Brief communication: Large changes in visible crevasse extent on Svalbard glaciers from 2017 to 2025, mapped with satellite embeddings

Schytt Mannerfelt, Erik; Schellenberger, Thomas

doi:10.5194/egusphere-2026-6305

Preprints

https://doi.org/10.5194/egusphere-2026-6305

Preprints

16 Oct 2026

| 16 Oct 2026

Status: this preprint is open for discussion and under review for The Cryosphere (TC).

Brief communication: Large changes in visible crevasse extent on Svalbard glaciers from 2017 to 2025, mapped with satellite embeddings

Erik Schytt Mannerfelt and Thomas Schellenberger

Abstract. Crevasses affect glacier dynamics, mass balance and travel safety, yet are rarely mapped across regions. We map visible crevasses on all Svalbard glaciers each year from 2017 to 2025 using AlphaEarth Foundations embeddings and a random forest ensemble, reaching 93±2 % accuracy on a withheld year. Of 817 glaciers with a significant trend, 616 lost visible crevasses, yet the regional total grew by 228 (181–281) km² a⁻¹. Surges cause the largest per-glacier changes, but glaciers without a recorded surge contribute comparably. Their decreases occur at low elevations, while increases concentrate in upper reaches where exposure by a rising snow line is a plausible explanation.

Received: 13 Oct 2026 – Discussion started: 16 Oct 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 2435 KB)

Supplement (8726 KB)

Download & links

Erik Schytt Mannerfelt and Thomas Schellenberger

Status: open (until 27 Nov 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-6305', Albin Wells, 27 Oct 2026 reply
This study maps annual crevasse extent across Svalbard using a high-resolution satellite-based dataset and a trained machine learning algorithm. The study focuses on changes in crevasse extents over the study period (2017-2025), and attributes many of the changes to the timing and prevalence of glacier surges. This brief communication is compelling and provides an interesting dataset that is of use to the broader glaciological community. I’d also like to add that the figures, in particular, are of very high quality. The work could be improved through (relatively minor) changes to make the writing more clear. Additionally, some of the statements and conclusions of the study appear to be overstated as currently written, given the presented data and analysis.

Main comments
My largest reservation about the study regards some of the claims made about surges and crevassing given the limitations of the study, i.e. that surges are the only process/factor the study assesses in relation to crevassing and that changes in snow cover implicitly impact interpretations of the results. Although the authors clearly state that snow cover is a limitation of the results, some statements need to be revised so as not to overstate conclusions. Furthermore, the study lacks an adequate comparison of potential factors that could affect crevassing, nor assesses crevasse changes by elevation(besides a brief mention of changes above/below 400 m asl) or equilibrium line altitudes, which could be used to infer potential impacts of snow cover change on the results. While it still seems reasonable to infer that surges impact crevassing, further analysis would be required to make more specific, targeted, and/or quantitative conclusions regarding this impact. As a brief communication, I understand that these changes may be beyond the scope of the current study, however, in the specific comments, I’ve pointed out a few sentences which I feel need to be re-phrased to appropriately reflect the data given their limitations.

Additionally, some of the writing could be more effective (in my opinion), particularly in the introduction. I’ve provided two examples at the end, although these changes are more stylistic and thus ultimately up to the discretion of the authors.

Specific comments
L10: suggest “increase”

L10: any rough estimates of the relative impact of crevassing on melt or dynamics?

L20: recommend citing the website, instead of “according to”

L23: I think this statement would be strengthened if you above include somewhere that crevasses are highly dynamic, and that they form and disappear over annual or multi-annual timescales (especially with rapid changes on Arctic glaciers, either from surge or thinning or melt-induced glacier flow pattern changes, etc). This would help motivate the need for these maps “every year”, which seems to be a big benefit of this study compared to previous studies or work.

L27: “explainability” … do you mean “process understanding”?

L27: recommend “We assess how… …and the impact of glacier surges on crevassing”

L29-30: This is very important to make clear, and I appreciate that it is so clearly stated here. However, I don’t think this is the appropriate spot for this statement. I think this paragraph should emphasize what you did (which you already nicely explain) and end with why it is important or a lens towards the future. I’d suggest moving this statement elsewhere.

L35: what do “positive” and “negative” refer to?

L35: Why are the years 2018, 2020, and 2022 skipped? Is this just because it’s labor intensive to digitize crevasses manually, so you select only every other year? This should be clarified somewhere.

L52: clarify what “within it” means

L55: This sentence is confusing. Please re-phrase to make it more clear.

L57: Fig. 2 is referred to before Fig. 1

L55-57: Is “the aggregate is positive at +228…” referring to the same thing as “the total across the archipelago increased” in the previous sentence? I think these sentences could be reorganized more cohesively to go present each result/pattern sequentially, instead of jumping back and forth, which was a little confusing to read (e.g., show patterns by glacier number/count first, then by glacier/crevasse area second instead of mixing these in both sentences).

L58: “Surging is responsible for the largest per-glacier changes in our data” I’m not convinced that this statement is fully justified, considering you didn’t assess many possible processes (please correct me if I’m wrong!). I’d rephrase to state that there is a significant correlation, without claiming that it is the greatest correlation.

L58: This would be more clear if you present the surge dataset briefly, something like “In total, there were N glaciers that showed a recent surge and an additional M glaciers that had an earlier surge.” Maybe this statement makes the most sense ~L51. Without it, I wonder how many of the active surge glaciers have an increasing trend vs. not having a trend or having a decreasing trend.

L60: “Among glaciers with a significant trend, surging glaciers have an 18 (15–21) times larger mean change magnitude than non-surging glaciers” Does magnitude refer to crevasse area change rate? I’m curious how much of this is based on glacier size. I think you may need to control for glacier area when comparing these rates, as the average area of the surging and non-surging glacier populations may be very different.

