the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 Mar 2026
META4.0: a new mesoscale eddy network atlas derived from altimetry
Abstract. This study introduces the new global Mesoscale Eddy Trajectory Atlases (META4.0), which provide eddy detections, trajectories, and interaction networks derived from satellite altimetry. Eddy detection relies on the pyeddytracker (PET) algorithm (Mason et al., 2014), further optimized by Pegliasco et al. (2022), and represents a substantial improvement over the previous META3.2 product (SSALTO/DUACS, distributed by AVISO+ with CNES support).
The main advance of META4.0 is the explicit identification of eddy merging and splitting events. By combining grouping, which links detections across consecutive days, and segmentation, which tracks continuity through interaction events, trajectories are organized into networks of interconnected eddies. This network-based representation complements the classical single-trajectory view of eddy life cycles by explicitly accounting for eddy interactions.
The paper presents both diagnostic tools designed to explore individual eddy networks (e.g., timelines, spatial trajectories, and eddy properties such as effective radius or shape error) and the results of a global statistical analysis over more than three decades. These analyses provide new insights into network properties, eddy lifetimes, and the spatial and temporal distribution of merging and splitting events. Clustering analyses reveal recurrent interaction patterns and identify regions where eddy networks are particularly active. Independent datasets, including sea surface temperature and chlorophyll concentration, are used for qualitative validation of selected events. Finally, Lagrangian advection of synthetic particles highlights coherent forward and backward transport signatures associated with interaction events, providing a physical validation of the reconstructed networks.
Overall, META4.0 offers a novel and physically consistent framework to characterize mesoscale eddy interactions and to better understand their role in shaping ocean dynamics.
- Preprint
(39818 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on essd-2026-108', Anonymous Referee #1, 01 Apr 2026
-
RC2: 'Comment on essd-2026-108', Anonymous Referee #2, 13 Apr 2026
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2026-108/essd-2026-108-RC2-supplement.pdf
-
CC1: 'Comment on essd-2026-108', Yikai Yang, 16 May 2026
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2026-108/essd-2026-108-CC1-supplement.pdf
-
RC3: 'Comment on essd-2026-108', Anonymous Referee #3, 04 Jun 2026
Dear Editor,
Thank you for inviting me to review this manuscript. This paper presents a new global mesoscale eddy network atlas, META4.0. Its main innovation lies in explicitly identifying eddy merging and splitting events on the basis of conventional single-trajectory eddy tracking, and in organizing related eddy trajectories into interconnected network structures. The authors present diagnostic tools for eddy networks, global statistical results, the spatiotemporal distribution of merging/splitting events, changes in eddy properties, qualitative validation with external data, and particle-advection-based coherence analysis.
Overall, this dataset has potential application value. However, the current manuscript still has several issues related to methodological definitions, parameter selection, independent validation, reproducibility information, and the interpretation of some figures and tables. In particular, the abstract is inconsistent with the validation presented in the main text; the definition of normalized overlap used for trajectory tracking is insufficiently clear; the 10% overlap threshold and the five clustering categories lack adequate objective justification; the external CHL validation is closer to a qualitative consistency check based on a small number of typical cases than to a truly independent validation of eddy boundaries; and some figures and case descriptions do not provide sufficient information for reproducibility and deeper interpretation.
For these reasons, I recommend that the manuscript undergo major revision before publication.
Major comments
1. The statements in the abstract regarding independent validation using SST and CHL are inconsistent with the main text
The abstract states that the study uses independent datasets, including sea surface temperature and chlorophyll concentration, to qualitatively validate selected events. However, in Section 3.3.5 of the main text, the external validation that is actually presented and discussed uses only CHL data, with no validation procedure or conclusions based on SST.
This creates an inconsistency between the abstract and the main text. I suggest that the authors either add validation results for typical merging/splitting events using SST data, including the data source, processing procedure, diagnostic criteria, and corresponding figures, or revise the abstract and related conclusions to avoid claiming that SST was used for qualitative validation.
