Both referees pointed our some serious concerns remain with the manuscript. I agree with referee #2 that without validation data, the significance is mostly technical in nature, and based on what is described in the manuscript, theoretical improvements are relatively small. The manuscript needs to be make a better case for what the benefits of the combined product are if it is to be suitable for publication in The Cryosphere. The authors should also carefully address referee #2’s comments about the uncertainty estimate.

With regards to the lack of proper validation raised by all referees. I agree this is a serious concern that must be better addressed. Here, because of the scant data available, and potential scale issues (briefly mentioned in the manuscript), I will offer my thoughts to expand on the referee comments:
(1) The lack of suitable validation data should not preclude publication of a potentially useful product (useful precisely because of the lack of such data) provided appropriate caveats are provided. In this case, for instance, the data is trained using a simple freezing degree day calculation of uncertain accuracy for the locations and times it is used. The authors could assess the potential errors from using this approach and its implications for the product uncertainty estimate (as implied by referee #2’s comments)
(2) There are some potentially useful thin ice observations, e.g. available at http://icewatch.gina.alaska.edu/ . One potential data source archived there is from the October/November Sikuliaq cruisein 2015, which spent significant time in thin ice.
(3), Lastly, however, I will note that the biases presented here are relative to the Huntemann et al (2014) SMOS product, which was compared against some independent data. As such, my reading is the combined product here is presented as a product that is consistent with a previous product. Then a reasonable approach might be to assess the uncertainty in the original product, and propagation of those uncertainties to the combined product.

I also agree with referee #3 that the presentation can be improved; it seems to me that the way the methodology is presented over several sections that it is difficult to follow. In addition to the reviewers’ suggestions, I would suggest the manuscript have a section summarizing step-by-step the processing methodology.

The authors should also take note of the recently published paper by Schmidt and Kaleshcke (2018) as pointed out be referee #1, which covers quite similar ground. This paper was published after submission of the original manuscript, but your manuscript would benefit from some discussion of the distinction and similarities to the methods and results described in that paper.

Based on the revised manuscript and your response to the last round of reviewer comments, I believe the manuscript is suitable for publication with a few minor corrections. I have listed some technical (mostly copy-editing corrections below, with a few small suggestions to help make the text clearer).

There are a couple areas where I believe minor changes to the text could still better address some of the reviewers’ concerns. The major concerns raised by the last round of reviews were that the improvements were relatively minor and it wasn’t clear why a combined product was useful. This is improved in the conclusion, but it would be helpful to the reader to know what improvements this product has in the abstract (e.g. marginally greater coverage, improved RFI, reduced TB RMSD relative to a prior merged product, etc).

Second, was the lack of in situ validation. You have addressed this in section 6 as I suggested. It does seem to show some relationship, although a lot of scatter. It is obviously a limited comparison, and not terribly conclusive. Certainly there is too much scatter to say anything about whether your algorithm is any better than any prior algorithm. The manuscript could benefit from a few sentences further discussion on this point. First, how does this compare to any prior published validation of SMOS and/or SMAP against in situ data? What does this new comparison say in general about the validity of SMOS/SMAP SIT (you do comment on this a bit)? Second, I think this is not so much the point, as you have shown that the new combined algorithm compares well to a previously validated single sensor product – so you should say this explicitly in the text (I think this is what your last sentence of the conclusions is trying to say, but it is not clear as written). Lastly, you do acknowledge Schmitt and Kaleshke (2018), but say nothing about it. You should explicitly discuss any similarities or differences between your results and theirs in your discussion.

Technical and other minor corrections:

P1L9 - “sea ice thickness” lower case

P1L20 “multi-angular” also “organized in approximately”

P1L22 – insert comma after “(Font et al., 2010)”

P2L6 – “on board”

P2L10 – “radio frequency interference” lower case

P3L22 – “at 6 pm”

P3L29 – “surface based”

P4L4 – “The new data version exhibits a”

P5L10 – As previously pointed out by a reviewer, you use bias to describe the difference between your version and the Huntemann et al. (2014) version. Of course, we don’t know which is more correct, so bias is not the best term. Perhaps just say the the RMSD is warmer in the new version.

P5L13 –This sentence is not very clear. I think you mean “The spillover produces an erroneous increase in TB over ocean areas adjacent to coastlines or ice edges, or decreases in TB over sea ice near the ice edge.”

P5L16 – please state what you mean by spatial ripples.

P5L20 – “at most a 3 cm difference”

P5L27 – be careful here calling this the error. You of course don’t know what the real error is, only that there is a difference of <1 cm thickness between the two algorithms.

P7L4 – change “As opposite” to “conversely”

P7L18 “Due to the fit computational requirement to have observations below 40°”

P7L20 “The decrease is around 1° in latitude” – this is not so clear. Where geographically is this? Do you mean around the pole? If so please indicate this.

P7L23 “around Iceland, Eastern Greenland and Vladivostok” – I do not understand why this is a positive point. None of these areas have sea ice at this time of year (unless you mean northeast Greenland, but I do not think you do, and Iceland never has sea ice). I do not see how this is an advantage. Any retrievals here are clearly in errors. I recommend either indicating that there are increased errors in some coastal areas (although these could be flagged by other data sources, such as AMSR2) or removing this discussion

P7L27 – “Figure 3, right” not Figure 4.

