This manuscript applies a recently published method assimilating local snow depth observations and able to propagate information in space to build a new 10-year reanalysis of snow cover based on the OSHD system over Switzerland. This dataset offers an unprecedented resolution and spatio-temporal coverage and is based on state-of-the-art components in terms of meteorological forcing and physical snow model. The dataset will be highly useful for the community to study the spatio-temporal variability of snow over the country and is going to become the new reference for that purpose. Unfortunately, the links currently provided in the introduction, conclusion and list of references are not working, so I was not able to check the availability of the dataset and its conformity with the description. This obviously has to be solved and checked by the editorial team before publication.

The introduction of the manuscript provides a robust analysis of the existing literature and products and their associated limitations, justifying the need for a new snow cover reanalysis over Switzerland. The manuscript reminds the main methodological principles of the simulation system which have been presented in various previous publications. The assimilation method is especially innovative compared to existing literature. The evaluations directly provided in the manuscript are limited to comparisons between snow depth simulations to non-independent observations. Although a previous publication suggests that leave-one-out experiments exhibit scores similar with simulations assimilating all observations, this choice is somehow questionable in terms of representativeness of the skill at large scale. The article also presents annual maps of peak SWE, Snow Melt Out Date and number of snow days obtained from the simulations. It illustrates well how the reanalysis can be beneficial for large scale snow monitoring. It could have been expected to compare some of these diagnostics simulations with satellite products as Snow Melt Out Date can be derived from optical imagery.

The language is perfectly clear and accurate and I did not find any typos in the document. The quality of the presentation of results is excellent. The article structure might be considered as not very standard compared to common literature because the discussion of the known limitations of the system is relatively short in the current version and is not a dedicated section after the presentation of the results. Nevertheless, this article is still easy to read and make the very valuable effort to make public a dataset useful for a potentially large community of users (although unfortunately the dataset can not be accessed from the preprint as mentioned before).

I recommend publication after solving the data access and considering the minor comments below.

L60 Crocus instead of CROCUS (it’s not an acronym).

L83, L325, L330 and L520 The link does not work and must be replaced by a doi which is the interest of zenodo.

L167 It is unclear how the kilometric liquid / solid precipitation fraction of ICON is downscaled on the 250 m grid.

L221-225 The authors explain that the assimilation procedure removes temporal discontinuities due to the switch from COSMO to ICON. However, from our experience (Vernay et al., 2022) temporal discontinuities in the assimilated surface observations might also result in discontinuities in reanalyses. Could the author comment the avaibility of the 444 snow depth stations over the 10-year period and implications in terms of temporal homogeneity of the resulting reanalysis ?

L232-236 In a publication presenting a new dataset, it could have been expected to present directly the main evaluations of the published version instead of relying on a previous publication based on a different version of the system. If possible, I think it would not be redundant (and even very useful) to incorporate evaluations of Snow Cover Fractions from optical imagery.

L242-249 I am not fully convinced that Figure 4 gives an accurate overview of the skill of the dataset as these observations are assimilated and despite the interpolation process, they can not be considered as independent evaluation data. Oberrauch et al., 2024 clearly show that the leave-one-out experiment exhibits a lower skill. I think it would be more fair to present the skill of the reanalysis from a leave-one-out experiment.

L250-260 These comments are very useful but they could be moved to a discussion section after the presentation of results in Section 4. I also think the discussion about limitations should be a bit extended. First, for the self-sufficiency of the paper, it would be nice to incorporate a short paragraph summarizing the main limitations of the assimilation procedure (as discussed in detail in Oberrauch et al., 2024) and a short paragraph explaining the limitation of the 250m resolution in terms of vegetation description (for instance based on the literature of G. Mazzotti). Then, the choice to not assimilate SCF products contrary to some references mentioned in the introduction could also be discussed : what are the main motivations : data availability ? challenges of spatial data assimilation ? and what are the future perspectives to assimilate remote sensing observations in future releases of this dataset ?

L303 It would be useful for the users to have an idea of the volume of the total dataset and to mention whether the compression facility of netcdf is used or not in the nc files (and with which compression level) or if compression is only achieved through the zip of all files.

L309-310 I am not sure if « version history » refers to the dataset or to the codes. Tagged versions of the different code components used to run the simulations will be a useful addition in terms of metadata for reproductibility.

L330 Even if the purpose of the mansucript is to describe a dataset, a description of code availability used to produce the dataset would improve again the agreement of this work with the FAIR principles.

This is a review of "A high-resolution snow dataset for Switzerland (2016–2025) combining physics-based simulations and in situ observations". The authors present an existing model setup and assimilation scheme to produce a high-resolution snowpack dataset for 10 years.

It is an absolute delight to read a well written and essentially ready-to-go manuscript. A few minor nits, noted below.

My main criticism is that this should include the NWP met forcing data, so the study could be reproduced. This dataset immediately strikes me as a reference dataset for comparing other numerical models against. And in that context, this would require the input data. I realize there might be redistribution license issues with the forcing data. But, if possible, the inclusion would elevate this from "very cool" to "exceptional community dataset for model validation".

I was able to download and spot check the data. The spot checked metadata looks good.

Minor points: 

L29: Add Canada and US example context

Figure 1: I realize this is just elevation. However, does the green regions correspond to the treeline? I, personally, find a low-bound of green problematic because it gives a sense of vegetation cover that may or may not be there. It would be excellent to have tree line noted in this figure

L98 I don't think m a.s.l requires the a.s.l

L102 Why is Alpine caps? Is it a proper name? If it is, place a map marker in Fig 1 as I don't know where this is.

L105 Same as above

L135 OSHD = define here

L140 "period of 24 hours" is this during the model run? Or some other analysis that uses a constant parameterization in the model 

L163 "station locations" it's a spatial model, why is a point-scale being noted here?

l176 "PF-based" for the heading I would not use the abbrv and would write out Particle Filter

L192 In the context of the distributions, it would be good to state explicitly which dist is used for each variable

L217 "notably improves…" how is this statement verified? 

Figure 5 (and surrounding) how is frozen soil infiltration estimated?

L299 is having these data in UTC+0 possible? How is DST handled? A fixed reference UTC+0 would make this easier to ingest (for model inter comparisons)

L308 Remove extra space in EPSG: 2056

L330 I strongly think that if at all possible, the input NWP fields should be included