This is a review of SNOWstorm (v1.0). Overall, I really like the idea of going after blowing snow with deep learning. Blowing snow is critical to include, and it's very tricky. However, I think this manuscript needs more work and contextualization to be a strong contribution.

I found that the manuscript layout to be difficult to follow, with the mix of results and discussion together making for a difficult read to really figure out what was tested and where. For example, the conclusion notes that direction and speed were tested against observations, but I'm struggling to find this in the text. The heavy use of model variables, e.g., SUBL_VI, SNOW_VI, makes parsing the paragraphs difficult and makes it easy to get lost. I would encourage the authors to have clear numerical model methodology, the tests, and the case study results, with a synthesis discussion section to tie them together. And then use this to tighten the text as much as possible. 

Broadly, I am left with an uncertainty as to how good this model really is. I find both the wind velocity and blowing snow analysis underwhelming. Specifically, I would really like to see figures showing model speed and direction, compared to both a model but also observations. I see figure 9, but I would prefer this as a wind rose and to see >24h of observation analysis. How are the ridges and lees /exactly/ compared ot the more complex model or obs? The lee is where the snow deposition will happen and if this is wrong, the deposition is wrong (which seems like the case). Can the model perturb the wind direction very much? It looks like it barely does, but this might just be a visualisation limitation. How large of a domain can this scale up to? How does this approach compare to other blowing snow approaches likes those of Snowmodel, ICAR, CHM, SnowPappus, and Alpine3D? For example, Figure 5 d-e shows a ~4x under estimation in transport rate. Figure 6 is completely missing deposition zones. This seems like a lot. I would also like to see side by side plots of the lidar snowdepth observations and the model outputs. Is the spatial variability we observe, that all models struggle to get 100% right, at least qualitatively present in the model outputs? Without these contextualizations, it is difficult to evaluate the manuscript.

I think having a core set of research questions that are answered throughout the manuscript would help tighten and refine the story. As is, this is a statement of what was done, and the clear common thread tying it all together, including the advancement versus existing approaches is lacking. Getting the wrong answer faster is not particularly useful unless the error in that approximation is well bounded, understood, and justified. 

Specific comments:

L1 "and (the) resulting" missing the

L22 e.g., add the comma after e.g. throughout

L 91, Figure1 these need scale bars, north/south arrows, and lat/lon reticules. As is, it's not at all clear what we are looking at. 

L95 The synthetic topographies are neat, but why go through the effort when global 30m DEM exist? I suspect it's for the periodic boundary conditions? If so, this needs to be stated up front and not as a buried lede

L 114 "to fit the requirements of the later model" where in the MS? what section?

L120 `a` term is what?

L145 "is used" -> "are used"

L155 Why are the particle sizes not a distribution as commonly noted in the literature?

L155 Is this domain masked for vegetation? how is that handled? I realize that the authors are interested in non-vegetated areas, but some description of how this is handled and impact on flow fields is needed

L 156 table 1, how are these sampled while maintaining physical coherence? Please cite ranges for these values

L174 "model run time" are these wall clocks or internal model hours?

L229 how is T changed but RH is not updated? this will dry/wet the atmosphere out and impact the sublimation estimates

L236 I would like to see clear wind direction eval here too

L308 has results in the overview

L320 how sensitive are the results to this light-ish density? Windslab layers are of course also missing here

L330-> would love to have S_WDI_W and other variables described or reiterated for the reader. By the time I got here, I had no idea what these were anymore. There is some description on L 325 but just they are inputs, of what?