P7L31 – remove extra “the”

P7L33 – change “with that thickness” to “with thicknesses falling within that bin”

P8L3 – change “Until 40 cm of thickness” to “for ice less than 40 cm thick”

P8L5 change “values is below” to “ice thickness differences are below”

P9L14 – you use bias here again to describe the difference between the SMOS data versions. As pointed out previously by a reviewer, this term is imprecise when used to refer to a difference between the algorithms when neither had been clearly validated against ground truth or other independent data (I understand the prior version has been compared to data in previous work, but it’s not immediately clear that version 6.20 compares any worse as this is not shown). I recommend here, and throughout the manuscript, that where you have used “bias” to describe a difference between two datasets where there is actual true value is not also known or compared to, that you use something like “difference” instead. If you do know that the prior version has been validated with zero bias against an independent data source, then it would be fine to use bias in your context, but only if you make it clear that the prior data set had zero bias and relative to what data (if the prior version had a bias of its own, then you would need to compare these to be able to define what your new bias is relative to).

P9L20-34 This section is not as clear as it could be. You jump between steps used in your algorithm and procedures used by Huntemann, so it is not clear to the reader if a particular sentence refers to your work or Huntemman’s. e.g. the first sentence on lines 24-26 appear to refer to your study, while the second refers to Huntemman, but by jumping between it becomes unclear. It would be better if you described what Huntemann did in its entirety first, and then any limitations (so as to let the reader know why you undertook your work), and then described your procedure after.

P9L33 – “both polarization, at least 1.3K higher than the one presented in this paper”. Also it is still a little unclear what you are saying here. Since this seems to be one of the main selling points of the paper (i.e. there is some advantage with respect to the Huntemann algorithm), please make sure your meaning is clear here. For example, it is not clear if you mean that the improvement in the RMSD is due to the calibration to compensate for ET contributions and warm “bias”, or if these are separate points. Please clarify this. It would also be useful to indicate here what the consequences of this reduced RMSD are for SIT retrieval – roughly how much improvement in SIT might you expect for a reduction of 1.3K in RMSD. This is again mentioned in the discussion, so clearly one of your main points, but you do not relate this to the SIT retrieval. I think it would be helpful to understand the significance of your paper to discuss this result in terms of its expected impact on SIT retrieval error, or quality (at least in the discussion).

P10L4-5 – this is an awkward sentence. Perhaps rephrase: “For the period from October 1 to December 31, 2015, the difference SITs between SMOS 40° incidence angle fitted TB retrieval and SMAP retrieval are small. Using only grid cells containing SIT < 50 cm and at least one of the two retrievals having SIT > 0 cm, the average difference for the SMOS SIT relative to the SMAP SIT is -0.2 cm and a 2.39 cm RMSD.”

P10L14 You have previously defined SIT as sea ice thickness, but you jump between the acronym and full wording throughout. Be better to define once and use SIT throughout.

P10L14-15 “which has greater coverage”

P10L19 – “this result means”

P10L20 – “is on average”

P10 L21 – be better here, and throughout to simplify dates, i.e. October 24, 2015, instaed of 24th of October, 2015.

P10L21 – “The greatest differences”

P10L22 “to over 50 cm”

P10L24 – Figure 6 (right panel)”

P10L25 “the average”

P10 L29 – “the data are limited to the region”

P10 L32 – what fit parameters? Do you mean for equation (3)? Since the main text is a bit dense for the casual reader, it would be better if each of the steps in this section where more specific as to what equation, or part of the paper the step is referring to. More examples of this below:

P11L7 – “the gridded SMAP TBs are converted to SMOS equivalent TBs” – how? By linear regression to SMOS TBs.

P11L9 – “are computed by error propagation”

P11L12 – “Polarization difference (Q) and Intensity difference (I) are”
P11L11 “are calculated from the”

P11L15 “the sea ice concentration (SIC) and cumulative freezing degree day (CFDD) error are included in the SIT uncertainty resulting from the previous step” I am not sure this (and other error calculations) should be listed as steps in the algorithm. My impression was the reviewer requested this list of steps so that the reader could see how the algorithm works, not every analysis step you performed in the paper. The error analysis I believe is to demonstrate the effectiveness of the algorithm and its sensitivity to errors, but this is not a step in the algorithm.

P11L16 this step should be deleted. I don’t see this as a necessary step to the algorithm.

P11L17 what do you mean by “them”?

P12L1 – be consistent with date format

P12L24 – “SIT below 20 cm thickness”

P13L30 “we take one day”

P14L1 – do you mean figure 9 here?

P14L13 “due to incorrect representation of the total..”

P14 L13 “due to the greater exchange of heat”

P14L14 “due to greater exchange of heat”

P14L20 again make sure you are consistent with date formats throughout the paper.

P14L32 “From the initial formation of sea ice…”

P15L4 – include a reference for the APSeCT protocol.

P15L4 – “During the day, this allowed for an estimate of ice thickness in an approximate radius of 1 km”.

P15L7 “ship-based”

P15L11 – “results in a slope”

P15L14 Fig ?? – which figure are you referring to here?

P16L10 – what do you mean be “refer to their comparisons” do you mean for the reader to refer to their validation of their product?


Figure 10 – “The red line shows the rolling average of the uncertainty”

Figure 11 – “is color-coded. The black line represents…”

The authors have sufficiently addressed the two main concerns from the prior reviews. The first related to the significance of the results. Here, while some of the improvements over prior efforts are small, the authors have made clear the differences with previous results and advantages. The second issue related to the lack of validation data. The authors have addressed this both by including a new comparison with in situ data (in the last revision), and clarifying that their results are consistent with a previously validated algorithm.