the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Crevasse distribution across Austrian Alpine glaciers linked to ice dynamics and surface topography
Abstract. Glaciers flowing over uneven bed topography or undergoing extensional flow develop tensile stresses, and where these exceed the fracture strength of ice, crevasses open. Their distribution expresses the surface strain-rate field and enhances local ablation. Yet observations of where crevasses occur across Alpine glaciers remain scarce, and the theoretical controls on their formation, established largely from individual glaciers, are untested across a diverse glacier population. Here we present the first country-scale inventory of visible crevasses, mapped across all Austrian glaciers from high-resolution remote-sensing imagery, and analyse their spatial distribution and structural controls across six mountain regions. Crevasse density varies strongly between regions, peaking on the steep glaciers of the Zillertal Alps and the fast-flowing glaciers of the High Tauern, and reaching its lowest values on the large, low-gradient glaciers of the Ötztal and Stubai Alps. Linking crevasses to surface topography and ice dynamics, we show that crevassing is favoured by convex surfaces, steep slopes, and along-flow changes in slope. Ice-flow velocity and driving stress instead behave as thresholds, distinguishing highly crevassed terrain only once exceeded. We find no relationship with along-flow velocity change, despite longitudinal extension being the classical mechanism, indicating that much of the regional crevasse signal is inherited rather than locally generated and raising questions of advection and suppressed crevasse healing under reduced overburden pressure. The inventory supports route planning in published hiking maps and provides an observational basis for representing crevasse-enhanced heat exchange in mass-balance models and calibrating damage-mechanics schemes in ice-flow models.
- Preprint
(2071 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-4768', Anonymous Referee #1, 06 Oct 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-4768/egusphere-2026-4768-RC1-supplement.pdfCitation: https://doi.org/
10.5194/egusphere-2026-4768-RC1 -
RC2: 'Comment on egusphere-2026-4768', Anonymous Referee #2, 21 Oct 2026
General comments:
This paper extends an existing mapping method to produce a country-scale inventory of visible alpine crevasses across Austrian glaciers. It then compares crevasse distributions with topographic data and ice dynamic variables. The coverage of the inventory and its open availability make this a useful contribution, with potential applications in route planning and glacier modelling.
My main concern is whether the previously reported accuracy of the detection method can be extended across the full study area. The manuscript should explain more clearly why that performance is expected to hold here and what the visual inspection establishes. Differences between the dates of the imagery, glacier boundaries, and velocity observations also need clearer discussion.
It is also not completely clear whether the study aims to describe where crevasses occur or explain why they form. The analysis identifies links between their distribution and the selected topographic and dynamic variables, but these relationships do not necessarily explain the processes responsible for crevasse formation. Please clarify this distinction in the objectives and ensure that the conclusions reflect what the analysis can show.
Overall, I consider this a worthwhile contribution. The main points that need addressing are confidence in the mapping and clarity about what the analysis shows. Some of the processing steps would also benefit from more detail.
Specific comments:
Transferability and validation
Please clarify the training, validation, and independent test datasets used to evaluate the model. The same edge accuracy of 98.7% is reported for training and validation. While identical results are possible, it would help to explain which datasets these results describe and how they relate to the model in this study.
The statement that consistent input quality and homogeneous acquisition parameters support transferability also needs qualification. Similar imagery provides a reasonable basis for applying the model, but it is not sufficient to assume that performance remains consistent across different glacier surfaces and imaging conditions.
You mention “visual inspection” (line 171). How does this inspection support confidence in the inventory, particularly regarding crevasses that the model may have missed? If missed crevasses were identified during the inspection, were they added to the inventory? A quantitative check on a representative sample would improve confidence in the inventory.
Differences between dataset dates
The glacier boundaries are from 2023, whereas the orthophotos were acquired between 2019 and 2022. Please explain the significance of this difference and how it was handled. Was the 2023 glacier boundary dataset the closest to the imagery dates?
