This paper adopts a set of methods to map grounded ice crevasses in the Amundsen Sea Embayment. The analysis shows the presence of surface crevasses, limited to a porous firn layer, that have migrated inland through time. The communication of the results is somewhat scattered, with some statements in the discussion and conclusions not fitting with the interpretation of your results. Figures need to be made clearer, data availability and repository need to be checked. Three major comments:

1. Provide some additional context/clarity on the methodology for identifying crevasses on grounded ice. The lack of basal stress present in ice shelves allows for an ideal and accurate assessment of the damaged state of the ice; this is the reason why previous work by Lai et al., 2020, Izeboud et al. 2023, Surawy-Stepney et al., 2023, focused their attention on ice shelves rather than grounded ice (no presence of grounded ice friction). Can you comment on that and add some clarification in the paper on why you focus on grounded ice? What are the problems with terrain shadows when assessing crevasses on grounded ice? It would be good to understand how this methodology compares to the other available crevasse maps (Lai et al., 2020, Izeboud et al. 2023, Surawy-Stepney et al., 2023)? As for the map of crevasses, these tools should be made available online as they would be useful to other researchers in this field, are you planning to do that? (the github repository is empty). In your conclusions, you state: “Finally, the ongoing collection of SAR images can also be processed rapidly enough using our automated framework that live crevasse detection in areas where researchers are conducting fieldwork is possible.”; how is this automated framework better than already available methods (Izeboud et al., 2023, Surawy-Stepney et al., 2023)?
2. Implications of firn crevasses: At the end of your discussion, you say, “Surface crevasses also affect estimates of ice flux across the grounding zone.” However, previously, you mentioned that your results suggest that surface crevasses expressed in satellite remote sensing datasets only weakly affect the bulk viscosity of ice shelves, and many times, throughout your text, you say that these features do not affect flow. It is important to remember that only those surface features that actively influence ice flow are pertinent to ice-flow dynamics and changes in grounding line flux. In the conclusion, you say: “The time series of crevassed-area evolution that we use to support these results presents a valuable target for models that incorporate near-surface fracture or continuum damage mechanics.” While I think they are useful maps, they do not necessarily help modellers when assessing damage, as they are such shallow features that do not affect the bulk ice viscosity. Moreover, If they were to propagate, ice shelves still bear enough buttressing capacity (Gerli et al., 2023a). I would discuss more about the local implications of these surface firn crevasse features (you already talk about the uncertainty in mass balance which is great to see). You could add something in terms of the local implications: increase of surface ice roughness, which enhances solar radiation and reflection in the surroundings and promotes atmospheric turbulent heat fluxes, all of which intensify melting at the ice surface, causing firn saturation, meltwater ponding and potential risk of hydrofracturing.
3. Perhaps future work? It’s a shame you don’t investigate more the presence of basal crevasses in the ice shelf. 

Specific comments

Line 131  – You use the F1 score to choose a threshold for the binary classification (greater than 0.8) ? Is a typical thing to do? Can you add references? I see the f1-score vs threshold plot in Figure S2; it would be good if you could add more information in the caption.

Line 237 – misspell Interpreted

Line 264 : Section 2.3?

Line 394-5: “These crevasses would not be expected to penetrate deeper into the ice column.” Expand and add references.

Line 458-9: “Our results indicate that changes in the crevassed area are generally correlated with increases in surface stresses in response to ongoing ocean-driven acceleration.” Why do you say that? Are you showing any correlation to ongoing ocean-driven acceleration in your results?

Line 465: “At present, the near-surface crevasse features appear to be a symptom rather than a driver of acceleration and retreat in the ASE.” Explain why is the case.

Figure comments:

Figure 3 : Would it be possible to outline the major crevasses or make a box-area for crevasses that are visible in SAR, and not visible in the panchromatic imagery? It would help the reader when you are comparing the two products.

