Steiner et al. present a community-led effort to identify and categorise lake-terminating glaciers globally, which is compatible with the Randolph Glacier Inventory version 7.0 (RGI7) and intended for future integration. The dataset has been primarily generated manually through a concerted and coordinated effort from the authors. Initial categorisations have been formed from existing global and regional glacial lake inventories, drawing upon and uniting a large portion of the glacial lake mapping and monitoring efforts. Error is estimated based on comparing classifications from two operators, which reveals low mapping conflicts (6.7 %) that is indicative of a low uncertainty in the dataset.

The dataset itself is logical and clear to a large extent, as reflected in the dataset description manuscript. The dataset description paper is well written and a thorough companion to the dataset. My feedback is mainly on the dataset itself and the accompanying Github repository, with my primary focus being to ensure that the dataset is unambiguous to users in the glaciology/cryosphere research community and beyond. Github repository pull requests and issues corresponding to some of this feedback have been included here, and I have added my comments early in the review process so that a dialogue can continue on the Github repository if needed.

In all, I would recommend acceptance after these revisions. I am looking forward to seeing this dataset integrated with RGI7. Great work!

Dataset comments

1. The naming of the lake-terminating glacier classifications

In the dataset and throughout the manuscript, the classifications to describe the relevance of lake presence are referred to as "lake level", "lake_level" and "lake level assessment". The term "lake level" is often used in reference to the water level of a lake, for example in remote sensing (e.g. Veh et al., 2025), modelling (e.g. Steffen et al., 2022), and in situ studies and monitoring efforts (e.g. Camassa et al., 2023).

I recommend that the naming convention is changed to something more suitable and less ambiguous. My suggestion would be "lake category" ("lake_catgy" in the gpkg field name), with Level 0-3 renamed to Category 0-3 (and capitalised throughout the manuscript).

2. Level 0 to Level 3 categories are not in sequential order 

The Level 0 to Level 3 relevance classifications are not in sequential order despite their numbering convention. Specifically, no lake contact (Level 0) is followed by > 50 % lake contact (Level 1), then < 50 % lake contact (Level 2), and then < 10 % and/or ambiguous lake contact (Level 3). 

The classification levels should follow the magnitude of relevance sequentially, therefore my suggestion is:

- Category 0: no lake contact

- Category 1: < 10 % and/or ambiguous lake contact

- Category 2: < 50 % lake contact

- Category 3: > 50 % lake contact

Where "Level" is replaced by "Category" in accordance with the recommendation above. The dataset, processing scripts, manuscript, repository readme, and statistical analysis should also be updated accordingly.

3. Ambiguous relevance classifications

The definitions of the relevance classifications (Level 0 to Level 3) differ between the ESSD manuscript and the Github repository readme, where the repo readme explicitly describes the relevance to glacier behaviour whereas the ESSD manuscript merely infers this. I would suggest amending the Github repository readme to align with the ESSD definitions, given that it is problematic to define an explicit connection between glacier-lake coverage and the certainty/amplitude of its impact on glacier behaviour. I have made a PR with these proposed changes: https://github.com/GLIMS-RGI/lake_terminating/pull/12. 

Additionally, there appears to be ambiguity surrounding the criteria for each relevance classification. At various points in the manuscript, the relevance classification signifies:

i) The portion of terminus in contact with lake (e.g. Line 70-72). 

ii) The perceived level of influence on the adjacent glacier based on visual indicators (e.g. Line 77-78) 

iii) The operator certainty of the classification/ice contact (e.g. Line 76-77, 113-115) 

Therefore, the relevance classification is ambiguous as it indicates more than one criteria. In future iterations of this dataset, I propose that criteria i) and ii) should be separated from iii), with a new field denoting the operator certainty. In addition, the criteria for the relevance classification should be revised and clarified in the manuscript (Line 63-84).

4. Dataset directory naming/structuring conventions

It is difficult to locate the dataset itself in the Github/Zenodo repository alongside the data handling scripts and documentation. I propose renaming the directory from "tables" to "dataset" in order to make this clearer, and ensuring that only the finalised dataset is in the top level of the "dataset" directory (i.e. moving all un-collated operator definitions to a sub-directory). I have opened a pull request to the Github repository (https://github.com/GLIMS-RGI/lake_terminating/pull/11) with these proposed changes.