A newer boundary could exclude ice that was present when an older image was acquired. You manually checked and excluded non-glacier features which should help address areas where the boundary includes exposed rock. Could some areas of ice visible in the older imagery have been excluded because they fall outside the 2023 glacier boundary? Please comment on this.
Please also explain why the 2020-2021 velocity dataset was used. Some of the imagery was acquired outside of this period, so a brief comment on whether this is a limitation should be included.
Model description
Please define HED-UNet at first use. People may be familiar with a U-Net, but not this implementation.
The description of the deep feature pyramid is difficult to follow. Please explain what the feature pyramid represents, why the model might overlook the lowest resolution feature maps, and how the training approach addresses this.
A mention of the original model’s use with SAR imagery would benefit the text, including the task for which it was originally developed, and how it was adapted for crevasse mapping in optical imagery.
Please provide a reference to work applying foundation models to crevasse mapping.
Boundary thinning and definition of density
You mention that the crevasse boundaries are thinned to one pixel. Please identify the method used and provide enough detail to reproduce this step.
You define crevasse density as the count of boundary pixels (line 207), but then later mention boundary length (line 420). Please clarify these measures and how the boundary pixel count is converted to (or used as a proxy for) physical boundary length and ensure it is consistent throughout. If density refers to crevasse boundary length per unit area, please state this explicitly with units. Please distinguish this from the proportion of the surface occupied by crevasses.
A box plot could help readers compare the reported density values, although this is a presentation suggestion rather than an essential addition.
Hexagonal basis
Please define “hexagonal basis” when it is first mentioned or provide a reference to the methods describing it. It would also be useful to explain why the H3 method was chosen rather than a square grid. Is this choice intended to reduce directional effects or serve another purpose?
Please clarify how hexagons that are only partly covered by glacier ice are treated. Is the density normalized using the full hexagon area or only the glacier-covered area?
The meaning of “resolution 11” also needs explanation. Stating the approximate physical dimensions of the cells would be useful.
Aerial imagery
A short description of the imagery would be useful as some context for the study, including how it is acquired and its repeat coverage.
Table and figures
Table 1: The caption states “the percentage of crevassed area that is classified as highly crevassed”. In the last row, 33.7 km2 / 118.3 km2 = 28.5%, which is different from the 10.8% in the table. The 10.8% value appears to use a different denominator. Please check the percentage calculation and caption to ensure the values and their descriptions are all correct.
Figure 5: The text is small and the panel headings could be clearer. Please increase the font sizes and make better use of the available white space. The values of Cliff’s delta could be coloured based on their magnitude or direction to improve comparability.
Figure A2: Please format the bar labels better on the x-axis. A contrasting colour for the percentage labels on the bars would make them easier to read.
Figure A3: Please explain why a linear fit is shown and what reader should understand from it. The data appears more complex than a linear trend. Consider whether this is needed or a different fitted curve would better represent the data.
Technical corrections:
- Line 148: “Ground truth predictions” is slightly confusing. Replace with wording that clearly distinguishes model predictions from reference labels.
- Line 156: Correct to “Ötztal”
- Line 185: Add “model” to “…a digital elevation…”
- Line 198: Replace “Velocity data an ice flow…” with “Velocity data and ice flow…”
- Line 222: Cliff’s delta of zero does not necessarily mean complete overlap.
- Line 234: Replace “Transvers” with “Transverse”
- Line 287: You mention 0.13 as negligible but also define at least 0.11 as small. Please apply these thresholds consistently.
- Line 390: Remove the repetition of “second” in “The second is initiation depth. Second…”.
- Line 410: Check the order of the precision and recall values in “lower precision than recall (86.3% versus 73.7%)”. The values appear to be in the opposite order.
- Line 554: Replace “…which potentially sustained…” with “…which are potentially sustained…”
Citation: https://doi.org/10.5194/egusphere-2026-4768-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 308 | 216 | 61 | 585 | 61 | 64 |
- HTML: 308
- PDF: 216
- XML: 61
- Total: 585
- BibTeX: 61
- EndNote: 64
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1