General comments

The manuscript describes a new daily snow depth product on Antarctic sea ice derived from satellite passive microwave radiometry. The proposed methodology builds on previous algorithms, which were mainly applied to Arctic sea ice, and the authors first evaluate the most useful frequency combination for Antarctic sea ice and then derive the corresponding parameters. The method itself (using a gradient ratio of passive microwave measurements and fitting the linear coefficients) is rather trivial and has been employed previously (Markus and Cavalieri, Rostosky et al.), but previous products were not very accurate over Antarctic sea ice, which makes this manuscript and the corresponding dataset a novel and unique contribution. The authors conduct fairly extensive and comprehensive comparisons to a range of in-situ data as well as the Comiso method and ICESat-2 laser altimetry data to validate their method. The manuscript also analyses and visualises seasonally averaged mean snow cover and trends in the data.

This new dataset is a very welcome contribution to science on Antarctic sea ice and beyond, as currently no operational snow depth product is available for the Southern hemisphere. Therefore, I expect this dataset to be useful alone and in combination with e.g. satellite altimetry to derive sea ice thickness. The dataset covers the full lifetime of the AMSR-E and AMSR-2 sensors of 2002-2020 and employs SSMIS data to bridge the gap in between the two, making it a complete dataset. Updating the dataset regularly in the future would be very useful though.

The manuscript is well structured, clear and it has a good length. References and citations are mostly complete. A recent related study by Kacimi and Kwok (2020, https://doi.org/10.5194/tc-14-4453-2020) could be added, though. Figures and tables provide detailed insights into the dataset and the comparisons made for the accuracy assessment. Overall, they are presented in a comprehensive manner. A few details are too small and hard to see though. A time series of (e.g. average and per section) snow depth over the full 18 years would round the paper up and add to its credibility.

The method description is kept rather short, but clear enough, as references to previous studies are given to point the reader to more details. However, it might be worth writing down how the gradient ratio is calculated either in the appendix, on the website or in the readme file. What strikes me is that the derived uncertainties (which are provided with the dataset) are much smaller than the RMSDs calculated with respect to all other datasets. To me, the calculated RMSDs seem more realistic and I would therefore suggest rethinking and adjusting the uncertainty estimation, as the uncertainties are the values that are provided with the data.

In terms of the discussion, my view is that the derived trend and the comparison to the ASPeCt data might not be significant. On the other hand, I am missing a comment and discussion on the bias between ICESat-2 and the proposed method. Please see more details in the specific comments.

The data is accessible via the given identifier and can be downloaded with an ftp client. Uncertainty estimates are included for each file. The data can be visualised and used for further analysis easily and the authors point to a range of open-source and standard software. The data description says ‘all values are in meter’, while snow depth is actually provided in ‘cm’ and so are the uncertainties. This might cause confusion, even if it should be clear from the data range.

Specific comments

Introduction: The study by Kacimi and Kwok (2020, https://doi.org/10.5194/tc-14-4453-2020) should be added, as it is quite related (same as general comment).

L. 42: You state that laser altimeters and a combination of laser and radar altimetry can be used to derive snow depth. However, also radar altimetry can be used stand-alone (without laser altimetry) to derive snow depth: See e.g. Guerreiro et al. (2016) 10.1016/j.rse.2016.07.013 and Lawrence et al. (2018) https://doi.org/10.5194/tc-12-3551-2018

L. 57-58: Please rephrase the last part of the sentence. One could misunderstand it as all of the snow cover in Antarctica is thicker than 50 cm.

L. 90: Delete ‘than AMSR-E and AMSR2’, otherwise the sentence makes no sense to me.

L. 93-94: How did you calibrate SSMIS? Please add more details.

L. 95-96: Were (a) only SSMIS data used to calculate sea ice concentration in this time period, or (b) did you calculate sea ice concentration for the full time (2002-2020) yourself? Please clarify in the text.

If (a) is the case: Why did you choose the ARTIST algorithm rather than sticking to NASA Team, which is used for AMSR-E and AMSR2?

If (b) is the case: You don’t need the downloaded sea ice concentration products mentioned before?! So, please modify the other text.

L. 109: ‘subtracted’ instead of ‘misused’

L. 126: Please rephrase ‘located in eastern and western Antarctica’.

L. 140: Why were these months excluded? Please clarify.

Fig. 1: In panel b the AADC 2007 data and 2005 ASPeCt data have almost the same colour (at least in my print out). Consider changing one of them.

Caption of Fig.1: I would mention AADC before ASPeCt to be consistent with the text and legend.

Eq. 1: Could be worth writing down how the gradient ratio is calculated either in the appendix or on the website / readme where the data can be found (same as general comment).

Also: Please specify (and cite) what you take as the brightness temperature over open water for the different frequencies.

L. 196-198 The calculated uncertainties are much smaller than the RMSDs calculated with respect to other datasets (same as general comment).

