Dear Veronika,

Referee #2 has had a new careful reading of your manuscript and acknowledge for the improvements since the last version. Nevertheless, there are still two important comments that should be addressed as well as few minor typos. Based on your reply and on the new version you will provide, I will take a final decision.

Regards,
Olivier Gagliardini

Dear authors,

I have now read carefully this new version of your paper as well as your reply to the reviewer comments. I found that a lot of the remarks were answered but the manuscript still presents a lot of typos and errors and could be improved on many aspects. I have edited in the supplementary file a number of them that should be fixed before your paper can be accepted in TC. Please have a careful reading of the paper before resubmitting.

The structure of the paper could also be improved. It is not always clear how the predictors are inferred, directly from a dataset (which one) or from a dataset but using some inverse or forward modelling. For example, it is not clear from what the stress are computed: observed velocity or forward model? This should be mentioned directly in the text for each predictor, and synthesised in Table 2.

Dear Veronika,

Thanks for this new version of your paper as well as your reply to my questions. I think that your paper can now be accepted in The Cryosphere after some minor corrections have been done (see the list below).

I my last comment, I forgot to ask about the way you are going to distribute the dataset you have build by picking up all these crevasses over Antarctica. One of the former reviewer has acknowledged that your paper should be accepted because of the interest of this dataset for the community. I strongly suggest adding a section "data availability" indicating how the data can be obtained (data repository, on request to you or whatever, see here for more information: https://www.the-cryosphere.net/about/data_policy.html).

Regards,
Olivier Gagliardini

Minor remarks:
- page 1, line 25: of Larsen C ice shelf](Mercer, 1978 -> of Larsen C ice shelf)(Mercer, 1978
- page 2, lines 25-34: this part is more already a conclusion presenting the main results. The introduction section should end with a presentation of the different sections that will follow. Also, regarding the comparison of the LRA results and the damage results, there is a bias that should be discussed. Your method is build on the observations you are using to evaluate the method whereas the damage method is not using these observations, and is doing a much more complicated job?
- page 4, line 17: this is not completely clear if this first dataset is included in the others? And also, my understanding was that the dataset only contains the crevasses location, but seems not the case from this sentence?
- page 4, line 21: (described in Section 4.2. The 450-metre resolution horizontal ice velocities were taken from InSAR (Rignot et al., 2011b, a)). -> (described in Section 4.2). The 450-metre resolution horizontal ice velocities were taken from InSAR (Rignot et al., 2011b, a).
- page 4, line 27: two-dimensional (2-D) and three-dimensional (3-D) ...
- page 5, line 23: mélange is more a mixture of icebergs and sea ice than snow and sea ice
- page 6, line 34: to the fact if there -> to the fact that if there
- Eq. (1): should be x^*_{ij} in this equation?
- after Eq. (5), is it a new sentence (then a dot after (5)), or not (then a "and" before Von Mises).
- page 12, line 4: fracture formation (described in Section 4.2.1): -> fracture formation (described in Section 4.2.1).
- page 12, line 8: the definition of the nodal function should arrive before the previous sentence. Also, this sentence is a bit technical and could be simplified? e.g. The calculation of these (not this) parameters is performed by computing the derivative of the bed and surface using the finite element nodal function in ISSM (I don't think you need to define them).
- page 13, line 7: only effective deviatoric stress -> only the effective deviatoric stress
- page 13, line 10: deviatoric or Cauchy principal stresses?
- page 17, line 23: a bit vague and it is really needed? This section about damage is mixing bibliography on previous works (that might have been presented in the introduction) and what is really done here, and all is a bit mixed so that it is not easy to follow. I suggest to move part of this material in the introduction to only keep what you have really used to infer the damage here.
- page 24, line 19: what do you mean by surface gradient? surface slope or strain-rate? (surface slope doesn't seem to be so important from Tables 4 and 5).
- page 25, line 9: fractures in or evaluation -> fractures in our evaluation
- page 26, line 4: I don't understand what you mean by "zero stress". Also: Mechanics, Continuum Damage Mechanics -> Mechanics and Continuum Damage Mechanics.
- Figure 7: Is there red dots only in panel c? This is difficult to see red dots over the damage or LRA coulour scale.
- caption of Fig. 8 should be: Labels are the same as for Fig. 7. and the last sentence removed.