5. The Greenland periphery glacier outlines .gpkg file is missing from dataset

https://github.com/GLIMS-RGI/lake_terminating/issues/10

Line-by-line paper comments

I don't have many line-by-line comments, largely because the language and communication of the findings presented in the manuscript are to an excellent standard. Therefore, my line-by-line comments are largely remarks, questions and figure/table queries.

Line 19-20: I am unsure about the general statement that calving "remains poorly constrained with scattered observations", especially given that two of the three references to support this are over 15 years old. Can the statement be amended to better reflect the advances in calving modelling and integration into system models over recent years.

Line 28-31: I did not realise that these two global inventories differed so greatly, therefore it is good to see this reported here. Do you know why the difference is so vast? Is this a reflection of the difference in classification approaches and/or discrepancies in manual intervention?

Line 49: Great to see the processing script openly provided for this.

Line 58: Was there any specific reason for choosing a lake size threshold of 0.01 sq km? Was this problematic in cases where an existing inventory only contained lakes with a higher minimum size threshold (e.g. Greenland, with a minimum size of 0.05 sq km)?

Line 79: Repetition of "these".

Table 2: I would like to see the entry type (i.e. string, integer, float) for each of these fields, mainly to guide users who are importing these using R or Python. Also, a short description for each field should be added, similar to those described in Lines 124-128.

Line 123-127: The AutoTerm field is not defined here, in Table 2, or in the Github repository. I am guessing this is a categorisation of the level to which an external glacial lake inventory dataset was used?

Line 149: "...(Table 3,Figure 5 and 6)." >> "...(Table 3, Figure 5 and 6)."

Table 3: I think the region name should be included here, if possible, rather than having to refer to Table 1.

Line 194-216: An additional table would neatly summarise and compliment these findings (i.e. "Table 4. Statistics from independent flag submissions of glacier classifications")

Line 218-220: Are these discussions openly available, for instance through issue postings on the Github repository? I think this could be a great approach to open, transparent discussion and resolutions in future iterations of this dataset. If you would like to use the repository as a user contribution portal then I would recommend: 1) adding a section to the readme on how to contribute; 2) adding an issue template to guide users in writing their contributions; and 3) adding a repo action to check the compatibility of user contributions (e.g. ensuring the submission is a .csv with all essential fields included). 

Line 228: The .gpkg information should be included when describing the format and contents of the .csv tables (Line 123-127), rather than at the end of the manuscript. In addition, the geographic projection (OGC:CRS84) and field descriptions (i.e. fid, IDs, aut_trm, lak_lvl, image_d, imag_dt, invntr_, cntrbtr, notes) of the gpkg files should also be included. Perhaps the field description names in the .gpkg files could be incorporated into Table 2.

Line 254-256: Normally ESSD publications require a section, or some comment, and how this dataset could be used in future work. I think a couple of comments could easily be added to the Conclusions here, tying back to the relevant literature highlighted in the introduction.

References

Camassa, R. et al. (2023) Extreme seasonal water-level changes and hydraulic modeling of deep, high-altitude, glacial-carved, Himalayan lakes. Sci Rep 13, 11705. https://doi.org/10.1038/s41598-023-37667-z

Steffen, T. et al. (2022) Volume, evolution, and sedimentation of future glacier lakes in Switzerland over the 21st century, Earth Surf. Dynam., 10, 723–741, https://doi.org/10.5194/esurf-10-723-2022

Veh, G. et al. (2025) Progressively smaller glacier lake outburst floods despite worldwide growth in lake area. Nat Water 3, 271–283. https://doi.org/10.1038/s44221-025-00388-w

This manuscript describes an approach to assign a “lake-terminating relevance level” to all glaciers in the Randolph Glacier Inventory (RGI). The study mainly uses published proglacial lake inventories to search for glaciers that are likely to be lake-terminating. The authors then use expert judgement to assign individual glaciers a level (between 0 and 3). I think that this is a great idea that will be a valuable addition to the RGI. Generally, I found the manuscript to be clearly presented. However, I thought that the execution could be improved. My main concern is that the requirements of the categories seem be used selectively (e.g. category definitions are also not strictly adhered to, reliance of single satellite images to infer “visual impact” of lake). Although the authors do a good job of showing general agreement between multiple experts, I think that this source of ambiguity will make the dataset difficult to update for another time period. Overall, I believe that this dataset will definitely be a useful contribution to ESSD but I encourage the authors to revisit the definitions and implementation of their categorization scheme. 

