Billecocq at al present an interesting study on refining the spatial resolution of meteorological forcings to feed a detailed snow model (ALPINE3D). The ultimate goal is to test whether this refinement improves the representation of snow microstructure as relevant to SWE retrieval algorithms from satellite. Results show improvements for the optical grain size and SWE for two seasons of data in the Glacier National Park in Canada.

Overall, the study is well conceived and the paper is well written and concise. The topic is relevant and within the scope of TC. At the same time, there are in my opinion a number of major and minor points that should be addressed before publication. Thus I am recommending a major revision.

The first major comment is that all snow evaluations are performed using simulated (not observed) time-series at only three locations over the study area. In the discussion, authors are clear on this being a limitation of their study (lines 312). However, I think this point should be better addressed throughout the manuscript as the main critical aspect of this work. Ideally, the best solution would be to include observations in this evaluation exercise, but it may be that such observations are not available at the considered study site. So I see two potential alternatives: (1) include results from other regions where such data are available, and/or (2) better discuss accuracy and precision of SNOWPACK simulations forced using AWS data using reference literature (e.g., https://tc.copernicus.org/articles/9/2271/2015/)

Second, results are promising with regard to snow depth / SWE, but quite incremental when looking at the optical grain size and density (see line 16 and then the results section). The same could be said with regard to weather forcing data, where a clear benefit of downscaling is evident (in my opinion) for radiation and humidity, while results for temperature and precipitation are mixed. While authors are again clear on this (see the discussion section for example), and while I totally see the main point of novelty provided by the authors (line 325), I am still wondering what is the significance of this work for the global audience of TC given these mixed results and the fact that authors focused on a comparatively small region and two years of data. To overcome this, I am proposing to (1) include a clearer justification regarding the choice of this study region, including why it is important for the global readership of TC; (2) significantly expand the Discussion section with much clearer statements of the main findings, implications, and future steps in view, and in the context, of the relevant literature; (3) ideally, include specific research questions in the Introduction to further generalize findings.

MINOR / SPECIFIC COMMENTS

- Abstract: in my view, the abstract focuses too extensively on background information (up to line 10). I would recommend summarizing this background information to focus on the main findings and implications

- line 8: this maximal resolution of 2.5 km is likely specific for Canada datasets (?)

- line 31: this statement on models yielding biased estimates of SWE at high elevation is likely too generic. Several correction approaches in this regard have been documented, but results are very site specific (which is in my opinion the actual main challenge here)

- line 49: I think that the main reason why AWS spatial interpolation in complex terrain is not accurate is because AWS systems undersample the real spatial heterogeneity of the processes (which is not mentioned here)

- line 51: AWS systems are also prone to undercatch and so underestimation of precipitation (this is one of the main reasons why I think it would be ideal to include actual measurements of snow properties in the evaluation).

- line 76: please avoid reporting units in italics

- Figure 1: consider including a DEM here

- Table 1 and all other captions: please consider defining acronyms in captions for diagonal readers

- line 88: please specify “most of Canada”

- line 100 to 110: correction factors for temperature, radiation, and precipitation are very succinctly presented, to the extent that repeatability of these experiments may be difficult. How was Eq. 1 derived (what data were used? What period? What optimization approach?). Same for Eq. 2. Why was Equation 1 used for dew point temperature too?

- Line 136: why did you first use a 20-m DEM and now a 100-m one?

- Section 3.3: the inflation approach is clear, and I generally agree with this. At the same time, microstructural parameters are (to some extent) dependent on SWE and HS (via overburden pressure and temperature gradients, for example). If authors agree with this, I would add some discussion on how this could impact these results.

Review of: Subgridding High Resolution Numerical Weather Forecast in the Canadian Selkirk range for local snow modelling in a remote sensing perspective, Billecocq et al., EGUsphere, 2023.

General comment

Billecocq et al. present a study introducing a subgridding method to downscale Numerical Weather Prediction model outputs to drive a detailed snowpack model, in order to produce spatially distributed estimates of SWE and snow microstructure properties at 100 m resolution. These estimates are necessary to provide a first guess of the snowpack structure to SWE retrieval algorithms in a perspective of remote sensing.

The paper covers a topic of interest for the cryospheric community and fits the scope of The Cryosphere. It is overall concise and reads easily, with helpful figures. It is an interesting contribution to that topic, but the following major issues need to be addressed. Please also see the specific comments for a more detailed description of these general issues.

