Poschlod and Daoz analyze snow depth from two high-resolution climate models and one reanalysis for an area in Southern Germany. The purpose of the study is not clear, since no research aims are stated. The title hints to “suitable for extremes and impact-related research?”, however, most of the manuscript is just model evaluation and little on extremes. Some of the impact-related variables are highly questionable. There are some major concerns on parts of the methodology, and the manuscript needs a clearer structure before of publication quality.

Major points

• I miss a description of why the research setup. What are your aims and/or hypotheses? Why snow depth? It is not a state variable, and you never discuss how different density estimates might impact your results. Why in-situ observation and not remote sensing? Again, the mismatch of point-vs-grid could be highlighted more clearly, and also the impact of resolution, since I guess a point is more representative for the 1.5km cell than for the 9km cell.

• L210: In case of heavy snowfall, compaction can be of larger magnitude than melt. (Also how do you derive this index in case of missing observation data - you allowed 30%, right?) Then for maximum accumulation, snowfall might be a better variable than increment in snow depth. And for maximum melt, SWE change (which is anyway the state variable in the model). I suggest removing these analyses. If you want to focus on extreme accumulation and melt, which of course are variables with significant societal impact, then you need to choose appropriate variables. And if you want to use snow depth as proxy instead, at least you have to prove it is meaningful compared to snowfall and SWE loss. Currently, they cannot be trusted, and therefore should be removed from the manuscript.

• Your study focuses on low-elevation snow cover, and your station coverage is rather limited. Especially, all stations are below 1000m. This is important to acknowledge in discussion and research setup.

• Consequently, your evaluations, as they are know, are heavily influenced by low snow amounts. For example, you can have high values of relative errors for irrelevant snow amounts. (this is less an issue for high-alpine sites, where you have large snow amounts, related L510). This needs to be discussed. Even better would be analyses that distinguish by elevation or snow amount, or summary metrics/table that take this into account.

• The discussion on snow albedo as driver of biases is good. However, the important figure is in the supplement and one that is of minor use in the main part.

Minor points:

• L20: improves relative to what? Numbers are somewhat in between.

• L23: “All models fail…” but observations do?

• L25: what limitations?

• L26: Winter climate is more than just snow.

• L26ff: Not sure. Abstract numbers suggest high accuracy, in fact. (after reading the manuscript the low biases make sense, because you only look at low elevations) For climate change research, another important factor is the boundary forcing from GCMs.

• L66: Please be more specific. Blowing snow has already been implemented offline (10.5194/tc-15-743-2021) and also online (10.5194/gmd-16-719-2023).

• L79: Less means there is something, what?

• Intro: Research aims are missing.

• Intro: Also, I would have expected something on snow studies in climate models, such as 10.3390/atmos10080463, 10.1007/s00382-012-1545-3, 10.5194/tc-17-3617-2023, etc

• Table 1: would be easier to read if references were put in caption or similar. And ERA5-Land is not statistical downscaling.

• Sec 2.4: The number of snow stations is quite low, compared e.g. to (10.5194/tc-15-1343-2021). Did you take only stations which had snow, temp, and precip simultaneously? Also the maximum elevation is rather low… Since your study is about snow, maybe it would make more sense to include more observed snow data (not necessarily with temp and precip).

• Fig4: I don’t see any large dependence between temp and precip (except for ERA5L), so maybe if your point is temp/precip bias depend on elevation biases, it would be better to put elevation bias as continuous x-scale (and not discrete point shape, which makes it hard to read).

• FigS2 is quite good and relevant, I suggest moving it into the main manuscript. Just make the elevation bins wider to reduce noise and have a constant line for ERA5L. Also elevation seems not to match between DWD and AMUNDSEN.

• Plots with obs vs. sim would benefit from a 1:1 line

• L315: since your prevalence is 3:1, “substantial” is overstated, since you need to put FP/FN in context to prevalence

• L343: not necessarily just resolution, might be bias in the forcing (too wet, too cold)

• L400: again, this is not only resolution!

• L474: overstated. You have overall accuracies between 70-90%, which is in range to the correlations for seasonal snow depth. So I assume also shorter-period analyses would be in the same line of accuracy, so you cannot distinguish here by length of the analysed period (only if you actually performed some analyses with your data for 1-2 week periods and then performed a comparison).

• L488: not new, there have been many SnowMIPs (Essery and co.) showing the same…

• L495: So what is better? Use the snow scheme from the climate model? Or take only meteo and apply higher complexity snow models? How does this fit with previous studies that used meteo forcing from climate models to drive snow models?

• General: Results have a lot of repetition on plots with maps and obs-sim scatter plots. You might consider aggregating the information to prove your point. For example, spatially averaged time series, summary by different elevation, etc.

General comments

In this manuscript, variables related to snow cover are compared between station measurements and four models in Southern Germany. The manuscript is mostly well-written with clear figures and presents interesting results. However, some points could improve the paper and strengthen the results.

I think the aims of this study and the justification of the research setup should be more clearly stated. The authors have decided to use only one variable (snow depth) instead of using e.g. SWE, even though SWE might be a better variable for estimating snow accumulation and snow melt. They state (L211) that snow melt is assumed as the main driving process for snow depth reduction without adding any reference. I think this should be further discussed. Also, I think the authors should state more clearly why this study area (Southern Germany) was chosen. It would be interesting to see the analysis cover also mountainous regions.

Why are only in situ observations used in the comparison and not remote sensing data? I think the authors should clearly state the justification for using only in situ data. As three of the used models show gridded snow cover estimates, using e.g. satellite-based gridded estimates as a comparison would make sense.

Specific comments

Table 1 is a bit confusing and it is hard to see which model is in which line, as the names that are used throughout the manuscript are not clearly listed. I suggest adding a column on the left with the model name (WRF, CCLM, AMUNDSEN, and ERA5). Also, consider organizing the table in the same order as the text (first Regional climate models, second ERA5-Land, and last AMUNDSEN).

Figure 3. I suggest adding the name of the model on top of each plot instead of just mentioning it in the caption. This applies to all figures throughout the manuscript.

Figure 3. Consider adding a diagonal line (from lower left to upper right corner) to each plot so it is easier to see whether there are biases in the models. This also applies to all similar figures throughout the manuscript.

Figure 4. Consider adding a bit darker horizontal and vertical lines at x=0 and y=0.

Figure 5. Please add somewhere (in the text or in this figure) that you use the acronym DWD for the observations.

Figure 10. Is the difference relative or absolute difference?

L381. Change 0.15 m to 15 cm. I think it is better to use the same units that are used in the figure.

I suggest the authors check the grammar carefully. The manuscript is overall very well-written but some mistakes (e.g. incorrect prepositions) exist.