2. Justify the use of 1/4° ADT data and discuss resolution sensitivity
This study uses global daily ADT data from the DT2021 reprocessing product at a spatial resolution of 1/4°. However, higher-resolution global L4 reprocessed products for sea level and derived variables are now available, such as the 1/8° × 1/8° product provided by Copernicus Marine. In addition, when discussing the spatial distribution of lonely eddies, the authors also note that some regions with a high occurrence of isolated eddies may be related to limitations in altimetric resolution.
Therefore, I suggest that the authors further justify why the 1/4° product is still used as the primary input dataset. Considering the large workload required to recompute a 30-year global atlas, the authors do not necessarily need to reconstruct the entire dataset. However, it would be useful to add a resolution-sensitivity experiment for representative regions or representative years. For example, several cases could be selected from western boundary current regions, the Southern Ocean, coastal regions, or equatorial regions, and the results obtained from the 1/4° and 1/8° products could be compared in terms of the number of detected eddies, the number of lonely eddies, the number of networks, the number of segments, and the number of merging/splitting events.
3. The area-overlap-based tracking method relies on a subjective threshold
The network grouping in this study mainly relies on the overlap between eddy polygons at adjacent time steps, with 10% set as the connection threshold. This threshold directly affects whether eddies are connected into the same network, and thus influences the number of networks, the number of segments, the number of merging/splitting events, and the resulting eddy lifetime statistics.
At present, the manuscript does not explain the rationale for choosing the 10% threshold, nor does it demonstrate the stability of the results under different threshold values. I suggest adding a threshold-sensitivity experiment.
In addition, I suggest that the authors compare the geometric-overlap tracking method used in this study with existing Lagrangian eddy tracking methods in the Discussion. The discussion could address the practicality of a geometric overlap threshold for constructing long-term global eddy networks, as well as its advantages and limitations relative to particle-advection methods in terms of physical constraints, computational cost, and parameter sensitivity. Relevant studies include Jones-Kellett A. E. and Follows M. J., “A Lagrangian coherent eddy atlas for biogeochemical applications in the North Pacific Subtropical Gyre,” Earth System Science Data, 2024, 16(3): 1475–1501; and Tian F., Zhao Y., Qin L., et al., “A Black Hole Eddy dataset of North Pacific Ocean based on satellite altimetry,” Earth System Science Data, 2025, 17(12): 7119–7145.
4. The definition of normalized overlap is insufficiently clear
The manuscript mentions normalized overlap around lines 125 and 135 and uses 10% as the network connection threshold, but it does not provide an explicit mathematical definition. It is therefore unclear how the overlap area is normalized. Different normalization schemes may substantially affect the identification of merging and splitting events.
I suggest that the authors provide a clear formula for normalized overlap before introducing this threshold, and explain how this definition affects the identification of the main eddy and secondary eddies in merging and splitting scenarios.
5. Table 1 lacks specific information needed to reproduce the representative network case
Section 3.1 introduces a representative anticyclonic network case and reports its number of observations, number of segments, number of merging/splitting events, lifetime, and longitude–latitude range. However, the manuscript does not provide the specific start and end dates of this case.
For a dataset paper, representative cases should be as reproducible as possible. I suggest that the authors add the start and end dates, as well as the specific data files corresponding to Figures 2–4. This would help readers verify the exploratory tools proposed by the authors.
6. Add local zoom-in panels for key interaction nodes in Figure 4
Figure 4 shows the temporal evolution of effective radius and shape error within the network, using marker size and color to represent eddy properties. Based on this figure, the authors interpret phenomena such as an increase in radius after merging, an enlarged radius before splitting, and elevated shape error near interaction events.
However, the current figure contains relatively dense data points, and the local changes before and after key merging and splitting events are not sufficiently intuitive. I suggest adding local zoom-in panels to Figure 4 to show the changes in eddy properties before and after typical merging and splitting events. This would provide more direct support for the interpretation that eddy interactions are accompanied by structural reorganization and morphological deformation.