1. The research plan remains somehow unclear in the paper. I would recommend to better define the research gap in the introduction, after a more complete literature review. What is the exact novelty of the paper? Almost no literature review is done on existing NWP downscaling methods and frameworks, while it seems to be identified as the main contribution of the paper. The definition of the research gap should then be followed by a clear exposition of the research questions that the study addresses. Discussion and conclusion could then be better structured to answer these research questions, based on the exposed methods and results. In the end, it could enable a better structure of the paper, enrich introduction and discussion sections and prevent reader’s misunderstandings (which may be the cause of some of my comments).
2. The validation of the snowpack simulations against AWS-driven snowpack simulations is problematic. First, they provide three point comparisons which cannot be representative of the domain. The lack of spatially distributed snowpack measurements and the absence of more point measurements are understandable, but the comparison cannot really be considered as validation. Indeed, because of error compensations, a better weather input (from AWS measurements) does not necessarily provide a SWE or HS estimate by the model closer to actual snowpack observations. Additional snowpack measurements should be included for validation, and if not possible, the authors should be careful with the used words (e.g., “validation”, “improve”, “perform better”, …). It also remains unclear if the HS measurements were used or not (i.e., are the reference station curves direct measurements or AWS driven simulations?). If so, it provides a first element of validation, but it should also be compared to the AWS driven simulations, that are considered as references otherwise.
3. The spatial variability study is incomplete: six arbitrary points are not representative. Instead, pixels of the whole domain could be aggregated by topographic categories.

I would therefore recommend a major revision of the paper.

Specific comments

• In general, the captions of all figures must be edited (in particular, from Figure 3 to the end). They need to be more descriptive of all represented variables.
  
  
  
  
• Abstract, l. 1-7: the part of the abstract about remote sensing could be shortened. It is only a context element, so it could be mentioned in one or two short sentences in the abstract.
  
  
  
  
• Abstract, l. 8-9 and Introduction l. 52-53: “too coarse to capture snow spatial variability in a complex topography”. It always depends on the scale of variability you need to resolve, and so mostly depends on the modelling goal. “too coarse” should then be related to the application. The introduction needs to justify why this scale of 100 m is chosen, and why it is the appropriate resolution to resolve the spatial variability required by the remote sensing application.
  
  
  
  
• l. 45-46: detailed snowpack models could be a bit more extensively described. It could also be through a more complete description of the SNOWPACK model in the methods (e.g., when mentioning Alpine3D).
  
  
  
  
• l. 48-50: also mention spatial representativity issues of AWS.
  
  
  
  
• l. 50-51: “high-resolution atmospheric models are known for their negative bias in precipitation”. This statement is too generalized: is it the case for all high resolution atmospheric models, in all regions?
  
  
  
  
• l. 55-57: Why should this particular downscaling be discarded, regarding snow microstructure? It needs more justification.
  
  
  
  
• l. 58-59: Many downscaling methods exist for driving snowpack models in complex terrain. The literature review about atmospheric downscaling could be largely extended (only one paper is cited). Secondly, the authors should justify clearly why existing downscaling methods are irrelevant for SWE and microstructure retrieval from snowpack models. The authors could be clearer about their research gap, to better identify the novelty compared to other recent studies. For example, Marsh et al. (2020) offer a modelling framework in Canada, including meteorological downscaling and the SNOWPACK model. Sharma et al. (2023) use dynamical downscaling of NWP with WRF within CRYOWRF (including the SNOWPACK model). A more complete literature review should enable the authors to better highlight the novelty of their study.
  
  
  
  
• l. 79: Why not also including the melt season?
  
  
  
  
• l. 81 and l. 272: “round grains”, “defragmented grains”. Please stick to the official classification of grain shapes by Fierz et al. (2009). Here, respectively: “rounded grains”, “decomposing and fragmented precipitation particles”.
  
  
  
  
• Figure 2: I assume a typo (“VWS” for “VW”)
  
  
  
  
• Figure 2: please use the full word “microstructure”
  
  
  
  
• l. 106: “Snow precipitation water equivalent” -> Snowfall
  
  
  
  
• l. 107-108: Were these precipitation boards located in areas free of wind-induced erosion or accumulation? This is worth mentioning.
  
  
  
  
• l. 112-122: This is all methods and results from Helbig and Löwe (2014), which should be cited. I would delete all these lines and equations and simply say Fsky is computed following Helbig and Löwe (2014).
  
