Dear Editor and Authors,

Please find attached my review of MacGregor et al. (manuscript number: ESSD-2024-578) with manuscript title “A revised and expanded deep radiostratigraphy of the Greenland Ice Sheet from airborne radar sounding surveys between 1993–2019”.



This paper is Version 2 of an already very important and successful paper/dataset by MacGregor et al. 2015, which has helped inspire and guide the Antarctic community to achieve something similar over Antarctica via the SCAR AntArchitecture programme. As for its Version 1 paper/dataset, MacGregor et al. have produced a robust dataset that is well described and presented in the accompanying paper, and that will undoubtedly be used extensively by the glaciology and modelling community as additional paleo constraints over Greenland. Knowing how much effort and time it takes to pick layers over different radar data across large areas, I can particularly appreciate the effort required by the authors to produce such dataset.

My main comments in this review pertain to the data formats used, as well as the content and structure of the abstract and conclusion sections. However, by in large, most of my comments are very minor and might just require a simple clarification or light re-structuring which I don’t expect will take very long as the paper and associated dataset are already in great shape. As a result, I would recommend this paper to be published in Earth System Science Data with minor revision, and I very much look forward to seeing the updated version soon.

With best wishes,

Julien Bodart

-------------------------------------------------------------------------

General comments

Line-by-line comments:

• Lines 38-41: Perhaps it would be relevant to cite the AntArchitecture paper (Bingham et al., 2024; in review) here to guide readers early to a review paper of isochronal stratigraphy and its uses/benefits. Sure, it’s based on Antarctica mainly but would fit in well nonetheless here.
• Line 47: Citation to Rodriguez-Morales et al., 2014 is missing in the reference list. Do you mean Rodriguez-Morales et al., 2013 (IEEE)? I have not checked the other references, but I would encourage the authors to do so just in case.
• Line 86: Sure, but the reason is because they had not been acquired yet when M15 was published. So, it is perhaps more accurate to say: “Values in parentheses for v1 are surveys that had not yet been flown and therefore traced by M15”, or similar.
• Lines 107-110: This is more for my own understanding, but I don’t understand why having repeat tracks in the dataset is an issue. Maybe it is clear to the authors who “see it” when they process the data and grid, but to me it sounds like a good thing: if it is a repeat flight with the same xy, the layers should be the same across both profiles and thus be an advantage rather than an inconvenience? Why would this be an issue for producing an “ice-sheet-wide radiostratigraphy”? Perhaps just a few words to explain this would help.
• Line 130 and Figure 2: You mention the word “set” – is this the same as “segments” in Line 70? If so, use a common word throughout (I personally prefer “segment”)
• Lines 156-158: This sentence is a bit confusing – could you rephrase it please?
• Line 256: “the reduced set” – do you mean the “greater set”? (i.e. the opposite of reduced?). Your previous sentence says that you relaxed your search radius which increased the number of core intersections?
• Line 261: “near” – could you be more specific (e.g. how many samples below or above)?
• Line 270: “paleoclimatic interest” – could you provide some references or be more specific to justify why those specific ages were chosen?
• Lines 307-308: Can you be a bit more specific as to how you account for the uncertainty associated with the interpolation/extrapolation of “age” here?
• Line 317: Refer to Figure 2 (third panel) here.
• Line 429: You could consider adding Sutter et al. 2021 (https://doi.org/10.5194/tc-15-3839-2021) here too.
• Lines 436 or 443-444 (when mentioning Karlsson et al. 2024 dataset): One could also highlight the higher level of uncertainty in the geolocation of the radar profiles which could introduce errors when comparing with this v2 dataset.
• Line 446: One could also add Bodart et al. 2021 (https://doi.org/10.1029/2020JF005927)
• Line 458: Again here, I would add Sutter et al. 2021 (even if it’s over Antarctica)
• Conclusion: Again, I would recommend that the authors add a bit more detail to this section, in a similar way than for the abstract. I found the conclusion a little underwhelming considering the achievement of this V2 dataset, and I believe it is worth highlighting again the key messages and figures shown in the paper. Perhaps this is also a further opportunity to encourage the modelling community to make use of this dataset to constrain their paleo simulations.

Figures and captions:

• Figure 1a: It is a bit difficult to see the difference between low and medium priority colours on Fig. 1a due to the white background. Could this background be grey, or could the colour scale be changed to something else (e.g. divergent)?
• Figure 2: Again, here for the “Date reflections” panel, it is a bit hard to distinguish on the radargram the different colours. Could you use a divergent colour scale? Also, perhaps it would be useful to name the sub-panels (a-d) and refer to each of these steps in the text. Finally, and still relating to panel 3 of this Figure, I find the information presented a bit confusing for several reasons: (a) I suspect that the numbers provided at the bottom of this sub-figure are for the whole GrIS, but it can be confusing as one might interpret that these numbers pertain to this specific segment; (b) a more complete caption would really help guide the reader, as it is not easy to understand what is meant by “overlapping reflections” and why there are two arrows between this step and “match to overlapping reflections”, beyond the obvious fact that it’s a closed loop. One of course can find this information in the text somewhere (or on Figure 5, which does help), but the figure and accompanying caption could help the reader more to get a quick sense of what is being presented without having to go find it in the text.
• I find Figures 6 and particularly Figure 7 very well made and informative. They are definitely the key figures of the paper for me, and the statements made in and around these figures (e.g. Lines 342-345 and 355-366) could serve as a basis for an improved version of the existing abstract (and conclusion), as discussed above.