Figure 4, enhance color visibility of panels b and c, red crevasses are barely visible. Panel d and e could have a higher contrast (especially panel d) for the detection of buried crevasses.

Figure 5 : Misspelling “in”

Figure 6: use a different colour scheme with better contrast; it is really hard to see the mapped crevasses

Figure 7 This is an interesting figure. I would be interested in seeing the location of these crevasses where stresses are greater than 200Kpa. Perhaps you can choose a year and represent their spatial distribution and colour crevasses as a function of stress?

Figure 8 red lines in c) are the same as the dotted lines in a and b?

Figure 10 subpanel c and d) I would colour the x-axis by the colour of the transect in subpanels a) and b) to make it easier to visualize. You can point out the major crevasse event in a) and b) that you talk about in the caption.

Figure S3 and S4 could be improved by adding some coloured lines to map features that correspond to both SAR and Optimal Imagery.

Figure S5 subpanels b) and d) have difficult readability; crevasse mapping needs better contrast.

Figure S6 a) and b) you could add the grounding line position for reference. c) and d) where are the black arrows?

I really like Figure S7, especially the radargrams of subpanels c) and d). Again just for reference I would add a grounding line position for subpanels a) and b) .

Line 346, 365 and 369-370 you mention figure S8 and S9, but I imagine you meant figure S7?  Or are figures S8 and S9 missing?

References

Gerli, C., Rosier, S., & Gudmundsson, G. H. (2023). Activation of existing surface crevasses has limited impact on grounding line flux of Antarctic ice streams. Geophysical Research Letters, 50(6), e2022GL101687 

Izeboud, M. and Lhermitte, S.: Damage detection on antarctic ice shelves using the normalised radon transform, Remote Sensing of Environment, 284, 113 359, https://doi.org/https://doi.org/10.1016/j.rse.2022.113359, 2023.

Lai, C.-Y., Kingslake, J., Wearing, M. G., Chen, P.-H. C., Gentine, P., Li, H., Spergel, J. J., and van Wessem, J. M.: Vulnerability of Antarctica’s

ice shelves to meltwater-driven fracture, Nature, 584, 574–578, https://doi.org/10.1038/s41586-020-2627-8, 2020.

Surawy-Stepney, T., Hogg, A. E., Cornford, S. L., and Hogg, D. C.: Mapping Antarctic Crevasses and their Evolution with Deep Learning

Applied to Satellite Radar Imagery, The Cryosphere Discussions, 2023, 1–32, https://doi.org/10.5194/tc-2023-42, 2023b.

Please see the attached PDF.

I very much enjoyed this article delineating crevasses in the Amundsen Sector. The delineation approach is quite novel, and the time-space analysis of crevasse changes is also quite novel. I share some thoughts on a few general points below. Beyond methodology and results, I also enjoyed the interpretation, although I think more support is needed before stating so conclusively that the crevasses are limited to the firn and have no dynamic impact on ice flow.

Firn-only -- The crevasses are referred to be “restricted to the firn” several times, including at abstract level. I would like to see further context for this conclusion. For example, what is the pore close-off depth in the region (i.e. firn-ice transition)? Simply put, how deep is the firn? Related to this, Figure 4e shows very deep crevasse tops (i.e. 130 m deep?). It is said that these may be off-nadir crevasses that are actually shallower, but in terms of a first-order calculation, how off-nadir would these crevasses need to be for the geometry to be projected to that depth? Is that reasonable? At present, the reader is uncertain how deep the firn extends, and whether crevasse tops are deeper than this.

C-band penetration -- As written, the paper seems to overestimate the penetration depth of C-band SAR, saying “C-band SAR imagery penetrates up to several dozen meters into the subsurface.” My understanding of the cited Rigot2001 is max 10 m penetration depth. Jezek only says “several meters” https://doi.org/10.3189/172756499781820969 . “Several dozen meters” seems to be quite on the high side, although I am not sure what impact, if any, the depth of penetration would have on analysis.