  
  
  
• l. 123-135: Once again, all of this is from Helbig et al. (2017), so not new. I don't see the need to reproduce all equations here if the authors simply say they use their downscaling method.
  
  
  
  
• l. 141: “ILWR was spatialized using IDW”. ILWR is strongly dependent on terrain elevation (e.g., Marty et al. (2002) found a climatological vertical gradient of - 29 W/m²/km in the Alps). I would assume a simple IDW is not sufficient to downscale ILWR in complex terrain, or am I missing something?
  
  
  
  
• l. 141-142: “All the algorithms mentioned above are a part of the MeteoIO library, which is integrated into the Alpine3D model”. It is a bit unclear here what is novel in this study (“We designed a logarithmic regression (…)”, l. 98) and what is already existing in Alpine3D. Clarification is necessary.
  
  
  
  
• l. 143: The snowdrift scheme is probably turned off since wind-driven redistribution is parameterized by a precipitation multiplier? It could be worth mentioning.
  
  
  
  
• l. 144: “considering the spatial processes affecting atmospheric variables”. Do you mean the atmospheric downscaling already described above? Please reformulate or clarify.
  
  
  
  
• l. 147-150: Why choosing individual points? How have they been chosen? They are not necessarily representative. Why not aggregating values by categorical topographic areas instead? What's the model slope at the chosen points? It can have a significant impact on ISWR differences between North and South.
  
  
  
  
• l. 152-153 and further: as mentioned in the general comments, I would not call a comparison of SGF-SNOWPACK vs AWS-SNOWPACK a validation of snowpack simulations. Some studies have shown that a "better" weather input could degrade the metrics of snowpack simulations, because of error compensations in the interplay of weather input and modelled snowpack processes. Please be careful with the wording “validation”, “better”, etc., to qualify the comparison. "Simulation A is closer to Simulation ref than Simulation B" would be more accurate.
  
  
  
  
• l. 186: as mentioned before, simply say “HRDPS tends to”...
  
  
  
  
• l. 192: “a mean layer-by-layer bias for density and OGS”. This needs to be clarified a bit. Is it a mean value for a snowpack profile where a 1 mm layer would weight the same as a 50 cm layer? Is there a weighting?
  
  
  
  
• l. 193: “Height of Snow and SWE were visually assessed”. Why not metrics?
  
  
  
  
• l. 195-197: The calculation of NSE may not need to be described here.
  
  
  
  
• l. 220-222: The bias should be computed as model - reference, so that a positive bias would mean an overestimation, and it should be the case for all variables. It would avoid unnecessary confusion.
  
  
  
  
• Figure 3: The vertical labeling is somewhat confusing. Perhaps simply write TA bias (°C), RH MAE (%), etc.?
  
  
  
  
• l. 225-226: It could be worth mentioning it corresponds to very shallow snowpacks.
  
  
  
  
• Figure 4, green curve: I would not call it Station, since it could be confusing for the reader assuming it's a station observation. More understandable labels could be, for example: SGF-SNOWPACK, HRDPS-SNOWPACK, AWS-SNOWPACK. Or is it actually the HS measurement in green? If so, the AWS driven SNOWPACK simulation should also be represented since it is the reference for the other metrics. Moreover, in Figure 6, the green curve is called SWE Station, even though there is no SWE measurement at the station (according to Table 1). This is very confusing and should be clarified.
  
  
  
  
• l. 239-241: Isn’t it rather related to the fact that the metrics are computed over very shallow snowpacks?
  
  
  
  
• l. 241-242: “the numerical analysis of the results was carried out starting on the first of December.” Is it also the case for the similarity metrics exposed a few sentences earlier? If so, please clarify.
  
  
  
  
• l. 242: “HRDPS and the subgridding framework are slightly overestimating OGS”. The concision of this paper is overall appreciated, but it might be clearer to use formulations such as HRDPS-SNOWPACK and SGF-SNOWPACK, because snowpack-related variables are not an output of HRDPS and the subgridding framework.
  
  
  
  
• l. 246-253: As far as I understand, the mean density bias is a mean of the biases of all layer densities, i.e., not equivalent to the bulk density bias. It would deserve to be better clarified. It would also be necessary to justify why this layer mean is chosen over a bulk density. It seems very dependent on the layering?
  
