The authors use two different satellite data products (one optical and one microwave) to estimate snowfall occurrence near one weather radar site in Finland in March 2018. This is a proof-of-concept study that demonstrates feasibility of the techniques for detecting snowfall occurrence, although both techniques struggle much more to detect non-occurrence. 

It is always somewhat difficult to know whether to recommend publication of proof-of-concept studies, since they invariably raise as many questions as they answer and demand further related work to realize their scientific potential. This article is no exception: it is well-written, scientifically thoughtful and well-reasoned, and the results are clear and intriguing. Yet, the sample and reference data are highly limited in time and space, and seemingly important caveats and limitations are highlighted but not investigated in sufficient detail. My instinct is that the study requires a bit more fleshing-out before it can be published as a standalone contribution. 

L63: Metamorphism that affects snow albedo is still possible at temperatures below freezing (e.g., Qu and Hall 2007). How does this affect your assumptions?

L85: Consider adding ERA5-Land SWE as a reanalysis-based reference product. It should perform about as well as ESA-CCI SWE and has many advantages.

L95: Typo of "Parameter"?

L140: The wind adjustment section was somewhat brief, and the reader couldn't assess how important this correction is. How big of an effect comes from the wind adjustment? What is the difference in snow rate before/after adjustment for wind? What is the sensitivity (if any) to assumptions about tree height and snow fall speed?

L257: It is not immediately obvious from the maps what consistitutes a pixel and what is a tile. Therefore, we can't tell where the one tile is that has some challenges in panel 4c. Perhaps the authors could add a grid showing the layout of the tiles as a reference, say on Fig.1 or Fig.4?

L256: The areas of misclassification in Fig.4b without forest correction do not appear to be the areas of highest forest canopy cover (Fig.1b). How do the authors explain this apparent contradiction?

L260: While the combined results show fewer misclassified pixels, they also have a considerable increase in missing or omitted values. So it is not completely straightforward to compare the performance of combined vs single-factor results. How is this difference in pixel number taken into account in the bulk metrics of Table 2?

L263: The section on SWE-based classification was incredibly brief. So much so that I found that it scarcely met the bar for demonstrating a proof-of-concept. What factor(s) are most important for explaining the (mis)classification? Presumably microwave emission is sensitive to the presence of surface water, topography, vegetation type/density, among other things? 

The authors undertake a proof-of-concept investigation into how accurately optical satellite observations, namely Sentinel-2 surface reflectance-based grain size and microwave-based snow water equivalent (SWE) estimates can detect snowfalls over the Arctic. The technique developed by the authors is capable to detect at least 77% cases of snowfall using optical measurements alone in a correct way, which is a success taking into account that the method can be further elaborated (especially selecting a better cloud detection scheme (thermal infrared measurements?)). I would advice to retrieve the value of L from Eq. (4) and use it in the analysis. This is related to the fact that the conversion of L to SSA may lead to additional errors, which are difficult to assess. On the other hand, this shortcoming may not influence the results of this paper aimed not to the SSA determination but to the detection of snowfalls. The authors may comment on this issue in the paper. I advice the publication of this paper.