Major comment

There is a mismatch in the category definitions and the implementation of the classification. For example, there is a strict requirement that the ice-lake interface must be >50% (Level 1), <50% and >10% (Level 2), and <10% (Level 3). But then there are also requirements for the lakes to have “visible impact” on the glacier. The two definitions appear to be used selectively. In L91-93 a glacier that has “some ice-marginal lakes are not in the terminus zone but show visible influence on glacier dynamics at scale” is just classified as Level 3 regardless of the size of the ice-lake interface. At L137-138, we are told that Level 3 represents “only some adjacent water bodies, but without significant interaction with the glacier ice”. Level 3 has a, somewhat specific, requirement for <10% ice-lake interface which seems to have been discarded. Finally, experts are asked to classify a glacier as Level 2 if it has a stream “cutting across” its terminus that also has “considerable impact on ice melt”. But I’m skeptical that anyone could determine the impact of stream on ice melt using a single satellite image.  

In my opinion, the classification relies too much on subjective judgement calls and does not incorporate ancillary data (such as DEM, multitemporal images) adequately. This is a problem because it will introduce uncertainties if this dataset is updated to another year. But I understand that this is the approach that was chosen. One suggested fix is to revisit some of the definitions and consider making it completely subjective (i.e. larger visible impact, some visible impact, no visible impact, no lake). This would involve removing the >50%, 10% thresholds which are already a little loose given that the experts don’t have access the actual length of the ice-lake boundary or the length of the terminus. Regardless of their decision, I would encourage the authors to tighten up the text a bit (also see specific comments) and be a little more transparent about the subjectivity so that others can understand the strengths and weaknesses of the classification approach.

Specific comments

L12: Could be a little more specific about “contribute to glacier velocity” to match the directional intent of the second statement in the sentence?

L15-16: I’m not sure cherry-picking one study that modeled one glacier in New Zealand is very compelling evidence for this statement. I don’t doubt that this is true but are there not observations from a sample of glaciers that could better support this claim? 

L19-22: It would be helpful to provide some more evidence for the statement about calving (L19). I see that the next sentence mentions 24 Gt a-1 from Patagonia but frontal ablation could all be submarine melt. It would also be useful to provide some context for magnitude of 24 Gt a-1 relative to total mass loss or something.

L20: What is meant by “scattered”

L24: Consider summarizing this paragraph with a sentence about the importance of lake-terminating glaciers.

L26: Not sure what “regional assessments and case studies” is referring to here. Contribution of lake-terminating glaciers to mass loss? Mass loss from frontal ablation at the ice-lake interface? Underestimation of mass loss from lake-terminating glaciers?

L28-29: This statement is at odds to the first sentence of the paragraph.

L28-31: The difference implies that 1) one study has large errors, 2) both studies have large errors, or 3) there was enormous growth in lake numbers and volume between 2018. Given that (3) is unlikely, I would encourage the authors to just come out and say that they suspect errors in these datasets, perhaps commenting on some possible causes.  

L31: “Both” instead of “All”?

L35-36: OK so some of the glaciers do have this information? I think the authors should describe the number of glaciers or regions which have (or don’t have) this information to more clearly motivate this study.

L38-39: Recommend adding some more background to this paragraph. For example, why was 2000 chosen as the target year? How many lake-terminating glaciers were identified? How does identifying lake-terminating glaciers improve the RGI?

L57: How were lakes <0.01 km2 manually identified?

L75-76: Poorly worded sentence

L88-89: Not sure how this definition is different to the “lowest end” defined by Cogley et al. (2011)? How was this achieved without the use of a DEM?

L91-94: But surely if the ice-lake interface is >10% then it should be classified as Level 2?

L116-119: It’s not clear how prior labelling reduces future subjectivity. Future efforts will have to use just as many subjective judgement calls.

L122: “following a simple structure” of what?

L130-131: So some glaciers in RGI7 are already classified as lake-terminating? If this is so, then that should be outlined in the introduction e.g. for which regions, how many glaciers etc.

L137-138: The definitions seem to have been discarded if Level 3 now represents “only some adjacent water bodies, but without significant interaction with the glacier ice”. Level 3 has a, somewhat specific, requirement for <10% ice-lake interface which now seems to be largely ignored.

L145-146: Not to be expected for an ESSD paper. Consider removing this sentence.

L168-170: Are there glaciers in the Canadian Shield?

L168-179: This is all pure speculation that only distracts from the main point of the manuscript (the dataset). Consider removing.

L183-187: The differences between the three datasets are important and should be explored further. Would be useful to make a figure showing some side-by-side comparisons.