7. Report network characteristics separately for cyclonic and anticyclonic eddies in Table 2
The manuscript states that network construction is performed separately for anticyclonic and cyclonic eddies. However, Table 2 only provides overall statistics for the META4.0-Networks dataset.
Since the construction procedure is polarity-specific, I suggest that Table 2 report statistics separately for anticyclonic and cyclonic eddies. These should include the number of observations, networks, segments, merging events, splitting events, and lonely eddies for each polarity, as well as the mean segment lifetime and network lifetime. This would help readers assess whether the two eddy polarities exhibit systematic differences in network complexity and interaction frequency.
8. The reason why segment lifetimes in Figure 9a are longer than META3.2 single trajectories is insufficiently explained
The central improvement of this study is that the network formalism connects multiple segments before and after merging and splitting events, thereby reducing artificial births/deaths in traditional single-trajectory methods and extending the effective lifetime of eddy systems. This conclusion is clearly reflected in Figure 9b.
However, Figure 9a shows that the lifetime distribution of META4.0 network segments is also longer than that of META3.2 single trajectories. This result is not intuitive. I suggest that the authors further explain why the segment lifetimes in Figure 9a are longer.
9. The diagnostic criteria for CHL-based external validation in Section 3.3.5 are unclear
Section 3.3.5 uses CHL fields to qualitatively validate merging and splitting detections, and Figures 18 and 19 show ADT and CHL background fields. The authors argue that these examples indicate that the merging/splitting events detected from altimetry can also be observed in an independent field.
However, the manuscript does not currently explain how CHL is used to diagnose interaction events. The eddy contours shown on the ADT and CHL panels in Figures 18 and 19 appear to be identical, suggesting that these contours are likely still eddy boundaries detected from ADT and merely overlaid on the CHL background field. If this is the case, CHL does not independently identify the same eddy boundaries, but instead provides a tracer-background response that is consistent with the ADT-based detection results.
I suggest that the authors clearly state whether all eddy contours in the figures come from ADT/META4.0, whether CHL data participate in boundary identification, or whether CHL is used only as a background tracer field. The authors should also specify the concrete criteria by which CHL is considered to support the merging/splitting detections.
In addition, the information in the corresponding network timelines in Figures 18 and 19 is rather crowded. I suggest retaining only the segments involved in the current interaction to improve readability.
Minor comments
- The two subpanels in Figure 10 are not fully aligned.
- It is reasonable to use normalized radius and normalized shape error in Figures 13 and 14 to analyze life-cycle stages before and after interactions. However, the normalized results cannot directly reflect absolute scale changes between parent eddies and resulting eddies. I suggest that the authors add corresponding statistics and figures for absolute effective radius and absolute shape error to support the statement that “merging leads to larger structures and splitting to smaller ones.”
- In the multi-panel composites of Figures 18 and 19, some subpanels are slightly misaligned, and the overall layout is not sufficiently tidy.
- In Figure 20, the colorbar labels in some subpanels are partially obscured or not fully displayed, and there are also alignment issues among different subpanels.
- French labels appear in Figure 23b and should be unified into English.
In summary, I consider the META4.0 eddy network dataset proposed in this manuscript to have scientific value and application potential. However, the current version still requires further improvement in several respects, including key methodological definitions, parameter selection, robustness of results, sufficiency of external validation, and reproducibility information for representative cases. Therefore, I recommend that the manuscript be considered for acceptance only after major revision. I hope the authors will further strengthen the transparency of the methodology, the robustness testing of parameters, and the experimental framework, thereby improving the reliability, reproducibility, and practical value of the dataset. Thank you again for inviting me to review this manuscript. I also hope that the comments above will be helpful for the authors in revising and improving the paper.
