Dear Dr. Lu and co-authors,

I have received the following comments from the two reviewers. They both acknowledge the efforts made to improve the manuscript but still have concerns about the presentation of the dataset made in the manuscript. Please carefully address the points raised and upload a revised manuscript along with a point-by-point response to the reviews.

Kind regards,
Baptiste Vandecrux
---
Report #1:
I commence the authors for a very thorough revision that significantly improved the manuscript and dataset. While all my previous comments have been addressed, further minor revisions are necessary which are outlined in my detailed comments below.

Abstract
Line 23: What does GrTDP stand for? Maybe write it out first and then add the abbreviation in brackets.
Introduction
Line 46: Currently, the Enderlin et al., 2014 and Mouginot et al., 2019 citation are the are incorrect as both discuss ice discharge and not frontal ablation. While similar, they are not the same. I would suggest changing “frontal ablation” to “ice discharge”.
Line 96: Add space between “glaciers” and citation
Line 99: Add space between “adjustment” and citation.
Line 101: Add space between “variability” and citation.
Line 110: Please write out the abbreviation when first used.
Figure 1: Maybe include in the caption for panel a) what the legend means as it might not be obvious to all readers. Also, include a citation for the Basins as I assume that they are the Mouginot et al., 2019 drainage basins.

Data and Methods
The numbering of the headlines is not correct in this section – It goes from 2.1 to 2.4 to 2.3 with 2.2 missing.
Line 148: GEEDiT is an abbreviation and should be written out the first time it is used. Also, the citation for GEEDiT (Lea, 2018) is missing.
Line 152: Add “and” between citation and “were manually downloaded”
Line 152: I am not sure the citation of Rignot and Kanagartnam, 2006 makes sense here. While they are using ASAR in 2004, they are using ERS1/2 for winter 1996 which is not within of the timeframe of the dataset presented here. Consider moving the citation to after the ASRA segment, and please add a citation to the ERS1/2 segment of the sentence.
Line 166: Change “was” to “were” as you are referring to satellite acquisitions.
Lines 177-178: Consider adding a citation to support the statement that glacier retreat at these glaciers is gradual and linear over monthly timescales.
Line 196: remove comma before semi colon.
Line 216: Change “terminus” to “termini”.
Line 230: What auxiliary dataset were used?
Figure 2: I am not sure what the box on the right under Quality Assessment is showing – Either include a description in the caption or consider removing it.
Validation
Line 278: Change “summaries” to “summarize”.
Lines 279 – 280: Repetition of Line 272. Consider removing.
Line 289: Remove “the” before Fig. 3a and c.
Lines 305 -308: The dashed horizontal lines in Fig. 3b and d are very hard to see. Consider choosing a different color. Also, both dashed lines appear to be at 0 so I do not understand how they can emphasize the difference between the two products.
Line 331: Add space between CALFIN and citation.
Line 340: Add space between “offset” and “(>1km)”.
Line 349: change “as the calving” to “at the calving front”.
Line 359: Change “well consistency” to “good consistency”.

Zenodo repository
- In the description, please use citations when referring to other datasets or tools e.g. GEEDiT, TermPicks etc.
- Rename “GrTPD_Centralline.gpkg” to “GrTPD_Centerline.gpkg”
- For Glacier ID101 the glacier name is “Spalteglet” which should be changed to Spaltegletsjer.

---

Report #2:
Thank you for the significant revisions to the manuscript. Please see the general and specific comments regarding this revised manuscript below.

General comments:

*Introduction*
Thank you for the changes. Still, I think the introduction does not do a particularly good job, given that the utility of calving front data is widely acknowledged. In my eyes, it could still use some streamlining and slight restructuring.

The long paragraph describing the need for land-terminating glacier fronts overlooks the fact that, from a mass balance point of view, these glaciers are not as relevant as marine-terminating glaciers due to their much lower velocity, discharge and also variability. This is why they were not the focus in previous datasets and automated extraction methods. I truly appreciate the inclusion of these land-terminating glaciers, but I would frame them as an added benefit of this product rather than a main selling point (looking here at the first sentence in the following paragraph). In fact, I think the introduction could read better without this paragraph (L90 to L109).

I think Table 1 would be more useful if the third column, "temporal resolution," were removed and replaced with calving front count. It might also make sense to include a column indicating the type of sensor used (optical, SAR, or both) to show whether the given dataset had winter coverage.

Please check the citations again. There still seem to be inconsistencies. For example, why is Otosaka et al. (2023) cited for the statement about glacier terminus positions (which is not part of their paper), yet not cited for the statement about mass loss at the beginning? There are also formatting issues, such as missing spaces.

Also, see the specific comments below.

*Validation*
The additions made regarding validation are highly appreciated and greatly improve the utility of this data product. However, there appear to be inconsistencies in the data supplement and in the corresponding Figure 3.