L186: Which results? What is meant by “relative importance”? Submarine melt? Ice flow? Mass loss?

L189-190: Long, wordy sentence, consider revising.

L190-191: Explain whether this is good or bad.

L204-205: This begs the question: why didn’t the authors use the same images as the ones used for the glacier outlines?

L242: I’ve made a similar point before but I’m struggling with the claim that the dynamics of all of these glaciers (Level 1 and 2) are significantly altered by lakes. A single multispectral image cannot provide that much information about dynamics.

L250: If Level 1 and 2 are collectively termed “lake-terminating” and Level 0 and 3 are “land-terminating”, then this reduces the need for four categories. One solution would be to use a smaller number of categories.

L255-256: The number of lakes does not necessarily imply that they have an important role. Consider adding some citations to studies that have demonstrated this. 

L259-260: I found this to be a little disappointing. The authors make a big deal about previous studies not determining whether lakes were in direct contact with glaciers (180-187). They then explicitly use “direct contact” as a requirement in their categorization. But only now do we find out that the analysis does not “does not consider the actual number of lakes in contact with an individual glacier or the length of the ice-water interface”. I think this should have been one of the primary goals of the present analysis. 

This study presents a global inventory of lake-terminating glaciers based on the RGI 7.0 glacier inventory and published datasets of glacial lakes. The key approach involves determining the relationship between glaciers and glacial lakes (direct contact or not) in different cases, considering the ambiguity in the published datasets, which map glaciers/lakes based on images with varying time stamps that may differ by several years or over a decade. The authors adopted a manual interpretation strategy, with clear guidance on different categories of glacier-lake relationships, each implying a different level of confidence on the impact of lake water on glacier dynamics. The manuscript is well-written and logically clear, except for a few instances of inappropriate naming (e.g., lake-level) and ambiguous implementation of different classifications. I believe that this dataset with consistent quality control benefits future investigation and modelling of glacier dynamics (velocity changes, mass balance) in response to climate warming.

Major comments

My main concern lies in the ambiguous definition of lake-terminating glaciers and how the current classification represents in the temporal span. If the lake-terminating glaciers refer to glacier terminuses that are in direct contact with glacial lake water, then the manual inspection is to rule out glaciers with supraglacial lakes and glacial lakes surrounding glaciers in the lateral zones. However, Figure 2d (lake-level 2) appears to include glaciers that are not directly in contact with the glacier terminus. It seems that lake-level 3 are not regarded as lake-terminating glaciers in the results and analysis (Line 135-137). In addition, it is also interesting to see how the data products of this study differ from a simple classification method, e.g., flagging lake-terminating glaciers by simply overlapping the buffering of glacier outlines with lake outlines. Such a comparison can help understand how the expert inspections implemented in this study improve the classifications. 

A second concern is about the stability of the current classification of lake-terminating glaciers, given the dynamics of glaciers and glacial lakes, particularly in certain regions such as High Mountain Asia, which show high increases in the number and area of glacial lakes in recent decades (Shugar et al., 2020; Zhang et al., 2024). I understand the alignment of the reference year 2000 for consistency; however, the ‘imperfect’ nature of current glacier and lake mapping, e.g., time consistency and resolution, may imply that such a classification needs to be representative over a time span, e.g., 1999-2005. This is also necessary for future studies to examine the different behaviors of lake-terminating glaciers over a relatively long time span (otherwise, we would need to determine the glacier type case by case, e.g., when the glacier becomes connected with lake water and when it detaches from lakes). From this perspective, I support the inclusion of some cases (e.g., Figure 3a) as lake-terminating glaciers, or by referencing lake outlines/images from nearby years. At least, the authors need to discuss the potential applications and the cautions need to be paid with the current classifications.

Specific comments

I have very few additional specific comments, as the manuscript is well-written, and some details have been pointed out in other comments.

Title: A title such as “A global inventory of lake-terminating glaciers” may be more appropriate as the work is essentially a classification of glacier types based on existing mapping results.

Line 132-133：marine-terminating glaciers are given a term_type value of 1 or 3? It is mentioned that they are assigned as type 0 in Lines 106-107.

Line 262-264: This implies that we could rely on the current classifications for assessing the long-term glacier mass balance of lake-terminating glaciers and others. However, the direct contact of glacial lake water with the glacier terminus can change rapidly, given the strong glacier terminus retreat and expansion of glacial lakes, which have been widely reported in the mountainous glacierized regions in the past decades. A single reference year (e.g., 2000) of the classification does not seem to be sufficient.