L. 282-284: It would be nice to indicate these sectors on the maps in Figure 1.

L. 283: Suggest renaming the ‘Pacific Sector’ or shifting the borders. 90 degrees is still in the middle of the Indian Ocean.

Table 5: Overall, the correlation coefficients seem to be scattered around zero and even in areas with comparably many grid cells negative coefficients occur. This makes me doubt a physical correlation between the two datasets and question to which extend this comparison actually adds to the credibility of the proposed method. I therefore suggest either clarifying and discussing what could cause negative correlations or consider leaving this comparison out.

Fig. 6+7: The numbers in panel c are extremely hard to read. Please increase the font size.

In the caption of Fig. 6 and 7 panels a and b should be mentioned explicitly.

Figures 6+7c: The PMW approach seems to estimate higher snow depths for all months. Do you have an explanation for this bias? As this is quite striking, it should be mentioned in the text and ideally discussed what might cause it.

L. 337: What is the uncertainty of this trend? Upscaling 0.13cm/year is still only 2.3 cm over the full 18 years that you looked at, and smaller than the uncertainties that you found, so I wonder if it is actually a significant trend.

Table 7: Instead or in addition to this table, I would find it really useful and interesting to see a time series of how snow depth develops over the full 18 years. You could for example plot the Antarctic wide mean and means from the sectors. Then you could also plot the calculated trend line on top. I think this would greatly increase the credibility of your dataset and the calculated trend, but could also verify the consistency of the three different satellites involved.

Fig. 10: The black dots are almost impossible to see.

Also Fig. 10: I would suggest flipping the colour map to make it consistent with previous figures and common practices where blue corresponds to low numbers (decrease) and red to high numbers (increase), but this might be a personal preference.

L. 367: ‘performed ONLY slightly better’

L. 467: Please check and update the link. ”https://nsidc.org/data/G10011/versions/2” leads me to “On-Ice Arctic Sea Ice Thickness Measurements by Auger, Core, and Electromagnetic Induction, from the Late 1800s Onward, Version 2”, which only contains data from the Arctic.

Data: The data description says ‘all values are in meter’, while snow depth is actually provided in ‘cm’ and so are the uncertainties. This might cause confusion, even if it should be clear from the data range (same as general comment).

Technical corrections

L. 195: ‘-‘ missing between ‘July-September’

L. 311: ‘time series’ instead of ‘times’?!

L. 430: … showed THAT no obvious …

General Comments:

This paper compared Gradient Ratios and selected the best 37/7 combination of passive microwave radiometer AMSR-E/2 to derive the snow depth over Antarctic sea ice with a linear regression equation based on the Operation IceBridge (OIB) airborne snow depth measurements. Then compared the proposed method with the Cosmic method proposed in 2003 and validated the derived snow depth with self-evaluation of OIB, a few in-situ measurements AADC, shipborne measurements ASPeCt, and the laser altimetry ICESat-2 derived snow depth. After that a daily snow depth data of Antarctic sea ice from 2002 to 2020 (with gap filled with SSMIS) was produced. This work is of great value to derive snow depth over the Antarctic sea ice for nearly 20 years. However, it can be greatly improved after a major revision. My detailed comments are as follows:

Specific Comments:

This work lacks specific month information for all data used throughout the paper, such as in the data description and validation sections.

About the growing and melting seasons, they should be clearly defined in the paper. I'm still doubt the suitability of the proposed method for both growing and melting seasons since the Eq.1 was derived from the growing season and the evaluation for supporting the suitability was with Aspect and ICESat-2 data. Furthermore, the model values have a clear overestimation than ICESat-2 in Fig.6 and 7.

The scale effect needs to be considered, since the scale of AMSR-E/2 is 25 km grid, ASPeCt is 1 km around the icebreaker, while the footprint of OIB is within 10 meters, and the AADC is even smaller. How would the different scales influence this study?

L74. This part lacks the introduction of ICESat-2

L87.The overlapping tracks and inter-comparison time period of SSMIS and AMSR-E/2 is not clearly introduced. The similar problem in Eq.2 and Figure 2b was also not clearly stated.

L97. Since the sea ice is continuously drifting, which can be several kilometers per day, Section 2.2 should show the overlapping strategy of different datasets from spatial and temporal perspectives. For example, were the OIB snow depth measurements averaged in the overlapping AMSR-E/2 25 km grid in the same day? Generally, how much time difference between the two datasets in the same day?

L117. Why was the OIB data in 2011 not used? Is it because there is no overlapping with AMSR-E/2 in the year? How about the validation of SSMIS-derived snow depth with OIB? I didn’t see it in Section 4 accuracy evaluation.

L148. Does Section 3.1 used both AMSR-E and AMSR-2 GR to correlate with OIB snow depth measurements? Please make it clear.

