The manuscript "Evaluating methods for reconstructing large gaps in historic snow depth time series" by Aschauer and Marty is an interesting technical study focusing on interpolation of snow depth time series. The topic is relevant for the community focusing on the snow modeling and time series analysis. The paper summarizes in an effective way concepts which are known by the community, by applying six different interpolation methods to a snow dataset from Switzerland. I think the manuscript represents a good contribution to the current scientific discussion on the topic. I have a few suggestions that the authors could consider in the revised version of the manuscript:

- Lines 43-45: I only partially agree with this statement. The fact that temperature index models require only precipitation and temperature data is the only reason why someone should apply them. In fact, they provide only a very rough physical representation of snow processes. Moreover, it seems that the authors do not consider in the modeling approach processes such as snow redistribution driven by wind or gravitation. These processes are generally not dominant if the choice of the location of the monitoring station is performed properly, but the authors may comment on it.

- Line 70: I think time series available for this dataset are generally longer than the period selected by the authors (1999-2020). Could they provide shortly a justification for this choice?

- Line 87: should “station or both” be “station of both”?

- Interpolation methods: I fully understand that it is not possible to test all available interpolation methods, but I was surprised that a standard interpolation method like Kriging with external drift or let’s say Kriging based methods were not considered. Could the author please motivate this choice?

- Line 178: Is a threshold of 1 cm really relevant for tourism?

- The Matiu/WNR method should be consistently named in the manuscript

- Line 230 should “derived form” be “derived from”?

- Line 237 I think that beside the value of HSmax it would be interesting to compare when HSmax occurs using the different methods.

- Figure 4: it is very hard to read the values of r2, RMS and BIAS, please move them to a Table

- Although I acknowledge the difficulty of summarizing the results of this work in a graphical form, I find the quality of Figures 4, 5 and A1 low. The black lines in the blue box plots are hardly readable and in general the size of the subplots it is too small allow the reader a quantitative interpretation of the results.

- Line 298: It would be nice to have a ranking of the methods to be applied in these circumstance.

- The authors correctly point out that data interpolation is an important step to have longer time series for climatological analysis. However, after data interpolation the next step would be to homogenize stations. Homogenization of snow depth time series is an actual research topic (e.g., Marcolini et al., 2019) and it would be nice to see in the discussion part some comments given by the authors about the effect of the different interpolation method on the quality of the resulting time series. I do not expect a quantitative assessment, since it would probably result in another paper due to the required amount of work, however some qualitative insights would be interesting to stimulate further research in this direction.

Marcolini, G., Koch, R., Chimani, B., Schöner, W., Bellin, A., Disse, M., & Chiogna, G. (2019). Evaluation of homogenization methods for seasonal snow depth data in the Austrian Alps, 1930–2010. International Journal of Climatology, 39(11), 4514-4530.

The manuscript presents the comparison of different methods to fill gaps in snow series. This is a task that has generated many doubts to snow reseachers and this paper provides very useful information for readers. The paper has a clear structure, is well written and conclusions are sound and clear. Therefore, I recommend the publication of the article, with just a few comments that authors may consider to prepare a revised version of the manuscript.

1- in my opinion, it would be interesting to present some analysis to show how differenet methods are suitable to fill gaps of different length, as probably there will be important differences among accuracy scores and methods.

2- As you can include other categorical variables in the Random Forest, authors can test or at least discuss other possible predictors that might refine the results. In example clasiffy if gaps occur in low/average/high snow years; or it existed different dominant weather types or atmospheric patterns in a given year when gaps must be filled.

3- Authors may discuss to which extent the use of more physically based (when possible) may improve the error estimators compared to the degree day model. Researchers from CEN uses adjusted crocus/safran simulation to fill gaps in snow series (see https://doi.org/10.1002/joc.6571 as example). In a similar way bias corredted ERA-land series could be used for some areas, or used as "virtual" best correlated stations.

Looking forward to see your revised manuscript,

Ignacio López-Moreno

In their manuscript “Evaluating methods for reconstructing large gaps in historic snow depth time series”, the authors compare different methods for filling large gaps in measured snow depth time series in Switzerland.

The manuscript is very well written and of high technical and scientific quality. The comparison of the presented methods is in general of interest for the respective snow hydrological community, however, in my opinion only for a very small number of real use cases. The authors show that already a very simple snow model approach using measured temperature and precipitation as input can yield more or less the same results. There are much more temperature and precipitation measurements available than snow depth observations, especially in data sparse region. For that reason, I don’t see very much applicability of the results.

As the authors state, HS is a good-natured variable for gap-filling. This holds true for measurements in terrain where the presented stations are usually located and for continuous snow coverage and typical seasonal, continuous accumulation and ablation dynamics. Therefore, it is quite obvious that good results can be obtained using statistical interpolation methods (more or less regardless of type) using neighboring stations of similar elevation. Much more interesting would be an extension of the analysis to terrain characteristics (lateral snow redistribution, steep terrain, slope, aspect, i.e. small scale heterogeneity in mountainous terrain). This could be tackled by connecting the presented methods to stations clustered not only by elevation and distance, but also slope, aspect, etc. However, I see that this is probably not possible due to the stations located at “representative”, flat, unobstructed terrain locations.

Regarding the results of dHS1, it would be interesting to see the same analysis for dHS10 or dHS5, i.e. a threshold for a snow day of 10 or 5 cm snow depth, as 1 cm is within a range of errors/uncertainties of all measuring and modeling methods. Probably the results will be much clearer using a slightly higher threshold.

As pointed out by reviewer 2, the study would highly benefit from an additional comparison to derivates or direct model values from, e.g. reanalysis products, which are readily available globally.

I support the idea raised in the other comments of including a table with particular strengths and weaknesses of the methods depending on the application and data availability.

Apart from the – in my opinion – rather low applicability of the presented results in other scientific use cases, the article presents a technically well performed study. The findings and conclusion are presented in a very clear and concise way.