L62: “indicating that our data could serve as a detection method for new surges in the future.” I think this is overstated. Probably best to rephrase something more along the lines of helping to diagnose or be an additional indicator that points towards surging, along with other potential evidence. I’m not sure changes in crevasse extent alone can prove this (which you state both in the next sentence), and also considering large impacts from changes in snow cover. If you want to make a statement like this, you need to show that, against the surge dataset, there is some crevasse-change criterion that allows you to identify surging glaciers (ideally without missing too many surges, but certainly without many false positives).

L65-73: Could be interesting to look at where the decline in crevasse extent occurs, e.g., at which elevations. This could help point towards or eliminate some of the candidate processes you mention, but, as you mention, may be beyond the scope of this study in particular.

L74: You could mention somewhere that, when it comes to glacier safety and travel, this doesn’t actually matter. This is obviously very important when correlating with surges and gaining a process-understanding of glacier change, but in terms of glacier travel (which is one of many motivations for the work) simply mapping the increase in crevasse extent that implicitly includes changes in snow cover is actually completely valid and useful.

L78: To me, this seems like the same trend. Am I misunderstanding something here? And, it makes me think that assessing changes in crevasse extent within absolute or normalized glacier elevation bins across Svalbard would be perhaps a very useful assessment for this study, and could provide as much insight as the per-glacier assessments. Particularly when trying to parse signals that are clearly independent of snow cover changes, and which might be affected by snow cover change (e.g., you could look at changes above/below reported ELAs). I understand some assessment of this was done in the supplement (the exact methodology for this in the supplement is a slightly confusing) but I think a supplemental figure showing this relationship would further be beneficial.

L82: Are there any studies that look at snowline / ELA changes across Svalbard recently, particularly how much interannual variability there is in the end-of-summer snowlines? You could possibly even ballpark estimates of changes up to ~100 m (e.g., Larocca et al. 2024), or whatever approximate range you deem appropriate based on literature. It would be useful to have an estimate of the glacierized area this change exposes based on the hypsometry, so get a sense of how much this could realistically impact your results.

L86: Given the conversation about embeddings here, it's worth (either here or earlier in the methods) explaining in 1 or 2 sentences what embeddings are.

L104: does this not contradict earlier statements?

Fig. 1: Very nice figure. The map makes me really wonder about the physical interpretation of the maps. What fraction of a pixel needs to be crevassed for the model to classify it as crevassed? I think this would be very insightful and beneficial for the interpretations and usage of these maps, and should be clearly stated in one sentence in the methods or results (correct me if I missed this somewhere!). The implications of pixel size on the size of detected crevasses should also be explicitly stated.

Fig. 2: Another very nice figure. I think there is enough space to write out the glacier names instead of just the first letter. Also just noting, in the paper units are km2 a-1 so I would keep it consistent in this figure as well.

It may still be beneficial to try to figure out a way to provide these maps. I think these maps would likely be much more beneficial and insightful from a glacier travel and a route planning perspective. Even if these are only provided for a selection of the most commonly-traveled glaciers in the region.

Sentence improvements

L8-10: “Crevasses affect how a glacier functions: they lead surface runoff to the glacier bed, they warm the ice through refreezing of water, they weaken the ice structurally, and they precondition calving events at glacier termini (Colgan et al., 2016). They also raise ablation because a rougher surface can trap more solar radiation (Cathles et al., 2011).”

Suggestion, something along the lines of: “Crevasses affect how a glacier functions, through both dynamic and mass balance processes. Crevasses impact glacier dynamics by facilitating surface runoff to the glacier bed, weakening the ice structurally, and serving as a precondition for calving events at glacier termini (Colgan et al., 2016). Crevasses have also been shown to affect glacier mass balance through the refreezing of water, and increasing ablation because a rougher surface can trap more solar radiation (Cathles et al., 2011).”

L22: “What is missing is a way to continuously map crevasses every year over a whole glaciated region with a simple method.”

Suggested, something like: “As such, there are no annual inventories of crevasse extents covering entire glacierized regions.”

Reply
Citation: https://doi.org/10.5194/egusphere-2026-6305-RC1

Erik Schytt Mannerfelt and Thomas Schellenberger

Supplement

https://doi.org/10.5194/egusphere-2026-6305-supplement

Erik Schytt Mannerfelt and Thomas Schellenberger

Viewed

Total article views: 444 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
285	128	31	444	65	35	30

HTML: 285
PDF: 128
XML: 31
Total: 444
Supplement: 65
BibTeX: 35
EndNote: 30

Views and downloads (calculated since 16 Oct 2026)

Month	HTML	PDF	XML	Total
Oct 2026	249	112	28	389
Nov 2026	36	16	3	55

Cumulative views and downloads (calculated since 16 Oct 2026)

Month	HTML	PDF	XML	Total
Oct 2026	249	112	28	389
Nov 2026	36	16	3	55

Viewed (geographical distribution)

Total article views: 444 (including HTML, PDF, and XML) Thereof 444 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 06 Nov 2026

Short summary

Crevasses are deep cracks in glacier ice that route water to the glacier bed and endanger travel, yet are rarely mapped regionally. We trained a machine-learning model on satellite data to map visible crevasses on all Svalbard glaciers each year from 2017 to 2025. Three in four glaciers with a clear trend lost crevasses in their lower parts, yet the total area grew, likely as thinning snow cover exposed crevasses higher up. Yearly maps can help detect glacier surges and plan safe travel routes.


Total:	0
HTML:	0
PDF:	0
XML:	0