the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 May 2026
PolarLakes: A Bi-Weekly Dataset of Supraglacial Lakes on Antarctic Ice Shelves from Multi-Sensor Satellite Observations (2015–2024)
Abstract. Supraglacial lakes in Antarctica are key indicators of surface meltwater processes and play a significant role in ice shelf stability and ice sheet dynamics. Despite their importance, the understanding of their spatiotemporal dynamics remains limited due to their high variability and the lack of consistent, large-scale monitoring approaches. To date, no bi-weekly, pan-Antarctic dataset exists that captures supraglacial lake extent over recent years. To address this gap, we introduce the PolarLakes dataset, generated using a deep learning-based workflow that automates the detection and mapping of supraglacial lakes across Antarctic ice shelves. This method integrates Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 optical imagery to enhance spatial and temporal coverage while improving detection accuracy under challenging conditions such as cloud cover and surface refreezing. The resulting open-access dataset provides bi-weekly records of supraglacial lake extents at 10 meter spatial resolution from 2015 through 2024. It reveals extensive ponding in East Antarctica until a pan-Antarctic peak in lake area in 2020, followed by higher melt extents on the Antarctic Peninsula, culminating in a secondary peak on the Antarctic Peninsula in 2023. The PolarLakes dataset offers a valuable resource for investigating supraglacial hydrology and provides essential observational constraints for hydrological and ice sheet modeling. The PolarLakes dataset is available at DLR’s GeoService at https://geoservice.dlr.de/web/datasets/polarlakes.
- Preprint
(1758 KB) - Metadata XML
-
Supplement
(174 KB) - BibTeX
- EndNote
Status: open (until 08 Aug 2026)
- RC1: 'Comment on essd-2025-569', Benjamin Davison, 30 Jun 2026 reply
-
RC2: 'Comment on essd-2025-569', Anonymous Referee #2, 06 Jul 2026
reply
Review of Baumhoer et al.: PolarLakes
Overall Baumhoer et al. present a convincing and competent study of lake extent, which a neatly organised file directory. Using a modified U-Net they advance on some previous lake-detection studies by increasing the frequency of retrieval and the overall year-range. Their goal is straightforward, and they straightforwardly achieve it through well-explained methodology and nice illustrations. Their findings have implications for future lake-sensing studies, and for reassessment of which years had the greatest lake extent in Antarctica.There are quite a few minor grammatical points and typos below (more may be hidden) but I would be happy to see this published follow these minor corrections.
Having gone through some of the data files I'm content that these show what they are meant to show, though I have of course not been able to quality check every folder. However, I am not really clear what the units in the annual mosaics represent. I looked at the George VI one and it went from 1 - 7, presumably years where it got to that extent? I don't know exacly what ESSD criteria are on this, but it would be good if the dataset was self standing, that is, could be made sense of without the attached paper. In that respect a short readme.txt with the most relevant information (such as units, projection info, etc.) could be a useful addition.
Medium points
The abstract and title make it seem as though the Ross and Ronne-Filchner ice shelves are included (well, in fact all Antarctic ice shelves). I don't think the title has to be changed, but it would be good for the abstract to include that it's rather a select subset of shelves.
Methods are presented very well, I appreciate the inclusion of the first methods paragraph.
In the discussion there is some coverage of how full data is only available from (I think) 2017 onwards. I think a short sentence or two on how to/the importance of replicating the methods could be useful. One could see a need in the future to compare say 2035 melt formation to the heady days of 2020 and keeping methodology uniform would be useful here.
Slush is omitted, when it's inclusion would of course be useful in interpreting lake extent and the slightly blurry line between the two. A bit more discussion about why this wasn't included would be nice (i.e. methodological impossibility, upcoming work, etc. ).
Minor points
Table 1. 'This study' could be added as an additional line to aid comparison. Where the dataset DOI is a dash, a brief comment could be added to say how/if the data is publicly accessible.
Table 2. Please make the scores as % or 0-1 consistent throughout the text.
Fig. 1 'identify errors in regions that were only ponded any four weeks' appears grammatically incorrect. Missing a word? Otherwise, this figure is really nice and I appreciate the file extensions at the end.
Fig. 3. As with the rest of the figures this is really nice, but the train and test sites are hard to spot. Suggest a different colour outline?
Fig. 4 Do you have a higher res version of this? It becomes quite blurry quite quickly. There is an errant grey line in the bottom right and the x-axis on b is painful to read. Otherwise really nice, though suggest decimal degrees rather than minutes and seconds for panel a. Usually two y-axis suggest different values are displayed, but both of these go 0-100 suggesting directly comparable? Unclear but if so only one necessary and the comparability info should go in the caption.
Fig. 5. I don't think the 1-1, 1-2, 1-3 notation has been previously introduced ? Also (%) technically needed as x-axis units.
Fig. 6 may be blind, but can't see any pink on this figure
Fig. 7 date axes are always hard to get right. There is scope here to simplify things by removing years from the lower x and just having months (as the years are above).
