General comments:

Congratulations to this very interesting paper! It brings the front of snow hydrological research one step further. The combination of the different modelling approaches is a valuable effort to combine methods, each of which as appropriate as possible for the scale, to ultimately integrate all relevant processes that determine the variability of snow depth in high mountain regions. Step by step we are coming closer to a snow hydrological model which allows robust prediction of snowmelt dynamics and, maybe even more important, of climate change effects on the snow distribution and its melting regime when combined with predictions from convection permitting climate models. This paper is an important contribution to this endeavour.

From my point of view, three issues desire some attention prior to finalizing the manuscript. The rest are minor comments. 

The English is very good, I only found few details in the text where I suggest an alternative formulation.

Specific comments:

1) I recommend the authors to add a paragraph for the integration of the models and their timing: how were the submodels parameteriized (wind-induced snow redisribution, avalanches)? Does this parameterization depend on the scale (model/DEM resolution)? What triggers an event (blowing snow, avalanche)? What is the order of the computations in a time step, does this play a role? If yes, why is the chosen order the better one? These are all interesting questions for modellers and should be presented at least briefly.

2) To my knowledge, SnowSlide updates the DEM surface elevation after each redistribution event with the accumulated mass of snow, thereby filling depressions and/or building up snow depositions in the runout zones of an avalanche. Isn’t this the feature in SnowSlide that controls the runout area size of the snow if another (one after the other, actually) avalanche flows down the same slope/couloir (apart from parameters like maximum accumulation per pixel etc.)? This should be discussed in chapter 2.3.2., together with the new „hysteretic feature“ (in a bit more detail).

3) it would be nice to (make an attempt at least to) to evaluate the results of the single process simulations: solid precipitation amount, the new layering scheme (wetting events, density of the modelled snow layers), the modelled wind-induced lateral snow redistribution and the modelled avalanches as well.

Technical corrections:

- 14-15: „… from the peak of winter to the end of the melt season“: but not before peak of winter? Why? This should be mentioned here

- Figure 1: the left panel should be larger (same size as the right one)

- 95: „mostly in open terrain“: what about forests, are these omitted here? There is probably a good reason for this, but it also should be expressed here

- 116, 142 and 151: I recommend to insert a table here with all existing FSM versions, including the original(s) by Richard Essery and all the follow-ups, including their names, references and main differences

- 152-177: it would be nice for the reader if you show the effect of the two processes by means of an example simulation for a small but typical sub-area of one of your domains

- 169-177: are you using a SnowSlide version that updates the DEM surface elevation after each simulated transport event (i.e., adds deposited snow to a new surface elavation so that the next avalanche flows over it) to prevent „endless“ increase of snow depth in depressions? See specific comment No. 2.

- 160: are these „adaptions and improvements“ that are discussed in the following? Maybe this could be made clear here

- 169: maybe better „using“ instead of „offering“

- 170: is the „snow holding thickness“ a snow depth threshold? The it should be mentioned here. A more general term would be „snow holding capacity“.

- 171ff: how did you tune the SnowSlide parameters? See my specific comment No. 1.

- 174: are the „few improvements“ the ones presented in the following?

- 176: „extent“: this means a larger deposition area, right? If yes, why not name it like this?

- 199-209: what can you say about the accuracy of the LIDAR-derived dataset? See my specific comment No. 3.

- 203: could you indicate explicitely earlier that you limit simulations to non-forested areas (see comment to line 95)?

- 204: 31 March 2017 is also covering the melting period?

- 209: evtl. better „aggregated to“

- 216: better „by“ Winstral et al. (2017) and Dujardin and Lehning (2022)

- 220: you have both „snow depth“ and „snowdepth“ throughout the text. The former one is correct

- 225: probably better „for“ subdomain B0 (all through the text where this occurs), and „while Fig. 4 shows subdomain“ … 

- 232: what do you mean with „spatialized“ snow depth measurements, an interpolation result?

- 233: better „produces too little snow“

- 236: does „deposit extent“ refer to area or mass, or both? I also think that it would better be „deposition“ than „deposit“

- 238: probably „accumulations“ should better be singular, because it refers to the general nature of the process; or do you mean specific events?

- 241: here „accumulations“ probably means „accumulated mass“?

