Comparison of Arctic Sea Ice Concentrations from the NASA Team, ASI, and VASIA2 Algorithms with Summer and Winter Ship Data

Round 1

Reviewer 1 Report

This paper presents correlation of sea ice concentration data between satellite-borne microwave radiometer observation and ship-borne in-situ observation.

This is a very informative paper and worthy of acceptance.

Minor editorial comments:

Page 5 Line 195: “tg” in the right end should be removed.

Page 5 Line 199: The explanation of X and Y is opposite. Also, suffix position of T-XY should be corrected.

Author Response

The authors are very grateful to reviewer for positive estimation of the manuscript and for attention to details. We corrected these parts of the text according to the comments.

Minor editorial comments:

Point 1: Page 5 Line 195: “tg” in the right end should be removed.

Response 1: We removed second “tg”.

Point 2: Page 5 Line 199: The explanation of X and Y is opposite. Also, suffix position of T-XY should be corrected.

Response 2: Suffix position was changed.

Sincerely,

Tatiana Alekseeva and Vasiliy Tikhonov and Co-authors

Reviewer 2 Report

Review of

Comparison of Arctic sea ice concentrations from the NASA Team, ASI and VASIA2 algorithms with summer and winter ship data

by

Alekseeva et al.

Summary: The paper deals with a mostly carefully planned, designed and conducted inter-comparison of satellite-based sea-ice concentrations with ship-based observations of the sea-ice cover, including melt stage. One novelty of this paper is the seemingly very valuable ship-based observations data set which in terms of spatial coverage outperforms currently used similar data sets. The second novelty of this paper: among the satellite sea-ice concentrations compared there are those of a fundamentally different algorithm which is purely based on physical modeling and does not require tie points (albeit it involves a lot of empirical formulas). The paper discusses results of the intercomparison separately for summer and winter using data from multiple years. The illustration and discussion include consideration of thin ice as well as melt stage on sea ice.

This is a re-submission of manuscript 483784 which I reviewed in April 2019.
I find the material well worth to be published but there is no doubt that the paper (still) needs substantial revisions before it can be published. I recommend major revisions.

GC1: I don't find a notion about how SIC from the different grids (polar-stereographic for NT and ASI, unknown for VASIA2) are co-located properly to the ship-based data. Since the grid-cell size of the polar-stereographic depends on latitude one needs to first convert the polar-stereographic data into cartesian coordinates to be able to carry out the co-location and the computation of the areal fraction of the ship-based observation corridor in the satellite pixels. This information should be added.

GC2: The description of both, the data on which the VASIA2 algorithm is computed from, and the algorithm details require further details to be understood correctly. See my specific comments.

GC3: A wonderful degree of detail is given for the ship-observations. Still, some information is lacking and seems also to be partly wrong. See my specific comments.

GC4: It appears to me that some additional analysis with regard to the actually observed thin ice fraction would enhance the papers' message. See my specific comments on that issue

GC5: While the figures have partly been described in sufficient and convincing detail there are others which deserve more attention, e.g. Figure 10 and 13.

GC6: The authors have an extremely nice and seemingly quite reliable ship-based observation data set at hand which they in addition handled with sufficient care to not mis-interprete their results. This should be highlighted more in this paper and, if possible, this paper could even serve as the platform to make these observations publicly available. I'd love to use this data myself.

Line 48-50: I suggest to re-write this statement. It is too harsh and even torpeds your own paper. This is certainly not a rule! Sea-ice parameter retrieval from satellite observations is over-all quite mature. Yes, I agree, one CAN get significant errors and yes, OFTEN ship-based and satellite-based observations do not agree favorably well. But it is often a question: What are we after. Passive microwave sea-ice concentration (and related parameters area and extent) are for large-scale / climate purposes, under ideal conditions also for meso-scale applications. But they cannot be used to aid navigation or prediction of ice load on marine infrastructures in the polar seas. This already makes clear, that any inter-comparison between ship-based and passive-microwave-data-based sea-ice concentration estimates cannot be ideal and needs always to be seen with a grain of salt.

Line 60-65: This paragraph reads as if weather filters were developed to improve algorithms' skills over sea ice. This is not the case. Weather filters were developed to further (!) mitigate noise caused be weather effects over open water which is not already dampened enough by a clever choice of channels and open water tie points. Please rewrite this paragraph accordingly.

Line 72: Yes, comparison with independent data, such as, for example, ship-based observations COULD help to explain differences in sea-ice area and extent derived by different algorithms. Please relax your statement accordingly and to not over-value ship-observations.

Line 76: I guess you can write [19-23] here, like you did further up in the paper. You might also want to check this paper in "The Cryosphere Discussions: https://doi.org/10.5194/tc-2019-120 where some work is dedicated to a comparison to ship-based observations.

Line 164-176: I doubt that mentioning how many versions of the ASI algorithm exist / did exist is helping for the understanding here. Wouldn't it be enough to state the current version? Please also consider deleting reference [33]. It is in German and might not be relevant for the current version of this algorithm - or, in other words, the paper by Spreen et al. (see [21]) contains the latest description of this algorithm. Even though [21] is for AMSR-E the description of the algorithm is the same.

Line 177/178: You seem not to use AMSR-E or AMSR2 data; therefore you can delete the sentence starting with "Only in the case ..."

Line 190: The main details and secrets of this algorithm are given in the [26] paper. This is very good, even though the agreement between the model and the SSMI measurements seems not optimal, particularly for multiyear ice. Still, particularly because this algorithm is fundamentally different from the others, I strongly suggest to write a few more lines about the motivation and type of this algorithms so that the reader gets a better feeling about what it is about. Please consider that not everybody reading your paper will also check [26] to learn what kind of an algorithm this is. At least the motivation to why you use tangents and what is the physical explanation behind this. Or, in other words, what is the physical background between the sea-ice concentration and the used tangents? I guess, only with this additional few lines of information the very detailed list of empirical equations given in section 2.3 will receive the credibility they deserve. I can imagine that one additional figure showing a sample brightness temperature distribution of one of the channel pairs, for example 85.5 GHz vs. 37.0 GHz, both vertical polarization, plotted for a typical winter case and a typical summer case from data of the entire Arctic would greatly assist in the understanding of your method.

Line 209/210: "The ranges of these characteristics were selected ..." --> This reads as if you included properties of snow which were also derived on land and/or glaciers. I'd say there is enough results about the parameters mentioned in lines 208/209 from observations over sea ice.
[44] Is in Russian and should be replaced by an English reference; [45] is about properties on the ice sheet and not on the sea ice. If it comes to snow the two most well-known English-language works are from Colbeck and from Sturm.

