Review of the manuscript  “Assessment of Snow Depth Retrievability from Passive Microwave Observations over Arctic Sea Ice: A Global Sensitivity Analysis” by Yan et al. 

This manuscript investigates the non-linear interactions between different key snow parameters on brightness temperature using radiative transfer model simulations with parametrization scenarios based on MOSAiC (besides others) observations.

The authors show that most of the variability with the TBs above 18GHz are mostly driven by non-linear interaction between the tested snow parameters.

Comments

Overall, I found this study well written, straight to the point and very interesting. The EFAST decomposition is a simple yet robust method to explore whether an individual parameter contribute more or less than its coupling with other parameter in a qualitative AND quantitative way. The paper does a really good job describing the coupling and non-coupling effect of the different snow parameters on the TBs and the different GRs. Given the fact that this paper relies on MOSAiC snow data to estimate the range of variability of the snow and ice parameters, I think it would be valuable to have results of this parametrization that fits the MOSAiC TBs to confirm that the proposed configuration still falls within the observations range.

I think this study is well suited for publication in The Cryosphere as it fits very well its scope, but some minor adjustments are needed prior publication.

Specific comments

Although the author mention in the text that the “global” term refers to the full parameter space, having it in the title can be quite confusing as it mostly focuses on Arctic snow. In general, the paper does a very good job describing the theoretical coupling between the ASMR2-3 frequencies, but given the title there is this feeling it would be more connected to observations. The Section 3.2 “Microwave radiometer configuration” is in data yet just describes the AMSR2 frequencies. If no observational TBs are used, this should be moved to the Method section.

Still in the section 3.2 or later in the text, I’m surprised nothing is said about the major footprint’s size difference between the 6GHz and the higher frequency channels (36 and 89 GHz. For snow depth retrieval from satellite measurement, this cannot be ignored as the 6GHz footprint is tens of km wider than the 36GHz, so the GR36/6 may also show something artificial due to the heterogeneity between the two footprints.

L255-260 and Tables 2-3: how is snow ice and superimposed ice modeled in SMRT? Are you using fresh ice layer? Snow with such high density (>500) can be tricky to configure in SMRT as some permittivity models can’t characterize snow with density that high. I would maybe suggest to better describe the permittivity model used for each layer (snow, ice, fresh, dry, wet etc..)

How saline the was the snow? A 5% of brine volume fraction isn’t something we can easily represent, as BVF is not something we directly measure on the field. It could be better to explicitly say the salinity value set in SMRT to get this BVF of 5%, as Figure 2 aslo shows the salinity of the snow in PSU.

Figure 3, 4 and 5: the author’s choice of underlying the TSI value is quite confusing and sometime makes the results not readable. It would be better to just write the the numbers (such as for Figure 3-b 18H where the author forgot to put the underline)

Figure 4 is mostly empty, I wonder if there would be a better way to present these sparse results.

For the figures showing the GR as function of snow depth and snow density (Figures 6, 7, 8 and 9), I would suggest removing the X-Y projection as it takes a lot of space and decrease the readability of the results and in a lot of case, they are showing no variability at all. They could be added as Supplementary material, according to the author’s decision.

L446: “The 89 GHz TBs are also susceptible to atmospheric effects…” This is also true for the 36GHz and in a lower extent the 18GHz.

I have reviewed the manuscript titled „Assessment of Snow Depth Retrievability from Passive Microwave Observations over Arctic Sea Ice: A Global Sensitivity Analysis “ by Yan et al..

Their study evaluates the retrievability of snow depth over Arctic sea ice for the AMSR2 frequency channels. I appreciate the effort of taking a more fundamental view on the question: instead of trying out different snow depth retrievals the authors use radiative transfer modeling to test sensitivities depending on background conditions.

The introduction is well written. The methods are clearly described but could benefit from a bit of additional information (see below) and more caution regarding the use of snow ice. The Data section describes the MOSAiC data but remains a bit too vague as to how the MOSAiC data is actually used later in the model setup. The Results section is well structured, describing first the results of the different snow setups and then the retrievable ranges, see some comments for improvement below. The conclusions provide nice guidance for further developments, highlighting also the importance of snow grain sizes. The figures are well done and I congratulate the authors for visualizing the information/coupling effects of such high dimensions, without oversimplifying it.

