Review of "High-Latitude Eddy Statistics from SWOT assessed by in situ observations" by de Marez et al.

The authors use ungridded, along-track satellite altimetry data from SWOT to detect mesoscale eddies in the Labrador Sea and compare individual eddies and eddy statistics compiled over a full year to eddies detected from in-situ, shipboard ADCP measurements, as well as gridded, lower resolution altimetry data. The study confirms existing knowledge about various types of mesoscale eddies in the Labrador Sea but a longer study period would be necessary to really assess this with confidence. However, the main advancement of the study is the methodolgy of detecting eddies from along-track satellite altimetry, allowing to detect smaller features than from gridded altimetry products. Both, this new detection method and its application to SWOT data make this study a significant contribution to our knowledge of mesoscale eddies.

The manuscript is well structured and written and I only have a few, relatively minor comments.

General comments

1. In the introduction as well as the discussion, the authors highlight SWOT's potential ability to investigate eddies in "ice-influenced" regions or "seasonally ice covered zones where altimetric data were previously unusable". However, their analysis excludes eddies in areas with >15% sea ice cover, so that it seems that this aforementioned potential is purely hypothetic. I do not see the value in emphasizing this unless the authors show how a seasonal ice cover can inhibit eddy detection in lower resolution altmitry data even at times when no ice is present.
2. The method of filling using "biharmonic inpainting" seems to be uncertain to me. It is not well explained in the text but looking at Fig. 1 this "filling" is an extrapolation over a distance of ~50% of the original width of the satellite swath. I assume that this introduces a lot of uncertainty even though only eddies that are mostly outside of those extrapolated regions are considered. I strongly suggest the authors do some sensitivty testing with different extrapolation methods, or present results if testing was already done. Even better would be to take a dummy dataset, i.e. a gridded SSH dataset or model output, subsample some artificial SWOT swaths and then detect eddies in those and compare to an eddy detection performed on the original complete, gridded dataset.

Specific Comments

Title: This might be irrelevant as I am not a native english speaker but "assessed by in situ observations" seems to not convey what the authors want to say. I suggest replacing by "compared to in situ observations" or something similar.

l. 2: I suggest adding "in the ocean" after "momentum" in the introductory sentence.

ll. 19-20: I find this slightly misleading as the automatic detection of eddies can be performed on mooring or ADCP data etc. What has been made possible through the gridded products is global detection of eddies.

ll. 23-24: I suggeset rephrasing to "The ability of gridded SLA fields to represent the eddy field has previously been questioned, as they have largely distorted eddy characteristics" unless this distorts the intended meaning.

ll. 33-41: The authors mention two environments where eddies are abundant, they then go on to describe the processes in the MIZ in detail but do not describe anything about the boundary currents. I strongly suggest to add some information on what the eddies do in boundary currents. Also see my General comment 1.

ll. 69-75: It is not clear to me how the reduction of overall energy does not have implications for the presented analysis. I do believe this is true but the authors should rephrase this part and include an explanation of why it is true.

l. 99: If possible, include those datasets in the references.

ll. 120-128: The explanation of the padding and filling is not clear to me from the text alone. I suggest the authors rephrase this in order to avoid that the readers have to switch back between the text and Fig. 1 to understand the process.

ll. 123-128: See General comment 2.

l. 168: Why was the SWOT-detected eddy from June 27 used while the one from June 25 would be closer in time to the ins situ-detected eddy on June 24? Is there an automatic algorithm that decides which of the duplicate detections to keep or is this chosen manually? I suggest the authors describe this process.

ll. 182-183: Are the "shallow eddies" here the ones that are detected in areas where the water depth is < 2000 m? If this is the case I suggest the authors rephrase this, as "shallow eddies detected above depth < 2000 m" gives the impression that the eddies themselves are above 2000 m in the water column.

l. 188: Global datasets of mesoscale eddies (and their statistics) are available from lower resolution altimetry data, e.g. Ioannou et al. (2024), extending to similarly high latitudes. While the higher resolution of SWOT certainly makes the presented statisitcs more valuable, it is not the first time that those statistics are derived. 

ll. 198-199: I suggest the authors add a reference that supports their statement that "AEs are less susceptible to steering by background currents and less prone to instability-driven decay".

ll. 204-205: The fact that the eddies are stronger near the slopes might also reflect the fact that those eddies are generated in those regions and are necessarily weaker away from them as they decay.