Lines 211-14: I have difficulties to understand these sentences. First of all I doubt that snow in winter contains liquid water at all. Secondly, once melt commences liquid water in the snow certainly starts at 0% and not at 9% as is implied by the "9-30%" notion. I am sure also, that in the many references you cited you will find a statement about when a snow cover is termed as being saturated with liquid water and is hence termed slush. Third, I don't understand what you mean by "with a wetness from 30% to melt pond)"

L217: I doubt the depth of melt ponds is required here. The penetration depth (or the equivalent of it) into liquid water is of the order of a few millimetres at the frequencies used in this paper. The depth of the melt pond therefore does not matter.

L219-220: This is a very interesting observation given that 85 GHz data are much more sensitive to surface property changes on the sea ice surface and more sensitive to atmospheric moisture, while 19 GHz data are more sensitive to internal processes within the snow or topmost ice layers.

L221: The open area (1) contains bare ice, ice with a dry snow cover and ice with a wet snow cover while the shaded area (2) contains sea ice with a very wet snow cover and melt ponds? Please clarify in the manuscript.
Please also note that regions (1) and (2) are given in Fig 1 C and not in Fig 1 A,B which you are referring to.

L223: Each of the regions "was averaged" ... While I can understand this for the open region (1) because its borders are almost parallel I have difficulties to understand this for region 2 which is triangle shaped and this "averaging" averages of a far larger range of "value of a tangent" at high than at low SIC. Please comment in your text about this step which seems to involve some additional assumptions?

L224: You mention linear dependences (=dependencies) and write that these are the dotted lines in Fig. 1 (A,B). However, the caption of Fig. 1 A,B says that these lines denote the medians of the open region. Please clarify and/or correct the text accordingly.

L233/234: What explains physically that tg(37-19 V) is so much more dependent on the presence of liquid water on the ice surface than the other two tangents? Please explain with the aid of typical emissivities.

L237: the "lower boundary of region 1" --> I am very puzzled. According to your description region 2 (!) is the region with SWM on ice. Here you write that "If the value of tg(37-19 V) at a certain SIC lies below this boundary [the lower boundary of region 1], then SWM is present on ice." This completely contradicts each other and needs to be clarified in the text. Possibly you wanted to refer to the UPPER boundary of region 1 and ABOVE this boundary.

L270-275: You obtained readily computed NT SIC but computed ASI SIC on your own. Is this correct? In this case, please provide a more detailed source for the brightness temperatures used and tell us which tie points you used. If this is not correct, then please provide the source from which you obtained the ASI SIC data.

L276-315: I have difficulties to understand the relation between the data source cited in L281 and the POLE-RT-Fields data base. The data source cited in L281 was created in April 2019. It seems hence much more novel than the POLE-RT-Fields which with reference [55] date back to 2007. In other words, the data cited with the reference in L281 seem not to be the data the system cited by [55] is based. This has to be clarified.
In addition, the data cited in L281 have all you need: These are daily gridded swath-based (either all ascending or all descending overpasses gridded together) data sets with the time of each overpass preserved. Grid resolution is 25 km for the lower and 12.5 km for the higher frequency channels. The grid is the equal area EASE grid, simplifying investigations involving metric distances.
Could it be, that there is this system cited by [55] but you have problems to trace back the various SSM/I and SSMIS data sources used so far in a transparent way?
- You are describing the POLE-RT-Fields in great detail. While this is very good and is a good advertizement of this freely (?) available data set I encourage you to shorten this part by deleting elements of the information given which are not relevant for this paper.
- Please provide also the information on which grid the VASIA2 SIC data are computed and on which grids the other two data sets (NT and ASI) are provided. My guess is that the latter two are provided on polar-steraographic grid which is not a true-area grid and hence the grid-cell dimensions and areas change with latitude.

L318-337 / Appendix A
- L321-322: There are three observers on the bridge all the time? Or: Continuous 24-hours sea-ice observations are carried out by a team of three observers taking 4-hours shift so that at least one observer is at the bridge all the time. ?
- Is there any material on board the bridge which helps the observer to estimate total ice concentration with 5% accuracy? Without I doubt that especially in the medium ice concentration range between about 30% and 70% observations are as accurate as 5%.
- Definition of homogeneous ice zones is, among changes in ice concentration and changes in surface parameters, depending on the ice thickness. How is thickness estimated according to the protocol mentioned? By the same token: How is snow depth estimated with an accuracy of 1 centimeter?
- I note that the melt stage is recorded but not the melt-pond fraction on the sea ice. Is this correct?
- I note that there is not distinction between ridges and hummocks. Is this correct?
- What is "repeatability of ice thickness"? (in L720)

L359/360: "in the region of ship motion" --> please try to use the same term for this throughout the paper. We have "along the sailing track" in L329/330 and "along the ship track" in L707.

L361: "may differ" --> "differ" ; there is no doubt that the size of the observation "footprint" of the the ship-based observations along the sailing track is substantially smaller than the satellite footprint.

L371: I suggest to add to the caption: "Note that discretization of the ship track and pixel size are not the scale; this is a schematic illustration."

L375-377: While this is an interesting information it seems not to be required as you stated above (L359-360) that "total SIC in the region of ship motion" was used. However, further reading of your manuscript unravels that you used the observations of the navigation area (e.g. Figure 5)

Figure 4: How is the visibility estimated? Is this an instrument such as a ceilometer, i.e. a laser device? Or this is estimated by eye? I am a bit surprised that the fraction of visibility < 0.5 km is so low.
I note, by the way, that the respective panel of the figure in the Appendix provides visibility with the unit "nm = nautical miles" while in Figure 4 you used kilometers. Can you provide the formula used to change the units?

L389: "25km x 25km" and "70-80%" coverage does not fit. Please revise.

Figure 5 and its interpretation:
- I suggest to add the information of the theoretical maximum coverage of a 25km x 25km and of a 12.5km x 12.5km grid cell by the ship-based observations.
- Figure 4 clearly showed that a substantial fraction of the ship tracks used in this paper suffered from visibilities < 5 km. Could it be that your choice of discarding all co-located ship-based observations with less than 40% coverage of the respective grid cell is also partly motivated by this limited visibility constraint? If so, please add respective text in your paper. In other words, the fractions shown in Figure 5 assumes 5 km or better visibility throughout the entire data set - which is not the case ... which is a good argument for the 40% threshold used.

L417/418: I suggest to delete "that are currently remain practically beyond reach of remote sensing techniques" because there has been a great deal of progress into this direction with both satellite and air-borne sensors. It is sufficient to mention that ship-based observations provide additional parameters.

L420-429: This is a good example for sure. However, the MODIS image clearly shows that the sea ice in the area where NT and ASI fail is possibly not the compact high-melt pond fraction ice of the central Arctic but highly-fragmented, brash-type sea ice in the marginal ice zone. While melt ponds may play a role here as well it could also simply be the wet surface. In addition, in the marginal ice zone weather filters are applied removing an unknown amount of ice.