Overall, the study is a valuable contribution for the scientific community and well suited for publication in this journal. However, before publication I have one bigger concern that needs to be addressed (minor comments are found below): how do the authors ensure representative sampling? From the method section I understand that the parameter are uniformly sampled, thus unrealistic input cases, that is unrealistic combinations of layer parameters or values that are rare (like very deep snow), are possible and could influence the sensitivity, is that right? Or is there a mechanism to ensure that the (observed) gradients of, e.g., temperature and density, are preserved or that the output is weighted according to the likelihood of occurrence? The use of the MOSAiC data here is not clear to me here.

Another general remark related to that: I think the confidence in the modeling results/snow configuration could be increased by a direct comparison, possibly in the supplemental material, of the modeled brightness temperature distributions to pan-Arctic satellite-measured brightness temperature distributions (possibly corrected for atmosphere), to see how modeled and measured TB variabilities compare (with no perfect agreement expected).

Below, I provide more specific comments:

- Section 2.1: I recommend that the authors follow the guidelines from the official SMRT documentation: https://smrt.readthedocs.io/en/v1.5.1/publish.html providing a table with additional information about the model setup (in the supplemental material) including the surface roughness assumptions, the version number of the model and the exact way the salty snow (snow ice in Table 2) is modeled (what permittivity formulation is used)

- L. 116 : is i going from 1 to N or from 1 to n (as suggested by line 113)?

- Section 2.3: While the general description of the the averaging kernel as matrix is of course correct, you could add that is is 1D in your case, am I correct?

- Section 3.1., Line 190: how are these measurements used exactly? To constrain the parameter ranges? Or to sustain realistic parameter combinations?

- Line 201/Table 4: you write about BVF for temperatures warmer than -2°C, but you only model 265 K ice. How did you come up with BVF of 5% in Table 4?

- Section 3.2.: I suggest to add AMSR2 footprint sizes

- Section 3.3.: snow ice formation: I suggest a bit more caution with using this term in the context of Arctic sea ice. To my knowledge flooding is common in the Antarctic (the Maksym and Jeffries paper quoted in line 217 and line 239 is also about Antarctica, which should be clarified), but the occurrence in the Arctic is still not clear. Also, the Merkouriadi study quoted later in line 360 talks about potential and not observations at large scale, please formulate accordingly. It is not clear to me whether the use of snow ice is backed by the MOSAiC observations. That is not to say that there is no salty snow (commonly observed e.g. on fast ice) but to be careful with the flooding hypothesis. Finally, the density values listed in Table 2 go up very high, and scattering formulations tend to break down at intermediate/high densities, see Picard, G., Löwe, H., and Mätzler, C.: Brief communication: A continuous formulation of microwave scattering from fresh snow to bubbly ice from first principles, The Cryosphere, 16, 3861–3866, https://doi.org/10.5194/tc-16-3861-2022, 2022. Please have a look whether any issues are observed with regard to that.

-Section 3.4/ Table 4: what is the source of e.g. the radius values and the BVF values? Is it based on literature or MOSAiC data?

- Figure 1: You could consider adding a panel giving the number of observations per snow depth/horizontal bars.

- Section 4, Line 307: why is the temperature not shown for wet snow? I would think it is the biggest contributor with wet snow being more similar to a black body?

- Line 383: “finite depth” please be more specific, I assume snow is always of finite depth

- Section 4.2: I could not follow on which simulation setup this section is based. Are these extra simulations? Is the single-layer snow used? But then, how can we get snow depths up to 1 m when the range was limited to 60 cm in the previous section? Is wet snow excluded? Please clarify.

- Figure 6: I am surprised to see no sharper gradient between 0 cm snow and the rest, is 0 snow actually modeled or is there always a thin snow layer present?