Sincerely,
Reviewer
-
RC4: 'Comment on essd-2026-108', Rémi Laxenaire, 02 Jul 2026
In this paper, the authors present META4.0, the latest version of the META database suite for eddy detection from altimetry. The novelty is that it accounts for merging and splitting of eddies to build networks of interconnected eddies, and the authors also provide a set of tools to analyze these networks. This database suite is clearly of great interest to the community, as shown for example by the around 200 citations of the reference META3.1exp paper (Pegliasco et al., 2022), and I expect META4.0 to be equally valuable.That said, this paper needs substantial work to reach the standard expected for ESSD. The review was complicated by the feeling of reading an internal technical report rather than a scientific paper. Data description is unclear (e.g., the covered periods only appear near the end, and the use of all sat data is only apparent indirectly through Figure 12), methods are not clearly explained (e.g., the Lagrangian scheme, or the exact differences with META3.X), and I would not be able to reproduce the results without digging into the shared code. There are also many errors in the figures, few references to dynamical features or related scientific literature outside the introduction, and claims about the database's robustness and effectiveness are not well supported by the figures or text. There is also no discussion of the possible weaknesses of the method, whether related to the use of altimetry itself or to the algorithm.I also feel the paper spends as much space presenting the diagnostic tool as the database itself, which weakens it: the tool's capabilities are listed without a clear discussion of results, giving the impression of unfinished work. I would suggest focusing on the database, with a few statistics demonstrating its importance and a clear description of what changed between META3.1exp (Pegliasco et al., 2022) and META4.0 and why. If the authors want to keep the diagnostics section, they could either trim it to a short list or expand it with a real discussion of results, but this last option would require a lot of work.I advise major revision, but I strongly encourage the authors to continue, as I think this database deserves to be released and supported by a solid scientific article.I took notes while reading the paper to help you improve it; I list them here:Title: you had "Global" in the Pegliasco et al. 2022 manuscript, and I think it's a good idea to keep this term.L7: "completes the classical single-trajectory view": as you clearly state later, there has been effort since at least 2018 to account for these events and consider networks, so it's not clear that "classical" still holds.L11: It's not clear why this is presented as a "new insight," since, as clearly stated in the introduction, META4.0 builds upon progress made by other methods. The first method to explicitly define eddy trajectories as networks was introduced regionally by Laxenaire et al. (2018), in parallel with Le Vu et al. (2018) (regional) and Cui et al. (2019) (global) who referred to multi-core eddies to assess merging and splitting. It's therefore not clear why, seven years later, you present this as a "new insight." Perhaps instead highlight that you provide a longer time series (three decades) or, if you have the evidence, more robust outputs, but "new insight" feels misleading as worded.General comment on Introduction: I was a bit surprised that you split the introduction into subsections, but since it's quite long, it does make it easier to read. It's also a bit odd that, after L70, you no longer refer to the bibliography of "1.4 Eddy network and interaction" but instead just describe META4.0 directly, so it might need a new subsection title.L44: Since, up to my knowledge, only META3.1exp was associated to a published article, we have to trust that the different versions "progressively improved." While I'm quite confident that's true, don't you have technical notes or documentation explaining the differences? This ties in with a related comment in the Data/Methods section: I think you need to say more about how META4.0 differs from META3.X and why those changes were made.General comment on Data and methods: Parameter choices are not well explained (e.g., why 10% for the overlap threshold, why no high-pass filtering of SSH when META3.1 with PET did apply one?). The way the steps are presented doesn't always make them easy to follow. For example, the overlap concept is introduced at L155 (without being fully explained), stating that it creates "chains of linked observations" (I assume this means using "best predecessor"?), and only afterward are the events described, before the reindexing is introduced. Is the reindexing a postprocessing step, by the way? For an ESSD paper, I would expect a clear definition of the algorithm and a discussion of the method's