L432-433: "the combination of channels is 221: 1 - visible, 2 - near infrared" --> This is not clear enough. Apparently this is an RGB-image and MODIS channels are assigned to color channels. It would be more useful, however, to explicitly mention which MODIS bands these are. Could be 2,3,4, ...

L440-441: "reveal the main regularities in the SIC retrieval" --> I don't understand. What do you mean? Perhaps you can skip this sentence and simply make the point that you now show results of the inter-comparison between SIC_SO and SIC_SMR?

Figure 7 to 9 and their interpretation:
- I did not find a notion that VASIA2 SIC is computed at 12.5 km grid resolution - which seems to be the case comparing the pixel numbers for ASI and VASIA2; please add this information to the respective section.
- While I am ok with you working with SIC in tenths in the figures, I strongly recommend to carry out the discussion in "normal" sea-ice concentration units with is fraction per 100 or fraction per 1. In most publications you will find fraction per 100. In order to have this paper been taken up properly by, e.g., the modelling community, I strongly recommend to switch from tenths or decimal fractions of tenths to fraction per 100, i.e. percent. That way NT would overestimate SIC for very open ice conditions (10-30%) by 10-14%, for instance. Since "we are in the percent world" with SIC there is also no danger that this is mistaken as the relative difference by readers.
- I suggest to add a notion about the relative contribution of the SIC-ranges used in L458-468 to the total number of pixels compared to stress how dominant the SIC ranges 0-10% and 90-100% are.
- I suggest to state in the context of L458-468 also that you first focus on the comparison to all ship-based observations, i.e. not those where the SIC by thin ice is subtracted. Alternatively, instead of explicitly stating this at the beginning of sections 3.1 and 3.2 you could add this information (i.e. the link to these sections) in L443.
- I note that you include very close and compact ice in your discussion for the summer season but not for the winter season. Please explain or change.
- L502-511 & Figure 11: I suggest to put this information into section 2; it belongs to the data description and it should be stated upfront that during late summer (mid August/September) thin ice is often found in the Arctic
- L500/501: I don't understand what you mean by "unreasonable to exclude the new ice and nilas from ...". Neither NT nor ASI exclude new ice or nilas ... and I don't think VASIA2 does? Please clarify in the paper what you mean.
- L513: I am inclined to say that this is even an effect for ice thicker than one wavelength of the radiation used. Ivanova et al. (2015) has one figure illustrating that thin ice with thicknesses up to 15 cm or even 20% can cause negative biases in SIC_smr. Please rephrase this statement accordingly.
-L516: Could this close-to-zero difference in the algorithm skills for very open and open ice during summer between SIC_co and SIC_so-ni also be caused by the fact that potentially it is more likely to observe new and thin ice in the vicinity of existing older / thicker / colder ice than it is to observe new and thin ice in the open ocean or when the fraction of the older / thicker / colder ice is quite low? Fig. 7-9 do not give any information about whether there have been fraction of new/thin ice among the SIC_so observations at all.
- L527-533: I have difficulties to understand the unexpectedly large differences during winter between a) SIC_so and SIC_smr and b) between the comparisons with SIC_so and SIC_so-ni. In order to have a change in the difference (error) between SIC_smr minus SIC_so and SIC_smr minus SIC_so-ni of 20% to 30% there has to be a very large fraction of thin ice in the corridor of ship-based observations ... possibly even 40 to 50% of thin ice. Can you confirm this in your ship-based observations? I mean, it totally makes sense if the ships were following leads most of the time. I (and the reader) would appreciate to have this issue explained in the paper. Would you could do to solve this riddle is to plot the difference between blue and green bars together with the fraction of thin ice (I'd recommend thickness < 20 cm) from the respective ship-based observations.
Assuming that my idea about the cause for the large step between blue and green bars is true, it would be clear that the majority of the differences indicated by the blue bars is caused by the thin ice, which has been shown in the past to result in negatively biased SIC_smr values (see e.g. Ivanova et al., 2015). It would at the same time point out that VASIA2 has superior skill compared to the other two algorithms.

Figures 10 & 13 and their interpretation:
- I note that there is almost no discussion of these two figures in the text. What I find worth noting is that VASIA2 has a lot more 100% and 0% values (than the other two products) coinciding with SIC_so and SIC_so-ni values spanning over almost the entire SIC range. This applies to summer but for 100% also to winter. This should be explained / discussed.
- I note further that for winter, in both these figures, SIC_smr does range between 0 and 100% a SIC_so or SIC_so-ni = 0. How do you explain such occasions? Most of these are actually observed for ASI, followed by VASIA2 and then NT.
- I also recommend to stress in the discussion of these figures that the majority of the data points is in fact in the SIC_so ranges 0-10% and 90-100% (compare figures 7 to 9). It is important to note that many of the data points shown in figures 10 and 13 are overplottet on top of each other. Only usage of a 2-dimensional histogram where the probability of a certain SIC_smr - SIC_so data pair is color-coded would solve this issue.
- Finally, I recognized that in figures 10 and 13 you wrote the squared correlation while in table 2 it the non-squared one. This could be harmonized.

L520: "correlation graph" --> I suggest to term this scatterplot.

L579: [36,37] --> It might be good to have a bit less of self-citations here but, e.g., also cite the work of people outside of Russia: Comiso and Kwok, 1996, Surface and radiative characteristics of the summer Arctic sea ice cover from multisensor satellite observations, J. Geophysical Research, 101(C12) or perhaps also Kern et al., 2016, The impact of melt ponds on summertime microwave brightness temperatures and sea-ice concentrations, The Cryosphere, 10.

Figures 14 to 16:
- Any reason why you use green shading in Fig. 14 and blue shading in the other two figures?
- Would it be a burden to include the mean values given in Table 3 in form of vertical black lines at the respective melt state in Figures 14 to 16? That way one could relate the average difference SIC_smr minus SIC_so better to the count given.

L587-589: What do you want to state here? Do you want to state that a SMR algorithm should have a larger error at more advanced melt-stage and/or melt-pond fraction? Theoretically perhaps yes, but but the algorithm developers not; they want to have 100% sea ice no matter how the surface looks like. So, I am not sure I properly understand your statement here. One might also ask why then the maximum underestimation is not occurring for melt stage 5?

L608/609: I guess you can delete the sentence "It is worth noting that positive errors over 1 tenth disappear, meaning no large SIC overestimations by microwave radiometry data occur in the presence of melt ponds." It seems mathematically impossible that for compact ice, when SIC_so is > 95% there is an overestimation by SIC_smr larger than 5% - unless you allow algorithms to retrieve SIC > 100% (which for some makes sense from the point the retrieval works even though a grid cell cannot contain more than 100% sea ice.
Instead of this sentence it might be a good idea to comment on the fact that for compact ice (as seen by SIC_so) VASIA2 is the only one of the three algorithms which exhibits mostly positive differences while both ASI and NT exhibit considerable counts on the negative differences side.

