wondering however, whether you are fine with the fact that for FYI the GR3719 is well below zero ... which is indicative of what? Figuring this out will aid you in your interpretation / discussion of your results which is not yet convincing.

L234-238: "The MYI tie points ... distributions." --> Your "well above" is something like 5 - 8 K which is not that impressive. Your observation that the modes of your MYI TB distributions do not match well with MYI tie points in the published literature could very likely be caused by the fact that you selected cases from March - a time of the freezing season where the MYI signature might not be representative of winter conditions. You could take this observation as a hint that your cases to derive MYI TB and sigma_0 samples are not well chosen.

- The comment you make in the last sentence does not really help here since Arctic MYI differs substantially in its physical and hence microwave structure from Antarctic MYI ice - simply because the melting process differs fundamentally. I therefore suggest to delete that last sentence.

L238/239: You state that your C-Band NRCS values for Antarctic MYI are higher compared to what is observed in the Arctic. Did you check with results from the Antarctic? Haas (Annals of Glaciology, 33, 2001) reported C-Band NRCS values of Antarctic perennial sea ice of -16.3 dB compared to about -10.7 dB during summer. Gohin (International J. Remote Sensing, 16(11), 1995) reported winter-time NRCS values of MYI of -14 dB. Finally, looking into Arndt and Haas (The Cryosphere, 13(7), 2019) I find C-Band NRCS value time series for FYI and MYI. In short: There is a lot of literature you could have taken a look into to illustrate how well (or not well) your MYI NRCS values agree with published literature in the Antarctic; there is no need to look into the Arctic.

L242: "the only source ..." --> How about NCEP/NCAR, MERRA-2, or JRA-55 re-analysis data? Are these not comprehensive and consistent as well?

L248: "Sea ice drift data ..." --> why is that? Please provide a reference which backs up this notion.

Figure 2: The color in the legend does not match with the description given in the caption and in the text. This has to be corrected. I guess you switched YI and MYI.

- Please refer to appendix A and the relevant tables of the regions used to define these distributions and also provide the total sample numbers of the four different surface types shown.

L285/286: "and the ice is more porous because ... scattering." --> Please cross-check this statement with the paper by Haas and Arndt and Haas I mentioned further up in this 2nd review of your paper.

L286: "we manually scaled ... classes." --> just a comment, no action required: This is another element of your work / paper which is not sufficiently transparent such that an interested student can redo the analysis the same way as you did it.

L287: "19 Sentinel-1 scenes" --> "Sentinel-1 SAR scenes"

- You must provide the source of this data.

- You write "in the Weddell Sea and near the Antarctic Peninsula". This applies to me that you used SAR images covering the Weddell Sea but also parts of the Bellingshausen Sea? ... As stated later on, a map would solve this lack of information (or ambiguity of the information provided).

Figure 3, L296/297: "Qualitatively, the ... is similarly good." --> What keeps you from showing a map into which you plot all 19 SAR images and superpose the ice type isolines as you did in Figure 3? One could get the impression that you picked the best example here and that in fact the agreement is less (even qualitatively) good for the remaining 18 images - simply because these might not focus that well on the wonderful MYI-FYI boundary shown in Figure 3. I know, it is just a proof-of-concept study and for this I do not expect a quantitative analysis of the SAR images used anymore and I do not expect something like a quantification of areas of misclassification (or good classification). But what I can expect is that you provide the reader with a sufficient amount of information that allows the reader to believe in the credibility of your results. In your case this means: show all SAR images (If you are in doubt of creating too many figures then --> merge Figs. 13 to 15 into one figure).

L305-307: However, no ..." --> Just a comment: This makes a lot of sense because there are places around the Antarctic coastline where sea ice advance kicks in as late as July.

Figure 6 / L322-324: I commented on that in my 1st review and do not see it solved here: The color scale used in Figure 6 implies that the total sea ice concentration for the majority of the areas downstream of the Filchner-Roenne Ice Shelf is less than 100%; the color suggests something between 80% and 90% with higher FYI concentrations occurring actually rather at the boundaries of the FYI areas than within the areas themselves. Therefore, please tell the reader that the partial concentrations shown in Figure 6 do not need to sum up to 100% (even though I don't understand why.). Otherwise the reader sees that there are areas where MYI concentration = 0%, YI concentration = 0% and FYI concentration = 80% in regions where the SoD maps clearly indicate 100% and would at least have doubts about your product. It is pretty clear that in these areas OW concentration cannot be 20% (and if so, then your algorithm is not producing correct results here).

L330/331: "The FYI areas ..." to 70%" --> I am not agreeing completely here because particularly in those regions where we have these strips FYI concentrations are often as high as 100%.

L354: "no areas of more than 40% YI concentration ..." --> Since you refer to YI concentrations in the northeastern part of the region shown in the previous example I suggest to detail this statement a bit better here.

L359-361: Is this rather global statement at the beginning of this short paragraph relevant and correct? I would say, how large icebergs are classified depends on whether their radiometric signature resembles that of MYI or any of the other ice types and this signature changes with the season. Your results do not support that icebergs are generally classified as MYI (compare Fig. 3 and 4 as well as 6).