- 243: what are the „new hysteretic features of the avalanche model“? Maybe the slope threshold application mentioned in Sect. 2.3.2.? This deserves a more detailed explanation (see comment to lines 169-177 and specific comment No. 1)

- 246: I think it should be „spring“ (lowercase; everywhere)

- 254: what do you mean with „resolutions … are irrelevant“? How can a resolution be irrelevant? Eventually you mean that the simulation results achieved for these resolutions do not properly reproduce redistribution processes …

- 258: is the reason for this the precipitation interpolation method the increase with altitude (the lapse rate)?

- 265: find something better than „over the whole subdomains“ (what exactly do you mean with it, areas with TPI≤200?)

- 272: what do you mean with „global“, maybe „regional“ or „in general“?

- Figure 3: better „Map of snow depth on 17 March…“, „for“ subdomain … and aggregated „to“. An image showing the difference between a) and b) would be very informative for the reader because it shows the spatial pattern…

- Figure 4: same as for the caption of Figure 3

- Figure 5: same as for the captions of Figures 3 and 4

- Figure 10: better „aggregated for the whole domain“

- chapter 5.2: see specific comment No. 3.

Review of the paper “Snow redistribution in an intermediate-complexity snow hydrology modelling framework” by Quéno et al. submitted to The Cryosphere.

This paper presents the development of a new modelling framework to simulate snow redistribution in mountainous terrain that can be applied over large domains with limited computational times. Such system is needed in the context of operational modelling of mountain snow hydrology in Switzerland. The authors give first an overview of the modelling system that accounts for wind-induced and gravitational snow transport. Its capacity to simulate snow distribution in mountains is then evaluated over a simulation domain surrounding Davos in Switzerland. Maps of snow depth derived from airborne LIDAR are used as a reference. The results show that model can generate realistic patterns of snow accumulation in mountainous terrain, including snow-free ridges, enhanced accumulation at the bottom of steep slopes, … Strong improvements are found in the distribution of snow depth close to peak snow accumulation and these improvements persist during the melting season. The main simulations of the paper were carried out at 25-m grid spacing. Additional simulations at 50- and 100-m grid spacing showed that improvements in snow distribution were also found at these resolutions, opening interesting opportunities for future operational system.

The subject of this paper is very relevant for the mountains snow hydrology community and the results shown here suggest that simulations including snow redistribution could be soon used in an operational context. The paper is well written, easy to follow and should ultimately be published in The Cryosphere. However, prior to publication, the authors should strengthen the results section to avoid statements that are not well supported by the figures and tables presented in the paper. This work would also benefit from a more quantitative approach relying on error metrics when comparing the different simulations and the observations. These two general comments are described first and are then followed by more specific and technical comments.

General comments

1. The results section of this paper starts with a comparison between simulated snow depth and observations from airborne Lidar (Section 4.1). This section is purely based on the visual comparison of maps (Fig 3 to 5) and probability distribution functions (PDF) of snow depth (Fig. 6 to 8). This section contains several statements that are not well supported by the results presented in these different figures. I recommend the authors to carefully revise this section and to remove the unsupported statements. Some of them can certainly be detailed introduced later in the text (in the discussion section for example), once more quantitative results have been presented (see my second general comment).

The first statement concerns the impact of combined snowdrift and avalanche modelling (P 9 L 230-232). I fully agree with this statement, but I find that it is not well supported by the results shown on the two maps discussed here. It could have been better illustrated by considering simulations that consider only avalanching or wind-induced snow redistribution. I think Figure 9 helps to illustrate this interplay and the authors could make this statement later in the paper.

A second statement is then made about the influence of the precipitation forcing (P9 L 233). At this stage of the analysis, it is not clear at all that the precipitation forcing can explain the underestimation of FSM2trans at the highest elevations. For example, Figure 3c does not suggest clearly that FSM2ref underestimates the snow depth at high elevations. A comparison of simulated and observed distribution of snow depth as a function of elevation could be used to show that FSM2ref (without redistribution) underestimates the snow depth at high elevation. This would strengthen the statement about the precipitation forcing. At this stage, it is not clear if this underestimation of snow depth is due to an overestimation of the intensity of wind-induced snow transport over exposed ridges in FSM2trans.