Crevasse width – I suppose it is implicit that the algorithm only detects crevasses wider than ~10m, or the Sentinel pixel size, or is it the resampled ~25m pixel size? These are clearly large crevasses. Is it possible that there are smaller, or narrower, crevasses that go undetected by the algorithm? Or put another way, can the authors say something about lower limit of crevasse geometry down to which they have detected? Presumably these maps would be a lower limit on the damage extent of ASE, if some scale of smaller crevasses and fractures has gone unmapped.

Firn Fracture Model – I like the idea of modelling firn fracture, as this is not often done. I guess there should be an explicit assumption stated that firn properties are constant over the 2017-2022 epoch of crevasse migration. Presumably, if the firn properties are changing through time (i.e. firn becoming increasingly brittle due to refrozen ice layers) then this can also impact apparent crevasse extent, without changes in the underlying ice stresses. This paper implicitly assumes that only dynamic stresses have changed, not firn properties. This is fair, but should be made explicit and perhaps discussed. 

Limited Direct Impact – I have some difficulty accepted that there is limited dynamic impact from these crevasses. The ice thicknesses only look ~300 m thick in Figure 4, and the authors have only delineated the top of the crevasses. Even if the crevasses only extend 50 m deep, that is still 1/6 ice thickness. By the provided van der Veen citation, it is conceivable they could be 100 m (1/3 thickness) deep. Presumably some ice dynamic model parameterized with and without such crevasse geometry/prevalence is needed to state so conclusively that such large crevasses are not impacting relatively thin ice?

Failure Envelopes – Figure 7a is too small to be useful. I’m not sure if presenting 32 failure envelopes is the best thing for the reader. Perhaps anomalies, by either the eight years and/or four seasons, might be more informative to highlight differences and change. In Figure 8, are you calculating the “crevassed” and “uncrevassed” pixels at the 25 m resample pixel size? Presumably there would be “uncrevassed” pixels between individual crevasses. So, are you averaging over some distance? Perhaps visualizing a binary crevassed/uncrevassed map would be a helpful inset here.

Tensile Strength – Figure 9: I guess there is more data than this available. See, for example, https://erdc-library.erdc.dren.mil/jspui/handle/11681/2698, which comes to mind. At the moment, misfit between remotely sensed crevasses and measured tensile strengths is attributed to “porosity dependence of near-surface tensile strength”. I wonder if there may also be some effects of anisotropic fabric, even in the near surface firn?

Fracture Mechanism – Anisotropic firn fabric would be most relevant for mixed-mode fracture. Perhaps firn reacts similarly to flow-aligned mode 1 opening across the study region, but if there is mixed-mode fracture, then crevasses aren’t necessarily always associated with mode 1 opening aligned with fabric. It would be helpful to have some velocity-derived flowlines on a map with delineated crevasses, so the reader than see that flowlines generally intersect crevasses at 90° (i.e. characteristic of mode 1 opening).

This study presents a method for mapping surface crevasses using a neural network image segmentation approach and satellite SAR imagery. It is applied to the grounded areas of several glaciers in the Amundsen Sea Embayment (ASE). The authors compare the derived crevasse fields against optical imagery to demonstrate the capability of mapping buried crevasses. They validate their results and interpret them in the glaciological context using ice-penetrating radar data and a geometric model for the tensile strength of firn. Thereby, the authors demonstrate that crevasses in the grounded regions of the ASE glaciers are confined to the near-surface, which suggests a limited influence of these crevasses on ice dynamics and ice shelf stability.

The various methods and data are nicely combined, and relevant implications are discussed. However, due to this multitude of approaches, some parts of the manuscript lack clarity and are not explained in sufficient depth. I suggest the manuscript be published after the following remarks have been addressed.

