The manuscript presents analysis of snow depth measuments made with airborne lidar in Alaska in boreal forest region. Lidar measurements were compared with in-situ snow depth measurements. Snow depth was estimated for three different canopy height classes. The study resulted statistically significant differences in snow depth between forest and lower vegetation.

General comments:

The manuscript is well prepared and easy to read. Figures and tables support the text well. Title is clear and abstract gives good summary for the manuscript. I have some general questions:

- How SWE values are calculated? Have you used density measurements? Please describe it with more details. 

- Could you estimate how large area was covered with the magnaprobe measurements?

- Have you considered accuracy of magnaprobe GPS in the analysis?

- How subsets were chosen?

- What was final resolution of the lidar data set after the reduction? 

- If I understood correctly, in-situ measurements were compared with lidar snow depths from the whole area. Why comparison was not made only with the lidar measuerments at the same area as the in-situ observations?

Specific comments:

Line 17: "and standard deviation SD= 15cm" so that SD is not mixed to "snow depth"

Line 21: Typically SWE is presented in mm, 200-300 mm (same comment for the chapter 5.2)

Line 147: Consider is it necassary to use DSM and DTM abbreviations instead of writing them open

Line 356: "Ten previous studies"

Lines 410 and 414: Is "this study" referring to your study or previous studies by others?

This is a review of "Analyzing vegetation effects on snow depth variability in Alaska’s boreal forests with airborne lidar".

Overall, this is a well written manuscript that was easy to read.

My main concern is with the very low r^2 values reported for the comparison of lidar SD to in situ SD. I am unclear on how a couple pieces of the analysis were done, and coupled with this low r^2 makes me think that perhaps there is a post processing error in the analysis. 

How, exactly, are the observed SD compared to the lidar SD? On L124-125 are the authors comparing the magnaprobe transect mean to a single lidar point? Or is the lidar transect averaged over the same length as the magnaprobe? If it's the individual magnaprobe point to a lidar point then there is a substantial scale mismatch. If it's an average of the magnaprobe transect to a lidar cell, then it's also a scale mismatch. The authors allude to this a bit in the discussion but it needs to be front and centre. And I think, will be a clue to exactly what is going wrong with the r2 values.

 Given both the forest (5b) /and/ the open (5d) have such poor R^2 values, I am skeptical it's magnaprobe GPS uncertainty. In figure 3, the treeless in the top left is very different than treeless in the bottom left, especially w.r.t GPS signal and SD uncertainty. Because both the forest and treeless has as poor a R2, this feels like maybe a post processing step. The SD is deep enough that I'd expect a lot of the small rocks/grass/etc to be buried in the treeless (as per fig3) and thus result in a pretty clean lidar vs in situ comparison. In Harder, et al (2020; https://doi.org/10.5194/tc-14-1919-2020) which is one of the few drone-lidar forest papers, the observation vs lidar is much more 1:1. The sub plots in Figure 5 might be interesting to look at on a per site (L124) basis, to see if there is a single site that is causing the major outliner points, or if this is from all (treeless / tree) sites.  

The OK aggregation step is, I think, correct, but the length scale is very short – 1 m. In canopy this seems reasonable but in the open it feels a bit short. Is there anything in this aggregation step that might be causing a bias in the lidar SD? e.g., it should be possible to extract the pre-OK high resolution lidar SD and see how the data compares to that to rule out any post processing of the OK aggregation step. Or look at larger averaging lengths to further remove sub-grid scale impacts. 

This is all to say, it is difficult to tell from the figures here, but I would urge the authors to take another look at the lidar and observation data, and really diagnose if there is an unintended bias or /something/ in the post processing that has caused this very low r^2. Break it up by site, etc and convince the reader this is a correct result.

As is, the in situ vs lidar looks like random values centered around a mean. 

Specific comments:

L64 Include the above mentioned Harder, et al (2020)

L75 pose this as a scientific research question

L110 Figure 1, it should be good to show the tree cover in (c)

L115 move ? into the " " and remove from end of the sentence

L118 on tundra

L163 "plausible SD" where were these plausible ranges? was this a subjective expert decision?

L176 “smaller sections” is section = area? if so, I would use area. otherwise explain section.

L177 “portioned” -> partitioned

L183 “resampled” I assume this was via OK method noted, but make it clear

L184 “canopy height resample” how was this resampled? bilinear? cubic? OK?

L194 note Arc version

L204 how were these data thinned? Is this the resampling above?

L207 what is the constant 12 in eqn 1?

L216 did the authors consider a 99% (0.01) range?

L240 how sensitive are the results to changes in this cutoff value?

L250 Figure 2 great plot

L255 Would be helpful to draw regions around the areas the labels apply to. Add sub-figure labels (a,b,c).