A third statement explains that certain features of snow accumulation are due to” the new hysteretic features of the avalanche model” (P9 L 243). How would they look without the new features? These features are described in Section 2.3.2 but the motivations behind this development are never explained in the paper. A figure that shows patterns of avalanche deposition in the default and in the revised version of SnowSlide would be useful to understand why the revised version should be used in step alpine terrain. It could certainly be added in the supplementary material.  

A fourth statement affirms that “FSM2ref can capture the average state of the snowpack over the subdomains” (P 9 L 253-254).  and it is not clear at this stage of the paper. Quantitative metrics are required to show that that the average state of the snowpack is indeed well captured by FSM2ref (see my second general comment). In addition, L 254 refers to simulations at 50 and 100 m whereas no result from these simulations have been presented at this stage of the analysis.

2. Figure 6 to 8 show very convincing improvements in the ability of the model to simulate snow distribution in alpine terrain. However, at this stage, the comparison is purely qualitative. A more quantitative approach would significantly improve the paper. It could be used when (i) comparing FSM2ref and FSM2trans (P9 L 250-255), (ii) comparing the results for the full sub-domains and for ridges only (P11 L 265) and (iii) discussing the impact of the model grid spacing (P 9 L 254-255; P11 L 268-273). The visualization developed for Figure 10 could be used to present the distribution of error metrics (bias or RMSE for example) as a function of the elevation and orientation of the grid cells.

Specific Comments

P2 L 57: note that Liston et al. (2020) have developed a multi-layer version of SnowModel.

P 3 L 66-70: it would be interesting to mention here the recent developments of deep learning methods to downscale wind in complex terrain and to provide forcing to blowing snow scheme. See for example Le Toumelin et al. (2023).

P3 L 75-76: Could you mention here feedback from users that have pointed out the limitations associated with the absence of snow redistribution in the operational model used at OSHD?

P3 L 79: a distributed version of SnowModel has been recently applied at 100-m grid spacing over the contiguous Unites States by Mower et al (2023). The paper is still in discussion, but I still recommend the authors to add a sentence or two about this new implementation of SnowModel.  

P4 L 97: what is the source of data used to generate the DEM at different resolutions?

P4 L 107: it would be interesting to add here a few sentences that describe how the OSHD version of FSM2 differs from the standard FSM2 version.

P 5 L 115: the authors have changed to layering in FSM2 to improve the simulation of surface snow properties and to better estimate snow erodibility. However, a change in the snow layering in a multi-layer snowpack model can also have an impact on the simulation of snow compaction, heat transfer and liquid water percolation through the snowpack, … Overall, can the authors comment on the impact of the new layering scheme on the simulation of seasonal snow evolution by FSM2? I guess it has been tested in the context of model development, especially if this version will ultimately replace the operational version of FSM2oshd.

P5 L 116: the readers need to understand the novelty of the changes made to FSM2. For this reason, I recommend adding a short description of the original layering scheme used in FSM2. It will allow the reader to understand why such a scheme was not appropriate to represent the properties of surface and near-surface snow that are crucial when simulating snow transport.

P 5 L 131: a few sentences describing the regridding steps (conservation of mass, energy, …) would be useful.

P6 L 153: I am not familiar with the code management of SnowTran3D but, if possible, I recommend adding the version number of SnowTran3D that has been used when implementing it into FSM2trans.

P 6 L 156-157: it would be interesting to add a few references describing the application of SnowTran3D at these resolutions.

P6 L 160-162: I am not sure to understand this sentence. Do the authors mean that the threshold friction velocity in the original SnowTran3D is computed using a constant density? Consider rephrasing this sentence.

P6 L 163-164: The default version of SnowModel described in Liston et al. (2007) includes a parameterization (Eq 18 in Liston et al., 2007) to simulate the increase of near-surface density due to fragmentation during blowing snow events. The influence of wind speed on near-surface density is also included in SnowModel through a wind-related density offset for fresh snow falling in windy conditions (Eq 16 in Liston et al., 2007). Is FSM2trans including these effects? If not, it should be explained clearly in the text. The absence of snow microstructure mentioned at L163 is not reason to justify the absence of compaction during snowdrift in FSM2trans.