Technical:

- Line 35: “most weather and all-sky conditions”: redundant

Review of “Assessment of Snow Depth Retrievability from Passive Microwave Observations over Arctic Sea Ice: A Global Sensitivity Analysis” – Yan et al. (2026)

Snow depth on sea ice is an important parameter that must be accurately retrieved to assess the Arctic sea ice and snow mass balance. While there exist multiple snow depth on sea ice retrieval algorithms using satellite radiometers, they are perhaps too simplistic in nature; ignoring non-linear interactions between different snow parameters (e.g., snow depth with density and grain size), therefore limiting their accuracy. In this manuscript, Yan et al. integrate typical snow and ice physical properties with radiative transfer modelling to understand their effect on brightness temperature (Tb) observations at the AMSR2 configuration. Not only is a parameter’s total contribution to Tb variance considered, but separation of its singular and coupling contribution is neatly achieved via the EFAST method. The results are extended to quantitively assess the retrievability of snow depth by using an averaging kernel, accounting for intrinsic variability and measurement noise.

The authors conduct a comprehensive analysis, finding that, for example, dry snow Tbs are heavily modulated by snow density and grain radius. They show that while Tbs have a measurable sensitivity to snow depth from 18 GHz onwards, the non-linear interactions between snow properties ultimately determine whether the snow depth signal is adequately retrievable. Acknowledging that the gradient ratio (GR) is commonly used to retrieve snow depth, the authors also demonstrate their method on this parameter, highlighting limitations using different GR combinations and ultimately making some optimal combination proposals. For example, GR(36/18) is better suited to fine-grained snow depth retrieval than GR(18/6), but GR(18/6) permits a broader range of snow depth retrieval if the snowpack is coarse-grained (i.e., from metamorphism). Moreover, the authors propose substituting GR frequencies to further optimise retrievals under specific circumstances, e.g., substituting 23 GHz with 36 GHz in late winter and spring to take advantage of its extended snow depth retrieval range in the coarse-grain regime. However, the authors also found that GR algorithms using 18 or 23 GHz are unlikely to achieve accurate estimations of snow depth on MYI if the grain radius is below 0.5 mm.

I find this manuscript to be well written and the methodology (linking SMRT with EFAST and an averaging kernel to quantify retrievability) to be well chosen for such study. The rigor of the analysis while working in a multi-dimensional parameter space is commendable. Overall, this manuscript delivers valuable insights into the complexity surrounding retrievals of snow depth on sea ice that is of great interest to the remote sensing community and has the potential to influence future snow depth algorithms. For this, I recommend this manuscript for publication in The Cryosphere following some revisions. In particular, I find that some specific details of the methodology require clarification and that there is little to no discussion placing the study into the context of existing retrieval algorithms or modelling studies, or discussion regarding the limitations of this study. Furthermore, I find the authors’ discussion around “snow ice” in the Arctic to be questionable. Please see below my comments and suggestions. These are divided into overarching, general comments and line-specific comments.

General Comments (GCs):

1: SMRT Configuration and Surface/Interface Roughness

The SMRT model is undoubtedly suitable for this study. However, by design, it is a highly configurable model. As such, user choices during model setup can influence the output. For example, the authors chose to represent the snowpack microstructure as sticky hard spheres (SHS). Several recent studies instead represent the snowpack using an exponential microstructure (e.g., Murfitt et al., 2024; Meloche, Sandells et al., 2024; Wivell et al., 2023; Meloche, Royer et al., 2024) as defined by a correlation length which can be derived using snow and ice densities and SSA. The cited literature on L95, Vargel et al. (2020), notably compares different SMRT snow microstructure configurations in reproducing observed brightness temperatures (Tbs). They found that a combination of the IBA model with snow layers defined by an exponential microstructure more accurately reproduced the Tbs than IBA with a SHS snowpack.

Because this study focuses on relative Tb differences, the absolute observation accuracy may be secondary. However, if Tb errors scale non-linearly, then these results may be a reflection of this specific model configuration. To ensure robustness, I recommend running a subset of the experiment using an exponential snowpack microstructure and checking consistency. This could be added to the supplementary material and briefly referred to in the main text.

