Please find my review in the attached PDF.

This manuscript describes a field experiment testing the feasibility of dark fibre DAS for the purpose of avalanche detection. The authors describe a few observations of triggered and natural avalanches, as well as road traffic-induced noise, followed by an STA/LTA and spectrogram analysis of two triggered avalanches. The main take-away that I distilled from the manuscript, is that in order for DAS-based avalanche detection to be sufficiently robust, road traffic signals need to be dealt with (attenuated or excluded from the detections).

The manuscript is well-written, and the story is straightforward. The analysis is maybe a bit too qualitative, and I see potential for a more quantitative analysis of the STA/LTA detection capabilities: how often would false detections be produced by road traffic? Does it help to stack the detection metric over e.g. 100m of cable? Or apply a low-pass filter in space, given that the main avalanche signals have a spatial footprint that is lager than that of individual vehicles on the road? Or apply a basic FK-filter to remove the characteristic vehicle speeds? Having that said, I must admit that I’m not a frequent reader of NHESS, and I don’t have a reference for the depth into which a study is expected to go. So, I leave it up to the editor to decide whether a more quantitative analysis is expected.

Aside from this comment, I only have a few tiny suggestions for improvement:

• The authors use “spiky” to describe some of the more localised signals. Perhaps “impulsive” could be a more formal alternative description, but I leave this entirely up to the authors to consider.
• I think that the readability of Section 3.1 could improve by splitting the big paragraph (lines 101-138) into a few individual paragraphs.
• The reference to Xie et al. on line 210 is missing a date. Also, there are several papers that describe DAS-based vehicle detection that are already published, so I would suggest citing one or two of those in addition to this preprint.
• A suggestion for future work, if the authors intend to remove traffic signals from their data/detections: the traffic signals in the 0.5-5 Hz frequency band are very simple, as they basically represent the spatial footprint of a point load indenting a surface (which translates in time and space; see the Methods section of Jousset et al., 2018; https://www.nature.com/articles/s41467-018-04860-y). It is possible to deconvolve this characteristic signal from the data, which yields high-resolution detections of each vehicle on the road (https://dl.acm.org/doi/10.1109/TITS.2023.3322355; preprint: https://arxiv.org/abs/2212.03936). The authors could use this approach to exclude STA/LTA detections associated with traffic, or to mask portions of the data that coincide with traffic signals. This is merely a personal suggestion for future work, I don’t expect the authors to expand on this for the present study.

Kind regards,

Martijn van den Ende

Kleine et al. describe a seismic data set of snow avalanches. The novelty is that the data were acquired with distributed acoustic sensing (DAS) using a pre-installed communication fiber running along an avalanche-prone road. In this regard, the presented work is highly relevant, because existing fiber optic infrastructure provides an unrivalled density and coverage of sensors for the detection of and possibly warning against avalanches and other hazardous mass movements. The manuscript is written clearly and the data are of good quality and benchmarked against independent acoustic records. On the other hand, the study makes use of only a small part of the available information: 1. It is not clear how many other avalanches were recorded that are included in the Vasom Regobs catalogue. 2. The signal analysis could be done in more depth. 3. Additional ground observations such as avalanche sizes and locations identified on the webcam or infrasound data could be taken into account for the analysis (based on the manuscript, these data sources exist). For a typical NHESS paper I would expect the authors to take at least one of the above steps beyond the current analysis state. For this reason, I recommend that this paper be rejected in its current form. On the other hand, expanding the paper along one or several of the suggested lines should be straightforward requiring mostly a deeper dive into what the authors have at hand and already partially done. From my point of view, it is not necessary to apply advanced data processing, location algorithms and signal classification, although this should also not be difficult to do. Alternatively, the authors may consider submitting the manuscript as a brief communication. For this, some consolidation of text and figures as well as the addition of more specific information would suffice.

Fabian Walter.

SPECIFIC COMMENTS

In its current state, the manuscript seems to be mainly a first glance at the data. To expand it towards a full-size paper it requires an exhaustive investigation in one of the above-mentioned directions. Here are some suggestions for further data analysis and discussion:

1. Spatial coherence: Are individual phases visible over longer distances? I would expect this if collisions with the cones are indeed seismogenic. How do the signals look like before and after crossing the cable? Are similar moveouts as for the explosion signal also seen in the natural signals? Assuming Rayleigh waves (often dominating high-frequency mass movement signals), different parts of the cable should show a different sensitivity to the approaching avalanche. The closest cable sections should be least sensitive assuming a perpendicular approach unto a straight cable section (Kennett et al., 2024 in GJI). Can this be observed? Are there relevant low-frequency signals that can be attributed to the avalanches’ weight? Can the signals be used to differentiate between avalanche paths (see specific comment below)?
2. Discussion: The first discussion paragraph suggests that the authors sorted their avalanches by size. How was this size information obtained? Currently, the reader cannot verify which avalanche records correspond to which size of events. A systematic comparison between size, signal strength, central frequency or other waveform characteristics would be interesting.