L624-633:
- [21] makes a clear point about why SIC_smr is larger than SIC_so in summer but not in winter. It has to do with the fact that i) leads are open in summer but closed in winter and ii) ships tend to follow leads, and iii) those ship-based observations are different in nature from those used here. This paragraph is ideally suited to underline the advantage of the AARI-type ship-based sea ice observations over the "classical" ASPeCt type of a circular area around the ship with 1 km radius. With your method you cover an area which is more representative to the satellite grid cell - provided that the visibility is sufficient and the fraction of the corridor observed by the ship within the satellite grid cell is > 40%. Please rewrite this paragraph, including the notion of [21] about the reason for the different signs of the difference and stressing your superior type of data.

L643-648: I suggest to rewrite this paragraph. For the summer part you might want to mention the footprint of the SMR sensors, which you mentioned further up as one of the reasons for SIC under-estimation in presense of thin ice. For the winter part you should take into account my comments regarding the blue bar to green bar shift and my suggestion to plot this along with the actually observed fraction of thin ice.

L655-659: I would not paint this as black and white as you did here because VASIA2 outperforms the other two algorithms seemingly during summer for compact ice (if one wants that the SIC is 100% no matter how many melt ponds are on top). Furthermore, during winter VASIA2 provides the smallest differences to SIC_so for the range 6-7 tenths to 9-10 tenths. I believe this is worth mentioning and you should not hide the positive things besides the negative things.

Appendix A

Typos / editoral comments:

L81/82: ... Appendix 1 ... check usage, because the appendix itself is termed "Appendix A"

L195/196: One "tg" too much.

L199: The sub- and superscripts description of what X and what Y is does not match with what is given in the formulas.

L350-351: "continual dedicated" --> "continuous"

L470: "either" ... "or" --> "neither" ... "nor"

Table 2: "12,5 km" --> "12.5 km"

L564 / L605: "charts" --> "panels"

L640/641: "SSMR" --> "SMMR"

L736 and L762: "3.5" --> "2.5"

References:
[25] Please add the year and a doi
[55] Is in Russian. Please consider changing that.

Author Response

The authors are very grateful to reviewer for comments and attention to details. We answered to all comments and marked changes in the manuscript by blue color, please see the attachment and the manuscript. Figures 1, 7-9 and Figure 1 to Apendix A were changed and attached.  

Sincerely,

Tatiana Alekseeva and Vasiliy Tikhonov and Co-authors

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This paper presents correlation of sea ice concentration data between satellite-borne microwave radiometer observation and ship-borne in-situ observation. Compared to the similar previous works, the discussions on the effects of initial ice and ice melting stage are interesting and useful information. I judge that this paper is acceptable with major revision.

Major comments:

1. I cannot understand why the authors omitted bootstrap algorithm, which is also widely used. The authors should comment something on this matter. Hopefully, they will add the comparison with bootstrap algorithm.

2. Define “summer” and “winter”. Are they defined simply by months, or by ice melting/production stages, e.g. by ship-borne observations?

Minor comments:

Page 2 Line 65: “Paper [17]” is a little unusual expression in a scientific paper. “Ivanova et. al. [17]” is more usual, I think.

Reviewer 2 Report

Review of manuscript remotesensing-483784, titled “Comparison of Arctic Sea Ice Concentrations from 2 NASA Team, ASI and VASIA2”

The manuscript addresses the common theme of comparison between three sea ice concentration algorithms, yet with uncommon validation set of ship observations. It also presents evaluation of the algorithms for different categories of concentration; ranging from open water through a gradation of concentrations such as closed, very closed and compact ice cover. This is important, needed and not usually done. Summer and winter evaluations are addressed separately. The manuscript also added VASLIA2 algorithm (which I never heard of).

While the study is relevant for this journal and would make a good contribution for sea ice concentration research, at least because it uses a unique data set for validation, it suffers from a number of deficiencies that make the paper not suitable for publication in the present form. The presentation style jumps between different points with no coherency. There are numerous writing mistakes, grammar, spelling, syntax. Some statements and conclusions are presented without support from the data or at least physical explanation. Detailed description on how to match the ship data with satellite data is lacking. The presentation does not reflect sound understanding of the sea ice concentration issues. It tries to use the two data sets; ship observations and satellite data hastily to develop some statistics of differences.

I believe that the authors have a unique data set that should be put to a better use for validating the concentration algorithms. Ship observations data have been woefully underemployed for assessing ice concentration algorithms.  So, the authors are expected to make best use of this data set, but I don’t see this happening in the present manuscript.

The authors should show better how they matched the two data sets and what the limitations of this task are. They should also provide strong evidences to support their conclusions. I do not believe that ASI and VASIA2 underestimate concentration by only 0.5 tenth in summer. Summer is most problematic season for ice parameter retrieval. 

In the following, I am pointing a few mistakes but they are many more. The number is so large that I don’t think the present manuscript meets even the standard of “major revision”.

Line 29: “In the paper, we also estimate the impact of ice melt stage and presence of new ice and nilas on SIC derived …”. In the summer there is degraded ice. I did not see this ice addressed though it really affects the calculation of ice concentration from remote sensing.

Line 45 “… however they allow retrieving but a narrow set of parameters” clumsy sentence

Line 60-62 “It was shown that all algorithms are less sensitive to hydrometeors in the atmosphere above close ice areas than above open water because of their lower content in the atmosphere above close ice”. First, it is not “close ice” but should be “close pack ice”. Second, the sentence can be reduced to be clearer. Third, there is no reference to support the statement “it was shown by eho???”

Line 55. The paragraph starts with “Paper [1] presents …”. This is not the right style.

Line 85: “When calculating SIC, nine known beforehand values of brightness temperature are used: nine tie-points”. The word “beforehand” is not the right word.

Line 94: The ASI algorithm …. Virtually nothing is mentioned under this title

Line 99: very insufficient description of the VASIA2 algorithm

Line 110: again, very brief and insufficient description of the passive microwave data. The reader needs to know the channels or the derived parameters that are used in the SCI algorithm.

Line 135: “A comparison was also made with the ship-based total SIC from which the initial types of ice were excluded (Cso-ni)”. What is initial type ice? Is it “ni” as suggested? But new ice exists only for short time in the fall.

Figure 1 shows that the authors have access to wealth of data. They should first sort all these marine routs and show us what sections they used (coastal, open sea, fast ice, etc). The accuracy of the algorithms varies in each section.

Line 154-160: needs better explanation on how the authors matched ship observations with satellite data.