limitations, as there is currently no discussion of the imperfections of altimetry, except when noting that short segments are removed. Yet we might expect that a long time extension combined with a low overlap threshold could, at least occasionally (e.g., for small eddies in regions with strong currents), link eddies that are not actually related in reality.L95: You could use the Ierapetra eddy as an example, as shown in Figure 1b of Ioannou et al. (2017), https://doi.org/10.1002/2017JC013158.L90: Altimetric "all-sat" or "two-sat"?L97: While it's true that spatial filtering is applied in Chelton et al. (2011), Chelton et al. (2011) did not use Okubo-Weiss; instead, they associated SSH pixels. They discuss the weaknesses of the Okubo-Weiss parameter in their Appendix B1 and present their own method in B2. In addition, META3.1exp (Pegliasco et al., 2022), which also used the PET algorithm to detect eddies, also applied a high-pass filter. This matters because you state that no filter is applied here, since you detect closed SSH contours rather than using Okubo-Weiss (or connected pixels as in Chelton et al.), but META3.1exp already used PET, so this justification is not fully consistent. Could you add some information or an example explaining this choice?L107 and following: There are many similarities with META3.1exp (Pegliasco et al., 2022). You could state clearly if there weren't many changes. What about the 2 mm step for contour extraction? Is Visvalingam new to PET? It's not clear what you mean by "facilitating further analyses and data colocation."L120-121: Very interesting: do you plan to one day use your algorithm to quantify these types of interactions?L125: What is a "normalized overlap area"? Is it close to the "similarity coefficient" from Pegliasco et al. (2022)? Why 10%?L126: A time extension is fairly common in eddy detection methods; you could reference other approaches to show this isn't a novel choice. Also, why 7 days specifically? Is it linked to the altimetric satellite orbit? Given that eddies with radii of several tens of km travel at around 10 km/day, don't you think a 10% threshold risks associating unrelated eddies? Maybe add a note on that.L133: If you find a matching eddy at day +1, do you still keep checking eddies up to day +7, or do you stop? If you continue, do you, in the algorithm, end up counting the same eddy multiple times (i.e., for a single trajectory without merging, if you associate eddies at t+1, t+2, t+3, ..., t+7 starting from t, then repeat the process starting from t+1, associating t+2, t+3, etc.)? I expect that, if yes, you then filter and count only eddies once, but isn't it time consuming?L143: Broken/incomplete link (URL longer than the page width).L145 and Figure 1: Are "trajectory segments" different from "eddy segments"? I found the description of segments and trajectories hard to follow. Are they the same thing? Segments start and stop at events, but if they "win" the event, do they continue rather than stop? For example, in this figure, segment 2 undergoes a splitting event (giving rise to segment 3) but continues. You give this example without explaining how you determine which segment "wins" during these events. Expert readers might infer it, but I think it could be confusing for others.L156: How is the overlap computed? Is it normalized? Does the larger eddy always "win"?L157: You write "segment (trajectory)." Again, does this mean the two terms refer to the same thing? In ToEddies, we also use the term "segment," but with a different meaning: a trajectory is composed of multiple segments, and segments always begin with a birth, death, merging, or splitting event. If "segment" means the same as "trajectory" here, I'd suggest not using the term "segment" this way, since it seems to add confusion without adding information beyond "trajectory." If trajectories and segements are different, that's of course not a problem if it doesn't mean the same thing as in ToEddies, but please clarify.L165: Maybe this reindexing step answers my earlier question. Using the same splitting example: two new segments are created, but afterward, using the "maximal polygon overlap" (L156), do you reindex one of the new segments to match the one before the split? What exactly do you mean by "continuity" here?L169-170: Why would connections involving several segments is never considered as detection mistakes, while connections to a single segment would be?L172: Why "trajectories" now (rather than "segments")?L174-176: Very interesting: do you have numbers for how often these cases occur? Perhaps add references showing this can happen in reality, or, if there's no direct observational evidence, insight from geophysical fluid mechanics theory. Alternatively, if you think this is actually an artifact of altimetric field reconstruction rather than a real phenomenon, it might be worth stating that explicitly.L187: "pyeddytracker" is