Review of Arndt and Fricker (2024) -A Framework for Automated Supraglacial Lake Detection and Depth Retrieval in ICESat-2 Photon Data Across the Greenland and Antarctic Ice Sheets 

This paper presents two algorithms that together allow the retrieval of supraglacial lake depths using ICESat-2. This method is scalable beyond the case studies presented and therefore a useful contribution to the community that I would strongly recommend for publication. 

Understanding surface melt, and lake depths is key for predicting ice shelf stability, and datasets of lake depths are limited. In situ data is scarce and remote sensing methods are necessary for providing validation and calibration for models as well as understanding the development of these lakes so this work will clearly be of use to the community. 

While there are methods existing to determine lake depths remotely e.g. using Landsat-8, this to my knowledge is one of only a small number using ICESat-2, the advantage of this particular method appears to be scalability (although see my comments below on this).  

The methodology and results presented are in my opinion sufficient for publication, and my recommendations for changes to the manuscript are mostly minor. 

General comments: 

It would be good to clarify the limitations of the algorithm e.g. max/min lake widths/ lengths/ depths detected clearly early on in the paper (and state how this compares to other methods). 

This method appears to only be scalable if you live in the US based on the information in the paper. Some comments on this would be useful e.g. can your methodology transfer to a reader’s local (i.e. institutional) supercomputer or does it need to be a National level facility? How possible would it be to do this?  

Detailed comments: 

Line 44: It would be good to have a more detailed comparison with the Datta and Wouters and Leeuwen methodologies to explain what the differences are here. Given the method presented here is promoted as scalable would it be possible to do a direct comparison to their results? 

Line 57: Most of the Greenland ablation zone rather than most of Greenland? 

Line 60: There are enough notable examples of melt away from the grounding zone (over the ice shelves of Larsen B, George VI) or on grounded ice (e.g. Corr et al. found more than a quarter or meltwater features on grounded ice https://essd.copernicus.org/articles/14/209/2022/) that maybe ‘mostly’ could be changed? 

Line 73: Acceleration isn’t always the case. Some of these references are fairly old, Davison et al. have a nice review (https://doi.org/10.3389/feart.2019.00010), although there may be updated literature. 

Line 75: I’m not sure this is strictly true it’s been observed, more than it’s been suggested as a possible mechanism. 

Line 97: If we’re providing estimates we can also model this (e.g. https://tc.copernicus.org/articles/12/3565/2018/tc-12-3565-2018.html for individual lakes) but remote sensing is quicker/ more scalable but it’s not technically true we have to rely on it. 

Line 152: Is it a good assumption that lakes are non-turbid? Certainly for sea ice ponds we can assume they are turbulent (of course they are shallower) but I wonder if e.g. it’s a very windy day when the measurements are taken how much this impacts the flatness. 

Line 207: How were these numbers determined (empirical observation is a little broad e.g. how many lakes were examined?) 

Line 259 onwards: What if the lake is e.g. 0.92m deep, would the bathymetric signal still get picked up? I think this is what you are saying in line 273 but you could clarify how you might discern the two, or if it is likely to be possible. 

Line 314: Can you determine why there is no signal from the lake bed here? Is it in the ice covered lake areas? 

Line 319: Not sure if the word ‘each’ here is a typo? 

Line 321: See my general comment about lake sizes, what does a ‘few’ mean here, please be more specific. Is this related to the 10 in equation 3? (Or if not where does that come from?) 

Line 350: I’m not sure what you mean here by ‘removing any lakes that fully overlap with another lake’. Does that not just make them the same lake (and is some information about that lake then lost)? 

Line 405: So is this a limitation on minimum lake depths that can be detected? 

Line 435: What is the situation where this could happen? Does this suggest a lack of confidence in surface retrievals or is the algorithm picking up something else (an ice lens?). 

Section 4.1.2: Could you compare with the Moussavi et al dataset? That goes up to 2020 according to the data description: https://www.usap-dc.org/view/dataset/601401 

Section 4.2.2: How many lakes are there actually in track 81? Is this something that could be determined manually to get an idea of false positives/ negatives from SuRFF? 

Line 630: ‘this’ satellite rather than ‘a’ satellite. Datasets exist for other satellites. 

Line 740: What were these equations based on? I understand you used trial and error but they are complex equations that must have had a starting point.