Line 181-184: there is key advantage of using the ship observations here, which has not been used in the manuscript. The ship observations include sea ice features such as ridging, melt state, snow depth, etc., all affect the retrieved concentrations. The users should link this information to the accuracy of the retrieval and use the information to explain the deviation from the ship data

Line 217: “For the summer season, total SIC calculated by all three algorithms is overestimated in areas covered with very open ice (1-3 tenths): NT by 1 tenth, ASI and VASIA2 by 0.5 tenths”. First, “all three … by 1-3 tenth, then NT by 1 tenth and ASI and VASIA2 by 0.5 tenth??  Second, these numbers do not look realistic to me. The error in the summer in this study area is expected to be larger. These numbers made me doubt the rest of the results.

L229: “The small errors shown in Figures 3-5 actually result from high positive and negative errors compensating each other.” Then, why didn’t the authors use the absolute values of the error. The arithmetic average is meaningless.

Line 270: the 5 points that mark melt stages are not explained. I do not know which number indicates which stage.

In summary, the manuscript planned to use a good data sets but it has not been put into good use. Example is the availability of melting information, ridging information and snow cover information - all can be used to interpret the differences between the algorithms results and the "true" ship observations.

I started reading this manuscript with interest but I was losing interest as I saw results addressing just differences between algorithms and ship data without any serious attempt to explaining the differences in terms of the physical environment of the observations. Moreover, it was difficult to follow the text in many parts as the structure was a bit unclear.

Based on these major criticisms, I cannot recommend this manuscript for publication (unfortunately). The decision is up to the editorial board but I think the authors may consider re-submission as a new manuscript and pay attention to what is needed to compose a scholarly article. I believe that the theme of this manuscript is important and has merit, but the methodology and the presentations need significant improvements.

Reviewer 3 Report

Review of

Comparison of Arctic sea ice concentations from NASA Team, ASI and VASIA2 algorithms with summer and winter ship data

by

Alekseeva et al.

Summary:
This paper deals with an inter-comparison between satellite-based daily gridded sea-ice concentration estimates of three different algorithms and ship-based observations of the ice conditions from aboard a good number of Russian cruises. These ship observation are very valuable and present a slightly different view on the Arctic sea-ice cover than more recent ship-based observations available through ASSIST/IceWatch. The ship-based data sets are binned into 0.1 degree segments and sampled / averaged onto the respectiv grid cells of the satellite sea-ice concentration grids. The comparison is done in form of histograms and is taking into account the potential influence of new ice formation as well as the impact of melt ponds; in fact the algorithms are also assessed with respect to stage of melt. Results of the inter-comparison are presented.

I have several major concerns with this manuscript which I have listed in my general comments. In addition to that there are a number of specific comments; these are partly covered by the general comments but also point towards required improvements in writing / formulation and presentation. I suggest that the authors take their time for a review of their elements of the analysis and improvement of it following the suggestions made befor they begin with the major revision of the manuscript.

General comments:
GC1 - about literature:
- Please make sure that you refer to issues in your paper with English literature. I myelf, working and living in Germany, would never include a German paper or book or report into an international journal like "Remote Sensing". I guess the readers of your paper would appreciate if you could follow this standard.
This applies e.g. to references [5] and [32].
- In addition to that I encourage the authors to use more recent literature where possible and also to use literature which is findable easily by other readers. I doubt, e.g. that [8] fulfils this. Is this an online document?
- I suggest that you could be more careful about whether the references cited indeed are appropriate for what is written in the paragraph / sentence where you use it. Using [6] in the context where you used it, for instance, doesn't seem appropriate to me. I'd probably delete [6] and add a reference which tells something about shipping in the Arctic (keyword northwest and northeast passages) and its relation to potentially changing sea-ice conditions.
- I finally suggest that the authors check their reference list and try to reduce self-citation to a minimum.

GC2: The paper would clearly benefit from a more clear motivation and presentation of the objectives.
It is not clear what the main purpose of the paper is. It goes from advertizing AARI ship-based observations over a simple algorithm inter-comparison and/or evaluation of a new algorithm VASIA2 to a discussion about main problems of PMW sea-ice concentration retrieval in summer.

GC3: The paper would clearly benefit from a more detailed description of the physics behind the 3 algorithms used and  of the interaction between sea ice in general and initial ice stages plus melt stages on the one hand and passive microwave data on the other hand. Currently, the results presented are kind of hanging in the air and no attempt is made to explain the results in a specific manner, i.e. taking into account the different nature of the three algorithms.

GC4: The paper would benefit from a more critical discussion of the validity of the ship-observations. I got the feeling that these are treated as the truth which they are not. They also have an uncertainty and can be biased and this should not be hidden in front of the readers.

GC5: The paper could be considerably improved on the technical side. Suggestion are
- inclusion of scatter plots,
- into account taking of actual numbers of observations per sea-ice concentration bin,
- description of the algorithms (see GC3) and the co-location (of which I have doubts that it is correct),
- improved description and interpretation of Figure 6
- constrained analysis (to high concentrations only) of the melt stages' relation to PMR sea-ice concentration.
- Clear definition of summer and winter
- Discussion of potential influence of the different satellite sensors used (SSM/I versus AMSR-E)


Specific comments:
Abstract: I suggest that you either avoid mixing quantitative, i.e. 1/10 and 0.5/10, and qualitative, i.e. open, close, etc., information about the results of the comparison - or make it more clear already here in the abstract which sea-ice concentration ranges are covered by your qualitative terms like "open", "close", et cet.

Keywords: I would add "Arctic" to make clear that this paper is not dealing with Antarctic sea ice. I also suggest to add "ship-based" to "visual observations".

Lines 35-40: L40: "today" --> Reference [6] already makes clear that satellite sea-ice observations date back into the 1980ties. This means that we have about 4 decades of satellite sea-ice concentration observations and therefore "today" is potentially not the right term. Please re-word accordingly.

Lines 54-59:
- L55: I suggest to add [18] to [9].
- L57: Please replace the "Discussion" verison of [15] by the final paper in "The Cryosphere".
- L59: Please replace [22] by the respective paper in IEEE Transactions in Geoscience and Remote Sensing, 53(4), 1985-1996, 2015.

Lines 60-64:
- L63/64: I doubt that you have inter-preted the results of [16] correctly. I suggest that you re-write this paragraph more carefully. Errors in the sea-ice concentration associated with atmospheric phenomana over open water or open ice are large - not necessarily at each of the frequencies and polarizations used - but they are large enough that so-called weather filters were developed (and applied) to discard false sea-ice concentrations over open water caused by weather effects. Errors in high-concentration areas can be moderately high during winter but are usually smaller than the above-quoted errors over open water. The only exception to this is the summer melt and fall freeze-up period because at these times the ice surface properties are more complex AND of course melt ponds are sensed by the microwave sensors as open water (they should at least) because the penetration depth of microwaves in liquid water is a few millimetres at most. Could you perhaps one more time take a careful look into [16] and also [18] to re-write this paragraph?