Additionally, SMRT permits adding roughness to interfaces. Surface roughness is known to affect the Tb as observed by satellites (e.g., Stroeve et al., 2006; Lee et al., 2018); however, this is not mentioned in the manuscript. As there is no mention of this, I assume the study chose not to add any rough interfaces in the SMRT simulations, meaning the retrievability conclusions all assume ice and snow layers are flat. While I think the added complexity of accounting for surface roughness is likely outside of the scope of this paper, which focuses on snow microstructure, I have 2 recommendations based on this omission. 1) Discuss neglecting snow and ice roughness as a limitation of the method, and 2) Flatten the interfaces of the ice layers in Fig. 1 and Fig. 3., so that the reader is not misled into thinking rough interfaces were modelled (if indeed they were not).

Murfitt, J., Duguay, C., Picard, G., & Lemmetyinen, J. (2024). Forward modelling of synthetic-aperture radar (SAR) backscatter during lake ice melt conditions using the Snow Microwave Radiative Transfer (SMRT) model. The Cryosphere, 18(2), 869–888. https://doi.org/10.5194/tc-18-869-2024

Meloche, J., Sandells, M., Löwe, H., Rutter, N., Essery, R., Picard, G., Scharien, R. K., Langlois, A., Jaggi, M., King, J., Toose, P., Bouffard, J., Di Bella, A., & Scagliola, M. (2024). Altimetric Ku-band Radar Observations of Snow on Sea Ice Simulated with SMRT. Copernicus GmbH. https://doi.org/10.5194/egusphere-2024-1583

Wivell, K., Fox, S., Sandells, M., Harlow, C., Essery, R., & Rutter, N. (2023). Evaluating Snow Microwave Radiative Transfer (SMRT) model emissivities with 89 to 243 GHz observations of Arctic tundra snow. The Cryosphere, 17(10), 4325–4341. https://doi.org/10.5194/tc-17-4325-2023

Meloche, J., Royer, A., Roy, A., Langlois, A., & Picard, G. (2024). Improvement of Polar Snow Microwave Brightness Temperature Simulations for Dense Wind Slab and Large Grain. IEEE Transactions on Geoscience and Remote Sensing, 62, 1–10. https://doi.org/10.1109/tgrs.2024.3428394

Vargel, C., Royer, A., St-Jean-Rondeau, O., Picard, G., Roy, A., Sasseville, V., & Langlois, A. (2020). Arctic and subarctic snow microstructure analysis for microwave brightness temperature simulations. Remote Sensing of Environment, 242, 111754. https://doi.org/10.1016/j.rse.2020.111754

Stroeve, J. C., Markus, T., Maslanik, J. A., Cavalieri, D. J., Gasiewski, A. J., Heinrichs, J. F., Holmgren, J., Perovich, D. K., & Sturm, M. (2006). Impact of Surface Roughness on AMSR-E Sea Ice Products. IEEE Transactions on Geoscience and Remote Sensing, 44(11), 3103–3117. https://doi.org/10.1109/tgrs.2006.880619

Lee, S., Sohn, B., & Shi, H. (2018). Impact of Ice Surface and Volume Scatterings on the Microwave Sea Ice Apparent Emissivity. Journal of Geophysical Research: Atmospheres, 123(17), 9220–9237. https://doi.org/10.1029/2018jd028688

2: “Snow Ice” or simply Saline Snow?

The authors note relatively high salinity in the bottom ~10 cm of the snowpacks sampled in the MOSAiC dataset and deem this to be “snow ice” on the basis that salinity is assumed to be negligible for Arctic snow (L234). I disagree with this conclusion and the associated terminology. Firstly, I think it is generally accepted that snow at the base of the snowpack on FYI is often saline; one cause being the upward wicking of brine rejected to the ice surface during sea ice growth (e.g., Domine et al., 2004; Nandan et al., 2017; Mallett et al., 2024).  Secondly, widespread snow ice formation due to flooding and subsequent refreezing of the snowpack is generally considered to be an Antarctic phenomenon due to the Antarctic’s higher precipitation amount and higher fraction of seasonal ice. Although snow ice formation has been observed in the Arctic, I do not believe it is known to be widespread. The references provided on L259 (Merkouriadi et al., 2017; Merkouriadi et al., 2020) state that their results are confined to the Atlantic sector of the Arctic which is heavily influenced by storms.  