L192. How about the uncertainty of intercept a and slope b over FYI and MYI respectively here?

L199. About “due to complex surface conditions here”, could you give some evidence to support this statement? Such as show the seasonal spatial distribution of GR.

L207-208. Why is it 6-year combination instead of 7 (2009, 2010, 2012-2014, 2016-2018, as stated in L117)? As we can see the data 2018 missing in Table 2, is it because the data in 2018 less than 80 matched points? I'm confused with the number of grid cells in Table 2. Is it the "matched points"? If it is, why were 2013 and 2017 included?

L217. About “the Comiso method”, please state the rationality to compare the proposed method with a paper published in 2003, nearly 20 years ago.

L243. About Figure 4, is this comparison of data in 2016? Please clearly state the time here. Is Figure 4c the mean snow depth in each year? Why does the year 2013 have an obvious valley? Based on the lower black dashed line in Figure 4c, I wonder why OIB cannot detect snow depth less than ~12 cm.

L293. We may see a clear overestimation for the proposed method in Figure 5, please explain it. We can also see this overestimation when comparing with ICESat-2 from Figures 6 and 7.

For Figures 6 (L316) and 7 (L321), I suggest to add grid information for each satellite and model values for column c.

L326. Figure 8a, please add model values for comparison. Figure 8b, I suggest to use grid-wise comparison instead of monthly mean scatter in order to show more details.

L334-336. There are several weird phenomena in Figure 9 as follows:

1. Snow depth increased from spring to summer during a melting period.
2. Snow depth in the Weddell West sector decreased from autumn to winter during a growing period.
3. The thickest snow depth for the East Antarctic and Bell-Amundsen sea is in summer, the hottest time for the Antarctic.

These issues should be discussed.

In addition, how did you decide the spatial extent of sea ice in Figure 9?

L354. How did you decide the spatial extent of sea ice in Figure 10?

L372. How does the scattering intensity lead to the overestimation? Is this a possible reason for the Figure 9b overestimation of snow depth?

Technical Corrections:

L11. I suggest to add “passive” to “microwave radiometers”. Please check it throughout the paper.

L20. Please specify the “previous method”

L34. The citation “Giles et al., 2018” work mainly uses ERS-2 to retrieve ice elevation instead of snow depth. Another paper of Giles may be more appropriate to be cited here:

Giles, K. A., et al (2007), Combined airborne laser and radar altimeter measurements over the Fram Strait in May 2002, Remote Sens. Environ., 111, 182–194.

L35-36. “long times” to “a long time period”.

L44. It is better to use “in the Antarctic” than “in Antarctica” since we focus on sea ice. Please check the phrase throughout the paper.

L45. “uploading” to “upper”

L59. The citation “Kern et al., 2016” is a wrong one. The right one concentrating on the factor value is in “Kern, S., et al. (2011), An intercomparison between AMSR-E snow-depth and satellite C-and Ku-band radar backscatter data for Antarctic sea ice, Annals of glaciology, 52(57), 279-290”

L60. “successor” to “its successor”

L65. “reliable for estimating snow depth” to “suitable for retrieving Antarctic snow depth”

L79. “in the Arctic and in Antarctica” to “in Arctic and Antarctic.” Please check similar phrases throughout the paper.

L82. About “pre-processing, bias correction and quality control were all applied”, specifically how? Any introduction or citations here would be clearer.

L90 “(lower than or equal to 19 GHz)” to “than 19 GHz”

L93. “in” to “to”

L103-105. The reference “Schenk et al., 1999” for urban elevation DEM measurements should be cited to show the elevation accuracy, thus it should be placed after "1 m and 0.1 m"

L109. “misusing” to “subtracting”

L160-164. Please give the equation of GR here.

L163. What weightings? It is unclear.

L168-169. Is “the root mean square residual” RMSD? If it is, I see it is different from that in Table 1. What is the “standard deviations of the derived regression coefficients”?

L178. “…GR (37/19), which ranked next to GR (37/7), as shown in Table 1” is an inaccurate statement. GR (37/19) was next to 37/11 instead of 37/7. You can say GR (37/19) was the best combination among frequencies no less than 19 GHz.

L185-186. The statement is too simple. Here lacks more spatio-temporal information for data used.

L229. “a factor of 2”, it is 2.3 exactly.

L250. Please show, in average, how many AADC points are used to compute a 25 km grid mean values.

L259. Why the number of grid cells different between the two methods? Please clearly state if data used in this table all from the melting seasons.

L279. “in other seasons” to “in the other seasons”

L286. About “most of the multiyear ice is in the Weddell West”, “the Weddell West is dominated by multiyear sea ice” is more appropriate here.

L357. “method” to “methods”

L424. Delete “is”

L425. “at” to “to”

L446. Add “than the Comiso method” after “performance”

L476. The wrong doi should be https://doi.org/10.5194/tc-13-2421-2019