Lines 65-68: Right, but could you please comment on why this connects to your paper and your goals?

Lines 69-79:
- L69/70: This statement is not true. Please check [22] (or basically the paper I asked you to cite instead of this conference paper). Another paper you might want to cite in this context is Worby, A. P., and J. C. Comiso, Studies of the Antarctic sea ice edge and ice extent from satellite and ship observations, Remote Sensing of Environment, 92, 98-111, 2004.
Once you checked these two papers, could I ask you to change your statement accordingly?

- In general I find this last paragraph a bit short as the finally presented main motivation why you carry out this study. It is not entirely clear. Did you do this study to evalute the VASIA2 algorithm cited with [25,26]? Or did you do this study to show-case how the potentially superior AARI ship-based observations can be used? Why did you include ASI? Do you also evaluate NASA-Team sea-ice concentrations? If so ... why? Please re-write this paragraph accordingly.

Lines 83-93:
- L87: "for each radiometer" --> If you are referring to the transitions between SMMR, SSM/I and SSMIS then this statement might be correct. Otherwise, it is often not the tie points which are changed but it is actually a brightness-temperature (TB)inter-sensor correction applied to match, e.g. SSMIS TB to those of SSM/I to re-use the same tie points. I encourage the authors to re-check this issue.
- L88: Are you sure that the tie-point values change? From my knowledge it is rather the brightness temperatures which do not match the tie points. In other words, as sea-ice tie point for first-year ice of 100% sea-ice concentration is one fixed value ... but 100% first-year ice can have a large variety of surface conditions which change the actually measured TB value. I guess this is exactly what you formulate in the following lines ... the only thing is, as I said, tie point are usually kept fixed.
- L90-93: Could you add "over open water" somewhere in this sentence, please? Then it describes the issue with the weather filter more clearly. And a suggestion at the side: These weather filters have been "invented" in the 1990ties or even earlier as well. Since you have been using so nicely the old literature when it comes to the description of the NASA-Team algorithm it would look good to also check out what the old/first weather-filter paper was. Might be something from Cavalieri et al., 1995, or so.

- What is missing in this paragraph is which version of the NASA-Team algorithm you are using. If you don't compute the sea-ice concentrations by yourself, but did download the readily processed NASA-Team sea-ice concentrations from somewhere, then it seems advisable to me that you include the information of section 2.4 relevant for the NASA-Team algorithm into section 2.1. Please make sure that you give the full web-address, access date and so on. Also make sure that you state: i) grid resolution and type of grid; ii) time period; iii) satellite sensors on which the data are based (together with the time periods).

- L94-98: This section is too short. I encourage you to write more about the ASI algorithm. I suggest to give a reference to the "father" algorithm by Svendsen et al., I suggest to give the native (first) reference to the ASI algorithm which is from Kaleschke et al. in 2001. Moreover, I ask you to be more specific. The ASI algorithm does not just use the same weather filters as the NASA-Team algorithm but it is a hybrid algorithm. After application of the weather filters all ASI sea-ice concentrations are set to 0 where the Comiso-Bootstrap is equal 0% (see equation 15 in Spreen et al. (2008)). Note that the implementation of the ASI-algorithm to AMSR-E data differs from the one used for SSM/I.
In addition please see my previous comments with respect to the content of section 2.4. It is not sufficient to just mention www.seaice.dk as the data source. You need to give more details.

- L99-109: If the main goal of the paper is to evaluate this algorithm then the description of the algorithm needs to include much more details. Perhaps one figure or two illustrating the dependencies of the mentioned tangents and the parameters would help. It is in particular not easy to understand why, given the penetration depth of microwave radiation at the used frequencies into liquid water of a few millimetres at most, you can kind of discriminate between open water and melt puddles. In particular when the algorithm is dependent on the theoretical emissions and radiation this seems surprizing and worth to again be written up in this paper.
Does the VASIA2 also require weather filters?
In addition please see my previous comments with respect to the content of section 2.4.

Lines 110-118:
- Please combine the content of this section with the previous three sections (as mentioned) and expand the information.
- Could you please comment on how one is able to have data from three different frequencies with footprint sizes between ~70km and ~15km on a grid with 12.5km grid resolution? Please provide more information about the POLE-RT fields.

Lines 120-132:
A comment from my side first: If I read what is observed here then it is quite close to what is also observed under the ASPeCt and ASSIST/IceWatch protocols. Yes, there are apparently more parameters recorded, this is true. It might make sense to stress this here - as a motivation for the developers of the two other protocols mentioned to increase their efforts.
- Please provide (an English) guide of the principles and observation guidelines mentioned
- "around the clock" --> what does that mean? Please specify. How about observations during nighttime when it is dark? I see that you answer the "dark" question in L166/167 - but that is specific for your paper - and here in this paragraph you need to describe these ship-based observations in general.
- Who are the observers? Are these volunteery scientists or have 2-3 experienced scientists been doing this during entire cruises for all cruises used?
- Please provide an estimate of the uncertainty of the sea-ice concentration observations.
- I understand that the visual observations are made for two areas, one immediately around the ship and the second one in the "navigational area". While the first area will be visible most likely also during poor visibility the other will not. How is this treated in the observations protocol and the data itself?
- "detailing of the spatial scale" and "time scale detailing" sound a bit vague and are - as somebody having performed observations after the ASPeCt protocol - difficult to quantify. Also the meaning is not clear. Of what are these the scales? How are homogeneous "ice zones" defined?
- Are there any auxiliary data / methods used, as e.g. the ships' radar? You write that "yes" in L164/165 but that is specific to your paper, right? Are ships' radar data used in general?
- What is the sampling with respect to sea-ice concentration? Is this a physical, quantitative parameter, i.e. tenths, or is this a qualitative measure, such as "open water", "very open ice", et cet.?

L134/135: Please specify what you term "summer" and "winter" in terms of the months.

L135/136: "initial types of ice" --> while it is intuitively clear what you perhaps mean here, could you nevertheless refer to the WMO ice types to describe which ice types are left out here?

Figure 1:
Please provide an additional Table where you list the dates of each expedition (like e.g. in Worby and Comiso, 2004).