I was not able to picture the avalanche/road/cable setting. I strongly suggest that the authors include one or more valley cross-sections, even if this is only showing a schematic not to scale. This would answer questions on how an avalanche reaches the road, the cable, the topographic step next to the parking lot or other sites.

MISCELLANEOUS COMMENTS

Avalanche events --> avalanches

Lines 21-22: Delete sentence starting with “Hence, …” (trivial content)

Line 28: delete “reliable”, since more is needed for a reliable tool than just independence of meteorological conditions

Line 31: specify characteristics; which “same test site”?

Lines 32-33: approaching which geophone? The reader needs more information, like avalanche size, source-station distance, …

Line 37: localizing --> locating

Lines 46-47: delete paragraph break

Line 68: rewrite “beginning and end”

Lines 72-73: Indicate infrasound system on Figure 1 or 2.

Lines 75ff: The described avalanche slopes should be labeled (or color-coded) on Figure 1.

Line 84: steepness --> slope

Lines 91-92: remove “below in the”

Line 92: remove “during data handling”

Section 3.1: Did you also record earthquakes?

Line 98: Check Varsom Regobs, there are two entries in the bibliography.

Line 99: Explain the infrasound measurements. Is this from antenna data?

Line 101ff: Paragraph is too long. Some numbers are spelled out, others are not. When describing channel sections and geographic directions, it would help to label channel distances in one of the plots, or at least refer to the color code in one of the figures.

Line 112: I would drop a hint here about the source of the four signals (even though they are discussed in more detail later).

Lines 124-125: The modeling and the “similar behavior” have to be explained. Especially the modeling requires details.

Line 128-129: Where does the information on the avalanche location come from? Why is this not shown somewhere?

Lines 134-135: The broadband character can have different reasons and influences. I would include more discussion here or leave out this reason.

Lines 137-138: I cannot verify the 16 s and the 20-100 Hz range. The figure seems to suggest a different burst signal in b) and a different frequency range.

Line 139: Avoid 1-sentence paragraphs. Also, what does “strong” refer to? A high signal-to-noise ratio?

Lines 150ff and 160ff: I could not confirm most of the exact time information.

Line 165: Can the different duration be related to different flow velocities?

Lines 182ff: It would help to see more quantitative information from the Surinach studies (avalanche sizes, source-station distances, sensor types, …).

Lines 192: The DAS signals should provide further hints with respect to the different avalanche paths. This is the kind of additional analysis that I strongly suggest (see general comments above), even if this is done on a qualitative level.

Lines 207ff: I strongly suggest discussing Kang et al., 2024, in GRL, here.

Line 210: No year for reference.

Lines 204-205: Is it really this simple to distinguish car traffic? What about cars leaving from the parking lot or moving across it?

Line 212: “Sobel operator” needs a reference.

Conclusion is a summary. I suggest deleting it and calling the Discussion section “Discussion and Outlook”.

FIGURES AND TABLE

Captions should not include paragraph breaks.

Table 1: The meaning of the gauge length is not explained. Either include it or remove the information from the table.

Figure 1 caption: I suggest rewriting “discriminability”.

Figure 2: I suggest labeling the cones, also in the inset; What is Saetreskarsfjellet? Is this mentioned/defined in the text somewhere? Caption: correct typo(s); “start of”  “entrance of”.

Figure 3: Panel b: It seems that the avalanche is visible on practically all channels. I suggest pointing this out or even amplifying the signal to make this more obvious. Panel d: I suggest decreasing the extent of the y-axis so that the explosion moveout can be seen here.

Figure 4: I suggest labeling the columns (rather than defining them in the caption) in their top panels (using titles or text boxes).

Figure 5: Here and perhaps in other figures, I suggest increasing the font sizes of some of the text. The precursory signal in b) (circled in white dashed) is not visible.

Figure 6: The caption seems to be equivalent to Figure 5, so I suggest replacing it by a respective sentence.

Figure 3 and elsewhere: the colors of the waveforms are difficult to discern, I am not sure the color scale serves its purpose.