Additionally, snow ice in this manuscript is treated as a snow layer, but in its conventional definition, it is a solid layer of ice that is often indistinguishable from sea ice without isotopic analysis (e.g., Jeffries et al., 1997; Granskog et al., 2017; Arndt et al., 2021). To avoid confusion with the conventional definition, I recommend renaming this layer to “Saline Snow,” or similar, and removing the suggestion that the properties of the basal snow layers in these profiles were a product of flooding, unless there is clear evidence to support it.

Domine, F., Sparapani, R., Ianniello, A., & Beine, H. J. (2004). The origin of sea salt in snow on Arctic sea ice and in coastal regions. Atmospheric Chemistry and Physics, 4(9/10), 2259–2271. https://doi.org/10.5194/acp-4-2259-2004

Nandan, V., Geldsetzer, T., Yackel, J., Mahmud, M., Scharien, R., Howell, S., King, J., Ricker, R., & Else, B. (2017). Effect of Snow Salinity on CryoSat‐2 Arctic First‐Year Sea Ice Freeboard Measurements. Geophysical Research Letters, 44(20). https://doi.org/10.1002/2017gl074506

Mallett, R., Nandan, V., Stroeve, J., Willatt, R., Saha, M., Yackel, J., Veysière, G., & Wilkinson, J. (2024). Dye tracing of upward brine migration in snow. Annals of Glaciology, 65. https://doi.org/10.1017/aog.2024.27

Jeffries, M. O., Morris, K., Weeks, W. F., & Worby, A. P. (1997). Seasonal variations in the properties and structural composition of sea ice and snow cover in the Bellingshausen and Amundsen Seas, Antarctica. Journal of Glaciology, 43(143), 138–151. https://doi.org/10.3189/s0022143000002902

Granskog, M. A., Rösel, A., Dodd, P. A., Divine, D., Gerland, S., Martma, T., & Leng, M. J. (2017). Snow contribution to first‐year and second‐year Arctic sea ice mass balance north of Svalbard. Journal of Geophysical Research: Oceans, 122(3), 2539–2549. https://doi.org/10.1002/2016jc012398

Arndt, S., Haas, C., Meyer, H., Peeken, I., & Krumpen, T. (2021). Recent observations of superimposed ice and snow ice on sea ice in the northwestern Weddell Sea. The Cryosphere, 15(9), 4165–4178. https://doi.org/10.5194/tc-15-4165-2021

3: Limitations

Although this study follows a thorough methodology, it is not without its limitations. However, the manuscript currently lacks a real discussion surrounding this. I strongly recommend adding a dedicated limitations paragraph or subsection.

For example, the MOSAiC dataset represents a spatially limited sample of Arctic floes due to the nature of the expedition being “locked-in” to the drifting ice. Additionally, the simulations are specific to observations at AMSR2 frequencies and 55-degree look angle. Furthermore, as mentioned in GC1, SMRT configuration choices can impact such simulations, and this study does not assess the quality of the absolute simulated Tb values. Perhaps another SMRT configuration (e.g., exponential snow microstructure) could be used to check for consistency, or the configuration could be evaluated by also simulating the Tbs observed by the ground-based radiometers on MOSAiC.

Finally, in the context of simulating AMSR2 data, it is important to note that observed Tbs do not come from a single, homogeneous snowpack. While the study investigates different ice types within a footprint, it does not investigate the potential spatial heterogeneity of the snowpack within a footprint. Because this is a fundamental study, it is acceptable to assume a homogeneous snowpack; however, it should perhaps be noted as a limitation when drawing conclusions for practical remote sensing applications.

4: Contextualisation of Study

I think the reader would appreciate some more context surrounding existing snow depth retrieval algorithms in the introduction and why they are limited, i.e., that they are typically parametric equations or neural networks trained on airborne or ground-based springtime data (often biased to the Western Arctic, or even Antarctica), therefore representing a limited snapshot of the metamorphic state of the snowpack and making their year-round applicability uncertain. Furthermore, there should be some introduction and discussion regarding prior modelling works that have sought to provide some understanding of the influence of snow properties on the retrievability of snow depth. For example, Markus et al. (2006), Powell et al. (2006), and Rostosky et al. (2020).