Fig. 4: I suggest the authors use a plotting algorithm without interpolation such that the 2x1 degree boxes are visually identifiable on the maps.

ll. 209-220: I suspect that the western and eastern regions got mixed up here as the West Greenland current is not in the western region and the eastern region is not the one with the weakest eddies. I suggest the authors check this and also make sure that the western and eastern region are correctly attributed to the columns in Fig. 5.

ll. 214-215: Does the average westward propagation speed of Irminger Rings support the connection between a winter maximum of their generation in the east and the presented spring-autumn maximum in the central region?

l. 227: I suggest removing "(surface-intensified)" as not all the types of eddies mentioned are surface-intensified (convective lenses).

l. 250: Not all datasets used in this study are mentioned here. If the code for the detection from SWOT data is available this should be added here as it is certainly of great interest to the community.

References: In general, I suggest the authors add DOIs to all references.

Minor Comments 

l. 25 add a comma after "to be resolved, ..."

l. 34 "sea-ice melt generates"

l. 68: "preserving the balanced"

l. 86: is a duplicate of line 84

l. 99: "Other datasets"

l. 106: some formatting issue with "no—KaRIn—measurements"

l. 115: "MIOST" has not been introduced

l. 149: I suggest replacing "in front of" with "using"

l. 160: "tracks"

l. 232: "detection"

l. 284: Use the final published version.

l. 341: citation has no journal

l. 346: Use the final published version.

References

Ioannou, A., Guez, L., Laxenaire, R., & Speich, S. (2024). Global Assessment of Mesoscale Eddies with TOEddies: Comparison Between Multiple Datasets and Colocation with In Situ Measurements. Remote Sensing, 16(22), 4336. https://doi.org/10.3390/rs16224336

Eddies are smaller in high latitudes due to the change of the deformation radius. The eddies in these latitudes are, however, important influencing deep-water formation, carbon uptake, and sea-ice melt. The manuscript shows that the new SWOT satellite enables robust, quantitative eddy detection in polar and subpolar oceans. I find the manuscript scientifically sound and well written. Only a few minor correction seems needed before publication.

(relatively major) comments

In Section 3.1 (methodology), the geostrophic balance is implicitly assumed. In scales much smaller than the deformation radius, it is not obvious that the velocity field is dominated by geostrophic flows. A short discussion to justify the geostrophic balance at these scales might be needed.

L.198, AE/CE asymmetry is mentioned here. Fig.3 of Chelton et al. (2011, https://doi.org/10.1016/j.pocean.2011.01.002) indicates that the asymmetry shows up after travelling > 2000 km. Is this the situation in the Labrador Sea (not obvious from Fig.4)? Or do you have any better reference? Later at L.211, it seems a local generation along the W.Greenland Current is suggested. Does this agree with the AE/CE asymmetry?

minor comments

L.95 "from classical a" → from a classical 

L.177 and Fig.3e, How is the Rossby number defined here?

Fig.2 caption, 3 lines from the bottom. The subscript 1 and 4 are connected by a strange character.

L.250, the website shown at "data availability" section is not a link to the data. Maybe this one? https://doi.org/10.24400/527896/A01-2023.017

General assessment

This manuscript presents a study using SWOT swath data to characterize mesoscale eddies in the Labrador Sea. The use of native-resolution SWOT observations combined with in situ SADCP data provides a promising framework to investigate mesoscale dynamics at high latitudes, where conventional altimetry has long been limited.

However, several aspects of the methodology, validation strategy, and interpretation of the results require clarification and strengthening.

