Grinsted et al. use ice-sheet-wide geophysical datasets to assess the applicability of a previously untested failure criterion, the Schmidt-Ishlinsky (hereafter S-I) criterion, to predict ice failure across the Greenland Ice Sheet. This is done with the intention of being able to develop accurate predictors of crevasse formation and presence in large-scale ice-flow and fracture models. The paper is timely and is a clear evolution of the ongoing large-scale work being done on ice failure in recent years. It is a bit of an “open secret” that von Mises (hereafter vM) is a poor predictor of crevasse failure and it is good to see work being done to find better approaches, especially involving large-scale datasets. I have some thoughts regarding the failure criterion chosen, and in particular the lack of wider assessment given the prime opportunity provided by the datasets collated.

As written, it is unclear what motivation or hypothesis led the authors to propose and test the S-I failure criterion. After the abstract, the criterion is not mentioned again until the discussion (L136), where it is defined but without any clear motivation as to why. At this point, it is argued that although both vM and S-I perform quantitatively identically (L144), if measurement noise was better S-I would likely be the better performer (L144-146). It is not possible for the reader to assess whether S-I performs better or worse than any other alternative criteria common in the literature, as this data is ‘not shown’ (L147-148). 

The S-I is not a criterion I have encountered before in the glaciological literature, and appears to be pretty niche outside the discipline as well -  indeed, googling the phrase ‘Schmidt-Ishlinsky failure criterion’ includes this preprint among the top results. As a result, I think it is important that the authors do more to contextualise and explain the criterion, and their motivation for choosing it. This is especially true as the authors disregard even comparing the criterion to other options, as previously noted (L147-148). Given the clear effort that has been put into producing and collating the various ice-sheet-wide datasets, it would be very interesting to see a comparison/EDA of all previously suggested criteria, and make it more convincing that the S-I criterion is (qualitatively) a better option.

In the absence of this comparison, a deeper theoretical concern I have is that the S-I criterion doesn’t appear to improve upon a key weakness of the vM criterion, which is that it is relatively insensitive to the direction of stress - indeed, it is noted by the authors that the difference between compressive and tensile ice strength is an order of magnitude, and that this isn’t consistent with the vM criterion (L38-40). As the authors note, the S-I criterion does not significantly deviate strongly from this assumption (L154). Although they highlight that the S-I criterion implies ice is 15% stronger in shear relative to the tensile stress (L142), as far as I can tell from Fig. 4 the S-I criterion implies that ice is *exactly as strong* in tension and compression. However, examining Fig 4 appears to show far fewer crevasses in the compressive sections of the figure - especially if the color scale is log, which these plots often are. Therefore, although I am excited by the datasets and study design presented by the authors, the paper (at least, in its current form) leaves me questioning that crevasse failure can be adequately described by any prescribed radius in the π-plane.

MINOR COMMENTS

L95/Fig 1 - It is not mentioned in the methods exactly how the high-elevation exclusion zone is exactly derived. Manually determined? If so, the mask could be included as supplementary data.

L46 - Chudley et al. (2021) also use this data to assess crevasse formation, which is probably worth including/contrasting/comparing in the discussion.

L80/Fig 4 - Although I understand that plotting on the π-plane is a key point of this paper, I imagine most will be more familiar with plotting on a simple τ_1/τ_2 plot following Vaughan (1993). I highly suggest including this alternative visualisation in the supplementary material to aid the interested reader in comparing and contrasting, as well as in understanding how this visualisation differs from Vaughan’s approach.

Fig 3 - Some indicator of y axis scale might be nice (unless normalized?)

Fig 4 - color scale needed for quantities.

L114-115 - Observational evidence of this can be found in recent papers (Harrington et al. 2017, doi:10.3189/2015AoG70A945; Hubbard et al. 2021, doi:10.1029/2020AV000291). I agree that it is likely that modelled MAT represents a lower bound of likely temperatures. Though for practical purposes, I don’t have a better suggestion of how this can be approached. 

L123-128 - This is absolutely fascinating. How could the seasonally varying regions be better represented? Is it a case of crevasse initiation being initiated at the maximum velocity/stress? Or limited by the minimum velocity/stress? 

L129-132 - Or limited by resolution/ability of crevasse dataset? The crevasse dataset is taken from another source and no limitations are discussed in the paper.

L136 - does the data have a hexagonal pattern, or is the data clustered around the shear components and the hexagonal plotting style gives the impression that this is the case?

See attached file.