Markus, T., Powell, D. C., & Wang, J. R. (2006). Sensitivity of passive microwave snow depth retrievals to weather effects and snow evolution. IEEE Transactions on Geoscience and Remote Sensing, 44(1), 68–77. https://doi.org/10.1109/tgrs.2005.860208

Powell, D. C., Markus, T., Cavalieri, D. J., Gasiewski, A. J., Klein, M., Maslanik, J. A., Stroeve, J. C., & Sturm, M. (2006). Microwave Signatures of Snow on Sea Ice: Modeling. IEEE Transactions on Geoscience and Remote Sensing, 44(11), 3091–3102. https://doi.org/10.1109/tgrs.2006.882139

Rostosky, P., Spreen, G., Gerland, S., Huntemann, M., & Mech, M. (2020). Modeling the Microwave Emission of Snow on Arctic Sea Ice for Estimating the Uncertainty of Satellite Retrievals. Journal of Geophysical Research: Oceans, 125(3). https://doi.org/10.1029/2019jc015465

5: Quoting Results

I find the results section to be well done and interesting. However, I do suggest integrating more specific quantitative values (e.g., of TSI and SSI values) into the text to support your points and save the reader from needing to continuously scroll back to the figures.

Line-specific Comments:

L30-31: The snow not only restricts oceanic heat transfer through the ice but also insulates the ice and ocean from atmospheric heat fluxes, both of which influence the sea ice mass balance as per the following sentence.

L36: I think “all-sky” here refers to day and night / regardless of the solar zenith angle. In which case, I do not think it should be hyphenated.

L38-39: Penetration depth is itself a function of frequency-dependent absorption and scattering within the system and perhaps listing it individually is redundant.

L43-45: Such variability does not inherently show that the algorithms are problematic, as snow depth is expected to vary in space and time. Consider rephrasing this sentence to highlight that biases caused by unaccounted for snow microstructural properties can lead to unrealistic variability, using appropriate references.

L45: I think further context on the “limited application ranges of existing snow depth algorithms” would be appreciated here (in line with GC4); explaining why current algorithms are limited, e.g., perhaps due to their simplistic parametric form that was derived using airborne campaign data that is spatially and temporally limited, reflecting a metamorphic state of the snowpack on Arctic sea ice that is unrepresentative of the entire winter. Overall, there is little explanation of how current algorithms “work” or were derived in this paragraph.

L64-65: Although this work may provide the most comprehensive study into the impact of microstructural properties on the retrievability of snow depth on Arctic sea ice and may be the first to consider coupling effects into the analysis, I do not agree there has not been “similar work done before.” Others have also modelled Tb as a function of snow and ice microstructure to assess uncertainty in snow depth algorithms, and I think these should be noted in this manuscript. For example, Markus et al. (2006), Powell et al.,(2006), and Rostosky et al. (2020), per GC4.

L90-91: I am a bit unsure of the phrasing here regarding “coupled snow and sea ice stratification” configuration. Presumably this is referring to the experiment in which the fraction of each ice type is varied? However, it makes it sound like the single- and multi-layer snow configurations are not coupled to a sea ice substrate. Consider rephrasing.

L93-95: Please see my GC1 regarding the SMRT setup and possible extension to ensure robustness using another model representation of snow microstructure.

L193: There is a minor inconsistency in this list. The instrumentation used is listed for all parameters but salinity, in which the location the sample was measured (the lab) is listed. Listing the lab is redundant and it would be better to list the instrument (presumably salinometer?)  to match the rest of the list.

L202/Section 3.1 and 3.2: Consider swapping the order of these two sections. The MOSAiC observation section flowing into Section 3.3, how it was used to create the layered snow scenarios, seems more natural.

L210: Perhaps reference Table 1 in Fuhrhop et al. (1998) directly to help guide the reader.

L216-217: While the description of snow ice formation is true, in accordance with GC2, it may not be relevant here.

L219: I recommend changing “freezes in the snow-ice surface” to “freezes at the snow-ice interface.”

Figure 2: Check punctuation in the panel descriptions (inconsistencies with colons and semicolons). Also, consider changing “the horizontal bars are the standard deviations” to “the horizontal bars represent the standard deviation”.