Major remarks:

1. The geometric model is not new and not properly used in the manuscript. Its derivation and the design of Fig. 2b-c closely follow Jelitto and Schneider (2018), which is not adequately stated. The authors only have applied the geometric model for the tensile strength of porous materials from Jelitto and Schneider (2018) to polar firn but not developed it. In the conclusion, the authors write, "Using this new model, we showed that the increase in the crevassed area on Thwaites Glacier is consistent with the increase in the area where effective surface stresses exceeding a critical tensile strength for the near-surface firn." I do not see how the model is used in this way. The model is only tuned to laboratory data but not used any further. For example, why is it not applied to available firn data from Gow et al. (2004)? This could give a prediction of crevasse depth for a given stress field.
2. The relation between increasing effective surface stresses and crevassed area is also elsewhere not clearly shown. Variations of flow speeds are only described in the text of Section 3.2. They could be displayed nicely along with data of the grounded crevasse area from Fig. 7. I assume the intention of Fig. 7c is to illustrate the flow dynamics, but the text never refers to it, and it is unclear why "grounded ice area where σ_e > 200 kPa" is chosen as the metric for illustrating this. Why is the threshold at 200 kPa and not 75 kPa, where crevassing initiates? I would suggest illustrating this relation by looking at the time series of effective surface stresses in areas where new crevasses open (e.g., cyan arrow in Fig. 6). This might also allow pinning down the threshold for the initiation of crevassing more precisely.

Minor remarks:

L30: I do not think a new paragraph is needed here.

L170-173: The trace spacing remains unclear without knowing the stacking. With no stacking, these numbers would give a very fine trace spacing of 1.6 mm. The horizontal resolution is also not directly set by the trace spacing, which only gives a lower bound, but it depends on the distance to the target and radar system characteristics.

L227: n is not introduced.

L294: The meaning of the upper 210 kPa envelope is unclear to me; from Fig. 8b, it seems that in general, no higher stresses are present at Thwaites Glacier. Do the 210 kPa only reflect this fact, or is there a deeper reason why higher stresses cannot build up, for example, because stresses are released by the formation of crevasses?

L330-331: More specifically than saying off-nadir, these could be crevasses that initiate next to the radar line and are recorded from off-track directions. Interestingly, these reflections do not show a refraction shadow, which supports this interpretation.

L331-333: Is there evidence that such large flaws can form deep in the interior of a glacier? How could this be explained? Is it also an option that these are former crevasses that advected down from the upstream crevasse fields?

L346-345: Figures S8 and 5e do not exist.

L358-361: Is there other evidence that the dipping of internal layers into the crevasse is an actual signal that can be attributed to a disruption of the flow? Or could these dipping layers also be caused by off-nadir reflections, for example, from hyperbolic reflections from the layering at the sides of the crevasse or from large crevasses that are not perpendicularly oriented to the radar line?    

L449-451: The discussion of the effect of crevasses on the firn air content is interesting, particularly the remark that it weakens the impact of horizontal divergence. However, I am not convinced by the estimated 2-8 m of unaccounted firn air content at lower Thwaites glacier due to crevasses. The total extent of crevasses cannot be directly identified by the crevassed area. For example, in Fig. 10c nearly the whole area between 7 km and 10.5 km distance has a high crevasse probability, but this of course only indicates the presence of many small crevasses in that region and not of a single big one.

Figure 4: The clarity of these figures could be improved by showing the radar lines (a-c) and the radargrams (d-e) in the same orientation, as it is done in Fig. 10. This also applies to Figs. S5-S7.

Figure 8: Typo in "envelopes".

Figure 10: The orientation of the radar profiles is not indicated.

References:

Gow, A. J., Meese, D. A., and Bialas, R. W.: Accumulation variability, density profiles and crystal growth trends in ITASE firn and ice cores

from West Antarctica, Annals of Glaciology, 39, 101–109, https://doi.org/10.3189/172756404781814690, 2004.

Jelitto, H. and Schneider, G.: A geometric model for the fracture toughness of porous materials, Acta Materialia, 151, 443–453,

https://doi.org/10.1016/j.actamat.2018.03.018, 2018.