spelled differently than at L106.L190: How was this network selected? Why not use a network already studied in a previous publication (from altimetry using merging/splitting events, and/or supported by observations)? A small inset map showing the location of the selected network on a world map could help orient the reader. We only see time/latitude (Figure 2) or longitude/latitude (Figure 3) later on, and it takes some effort to figure out where we are.L192 (and elsewhere in the manuscript): The extensive use of numbered lists and dashed bullet points gives the impression of reading a technical report rather than a scientific journal article. This doesn't necessarily need to change, since it doesn't prevent understanding, but making an effort to write this more smoothly could improve readability in places.Figure 2 / L196: You write in the text that segments are ordered by segment identifier, while Figure 2a labels this "Order of Segments." This seems misleading, especially since the notion of "order" in eddy networks was introduced by Laxenaire et al. (2018), and doesn't match what you show here: they defined it as a way to indicate the number of merging/splitting events separating a trajectory from the origin trajectory of the network. For example, trajectories 4 and 5 in Figure 2a would have the same "order" under their definition. If I understand correctly, what you call "Order of Segments" here actually corresponds to the segment index as defined in your Data section. There's no problem using the word "order" with a different meaning than Laxenaire et al. (2018), but you should be clear about what it means here, as it could otherwise mislead readers familiar with their work.L203: The shape error needs more explanation of its role: okay, it quantifies "the deviation from an idealized circular eddy," but how? I saw a bit more information in L113 but that still very short. This parameter, presented as-is, is a clear example of how the presentation differs from META3.1exp. Indeed, for reference, Pegliasco et al. (2022) state: "For the outermost closed contour encompassing only one extremum, a shape error test is performed. This test, similar to Kurian et al. (2011) verifies that the ratio between the areal sum deviations of the contour from its best fit circle and the area of this best fit circle is below a certain value. This specification aims to avoid the selection of eddies with shapes too different from circles, where rotation is not possible, as for banana shapes for example. In Mason et al. (2014) and Kurian et al. (2011), the shape error was limited to 55%. We increase this value to 70% to ensure that elongated eddies are detected, a case often visible in highly dynamic regions and when eddies are interacting."Figure 4: This is not "latitude" on the y-axis.Table 2: You give numbers but I couldn't find the exact covered periods... and again, "all-sat" or "two-sat"?L230: Do you have information on the size and intensity of these "lonely" eddies? Are they small/weak eddies? Are eddies discarded because they belong to short segments (postprocessing step, L169) later becoming "lonely" eddies? If they are small and weak, this could be linked to eddy detection errors as discussed in, for example, Stegner et al. (2021), https://doi.org/10.1029/2021JC017475.Figure 5: Why not separate anticyclonic and cyclonic eddies?Figure 6: Not very clear in the pdf. Could the resolution be improved? And again, why not separate cyclonic and anticyclonic eddies? Some studies suggest anticyclones are more persistent, do you find the same?L240-241: "non-negligible number": how many?Section 3.3.2: Could you provide a table similar to Table 2 for META3.2 (and/or for META3.1exp, which is published)? Of course, there's no network information for META3.2, but if I understand correctly, segments/trajectories in META4.0 correspond to segments in META3.2? Perhaps normalize by year, or give a daily eddy count as in Table 2 of Pegliasco et al. (2022)?Figures 7 and 8: Why not separate anticyclonic and cyclonic eddies? You give observations/year, which is good, but do you use exactly the same period for both datasets? I'm not saying this would change the results, but it should be stated somewhere. Also, would the number of birth+splitting in META4.0 correspond to the number of birth in META3.2, and would death+merging in META4.0 correspond to death in META3.2? If so, this could support the sentence "this difference arises because..."L249: "as both datasets are derived from the same...": I understand from this that the PET eddy detection used in META4.0 is exactly the same as in META3.2. So, for example, was the high-pass filtering used in META3.1exp no longer used in META3.2? This is the kind of question I would expect to already know the answer to from the Data/Methods section, rather than encountering it for the first time in the Results.L250: "it reduces artificial