The median AMD between GrTPD and TermPicks/AutoTerm differs immensely in some cases. For GID 12 (Zachariae Isstrom), the median AutoTerm AMD is 940342 km. Many of the AutoTerm extractions for this glacier are incorrect; still, the values are far too high, resulting in Figure 3(d) showing little else.

Examining the data file provided as a supplement reveals more of these inconsistencies. For example, the mean AMD calculated for Rink Glacier (GID1) in regard to AutoTerm is 183 km with a median AMD of 75 m. How is this possible?

Differences between GrTPD and TermPicks for GID 89 are in the range of kilometers according to Figure 3b (data for GID 89 is missing in the file, please check), but the calving fronts seem quite close when examined in QGIS.

It seems to me that there is a problem with the calculation of this AMD, resulting in some unrealistically high outliers. I really appreciate the efforts here, but these results need to be double-checked and revised.


*Examples images (Fig. 4-8)*
In L419 the authors state that “ the temporally dense sampling illustrated in selected examples does not represent the sampling characteristics of the entire dataset”. This raises the question for me of what the utility of these figures is at all. This is particularly relevant since they take up a lot of space in the manuscript. Perhaps it would make sense to only display one or two (actually representative) time series and include the rest in the supplement. The same applies to Figure 10, which does not seem to add much.


Specific comments:

L16: Calving fronts are definitely a key indicator of ice-ocean interaction. I wouldn't include atmosphere here.
L18: To my understanding calving fronts do not play a critical role for quantifying dynamic mass loss. Discharge is typically calculated using upstream flux gates.
L22: Akronym needs introduction.
L24: What does seasonally targeted mean?
L29: What is the office name? I can't find it in the files.
L42: Consider using more concise language, such as adding IMBIE numbers for ice mass loss or rankings for sea level contribution.
L66: I recommend removing this because this dataset does not address the issue (nor does it have to).
L73: "Strong potential" seems like an understatement when you consider that ML is used to derive the largest datasets for Greenland, Antarctica, and Svalbard.
L79: What is performance transferability?
L83:Does "for example" refer to the previous sentence? If so, it does not fit very well.
L86: The increase in extracted calving fronts between CALFIN and AutoTerm is mostly due to the mult-sensor approach of AutoTerm (whereas CALFIN is only using Landsat).
L90: I don't agree with this argument. Automated methods did not focus on marine-terminating glaciers because no other data was available. In fact, many studies created their own reference data. The focus on marine-terminating glaciers was primarily driven by their high variability and importance in ice dynamic mass loss.
L100: Is this accelerated in frontal retreat reported on in all three references?

Figure 1: Why are there glaciers outside of the ice sheet mask that are not indicated as peripheral?

L119: Is there a reference for the NunaGIS geographic map?
L125: Peripheral glaciers can be marine-, land-, or lake-terminating. Lake-terminating glaciers are a type of marine-terminating glacier.
peripheral glaciers can also be …

Table 2: The primary data provider for the Sentinel missions is Copernicus, not ESA.
Also, see the suggested citation guidelines from the USGS, ESA, NASA, and CDSE websites.

L177: "Over monthly timescales, terminus migration at these glaciers is generally gradual and approximately linear". Please add a reference for this.
L181: Reference for arctic DEM.
L185: Why is EPSG:4326 used instead of a polar stereographic projection, such as EPSG:3413?
L243: I disagree. The validation in this paper also shows this. The maximum accuracy (i.e. the best possible manual delineation) may correspond to the resolution of the source image.
L264: Since AutoTerm was trained using TermPicks data, I would remove the “independent”.
L303: greenland-wide?

L512: Once again, I don't see calving fronts as essential for quantifying ice mass loss. The three methods we are currently using do not critically depend on calving fronts.
L528: Are the centerlines included in the data? I might have overlooked them.

Dear Dr. Lu and co-authors,

Thank you for the thorough revision of your manuscript and for your response to the reviewers.
I am happy to accept the manuscript for publication.

I have a few minor comments regarding your response to Erik Loebel’s comments. Your manuscript will automatically proceed to production once you have addressed these points.

Yours sincerely,
Baptiste Vandecrux

-----

- Erik Loebel, in the section “Example images (Fig. 4–8)”, discusses the use of glaciers for which the “sampling density is not representative of the entire dataset”.
I understand that you want to showcase those glaciers. However, it is quite surprising for the reader to see figures that are “not representative” of the dataset. Instead, you could make a more quantitative statement such as: “Note that these well-studied glaciers with good data availability have a temporal resolution of X termini per year, higher than our dataset’s average of Y termini per year across all glaciers.”

- Later, in response to the comment: “4. L24: What does seasonally targeted mean?”, you reply:
“Here, ‘seasonally targeted’ refers to the goal of delineating glacier terminus positions at seasonal resolution.”
Here, and whenever you mention “seasonal resolution”, please clarify what you mean by seasonal: at least twice per year, or four times per year (spring, summer, autumn, winter)?