Major comments

1. Detection on SLA field

It has been previously assessed that the detection should preferably be done on ADT field rather than SLA field as in Pegliasco, C., Chaigneau, A., Morrow, R., & Dumas, F. (2021). Detection and tracking of mesoscale eddies in the Mediterranean Sea: A comparison between the Sea Level Anomaly and the Absolute Dynamic Topography fields. Advances in Space Research, 68(2), 401-419. Most recent global eddies atlas perform the detection on the ADT field as in ToEddies (Ioannou, A., Guez, L., Laxenaire, R., & Speich, S. (2024). Global assessment of mesoscale eddies with TOEddies: comparison between multiple datasets and colocation with in situ measurements. Remote Sensing, 16(22), 4336.) or META atlases (Gamot, J., Delepoulle, A., Nencioli, F., Pujol, M. I., & Dibarboure, G. (2026). META4. 0: a new mesoscale eddy network atlas derived from altimetry. Earth System Science Data Discussions, 2026, 1-31). This can, for example, explain the AE/CE asymmetry.

I strongly recommend performing the detection on the ADT field as prescribed by previous studies.

2. Choice of reference altimetry product

The comparison with mapped altimetry relies on the CMEMS 1/4° product (as performed in Bendinger et al., 2025). However, more recent higher-resolution products (e.g., DUACS DT2024 1/8°) are available for this region. Given that eddy detectability is strongly resolution-dependent, this choice may artificially enhance the contrast between SWOT and CMEMS.

In addition, eddy detection strongly depends on the algorithm and parameter settings. The detection method applied to the CMEMS dataset should therefore be clearly specified, and the results interpreted in the light of these methodological choices.

Please justify the use of the 1/4° product, specify the exact CMEMS product and detection configuration used, and discuss whether using higher-resolution products (e.g., 1/8°) would affect the conclusions. The use of the last altimetry product should be considered in order to improve the quality of the comparisons.

3. Missing key parameter: shape error

The shape error parameter in pyeddytracker is not documented. This parameter strongly controls the acceptable deviation from circularity, and the sensitivity of detection to noise and interpolation artifacts.

Please provide and justify the value used, and indicate whether sensitivity tests to this parameter were performed. This would enhance the reproducibility of the method.

4. Value of the amplitude threshold

The amplitude threshold parameter in the detection procedure is set to 1 cm, while in other and global eddy atlases it is set to 0.4 cm (META), 0.25 cm (TIAN) and 0.1 cm (ToEddies) (as reported for instance in "Ioannou, A.; Guez, L.; Laxenaire, R.; Speich, S. Global Assessment of Mesoscale Eddies with TOEddies: Comparison Between Multiple Datasets and Colocation with In Situ Measurements. Remote Sens. 2024, 16, 4336. https://doi.org/10.3390/rs16224336"). For submesoscale eddies, this amplitude threshold appears to be a bit high and may affect the detection rate.

I recommend using an amplitude threshold closer than the ones used in other atlases (1 to a few millimeters appears to be more appropriate for high-latitudes eddies).

5. Justification of tiling strategy

The choice to divide SWOT swaths into overlapping 128×128 tiles is not justified, while pyeddytracker can operate directly on full grids. This tiling strategy may introduce artificial boundary effects, and additional complexity in the merging procedure.

Please justify why detection on the full swaths was not used and discuss the potential impact of tiling on detection results.

6. Inpainting method

The choice of the biharmonic inpainting method is critical, as it directly affects the reconstructed height fields and therefore the eddy detection.

Please justify the choice of this specific method, indicate whether alternative inpainting approaches were tested and discuss its potential impact on eddy properties (e.g., amplitude, radius, detection rate).

Indeed, this gap filling process may induce many erroneous closed contours, especially at the edges of the swaths where the field is extrapolated.

The inpainting method (ideally) or the eddies concerned by this bias should be validated. At least, this should be discussed in the article.

7. Duplicate removal methodology

The duplicate removal procedure is custom-built, with several criteria, whereas pyeddytracker provides a native eddy matching algorithm based notably on an overlap score threshold. In addition, it is unclear which criteria are used to select one eddy among duplicates (both between overlapping tiles and between swaths within the same cycle), and how this choice may affect eddy characteristics.

Please clarify why the built-in matching routine was not used, and describe the selection criteria for retained eddies, and discuss the sensitivity of the results to these choices.

Moreover, Figure 1 does not clearly illustrate duplicate detections before and after merging, particularly for overlapping tiles and multiple swaths within a cycle.