discontinuities and allows for more consistent tracking": most likely true, but this could also link unrelated eddies.Figure 9: Why not separate anticyclonic and cyclonic eddies?Figure 9a: I would have expected closer agreement between the two datasets: if the same PET detection is used and trajectories equal segments, I would expect the (nearly) same curves. I'm also not sure I understand the counting method: does 5x10-7 at 3500 days in Figure 9a mean 30 eddies out of 60,000,000 have lifetimes longer than 2000 days?Figure 9b: I'm not sure I understand this one either: it looks like 5x10-7 at 7000 days, implying a 20-year-long network? And is the fraction based on number of networks or number of observations? Sorry if I'm missing something, but the numbers aren't clear to me.L261-263: You state that, as META4.0 detects and connects a larger number of eddies, the visual comparison "confirms" that the network approach gives better results, but this may not be strictly true. For example, you state that 10% is the minimum overlap used, but with a 1% overlap threshold you would likely detect and connect more eddies. Would you say that results using a 1% minimum overlap "confirm" the network approach too? I expect not, since you selected (while I do not know why) 10%. I know this is hard to confirm visually and hard to find independent data, but at minimum, you could support the claim using established results from the literature.L265: Same remark: you state that these diagnostics provide a "more robust representation of mesoscale eddy dynamics," but nothing demonstrated here actually confirms that. What you've shown is that the representation is more complex, not necessarily more robust.Figure 10: What about the short networks? You should show where they are located, as this could be informative. Also, I think a figure showing the total number of eddies is missing (similar to Figure 5, but for all eddies that are part of a network), to show that the spatial coverage differs from that shown in Figure 10a. Legend not easy to read, and even incomplete, in Figure 10b.L269: You clearly indicate here that you count pixels within eddy contours. Did you do the same for Figures 5, 7, etc.? If not, why the change in counting method? I don't understand why this would "emphasize the spatial regions": the difference between counting by eddy center versus by contour is that the latter accounts for both position and size (large eddies get more pixels, so more weight, than small ones), whereas counting by center weights large and small eddies equally. Also, in the case of merging, do you count the merged eddy, or the two (or more) eddies that merged? What about splitting?Figure 11: Why not separate anticyclonic and cyclonic eddies? Panels a and b are, as you note in the text, very similar. Wouldn't you find the same pattern in a version of Figure 5 that included all eddies in networks, rather than just lonely eddies? I agree this is because the region is dynamically active, but that just reflects there being many eddies there. It would be more interesting to highlight where the two panels differ. For example, I see more merging in the Mozambique Channel, and more splitting upstream of it and in the Madagascar retroflection region. If you can identify a dynamical explanation for this, it would support the idea that the merging/splitting hotspots identified by your network approach are dynamically meaningful, which would strengthen confidence in the method.Figure 12: I think this is the first place where the exact period covered is clearly stated! Why not reproduce Figure 4 of Pegliasco et al. (2022) to show the period covered by satellite missions, rather than just their launch dates?L276: What are the "four points per year"? Seasons?L280: "Pegliasco et al. (2022)" instead of "(Pegliasco et al. 2022)" (citation format).L285: This normalization isn't clear to me. For example, in the case of a splitting event, one segment/trajectory continues while a new one begins. Are the eddies before and after the split, within the continuing segment, normalized by the same value, while the newly created segment is normalized by the maximum value it reaches after the split? I was actually a bit disappointed reading the text, because looking at the figure, I noticed that shape-error values before and after 20 days around an event look comparable; yet merged eddies are on average larger than eddies that merge, and the opposite is true for splitting. Showing this kind of behavior more explicitly would strengthen the case that these are genuine merging/splitting events, since artificial separations caused by altimetric error would not be expected to show notable size differences 20 days before and 20 days after the event. Also, how do you handle eddies that are not active 20 days before or after the event? What about successive merging/splitting events occurring within less than 40 days of each