The authors present a thorough analysis using Random Forest (RF) to extend (or connect) lidar snow depth snapshots through time. Physiographic variables (elevation, northness, etc.) and dynamic variables (snow depth, air temperature, etc.) are used in training and prediction. The target variable is difference in snow depth at each point in the domain, calculated relative to depth at a (temporally-continuous) snow depth station. The authors complete an experiment replacing local/in situ measurements with coarse ERA5 dynamic variables. All analyses and experiments are reasonable and carefully described. The lidar data (and validation data) is extensive and covers two contrasting sites. The figures and tables (mostly) convey the results. Overall, the manuscript contributes to the growing body of research demonstrating how machine learning (specifically RF) can be used to extrapolate snow observations, with potential implications for operations. I feel the manuscript could be improved; emphasizing the unique contributions of the research in a more concise, direct format with improved figures.   

First, the introduction (and other sections) should describe similar efforts – and then focus on the differences in this study. First, the study of Geisler et al. is rather similar to what is presented here (Geissler,J.,Rathmann,L.,&Weiler,M.(2023).Combining daily sensor observations and spatial LiDAR data for mapping snow water equivalent In a sub‐alpine forest. Water Resources Research,59(9), e2023WR034460.https://doi.org/10.1029/2023WR034460), yet this paper is not cited or discussed. Geissler et al. develop an approach to fill in the gap between lidar flights using daily depth observations from stations. The scale of the Geissler study and this one are also similar. Second, the methodological framework used in this paper is the same as that presented in Herbert et al. (2025), which used more than 50 lidar acquisitions across 8 basins in Colorado and California. While Herbert et al. (2025) is cited in limited contexts (e.g., around line 284 when defining the target variable, and around line 650 when discussing the number of lidar acquisitions), the manuscript does not clearly acknowledge that the core problem formulation and modeling approach closely follow that prior work, including (i) casting snow depth estimation as a residual (relative-depth) between lidar and a temporal driver, and (ii) using a random forest model to reconstruct spatial patterns from those residuals.

The contribution of this paper lies not in introducing a new framework, but in extending and testing an existing approach in a different setting. Emphasizing these differences, rather than presenting the method as newly developed, would clarify the contribution and better situate the work within the existing literature.

Second, the manuscript includes a fair amount of extraneous text that distracts from the main points. The manuscript would benefit from tightening to better emphasize its core contributions. For example, the first paragraph of the introduction could be shortened to two sentences, and the text around line 530 and following could be greatly abridged. The discussion includes sections that, while interesting, are not directly tied to the study design or results. For example, lines 604–610 do not relate directly to the results and could be shortened or removed. The conclusion reads more like a short discussion than a set of clear takeaways. I recommend focusing on the text needed to support the main findings and removing or condensing the rest.

Third, the figures could be improved for clarity and readability. Some appear fuzzy in my version, and several rely on shades of gray where color would improve interpretation. In Figure 4c and 4d, the dark histograms appear to show the distribution of differences between each pixel and the in situ site (line 450), which represents basin-wide snow depth referenced to the station, rather than model error as described. In Figure 2, it would be helpful to show individual yearly traces in different colors, similar to a standard SNOTEL plot, rather than a gray band. In Figure 7, the grayscale is difficult to relate to RMSE values. Overall, improved figure clarity would enhance readability.

Fourth, many of the dynamic variables appear to have limited influence on prediction (Figure 5: “While dynamic predictors provided minimal predictive value…”). The authors should exclude variables with little predictive value, particularly given the large number of predictors used. This would simplify the model and improve interpretability and transferability. Alternatively, if there is a reason to retain these variables, this should be clearly explained and justified.

Minor comments:

L63: and perhaps equally important, lidar only provide snow depth, not SWE

L134: this could be shortened, at least for the content included. I think it could be helpful to show variability from year-to-year in figure 2 - a series of small subplots perhaps.  each with the "average" vertical lines in the text superimposed. This would show how these date compare to the year-to-year evolution of the snowpack. This would be helpful to evaluate the results describing transferability from early to late season, etc.

L230: How was this choice made. Why not reverse the procedure (predictor versus validation) to evaluate robustness of results?

L512: not monotonically, maybe roughly or generally?  it does bounce around a bit, so not strictly monotonically.

L573: but would it be the same "everywhere", or would you need to make the "on site" comparison, then determine the bias?  I.e., is it transferable without local information as a test?

L584: why not color?  the gray line is very difficult to see.  the who figure is fuzzy.

L614: How is this result supported by the experiments?  Rewrite or clarify.

L672: this section seems valuable, given the west-east contrast is central to the study, and unique from previous studies.

L746: the conclusions feel a bit redundant with the discussion. The conclusions could be more "conclusions" with the main points from the study.