MacGregor et al. describe a dataset of revised and expanded deep radiostratigraphy of the Greenland Ice Sheet (GrIS) from airborne radar soundings collected by KU/CReSIS/NASA/NSF between 1993 and 2019. The dataset is an improved version of the initial version from MacGregor et al (2015). The authors describe the methods to generate the data, methodological improvements to the initial dataset version with the aim of developing a more complete radiostratigraphy of the GrIS with reduced uncertainties. Along to this manuscript, the authors publish the data as a gridded product for modelers and point data in different data formats.

In my opinion, this dated radar stratigraphy dataset (along with the previous version by MacGregor et al., 2015), represents one of the most important and comprehensive datasets of the Greenland Ice Sheet (GrIS). It is particularly relevant for paleo ice sheet modeling and deciphering past dynamics of the GrIS and deformation history its englacial architecture. A homogeneous dataset of radiostratigraphic information of this scale is unique and demonstrates an immense amount of work and foresight in publishing it in such a consistent form as presented in this paper.

The improvements to the previous dataset are very useful, and I highly appreciate that they have been made. Additionally, the thorough documentation of these improvements (e.g., Table 2) is well-presented and well-justified. The description of the tracing philosophy, methodology, and procedure is detailed and very useful for anyone working with internal reflection horizons in radar data. The coverage, usability, and dating approaches are well-explained and demonstrate that the data have been sufficiently validated and are robust. I also find the discussion very useful, particularly, the part on machine learning methods for accelerating the tracing of radiostratigraphy.

I appreciate that this dataset has been submitted to ESSD and recommend its publication, taking into account a few minor comments and questions.

________________________________________________________________________

Main points (manuscript)

• L375-381: I agree that your gridded data product is difficult to compare to those released in Franke et al. (2023). My additional remarks here are that they are particularly difficult to compare because the “FINEGIS” and “NEGIS” in northeast Greenland data sets are based on AWI radar data alone. Hence, in addition to the different ages of the AWI IRHs, the data basis is also different.
  
  The Peterman and central Greenland datasets in Franke et al. (2023) are constructed from CReSIS data, however the segment-wise interpolation along fold anticlines and synclines to maintain fold orientation introduce also differences in the gridding products.
• L437-438: Consider citing rather the original studies of the AWI surveys (e.g., Franke et al., 2022 and Jansen et al., 2024) instead of the data description paper.
• L438-444: Additionally, one could mention here that a large portion of the AWI radar data in Greenland have a similar range resolution as the CReSIS data: ~ 5 m for the EMR system with the 60 ns burst and ~ 4.3 m for the AWI MCoRDS 5 system in narrowband mode (180-210 MHz).
• L467: Code and data availability:
  
  Please provide the .mat file version, as for some non-Matlab programs or libraries this is important to know.
• Table 3: Is somewhere documented that the stratigraphy.twtt in the mat files is two-way travel time below the surface reflection twtt? I can’t remember finding this info in the text, but I might have overlooked it.
  
  At least, when I tried to plot the stratigraphy.twtt on top of the radargrams (example: 19990507_01_001-003) they needed to be below surface twtt to appear correctly.
  
  Regarding stratigraphy.int Reflection relative echo intensity: How is this determined? Is it the value in dB of the pixel where the pick is allocated or a cumulative value within a certain window around the pick?

________________________________________________________________________

Main points (dataset)

• I made a short test plotting the stratigraphy of one segment for Greenland_radiostratigraphy_v2_1999_Greenland_P3.mat – Segment: 19990507_01_001-003 and my impression was that working with the data was user-friendly and I find the information in the matfiles very useful and their structure very well organized. I think it is ok to have one file per season and not one huge file or hundreds of smaller files.
• I have read Julien Bodarts review, and I agree that publishing the extensive data version as .mat files only (I am aware that you are also providing a GeoPackage version with reduced information) can be problematic for non-matlab users (or those without a license). For Matlab users, this is format and structure is wonderful. For Python users you mention the mat73 package, and it is probably the only python library (that I have found) that can easily read the matfiles provided here because it directly translates it into a python dictionary with nested lists and dictionaries and keeps most data types intact. I’ve tried to load the matfiles with h5py, which I believe is more common to read HDF5-based data formats in python, and which should work also, but here I experienced problems accessing the nested data structures and resolve the data types.
  
  However, I admit that I have not invested an enormous amount of time in this, and therefore, this comment should be seen more as the perspective of a potential user who is trying to use the data. I also don’t know how data access works for users who neither use matlab or python.
• I have mixed feelings about the idea publishing the matfile information as tabular data and I understand the dilemma, because due to the nested structure it will probably end up either huge files with tens to hundreds of columns or in many many single files. I think one idea for a more universal access to the information provided in the matfiles could be to save them in the same or similar structure as netCDF files. However, I cannot say for certain whether I, as a reviewer, can demand that this format should be provided. This could and should probably be alongside the mat files and not replace them, as for Matlab users and for the CReSIS/KU community they are the most useful data format.
• Regarding the Geopackage files: I was unable to retrieve the metadata on the ages of the different depths. Could you provide either an instruction how to access this metadata in the geopackage files or (as Julien suggested) add the ages to the “depth_*” variable names?

________________________________________________________________________

All in all, thank you for the interesting read and I am looking forward to see this paper and data published.



Best wishes,

Steven Franke