the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Jul 2026
Global submesoscale eddy identification and characteristic analysis based on multi-source remote sensing data
Abstract. Research on oceanic submesoscale eddies has long been limited by the spatiotemporal resolution constraints of satellite altimeters. Relevant studies remain insufficient and have mostly focused on regional waters. Global investigations on submesoscale eddy detection, dataset construction, and distribution characteristic analysis based on multi-source remote sensing data and multiple methods are still lacking. In this study, we first develop a submesoscale eddy detection method by integrating high spatiotemporal resolution ocean color data, deep learning algorithms, and digital image processing techniques and construct a global submesoscale eddy dataset based on chlorophyll-a observations. Furthermore, we design a multi-scale eddy detection framework using altimeter data and establish two global datasets: a submesoscale eddy dataset derived from SWOT satellite altimeter measurements and a multi-scale eddy dataset generated from merged altimeter data. Finally, we statistically compare the scale, seasonal, and geographical distribution characteristics of eddies from the three constructed datasets and systematically analyze the strengths and limitations of different data sources and algorithms for submesoscale eddy detection. This study effectively compensates for the scarcity of existing submesoscale eddy datasets, provides a valid verification approach for conventional eddy products, and offers a feasible guideline for data selection in diverse submesoscale eddy research scenarios.
- Preprint
(4101 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 08 Oct 2026)
- RC1: 'Comment on essd-2026-475', Anonymous Referee #1, 27 Aug 2026 reply
-
RC2: 'Comment on essd-2026-475', Anonymous Referee #2, 09 Sep 2026
reply
The authors have integrated multi-source remote sensing data, digital image processing technologies, deep learning methods, and the Sea Surface Topology method to establish three global submesoscale eddy datasets. Then, analyzed their size distributions, spatial patterns, and seasonal variations. For the first time, this study constructed intercomparable global submesoscale eddy datasets using multiple input sources and detection algorithms, and assessed their respective advantages and drawbacks in submesoscale eddy detection. This development will make a profound impact on the field of oceanographic research. However, there are still several issues that need to be addressed, and I recommend minor revisions. The following points should be considered:
- The number of CEs and AEs in the training set for deep learning varies greatly. Could this imbalance affect the results of eddy recognition?
- Was the range of the colorbar in Figure 2(b) deliberately chosen to be large, causing the eddies to appear less prominent, in order to highlight the advantages of your method as shown in Figure 2(c)?
- How is the eddy center determined in Figure 5? Is it based on the equivalent circle center or the centroid? Please clarify the methodology used to determine the eddy center.
- The primary contribution of this paper is the establishment of a global submesoscale eddy dataset. The authors should elaborate on how users can effectively utilize this dataset and provide any necessary precautions.
- To improve reader understanding, the authors may consider adding a scale bar to the map to visually represent the size of the eddies.
- Some submesoscale eddies are generated within the ocean when the kinetic energy of the mixed layer is relatively strong. These eddies may not show height anomalies on the ocean surface and thus cannot be observed by altimeters. Can these eddies be identified using ocean color data?
- The authors should include a geographical distribution map of the eddy radius to provide a clearer understanding of the spatial patterns.
- The eddy dataset established based on chlorophyll data should be expanded to highlight its innovation, and it should be continuously updated to ensure its practical value.
- In addition to analyzing the distribution of eddies in terms of time, geographical seasons, etc. The author should further discuss more features of the eddies, such as the variation of eddies in latitude, interannual variations, etc.
- Due to significant differences in spatial coverage among the three datasets, such as the fact that chlorophyll data is prone to interference from clouds and SWOT data only covers a portion of the area each day, it is recommended to add a normalization analysis to mitigate the impact of spatial coverage.
Citation: https://doi.org/10.5194/essd-2026-475-RC2
Data sets
A Global Sub-Mesoscale Eddy Dataset Using SWOT SLA data Meng Hou https://doi.org/10.5281/zenodo.20714410
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 197 | 73 | 73 | 343 | 80 | 72 |
- HTML: 197
- PDF: 73
- XML: 73
- Total: 343
- BibTeX: 80
- EndNote: 72
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript presents three global eddy datasets derived from Sentinel-3 chlorophyll-a observations, SWOT sea surface height data, and a 0.125° merged altimetry product, and analyzes their size distributions, spatial patterns, and seasonal variations. The topic is generally relevant to Earth System Science Data, and the resulting data products may have potential value for the oceanographic community. However, the manuscript still has several substantial issues regarding the novelty of the dataset collection, the comparability among the three products, the cross-validation strategy, and the depth of the characteristic analysis. These concerns affect the core design and main conclusions of the study and should be carefully addressed.
1. The chlorophyll-based global submesoscale eddy dataset constructed using Sentinel-3 observations and the DeepLabv3+ method appears to overlap substantially with the authors’ previously published work in terms of the observation period, identification method, model performance, number of detected eddies, and major statistical analyses. The authors should clearly distinguish the new data, methods, and analyses presented in this manuscript from those already published, and further clarify the added value of integrating the three datasets in the present study.
2. The three datasets differ substantially in their temporal coverage and sampling characteristics. The chlorophyll dataset mainly covers 2020–2022, whereas the SWOT dataset covers 2023–2025, with no common observation period between them. The merged-altimetry product covers a much longer period. In addition, chlorophyll observations are strongly affected by clouds and illumination conditions, SWOT provides discontinuous swath sampling, whereas the merged-altimetry product provides regularly gridded fields. Therefore, direct comparison of radius distributions, spatial detection counts, and monthly mean numbers may reflect differences in observation period, sampling strategy, effective resolution, and data processing rather than actual differences in eddy characteristics.
3. If multi-source remote-sensing comparison is intended to be one of the main contributions of the manuscript, the current cross-comparison is insufficient. Quantitative matching should be performed within common temporal and spatial domains. For example, SWOT and merged-altimetry detections could be compared in terms of eddy-center distance, radius difference, polarity agreement, boundary overlap, and matching rate during their overlapping period, with similar analyses conducted between chlorophyll and altimetry products. In addition, agreement among remote-sensing datasets should not be regarded as equivalent to independent validation. More independent observations or reference data are still needed to assess the reliability of the identified eddies.
4. The current “characteristic analysis” mainly consists of eddy-radius distributions, spatial detection counts, and monthly mean numbers. These analyses remain relatively basic and appear insufficient to fully support the emphasis on “global submesoscale eddy characteristic analysis” in the title and conclusions. More comprehensive analyses could include regional and latitudinal variability, cyclone–anticyclone asymmetry, normalized occurrence frequency, eddy intensity, and interannual variability. Otherwise, the corresponding statements in the title, abstract, and conclusions should be moderated.
5. The three products contain markedly different eddy-size ranges, and it is therefore necessary to clarify whether they represent the same class of physical features. In particular, some of the SWOT-detected eddies have sizes close to the nominal grid spacing of the product, and their reliability requires further justification. Moreover, because the three datasets have very different effective observation opportunities, raw detection counts should not be directly compared and should instead be normalized by valid observation area and observation time. If no eddy tracking or temporal de-duplication is performed, the reported numbers should more appropriately be described as eddy detections or eddy snapshots rather than independent eddy events.