L221/Fig. 2 caption/306/419. British English “colour” is used, however the rest of the manuscript is consistent with American English (e.g., “utilizing”, “summarized”, “parameterized”). I recommend changing to the American English “color”.

L233-239: Please refer to GC2.

L244: The choice of parameter space is key to this study. However, it is not clear exactly how the “physically plausible ranges” were derived. Was there a quantitative selection process, or were they selected based on the authors’ judgement? Furthermore, how were the MOSAiC measurements combined with the snow properties listed in Fuhrhop et al.(1998) to define the study’s ranges? Please add a sentence or two detailing the range selection process more clearly.

L283-286: The MYIF study is a great addition due to the known challenge of PM snow depth retrievals over MYI but the unknown nature of how this may couple with snow microstructure properties and the fraction of ice types within a footprint. However, please consider adding a brief 1-2 sentences on why the MYI bias is perceived to exist (i.e., scattering within the porous upper ice layer that mimics the scattering of a snowpack). Additionally, it should be stated what snowpack is assumed for the MYI here – I assume it is the same as the FYI dry snow single layer?

L302: I suggest changing the section header to “Results and Discussion.” Discussion points are interwoven into this section and there is no stand-alone discussion section.

L359-361: See GC2 regarding Arctic snow ice and its occurrence.

Figure 4: What determines when TSI and SSI values are overlaid on the figure? There are clearly valid instances for a given channel, layer, and physical parameter that do not have a corresponding number set overlaid. Are TSI and SSI values over a certain threshold deemed to be of interest?

L413: Why “based on V-pol TBs”? I assume this is because V-pol is the typical choice for existing snow depth algorithms, however the reader may not know this without context (see GC4).

L495: Consider adding an “e.g.,” before Markus and Cavalieri, 1998 as it is not the only study demonstrating this.

L496: Perhaps an opportunity to relate this result (GR(18/6) extending the snow depth retrievable range under large grain snow conditions) to the studies of Powell et al. (2006) and Rostosky et al., (2018).

Powell, D. C., Markus, T., Cavalieri, D. J., Gasiewski, A. J., Klein, M., Maslanik, J. A., Stroeve, J. C., & Sturm, M. (2006). Microwave Signatures of Snow on Sea Ice: Modeling. IEEE Transactions on Geoscience and Remote Sensing, 44(11), 3091–3102. https://doi.org/10.1109/tgrs.2006.882139

Rostosky, P., Spreen, G., Farrell, S. L., Frost, T., Heygster, G., & Melsheimer, C. (2018). Snow Depth Retrieval on Arctic Sea Ice From Passive Microwave Radiometers—Improvements and Extensions to Multiyear Ice Using Lower Frequencies. Journal of Geophysical Research: Oceans, 123(10), 7120–7138. https://doi.org/10.1029/2018jc014028

L522-525: It is my understanding that the lower frequency is typically considered to be the ‘anchor’ for GR-based snow depth retrievals because it is conventionally regarded to be less sensitive to volume scattering within the snowpack. Check terminology.

L525-528 and L566-568: The discussion surrounding replacing the 36 GHz (L525-529) and 18 GHz (L566-568) channel with 23 GHz for these applications is interesting. But it may be worth noting the ~23 GHz channel is seldom used for sea ice applications, I believe due to its strong sensitivity to atmospheric water vapour. If this channel were to be used operationally, it would likely require an atmospheric correction using a radiative transfer model. On this topic, I appreciate the supplementary material demonstrating that the higher frequency channel observations from AMSR2 can correlate better with in situ data following an atmospheric correction (depending on the atmospheric state) – perhaps this operational requirement could be reiterated in your conclusions?.

L543: Reiterating that Sect. 3.4. lacks clear information of the snowpack set up for this ice type experiment. Were both types considered to have the same snowpack here?

L544-545: It may be worth noting that the spatial response of the AMSR2 antenna is not uniform. Because the antenna gain function is approximately Gaussian, a floe’s actual contribution to the measured Tb depends on its relative position in the footprint. Therefore, a simple linear weighting is a simplification.