Please include a concrete example showing duplicated eddies, and how they are identified and removed in the final detection.

8. Biases introduced by gap-related filtering criteria

The exclusion of eddies with more than 40–60% of their contour located in missing data regions likely introduces a systematic detection bias. Indeed, this approach tends to remove eddies crossing the nadir gap, remove large eddies spanning swaths and favor small, compact eddies fully contained within a single swath. This creates a structural asymmetry with altimetry (e.g., CMEMS) products, which tend to detect larger and smoother eddies due to their lower resolution.

Please explicitly acknowledge this bias, and assess its potential impact on the statistics presented in Figures 3, 4, and 5.

9. Validation strategy

The validation is interesting but remains largely qualitative:

• Limited sample size: only 4 eddies are used for direct comparison;
• No quantitative metrics (e.g., bias, standard deviation, RMSE, correlation…) ;
• Temporal mismatch between SWOT and ADCP observations

Please include simple quantitative metrics, discuss the uncertainty associated with temporal mismatch, and moderate statements such as “excellent agreement” in the light of these limitations.

10. Missing visual intercomparison (SWOT vs ADCP vs CMEMS)

The manuscript lacks direct visual comparisons of the same eddies across datasets.

Please include case(s) study showing matching and mismatching eddies from the several datasets.

Suggestion: As mentioned above, pyeddytracker includes built-in matching tools that allow for systematic comparisons between datasets. These tools can be used, for instance, to identify eddies detected in multiple datasets in front of those detected in only one dataset, and to compare their spatial distribution and dynamical characteristics.

Such an analysis would provide a more quantitative assessment of the similarities and differences between SWOT, CMEMS, and in situ detections, and would help better identify the respective strengths and limitations of each dataset.

11. Missing of CMEMS curves in Fig. 3 (panels a & e)

CMEMS results are discussed but not shown in Figure 3 for eddy polarity and Rossby number.

Please include these curves or explain their absence.

12. SWOT Cal/Val phase validation

The SWOT Cal/Val phase provides higher temporal sampling, which offers an opportunity to assess whether the same eddy is consistently detected across successive passes and evaluate the sensitivity of the detection to variations in SWOT sampling (e.g., swath geometry and data gaps).

Please explain why this dataset was not used for validation or for assessing detection robustness, and eventually use it to complete the validation.

13. Interpretation of seasonal results

The seasonal variability analysis is based on a single year (2024).

Please explicitly acknowledge this limitation and interpret the results accordingly.

14. Suggestion on the dataset availability

Would it be possible to share publicly the eddy detection dataset? This would allow users to perform comparison, validation and statistics with other atlases or personal data.

Minor Comments

General: The statement “first statistical description” should be nuanced. Previous studies based on in situ observations and numerical models have already provided partial statistical characterizations of eddies in the Labrador Sea.

Please clarify that the novelty of this work lies in providing a SWOT-based, high-resolution, etc statistical description, rather than the first statistical description overall.

Note that I am not an English native speaker so some of the following remarks can be irrelevant:

l.34 : generateS

l.60 : “see some background in e.g., Morrow et al., 2019” to “see, e.g., Morrow et al. (2019) for background” or something similar (for the use of e.g.)

l.71 :     “this has no implications for our analysis” to “this is not expected to significantly affect our analysis” or something similar to nuance the sentence.

l.78 : which represents

l.124 : “the latter reconstructs”

l.135 : “sensitivity tests to this parameters were done”: “sensitivity tests for these parameters were performed” ?

l.136 : WAS obtained

Figure 2 : the units in the table should be given.

Conclusions

In summary, this study is promising and addresses a relevant topic with a novel dataset. The approach has strong potential, and the results are encouraging. However, the current version of the manuscript lacks sufficient methodological justification and quantitative validation to fully support the conclusions.

I recommend major revisions, with particular attention to:

• performing the detection on the ADT field rather than the SLA;
• clarifying and justifying methodological choice;
• strengthening the validation and results;
• and moderating some interpretations in light of the current limitations.

Addressing these points would significantly improve the robustness and impact of the study.