Lines 154-172:
- In order to discretize the ship-based observations along the segments of the ships' track with 0.1 degree step size you need to have a projection or grid, don't you? Which projection did you use?
- How did you take into account that the distance (in kilometres) along longitude coordinates varies with latitude?
- I have the impression that by using your approach you have comparably fewer ship-based observations in 0.1 degree segment (or even 0.1 degree by 0.1 degree grid cell?) in, for instance, the southern Kara Sea than, for instance, near the North pole.
- You are very specific about the length of the ships' tracks in the respective grid cells and also about the approximate fraction of the grid cell that you estimate is covered by the observations. I have several concerns here and encourage you to write more details and/or potentially even revise the methodology.
A) An average observation radius of 8 km sounds very optimistic - particularly during summer conditions. Please provide evidence for this estimate from the observations. Potentially, visibility is reported together with the sea-ice observations and you can provide a histogram plot.
B) How did you compute the average distance a ship traveled in the respective grid cells? I am asking because I would have thought that within a 25 km by 25 km grid cell the minimum track length is 25 km (and not 20km as given in L161); also the maximum track length would theoretically be the length of the diagonal of the grid cell and hence something like 37 km. Similar considerations apply to the 12.5 km grid cells. Could you specify in more detail how you ended up with these, kind of unexpected numbers?
C) All three algorithms considered use gridded sea-ice concentration data instead of swath and/or single footprint data - for good reason - because you want to evaluate the products. But gridded data contain a mixture of single footprints with different degree of overlap in that grid cell with contributions from up to 3 successive scan lines. Hence the information that is "merged" in just one grid cell is the information from several footprints. Therefore, I am in doubt that such a detailed description as you have done has the expected quantitative value. If you would compare sea-ice concentrations computed at the scale of a single footprint, then such a quantification would be really valuable. I am therefore wondering whether you would perhaps consider to revise this writing to a level where the information of an estimated (!!!) field-of-view around the ship is not over-interpreted with respect to the satellite data.
- Please note in your text what you think is the relative fraction of the "close-to-the-ship" observations relative to the "navigation area" observations. I rate this as an important information required to explain your statement in L248/249.

Lines 180-192 & Figure 2:
- In Figure 2 you are including a sea-ice concentration product which is not among those you are comparing. I suggest that you either find a similar example out of the time-period you aim to investigate and then also select one of the ASI products which match the grid resolution you are investigating in your paper - for consistency.
- In addition, Figure 2 would be more easy to interprete if you would color the ASI sea-ice concentration data the same way the AARI ice charts are colored. In other words: I suggest to classify the ASI sea-ice concentration using the same sea-ice concentration ranges as are used in the AARI ice chart. Would that be possible?
- You need to provide the last access date for all images of Figure 2.
- How are the AARI ice charts generated? Which data, collected over a how long time-period (a week?) are used to create the AARI ice charts?
- Please also provide the time of acquisition for the MODIS image.
- Please provide information about what kind of an image this is (RGB composite of channels xy? ... just one channel?)
- Even though you state which areas are potentially cloud covered and which not it would help to mask the clouds by the same way in all three images of Figure 2. Currently this is not the case.
- Please provide more information about the sea-ice concentration ranges used in the AARI ice charts. For instance "1-3", is this 5% to 35% or 10% to 30% or 10% to 39% ....?
- "air temperature ... was +3degC" --> Is this really a relevant information? i) From where is this information taken? ii) More relevant for the melt progress is the water temperature and the net-radiation balance at the surface. One could imagine that "during melt conditions" is enough information.
- At the end of the paragraph you could add a sentence stating that wet ice and/or slush / brash ice floating in water, being almost saturated with sea water is very challenging to be observed properly by SMR; this has been illustrated in a few publications.

Lines 198-201:
- L199: "gradations" --> Do you mean "grid resolution"?
- I am surprised to see that you put the difference C_smr minus C_so-ni under the summer cases. My understanding would have been that young ice (or the mentioned initial sea ice stages) are more relevant during winter months. Since you have not yet specified which months belong to which season (in this paper), it is difficult to further assess the results in the summer images of the three algorithms shown in Figures 3 to 5.

Figures 3-5:
- What are the polynomial fits good for? I believe that these bar graphs provide enough information already.
- I can see that your first bin [0-1] includes open water. Was this done on purpose? Wouldn't it have been more meaningful to look at the pure open water cases separately?
- In Figures 3-5 please provide information about the number of data pairs falling in each of the bins. I would suspect that the fraction of high concentrations is much higher during winter than summer.
- I encourage you to, for each of the histograms shown (summer, winter, all three algorithms), also provide a scatterplot of the data pairs. I am sure the readers would be interested in the amount of false detection of open water by the satellite data but also by the ship-observations - which is difficult to asses from this kind of histograms.

Lines 217-234:
- I am wondering whether the fact that you are describing the results only in full tenth ranges has to do with the accuracy of the ship observations. I am asking because I guess that both the PMW sea-ice concentration products and the "gridded" ship observations are not provided with 1/10 precision but with 1/100 precision. And also Figure 3-5 provide the differences with 1/100 precision. Therefore it could be a good idea and would appear more consistent with your analysis to show the results at the precision given by the histograms.
- L220: I find 1.5 tenth as too large looking at Figure 3 to 5: I'd say it is rather 1 to 1.5 tenths for ASI and 1 tenth for VASIA2.
- L222/223 and L225: "remains negative ..." --> I suggest to describe the figures using the same style, i.e. either you write "over-estimation" and "under-estimation" or you write about positive and negative differences.
- L224: "considerably overestimates" --> please, like for the examples so far provide the actual amount of overestimation.
- L226-227: Here it would be very informative to know the actual number of observations in this sea-ice concentration range (see my comment above).
- L230: "high-positive and negative errors compensating each other" --> I am inclined to believe what you write here. Other readers might not. Therefore it would be very instructive if you would show scatterplots (see my comment above) or, alternatively, if you would add the standard deviation of the differences shown in Figure 3 to 5 in the histograms as an additional error bar.
- L234: "ice gradation" --> not clear, perhaps you meant "sea-ice concentration ranges"?

Table 1:
- How did you compute the correlation coefficients and the overall standard deviations and overall average error? Did you average over the values obtained for each of the bins or did you average over all single data pairs?
- It is a bit inconsistent to give the average error in unit "tenths" and the standard devation in "percent".
- Please change the "decimal ," to a "decimal ."

Lines 244-247:
- Please provide the unit for the differences quoted here, i.e. tenths, or switch from tenth to 1/100.
- Do the observed changes in the difference make sense to you? Please discuss. What would be the influence of a lot of thin ice on the retrieved sea-ice concentration - from the algorithm side?

Lines 248-249:
This is a strong statement and not easy to understand without further information about i) which months belong to winter, which to summer? ii) what is the fraction of the near-ship ship-observations (which cover a substantially smaller fraction around the ship which, paired with the fact that ships typically follow easy-to-navigate ice conditions, can lead to an over-estimation of the actual fraction of young ice / nilas)? iii) what sign had these errors and is the reported value of "reaching 2.5 tenths" the overall mean or a value from the separate sea-ice concentration bins? I strongly suggest to also show the results of this intercomparison as green bars in Figures 3-5.