L364: "increase around July" --> In view of Fig. 11 this is clearly an underestimation of the situation. If one would compute a running 30-day mean of the time series shown in Figure 11 one would get an upswing of the MYI area that begins in May/June and ends in August before again an increase in MYI area kicks in during October. I recommend to state the situation in a light that is more in line with what the figures shows.

- Did you check the minimum sea ice area at the end of summer in Feb/March 2018? Is your initial MYI area in line with the minimum sea ice area?

- Before you begin with explanations what might cause this increase during winter I recommend to first briefly explain how you expect this curve should look. I am wondering in this context whether the intial decrease of the MYI area during March and April is not far too large and/or rapid.

- Finally, a piece of information that is (still) clearly missing is how accurate and credible your intitial definition of MYI is. You did not come up with this and your footnote 2 a few pages up is not sufficient to understand how you regionally defined the starting point for your MYI occurrence. From what you wrote I can only speculate that the dates from which onwards ice present in a particular region is classified as MYI differs between the regions and that this could have had an impact on how Fig. 11 looks.

L377/378: "Usually, snow ..." --> Please be more specific about the processes that can occur during winter on sea ice relatively close to the ice edge. It is deformation, increasing roughness. It is snow thickness increase due to snow accumulation, resulting in a decrease of the GR3719 towards values that resemble MYI (but I note that your FYI samples exhibit a quite negative GR3719 anyways ... unlike in the Arctic, see your Figure 2). It is snow metamorphism due to air-mass change induced melt-refreeze cycles. It is flooding of the ice-snow interface with formation of slush at the basal snow layer which eventually refreezes to become snow-ice - an ice type widespread in the Antarctic which radiometric properties not well studied.

Figure 12: Also for these maps you must detail why the sum of the ice type concentrations does not need to sum up to 100% but can have any arbitrary value between - say 80% and 110 or even 120% - for areas with 100% sea ice concentration.

L379/380: What about the emissivity of pancake ice? Backscatter is just one out of the four input parameters.

- Also, I note from your sample maps of the YI distribution in Fig. 4 but also here in Fig. 12, that there is a fringe of elevated YI concentrations along the ice egde being indicative of the presence of pancake ice. It won't be any of the typical sheet ice types such as nilas or grey ice - these ice types are only interspersed in pocket like structures in the otherwise dominant pancake ice cover (see e.g. Ozsy-Cicek et al. 2011, Annals of Glaciol). Hence I would guess that your approach deals reasonably well with pancake ice.

L393: The third reason can simply be that your correction approaches do not work the same way as they work in the Arctic because the processes that change sea ice and snow physical and hence microwave properties are more vigorous and different from those in the Arctic. Also their impacts could either be lasting longer or, what is possibly more probably, their frequency is too high to capture them properly with your correction method.

L397/398: "The fact that ... polynyas" --> see my comment above about that your YI concentration maps seem to resemble the pancake ice areas well.

L400-407: I doubt that in the weekly SoD charts of September and October there is an oscillation of areas classified as FYI or MYI in the Antarctic. If so, then you should perhaps consider not to use the SoD charts at all. I have doubts about this oscillation because these SoD charts are always analysed taking into account the distribution of SoD from the previous week, including knowledge about the past ice conditions and their persistence. In a way, your statement also misaligns with your earlier notion that SoD charts are developed by well-trained, well-experiences analysts.

- Both, September and October are months of freezing conditions. Melt conditions occur, if at all, still episodic and in association with warm air advection ahead of low pressure systems - at it does all year / winter. The same applies to some extent to November. In my eyes, the most likely reason for the misclassification is that your MYI cases used to train ECICE are from March only and very likely do not reflect the deeper, more layered snow pack conditions encountered in September / October - the months where also the NASA-Team algorithm has severe difficulties to correctly retrieve the sea ice concentration of Antarctic sea ice - because of the layering of the snow (see Comiso et al., 1997 in Remote Sensing of Environment, a paper that initiated the development of the NT2 algorithm to get rid of the substantial biases). In addition to the samples used for MYI you might also consider that the FYI samples might not represent the deeper snow pack conditions often encountered in late winter / spring - aka September / October.

- The strong decrease in MYI area shown in Fig. 11 during September / October could also simply be the rebound to "normal" conditions with all the artificial MYI area discussed in the context of Fig. 12 vanishing. What puzzles me more is the increase in MYI towards spring, i.e. October / November.

- In conclusion, I strongly recommend to revisit your strategy to choose sampling location and their time periods, take a look into publications that describe what we know about the physical processes of Antarctic sea ice and snow during the course of winter and how these influence AMW and PMW properties of the sea ice; I recommend to again read Voss et al., 2003; Arndt et al., 2016; Haas et al 2001; Arndt and Haas, 2019; Willmes et al., 2014, Willmes et al. 2011, and others and based on what you learn from studying these publications critically discuss what might be wrong in ECICE as used in your manuscript. As you stated 1-2 paragraphs further up in the context of Fig. 12: It is not the correction which partly jeopardize your results but the complex nature of the Antarctic sea ice and its snow cover that is not adequately taken into account into "tuning" ECICE for Antarctic conditions by simply using local TB and NRCS distributions.