Dear Veronika,

Thanks for this new version and your reply to my comments.

I nevertheless think there is still missing information about the way the 46 datasets are constructed. Page 5, it is mentioned that the first dataset includes combined information from 35 regions, and the 45 others information for each individual region. Is the first dataset including all the information of the 35 regional datasets (i.e. is it just the sum of the 35 individual dataset?)? Some words should be added about the choice of these 35 regions? How they were chosen? Randomly? How the results are influenced by the choice of the 35 regions included in the first dataset? The 35 regions chosen to calibrate the model should be tagged in Table 6?

This bring a second question, that was already in my last comments but that you didn't answered : "Also, regarding the comparison of the LRA results and the damage results, there is a bias that should be discussed. Your method is build on the observations you are using to evaluate the method whereas the damage method is not using these observations, and is doing a much more complicated job?". This is even more true if the method is evaluated against data 35/45=78% of which have been used to calibrate the model?

Regarding my remarks on the wording "surface gradient", that you have changed in "surface change": you should check in all the manuscript that the same wording is used for all the predictors all along the manuscript. For example, the name (and number! 14 in Table 2 and 16 in Tables 4 and 5) of predictors in the first column of Table 2 should be the same as those in the first line of Tables 4 and 5 (even is some of them are only relevant for grounded part or floating part, they should be tagged as irrelevant). Typically, "surface change" is not listed in Table 2 as a predictor and this name appear for the first time in the discussion page 24, where it is given as one of the most important (and its definition is only given in the discussion, why?). All the predictors (16 as in Tables 4 and 5 or 19, as stated at the beginning of 4.2) should be defined in Table 2 and in the text in Sect. 4.2. Also, the order of the predictors in Tables 2, 4 and 5 should be the same for an easiest comparison. Names should be exactly the same: are "strain-rate gradient" in Table 2, "strain change" in Table 4 and "strain-rate change" in Table 5 the same predictor?

Regards,
Olivier Gagliardini

Dear Veronika,

Thanks for this new version and your reply to my comments.

After this last reading, I have still some technical points that should be corrected before moving to the publication stage (see below).

Best regards,
Olivier Gagliardini

Technical corrections:
- page 5, lines 6-7: are these two BC rely needed for the statistical model? I think it is for the ice flow model?
- page 5, line 15: on these 35 ice shelves/glaciers, completed by an additional 10 ice shelves/glaciers (should be mentioned here already).
- page 5, line 21: (described in Section 4.2. The 450-metre resolution -> (described in Section 4.2). The 450-metre resolution
- page 6, line 10: regions to evaluate the predictor factors (?)
- page 7, line 15: Therefore, we use our calibration dataset to train our LRA. -> Therefore, we use the calibration dataset to build the LRA model (?)
- page 11, line 2: as described in (Borstad et al., 2013). -> as described in Borstad et al. (2013).
- Table 1: what does the horizontal black line mean? the 2 2010 references could be on a single line.
- Table 2: \mu and \lambda have never been defined. To be removed
- Table 3: why glaciers 4 and 29 are in red?
- Table 4: in the note: it's -> it is. Also, add "No value indicates that the predictor is not included in the model."
- Legend Table 5: Same as Table 4 but for floating ice
- Table 6: why glacier 4 in red?
- Figure 1: replace "success simple set" by "effective deviatoric stress", etc.. ; replace "success LRA" by "LRA"
- Figure 2: in panel b, where success Damage is larger than LRA, LRA is not shown. Same if success damage is lower than failure LRA. What is the problem for Glacier 4 (all success lower than failure I presume?)?