P6 L 170: Is the snow holding capacity considered in FSM2trans applied to the snow depth (measured vertically) or the snow thickness (measured perpendicular to the slope)?  Are the authors using the default formulation from Berhnard and Schulz (2010) for the holding depth?

P 7 L 187: Was a cosine correction applied to adjust precipitation based on the local slope of the grid cell for mass-conservation purposes (Kienzle, 2011)?

P 7 L 190: Which formulation is used to split between rain and snow?

P 8 L 194: Was the wind downscaling done at model runtime? Or did the authors prepare downscaled wind fields for the whole season that were then used to drive FSM2trans and FSM2ref? It would be interesting to add a few sentences about the numerical cost of the wind downscaling since the main objective of this paper is to present a system that can be used in an operational context. The wind downscaling is a crucial step for the success of any modelling of snow redistribution in complex terrain.

P 8 L 209: How are treated the data that were masked out (glaciers, lakes, outliers) when computing the averaged snow depth at different resolution?

P 12 L 278: to better understand the maps shown of Figure 9 it would be interesting to have one or two sentences describing the dominant direction of the main blowing snow events in the region.

P 16 L 337: It would be interesting to add information about the numerical cost of the generation of the wind fields at different resolutions. Marsh et al. (2023) (Section 4.4) have shown that the stand-alone version of WindNinja can have a large numerical cost compared to a method based on pre-computed wind library.

P 17 L 350-353: A figure illustrating the evaluation of wind speeds downscaled by WindNinja would be useful for the readers since the wind forcing is crucial when talking about wind-induced snow redistribution in complex terrain. What is the quality of the simulations for strong wind events that are driving wind-induced snow redistribution? I believe that in the context of this work a bias computed over a full month is less relevant than statistics about strong wind events.

P 17 L 360-365: Mott and Lehning (2010) found a similar overestimation of snow redistribution for a crest of the Swiss Alps using the Alpine 3D model running at 25 and 50 m grid spacing. They showed that increasing the model resolution finer than 10 m increased snow accumulation on the windward side due a more accurate representation of small-scale terrain features trapping snow on the windward side. Therefore, I am not sure that the lack of snow on ridges is only explained by a bias in the precipitation forcing. It can also be associated with limitations in the snow redistribution module.

P 18 L 383: on this figure, are the authors comparing snow depth (measured vertically) or snow thickness (measured perpendicular to the slope)?    

Technical Comments

P1 L5: maybe add “the models” or “the module” before “SnowTran-3D and SnowSlide”

P1 L8: Use superscript for km2

P4 L100: Paragraphs made of one sentence should be avoided.

P 5 L123: New snow that accumulates from avalanches cannot be considered as fresh snow. Please rephrase the sentence.

Figures

Figure 1: The contours of Switzerland are hard to see on the first map. The contour of D2 in light green are also hard to read on the main map.

Tables

References (used in this review and not present in the initial manuscript)

Kienzle, S. W.: Effects of area under-estimations of sloped mountain terrain on simulated hydrological behaviour: a case study using the ACRU model, Hydrol. Process., 25, 1212–1227, https://doi.org/10.1002/hyp.7886, 2011. 

Le Toumelin, L., Gouttevin, I., Helbig, N., Galiez, C., Roux, M., & Karbou, F. (2023). Emulating the Adaptation of Wind Fields to Complex Terrain with Deep Learning. Artificial Intelligence for the Earth Systems, 2(1), e220034.

Liston, G. E., Itkin, P., Stroeve, J., Tschudi, M., Stewart, J. S., Pedersen, S. H., ... & Elder, K. (2020). A Lagrangian snow‐evolution system for sea‐ice applications (SnowModel‐LG): Part I—Model description. Journal of Geophysical Research: Oceans, 125(10), e2019JC015913.

Mott, R. and Lehning, M.: Meteorological modeling of very high-resolution wind fields and snow deposition for mountains, J. Hydrometeorol., 11, 934–949, https://doi.org/10.1175/2010JHM1216.1, 2010. 

Mower, R., Gutmann, E. D., Lundquist, J., Liston, G. E., and Rasmussen, S.: Parallel SnowModel (v1.0): a parallel implementation of a Distributed Snow-Evolution Modeling System (SnowModel), EGUsphere [preprint], https://doi.org/10.5194/egusphere-2023-1612, 2023