the manuscript by Herla et al. introduces a novel method that allows the synthesis of a large number of simulated snow cover simulations resulting in an average profile, which can further be queried if an in-depth analysis is of interest to the user. The proposed method builds upon and expands previous research in this direction. Furthermore, the presented algorithm provides a solution to facilitate the interpretation of snow cover simulations for regional avalanche forecasting.
The manuscript is well written, concise, but still easy to follow. The figures are of high quality, supporting the understanding of the described workflow (Fig. 1) and the visualizations obtained with the algorithm (Fig. 2 and 3).
I have three general comments, which I believe would further improve the manuscript:
On l. 104, the authors state that their testing has shown that the rules applied for initiating the algorithm consistently produced reasonable results. While the described rules (l87-95) do indeed sound plausible, no further detail regarding the testing is provided. - Please elaborate on this testing. For instance, provide a reference if the tests you have made are described elsewhere. What do you mean when you say that "reasonable" average snow profiles are produced? Does reasonable mean that you compared these profiles with observations or is this based on feedback from avalanche forecasters?
Section 3.2 and Figure 3 show an example of an average snow profile over the course of a season. This example is helpful as it nicely illustrates the potential of the presented algorithm for the analysis of snow-cover simulations at a regional scale. However, from the perspective of a potential user of this algorithm, it would be useful if you could address the following two points: (1) In this example, the early part of the season is presented, but the melting season is missing. This makes me wonder whether the algorithm works equally well in spring when the snowpack height and the number of simulated layers decrease with increasing wetting. As the first wetting of the snowpack is highly relevant for forecasting wet-snow avalanches, snowpack characteristics like the advance of the wetting front are very important pieces of information (e.g. Wever et al., 2018). - I suggest expanding the average profile shown in Figure 3 well into spring; or, in case the algorithm is less reliable during the melting period, to mention this limitation. (2) I personally would have greatly appreciated if this example would have been supported with the (observed) weak layer summary in the region. From what I remember, the main author presented such a comparison at a conference last year (Herla et al., 2021), showing that the average profile captures most of the weak layers tracked by the field observers. While not a full validation, this would help the reader to understand that the average snow profile, synthesizing snowpack simulations driven with an NWP model, captures the most important snowpack features in the region.
Minor comments (line numbers are indicated):
1: ...a way that is... or ...ways that are...
27-30: rather long sentence. consider splitting
28: ...a well-established algorithm...
31: ...their medoid approaches... - does "they" refer to the two references introduced before or just Herla et al.?
I hope these comments are helpful when revising the manuscript.
References: Herla et al. (2021): Herla, F., Horton, S. and Haegeli, P. Creating regional snowpack summaries from model simulations and starting a large-scale validation project. Presented at Colorado Snow and Avalanche Workshop 2021 (virtual)
Wever et al. (2018): Wever, N., Vera Valero, C. and Techel, F. Coupled snow cover and avalanche dynamics simulations to evaluate wet snow avalanche activity
the presented brief communication by Florian Herla and collegues describes the use of a specific averaging technique, the Dynamic Time Warping Barycenter Averaging (DBA) in the field of Dynamic Time Warping (DTW) with pure focus on analysing modelled snow stratigraphy. While the appraoch and methods are not novel, it is the first time that these set of methods was applied to modelled snow stratigraphy and to the field of avalanche forecasting. Large parts of the devleopped DTW method for modelled snow stratigraphy were presented in an earlier manuscript by Herla et al (2021). The focus and added value on the presented manuscript compared to the already published content by Herla et al (2021) is on (1) the newly added averaging technique DBA (section 2), (2) some added features on the layer matching appraoch (lines 73-79) and (3) two newly presented sets of figures (Fig. 2 and 3) for better communicating the obtained results.
The text is well written and most of the prestented Figures are clear, easy to understand and enjoyable. Sometimes explanations are a bit to short and due to the nature of a brief communication explanations are sometimes not easy to grasp for an uniformed reader. In addiation, I would suggest improving Figure 1.
Even though parts of the content were already described in Herla et al (2021), I like the idea of this brief communication since the authors focus more on the quality of the results while within the other publication the architecture of the algorithm covered most parts of the reading. Nevertheless, I would expect a little more quantitative presentation on some of the descriptions, which leads me to my four general comments that may improve the quality of the manuscript:
As stated, Figure 1 is a bit confusing and hard to understand. Could you maybe use less profiles in between and describe the workflow a bit more in detail within the graph. In addition, add some more description within the caption.
You state that for the DBA it is essential to start the interation by choosing initial condition profiles strategically (line 89). How influential is that condition of the initial profile? Or with other words, if I miss to chose my starting position carefully, does the algorithm support me and is able to find weak layers that I just missed when picking the starting conditions. Can you quantify that by adding some noise to your initial profile? Related to that are your statements on the testing. I would like to see some more quantitative results and more in-depth description on how you did that. Reading phrases like (...)consistently produce reasonable average snow profiles suitable for avalanche forecasting. (Line 105), are with low support and not helpful for the interested reader or an avalanche forecaster that wants to apply your findings. In addition, I would be curious what you think is suitable for avalanche forecasting and what is not ;-).
I like Fig. 2 very much. It will be very helpful in daily routines of avalanche forecasting centers. However, I have some issues with how the content of Fig. 2b was produced. You basically applied the approach by Schweizer and Jamieson (2007) which turned out to be inpropriate or at least less helpful when applied to simulated snow cover data (Monti and Schweizer, 2013). Main reason for that is the fact that the thresholds by Schweizer and Jamieson (2007) were obtained with statistics based on observed snow stratigraphy parameters which may differ compared to simulated ones (especially grain size). That's why Monti and Schweizer (2013) introduced the relative threshold sum appraoch and I would love to see if there are particular differences for the presented example. In fact, I would expect, e.g. the facets below the thick layer of RGs (I assume this to be the slab) to give more indication towards instability. This in turn would give you the option to included FCs as weak layers as well. At the moment the representation of Fig. 2b is heavily driven by grain size only, since the used underlaying snow cover model classifies the weak layer DH and SH mainly based on their size.
Can you please give some more insights of the model behind the modelled snow stratigraphy data? Are you using SNOWPACK or Crocus?
Finally some minor comments:
The algorithm seems to work dry snow conditions only? Can you comment on that?
1: ...a way that is... or ...ways that are...
28: ...a well-established algorithm...
I hope the comments are helpful. Congrats to this delightful work.
References:
Herla, F., Horton, S., Mair, P., and Haegeli, P.: Snow profile alignment and similarity assessment for aggregating, clustering, and evaluating of snowpack model output for avalanche forecasting, Geosci. Model Dev., 14, 239–258, https://doi.org/10.5194/gmd-14-239-2021, 2021.
Monti, F. and Schweizer, J.: A relative difference approach to detect potential weak layers within a snow profile, in: Proceedings of the 2013 International Snow Science Workshop, Grenoble, France, pp. 339–343, https://arc.lib.montana.edu/snow-science/item.php?id=1861, 2013.
Schweizer, J. and Jamieson, J. B.: A threshold sum approach to stability evaluation of manual snow profiles, Cold Reg. Sci. Technol., 47, 50–59, https://doi.org/10.1016/j.coldregions.2006.08.011, 2007.