  
  
  
• l. 255-256: "As a result (…)". This logical link is unclear. The SGF could also reduce the PSUM bias. Please reformulate or clarify.
  
  
  
  
• l. 263: The observed altitudinal temperature gradient is the reflect of the lapse rate chosen for TA downscaling (l. 137). There is no proof here it is realistic.
  
  
  
  
• l. 264-265: “slightly lower temperatures in the north aspects”. Figure 7 shows the contrary for TA in Alpine area (warmer TA on North slopes vs South slopes, consistently throughout the season). Any explanation? Or is it a plotting error?
  
  
  
  
• l. 268-269: “the altitudinal precipitation rate gradient is also respected by the subgridding framework, with precipitation rates getting higher with elevation”. Is there any incremental improvement compared to the original HRDPS gradient?
  
  
  
  
• l. 270: What is the reason for simulating the snowpack in forested areas (in a remote sensing perspective), if the forest snow processes, which have a strong impact on the snowpack, are not represented?
  
  
  
  
• l. 271-272: “The wind erosion effect on the snowpack is also well represented, as dominant winds are blowing from the South / South-West. As a result, the south aspect profiles show more defragmented grains (dark green) on the surface”. I am not sure I understand this cause-consequence. As far as I understood, wind-induced snow transport is represented by a precipitation multiplier. Consequently, associated effects of snowdrift on snow microstructure are not represented. Or am I missing something? Please clarify.
  
  
  
  
• l. 274: “slower settlement”. This needs to be proven, it is not obvious when looking at the figure.
  
  
  
  
• l. 274-275: An extension of the simulations to Spring would be interesting to assess the melting processes and differences between North and South slopes.
  
  
  
  
• l. 280: “a considerable improvement”. With respect to the results exposed in Figure 3, this assertion could be more nuanced.
  
  
  
  
• Figure 8: Please provide the label for grain type color codes.
  
  
  
  
• l. 290-295: It could be clarified.
  
  
  
  
• l. 300: The discussion could explore further the reasons why and how simulated microstructure parameters are modified, with the modified input.
  
  
  
  
• l. 306-310: The authors could provide typical snow OGS values to give an idea of the relative change.
  
  
  
  
• l. 311: “wind transport in the alpine is likely exaggerated”. This statement needs to be justified.
  
  
  
  
• l. 314-323: In the current state, this paragraph is more a perspective for the conclusion section.
  
  
  
  
• l. 325-326: See previous comments about clearer identification of the research gap.
  
  
  
  
• l. 341: The word "stabilize" may be misused, since the SNOWPACK simulations are probably not strictly speaking "unstable"?

References

Fierz, C., Armstrong, R.L., Durand, Y., Etchevers, P., Greene, E., McClung, D.M., Nishimura, K., Satyawali, P.K. and Sokratov, S.A. 2009. The International Classification for Seasonal Snow on the Ground. IHP-VII Technical Documents in Hydrology N°83, IACS Contribution N°1, UNESCO-IHP, Paris.

Helbig, N., and Löwe, H.: Parameterization of the spatially averaged sky view factor in complex topography, J. Geophys. Res. Atmos., 119, 4616–4625, https://doi.org/10.1002/2013JD020892, 2014.

Helbig, N., Mott, R., Van Herwijnen, A., Winstral, A., and Jonas, T.: Parameterizing surface wind speed over complex topography, J. Geophys. Res. Atmos., 122, 651–667, https://doi.org/10.1002/2016JD025593, 2017.

Marsh, C. B., Pomeroy, J. W., and Wheater, H. S.: The Canadian Hydrological Model (CHM) v1.0: a multi-scale, multi-extent, variable-complexity hydrological model – design and overview, Geosci. Model Dev., 13, 225–247, https://doi.org/10.5194/gmd-13-225-2020, 2020.

Marty, C., Philipona, R., Fröhlich, C., and Ohmura, A.: Altitude dependence of surface radiation fluxes and cloud forcing in the Alps: results from the alpine surface radiation budget network, Theor. Appl. Climatol., 72, 137–155, https://doi.org/10.1007/s007040200019, 2002.

Sharma, V., Gerber, F., and Lehning, M.: Introducing CRYOWRF v1.0: multiscale atmospheric flow simulations with advanced snow cover modelling, Geosci. Model Dev., 16, 719–749, https://doi.org/10.5194/gmd-16-719-2023, 2023.