Fig. 8. I would say this is much more legible than 7, but could be made even more so by removing the year from the labels in the lower x axis, as this is included in the shading at the top, though note that your years here are not correctly aligned.
Fig. 9. Is the bi-weekly recurrence in days? Does a bi-weekly recurrence of 15 or above make sense in this context?
l1. Does EGU have any guidance on capitalised titles? I was under the impression full capitalisation was American.
l17. The pedant in me notes that if there are no bi-weekly records, there could still be weekly records (which would be better). Suggest stating what the current best is and then saying you do better.
l30. Good to specify very briefly what these variations are if possible (space, time, etc.).
l31. To *this* end?
l35. References should go year and then alphabetical, at present they are just placed at random.
l38. And at all in winter??
l43. Scope here to talk about general ice shelf fracturing in addition to full ice shelf collapse. See for example Banwell et al. (2024, https://doi.org/10.1017/jog.2024.31) for more localised and Banwell et al. (2013, https://doi.org/10.1002/2013GL057694) for chain reaction drainage.
l46. This has been quantified for Greenland, but it isn't actually so important overall (Ryan et al., 2025, https://doi.org/10.1038/s41467-025-62503-5).
l60. I would try to draw more comparisons to Greenland literature, which is more advanced in terms of lake formation. Some citations would be... TODO
l62. The above comment applies to this paragraph, too.
l65. The Scambos reference would be better suited in the paragraph above.
l70. I would just like to use the references on this line to emphasise that citation order should be sorted out throughout this paper
l71. ration
l74. TOA has not been introduced, and is only used once. I suggest just putting top of atmosphere.
l87. Would 'AnarcticLakes' not be a bit more descriptive?
l92. How, exactly? I think this is a broadly useful dataset, but I'm generally not a fan of papers saying things can help improve ice sheet modelling unless there is actually a clear and direct link.
l92. Could be rounded out to say, i.e. 'when combined with observations/models of ice shelf collapse and ice sheet velocity, PolarLakes can provide critical insights into...'.
l107. A bit of further information regarding how 'frequent surface melt' is obtained would be appreciated.
l121. Good to specify which package the Apply Orbit File function comes from. Ditto with how thermal noise is removed and with the speckle filtering.
l127. Ok I see this is introduced here, good to put this earlier.
l131. AOI = area of interest?? Only used once so suggest not using the acronym. Citation shouldn't be fully parenthesised.
l133. If we turn digital numbers into an acronym then nothing is safe from this treatment. Also only used once.
l140. Furthermore, and doesn't make sense
l150. Good to briefly say what classifies as pronounced surface melt. Also good to know what % of overall data is used as training data.
l166. A brief and parenthetical summary of these rules could be included following this sentence.
l185. suggest 10^x for these numbers for improved legibility.
l193. Generally style guides suggest putting the % directly after a number (as with deg C). Unsure which one essd follows.
l200. The Ross ice shelf is very large, could it have made sense to include a subset of it that did fulfil your requirements?
l203. Including this information was one of my previous comments. Good to foreshadow this section by saying (see Section 2.5.1 for site selection...).
l210. Great, this covers a few of my earlier comments, but as above, good to foreshadow this section, or even move it all up into methods.
l241. be-weekly
l259. %s
l260. This indicates that,
267. The pedant within me dislikes that here scores are given 0-1 and above in %. Please choose wither or.
l310. Nice
l337. more strongly
l347. Not clear what 'less significant' means in this context. If you're finding 2020 as a larger melt year than 2017 using your methodology that could be something worth mentioning in the abstract.
l356. ,,
l357. and as such these are not included
l360. Is there a reason you didn't include slush as well as lakes? May have helped with some uncertainty statements.
l391. This is a good and thorough limitation section, but foreshadowing could be peppered throughout the manuscript. For example, the issues with consistent time series data is worth mentioning earlier when the data sets are introduced.
l406 As in Dell et al.... ? and how (rather than just giving the citation)
Citation: https://doi.org/10.5194/essd-2025-569-RC2 -
RC3: 'Comment on essd-2025-569', Anonymous Referee #3, 13 Jul 2026
reply
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2025-569/essd-2025-569-RC3-supplement.pdf
Data sets
PolarLakes - Bi-weekly supraglacial lakes in Antarctica C. Baumhoer et al. https://geoservice.dlr.de/web/datasets/polarlakes
Model code and software
The PolarLakes Workflow: A Reproducible Framework for Multi-Sensor Supraglacial Lake Detection and Dataset Generation Using U-Net on Sentinel-1 and Sentinel-2 Data C. Baumhoer et al. https://zenodo.org/records/17467230
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 318 | 118 | 18 | 454 | 113 | 18 | 14 |
- HTML: 318
- PDF: 118
- XML: 18
- Total: 454
- Supplement: 113
- BibTeX: 18
- EndNote: 14
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Please see attached pdf.