Lines 250-264:
- I strongly suggest to replace reference [32] by an English one and by a more recent one - see GC1.
- I suggest to re-write the first paragraph after the other comments have been taken into account by you. In this context it would be good to also decribe why you explicitly are referring to a thickness of 5 cm. I don't actually understand your conclusions in the last sentence of the first paragraph.
- The same applies to the second paragraph. What is missing here in your explanation is the physical reasoning. It would be important to discuss why thin ice could have an influence on the sea-ice concentration retrieved from passive microwave brightness temperatures and in which direction retrieved sea-ice concentrations do differ compared to the actual sea-ice concentration.
- Looking at Figures 3 to 5 it would be interesting to discuss the differences between the algorithms / products with respect to inclusion / exclusion of thin ice. I suggest you focus on those sea-ice concentration bins where the differences between the green and blue bars are (potentially) larger than the noise, i.e. 3-5% (0.3 to 0.5 tenths). If one does this then it seems obvious that both NASA Team and ASI algorithms behave similarly and have the largest differences between blue and green bars for bins 60-70% through 80-90%. For VASIA2 differences seem to be larger in more bins.

Lines 266-269 / Table 2:
- Could you also provide a similar table for multiyear ice? Or do you think that this is not relevant?
- When does melt actually start? When you state "stage of melt" equals 0: "no melt" --> does this include already melting of the snow? Melting of the snow results in the snow pack getting moist and finally wet - which has an impact on the observed microwave brightness temperatures already. I am wondering whether you have taken this into account in your analysis?
- Albeit I agree that the stage of melt observation is a qualitative one there have been cruises into the Arctic where also the percent melt pond fraction on top of the sea ice is reported in tenths; this is done, e.g., within the ASSIST/IceWatch protocol. Would it be possible to convert this qualitative information into a more quantitative information post analysis?
- Intuitively, I have a problem to relate the results of Table 2 to Table 3, particularly for melt stages 3 and 4. Why? Because melt stage 3 already says: "Ponds are everywhere" - to me this sounds like that at least 30 percent of the sea ice is covered by melt ponds. It is very difficult and vague to properly relate the melt stages to difference or changes in sea-ice concentration.
- Looking at the melt stages I am wondering, whether the estimates total sea-ice concentrations from the ship-observations are not becoming more and more inaccurate the more ponds have melted through and the darker the ponds are. I am inclined to say that an observer will have a very hard to decide whether a dark patch in 3 km distance is in fact open water or a dark melt pond. Did you take this into account in the interpretation of your results? If not then I suggest to discuss this as an element of the uncertainty of the ship-based observations.

Lines 270-279:
- L274/275: "This is explained by ..." --> Please check this sentence and re-write it. It would possibly help if you delete the "point" or "points" in the context of referring of the melt stage. This applies also to the paragraphs further down where you used "points" in your formulation. I recommend that you replace expressions like "melt stage of 1-2 points" by "melt stages one and two", as an example.
-L277: "real" --> perhaps better "actual".
-L278: "melt through the ice" --> I suggest to make clear here that this means that the actual sea-ice concentration decreases more and more and what formerly was a melt pond on actually present ice, being under-estimated by the algorithms, has become open water which is correctly seen as open water.
- It appears to me that for the analysis presented in this paragraph, you did include all sea-ice concentrations ("From the entire array of ship observations ..."). Did you perhaps also think about only using cases with sea-ice concentrations close to 100%, i.e. > 90% or > 95%? It appears to me that otherwise the differences and results you report about in Table 3 are not only influences by the melt stage differences but also by the differences between satellite and ship-based observations of the sea-ice concentration across the entire ice-concentration range.

Table 3:
- How did you compute the average melt stages? I am asking because during the track the ship encounters different ice types which are represented by the observations. The stage of melt differs between ice types. Hence, when computing the average stage of melt given in Table 3, one needs to weigh the different stages of melt with the fraction of that ice type. I am sure you did this and I strongly suggest that you include the equations for this in your data and methods section.
- Already the differences for melt stages between 0 and 1 are different from the typical "average" differences for summer (compare Table 1). Why?

Figure 6:
- The x-axis needs an annotation.
- How is the relationship between Table 3 and Figure 6? Melt stage ranges in Table 3 are 5, greater or equal 4 to less than 5, greater or equal 3 to less than 4. If I look at Figure 6, then cells of this 2-dimensional histogram have size 0.5 melt stage and 1.0 tenth (=10%) sea-ice concentration difference and - more importantly - the cells are centered at stages of melt 5.0, 4.5, 4.0, and so on, actually covering the range 4.75 to 5.25, 4.25 to 4.75, and so on. Please advise the reader how to interprete.
- It seems to me that the NASA-Team 2-D histogram has much fewer counts in total than the other two algorithms. I guess this can be explained by the different grid resolution - which should be mentioned here.
- Please provide the same color range for algorithms with the same grid resolution. Currently the scales for ASI and VASIA2 are different.
- Did you check how the results of Figure 6 compare with the results given in Table 3?
- I am surprised to see very large deviations which seem to go to 7 or even 8 tenths negative AND positive differences for VASIA2 but to some extent also for the other two algorithms - for instance between melt stages 2 and 3 and at melt stage 3. I have difficulties to understand this; at these melt stages I would think that melt pond fraction of 70 to 80% are rather unlikely. Perhaps this is an artificial result from the weighing according to the different ice type fractions when computing the total mean melt stage ... or from the fact that apparently you did not just focus on sea-ice concentration above, say, 95% in this analysis.

Lines 304-316:
- What you write about the results of [21] is not correct. Neither was the ice cover "open" and "very open" --> see Figure 4 of [21] which shows sea-ice concentrations mostly above 90% for the March/April and above 80% for the July/August cruise - nor was the difference between ship-observations and ship-based observations "12%" --> it was between -1% and -4% in March and between +10 and +12% in July/August. Please provide the correct information! Also, it might make sense to i) mention from which satellite data the sea-ice concentrations were retrieved in [21] and ii) whether the ship-observations they used are similar (in terms of the size of the area observed around the ship) to your data. Otherwise this comparison to previous work is not particularly valuable.
- In case of what you write with respect to the results of [20] it would help as well to i) mentiond the satellite sensor these data are based on and, more importantly ii) make clear that this is Antarctic sea ice.
L314: References [33, 34] are not in the reference list. In addition, their comparison is completly different from yours. Statistically, and also regarding the spatial coverage, a comparison between two remote sensing images and/or techniques, as done there, is not the same as a comparison between PMW SIC estimates and ship-observations. I recommend to delete this part of this paragraph.

Lines 317-322: This paragraph needs to be re-written - pending the comments I gave further up. In particular an attempt to explain your observations physically would really help in the understanding.

Lines 323-328: I suggest to rewrite this paragraph as well - again for the same reason as stated in my previous comment.