L418: "A first check is ... total sea ice" --> Certainly this is a good idea - however, the examples you present in your manuscript do not credibly support such an approach since - at least from visiual inspection of the maps shown - even for 100% sea ice cover the ice type fractions do not add up to a total sea ice concentration of 100%. This needs to be clarified better / corrected before you can consider to carry out the step described in this line.

L421/422: I don't understand what you mean by "as the standard one" and it certainly does not surprize you that I suggest to add to the plot at least one time series of a sea-ice concentration product based sea ice extent, e.g. OSISAF (not the one from Tonboe et al., 2016 but the newer one from Lavergne et al., 2019, i.e. OSI-450 / OSI-430-b).

L425: You sum up these areas using all non-zero concentrations, am I correct? What I mean by this is that you do not use a 15% threshold.

L429: "mainly caused by ..." --> I suggest to re-formulate this sentence because it is very evident in your manuscript that this correction is most likely not the main reason but the fact that your choice of sampling areas and periods for MYI but potentially also for FYI is not representative enough for the full range of conditions encountered during the course of the winter. I suggest not to blame the correction at first place.

L431/432 / Fig. 15 (and also Fig. 14): What is the minimum sea-ice area of the respective years?

- A "." is missing at the end of the sentence.

L452/453: "The data become unstable ... underestimated." --> Why is that? Shouldn't the MYI area in September / October be smaller than the MYI area you begin with in March? I'd say that a substantial amount of the MYI found in the Weddell Sea end of summer has melted by then in the Southern Ocean facing the Atlantic and should continue to melt in November. Hence the low September / October MYI area does not seem unrealistic to me but the increase in MYI in November does.

L459-462: May I propose a slightly corrected version of this? For (1) I clearly see the need to incorporate also again MYI into the revision of the distributions. Step (2) should be a thorough evaluation of these TB and NRCS distributions with values available in the literature. Step (3) should be a quantitative evaluation of the ECICE results before any correction with a considerably enhanced and better analyzed set of evaluation data which may or may not contain in-situ and ship data (I would say the latter are most useful for the identification of pancake ice locations). Then, step (4) could be an application of the corrections with more up-to-date data. Step (5) would be a final evaluation. Only step (6) can then be an extension of the current data product using AMSR-E and respective AMW sensors plus - again - an evaluation of the extended product.

Typos / Editoral comments:
L43/44: "In the absense of ... still unknown" --> Are you referring to different surface topography and hence impulse fluxes or to different thicknesses of sea ice and its snow cover and hence heat fluxes?

L47: "and thus" --> "and the ice-snow interface thus"
And further: "... flooded. The slush layer resulting from flooding of the basal snow layer with sea water eventually refreezes, becoming snow ice."

L50/51: "to higher wind that triggers" --> "to more variable wind conditions that cause"

L98-103: Please stress more that the following two sections, 2.1 and 2.2, are valid for the ECICE as developed in the Arctic - including the two correction schemes.

L302: "and reconnaissance" --> is this air reconnaissance? Then please add this important detail.

L312: "plus correction schemes" can be deleted as you write "corrected MYI concentration"

L332: "one sea ice type class" --> add "such as in the SoD charts"

L333: "retrieval where" --> "retrieval such as ECICE where"

L358: "given the limitation ... yet." --> I suggest to delete this part of the sentence; it does not add relevant information.

L364: "form" --> "from"

Figure 13: You can remove the title. Both, the y-axis title as well as the legend state what is shown.
A "." is missing at the end of the caption.

Figure 14: As in Fig. 13 you can remove the title.

- You could also plot the total area and compare it with the total area derived from the OSI-450 / OSI-430-b product. This would lend the figure more credibility.

Figure 15: The figure is cropped at the left side. Like in the previous two figures the title can be removed.

L434: What is "maxima of total ice" --> please be more specific in your wording. What you mean are "maxima of the total sea ice concentration".

L455: "i" --> "in"

L457-459: This sentence should be merged with the next paragraph with (currently) point (4).

L468/469: One mentioning of "at 1.4 GHz" seems enough here.

Hide

ED: Reconsider after major revisions (further review by editor and referees) (17 Aug 2022) by Yevgeny Aksenov

AR by Christian Melsheimer on behalf of the Authors (12 Oct 2022) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (28 Nov 2022) by Yevgeny Aksenov

RR by Anonymous Referee #1 (29 Nov 2022)

ED: Publish as is (29 Nov 2022) by Yevgeny Aksenov

AR by Christian Melsheimer on behalf of the Authors (08 Dec 2022)

Short summary

It is necessary to know the type of Antarctic sea ice present – first-year ice (grown in one season) or multiyear ice (survived one summer melt) – to understand and model its evolution, as the ice types behave and react differently. We have adapted and extended an existing method (originally for the Arctic), and now, for the first time, daily maps of Antarctic sea ice types can be derived from microwave satellite data. This will allow a new data set from 2002 well into the future to be built.