other?L297-298: Very good that the absolute values confirm this! Is that also the case 20 days before and 20 days after? I think this is a strong point in favor of your merging/splitting detection; I would suggest presenting this as a figure (or numbers), even if only in the appendix.L303-307: The merged altimetric product is constructed via optimal interpolation. Wouldn't this tend to make structures appear to grow or decay slowly before and after the periods where the eddy is actually captured by altimetric tracks? Are similar tendencies also found in, for example, numerical models? Or, if you repeat this analysis using only long-lived trajectories (which should be less affected by optimal interpolation at the start/end of their lifetime), do you get the same result? I think Figures 16 and 17 suggest that the answer is no since after 15 days the evolution appears to reach a plateau.L322: You don't give any information on the CLS CatSat chlorophyll product: no DOI, no reference. At minimum, please explain it, or use a freely available product instead.Figures 18 and 19: No colormaps shown... I can't see the latitude. Are these known merging/splitting eddies?L328: "are also visible": not sure this is accurate. You only show that, at least in two examples, they can be seen.Figure 20: Why no distinction between anticyclones and cyclones?L340: Did you try computing the maximum latitude and longitude differences separately? It could be interesting to identify zonally versus meridionally oriented networks. I have to say I like this clustering method. I think that, with careful choice of the parameters, it could indeed be a nice way to find patterns in such a complex database of eddy networks.Group 2: There's no constraint on latitude. Why do you think these are found near the equator? Perhaps because of the very large Rossby radius of deformation there? Are you confident that the geostrophic assumption holds in this region?Figure 21: You define groups 0 to 4 in the figure, but the text and Figure 22 use groups 1 to 5.3.5 Lagrangian detection of eddies: Lagrangian eddy detection is itself an active research field; maybe give at least a few references. Researchers from that community often argue that Eulerian detection methods don't guarantee that the detected eddies actually trap particles. It's a good idea that you made the effort to check this, but you don't provide enough detail on your methodology to make it reproducible. For example, at L378: how do you advect the particles? Do you use a specific tool such as Parcels? If not, please provide some information about the numerical scheme used. Also, do "coherence" and "auto-coherence" mean the same thing?Figure 24a: Nice figure, and I understand why it's placed here to introduce M, M1, M2, etc., but you could show it earlier in the text, even if the naming convention would only make full sense at this point. I think Figure 24 does a reasonable job showing that merging/splitting events are fairly well detected (though it could also simply reflect that merged eddies M are larger than M1/M2). Since the Lagrangian method itself isn't well explained, however, this isn't entirely clear.General comment on the Conclusion: The conclusion isn't bad in itself, but it's long while not always conveying clear information. The final paragraph on future developments is of great interest, but it isn't linked to existing related work, whether using numerical models or Argo profiles. Additionally, you mention "next-generation sea surface height products" without mentioning the SWOT mission.I am, of course, available for a next round of review and really hope the authors will make the effort to improve the manuscript, as I strongly support the publication of this database in ESSD.Rémi LaxenaireCitation: https://doi.org/
10.5194/essd-2026-108-RC4
Data sets
META4.0 Juliette Gamot, Antoine Delepoulle, Francesco Nencioli, Marie-Isabelle Pujol, and Gérald Dibarboure https://doi.org/10.24400/527896/a01-2026.001
Model code and software
pyeddytracker Anoine Delepoulle and Evan Mason https://github.com/AntSimi/py-eddy-tracker/tree/v3.6.1
Interactive computing environment
PyEddyTracker documentation Antoine Delepoulle and Evan Mason https://py-eddy-tracker.readthedocs.io/en/latest/index.html
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 512 | 291 | 35 | 838 | 24 | 27 |
- HTML: 512
- PDF: 291
- XML: 35
- Total: 838
- BibTeX: 24
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Gamot et al. introduce and describe a new Eulerian-based eddy tracking dataset, META4.0, that can detect eddy merging and splitting (EMS) events based on the temporal evolution of sea-surface height contours. Notably, they a posteriori test their dataset by measuring Lagrangian coherence and find that the EMS events they detect are largely coherent in the Lagrangian sense. I do not have any comments and I recommend the manuscript for publication as is.