the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 26 Jun 2026
Eulerian and Lagrangian Assessment of Arctic Surface Current Products Using Drifter Observations
Abstract. Accurate Arctic surface current estimates are needed to quantify freshwater redistribution, upper-ocean connectivity, and Arctic–North Atlantic exchanges, but validating them is difficult given sparse observations, sea-ice cover, and overlapping dynamical processes. This study evaluates six widely used products against drogued drifters from the Global Drifter Program (2011–2021): AVISO altimetry-derived geostrophic currents, NeurOST machine-learning reconstructions, GlobCurrent and OSCAR multi-observational/mixed-layer total currents, and the GLORYS and TOPAZ reanalyses. Skill is assessed with complementary Eulerian point-wise metrics and Lagrangian trajectory diagnostics, including a Lagrangian Uncertainty Quantification (LUQ) framework. Eulerian results show marked regional and seasonal contrasts. NeurOST and OSCAR agree best with drifter velocities, GlobCurrent and GLORYS show intermediate skill, TOPAZ has weaker near-surface speed agreement, and AVISO underestimates current intensity and variability. Products incorporating additional observational constraints, wind-driven contributions, or mixed-layer dynamics outperform altimetry-only geostrophic estimates, especially in regions shaped by mesoscale variability, bathymetry, sea ice, and narrow gateways. The Lagrangian analysis shows that good local velocity agreement does not guarantee realistic transport: small velocity differences accumulate along trajectories, producing large deviations in pathways, retention, and connectivity. LUQ diagnostics indicate part of this mismatch reflects intrinsic transport variability and initial-condition sensitivity rather than product error alone. The authors conclude that Eulerian validation must be complemented by Lagrangian, uncertainty-aware diagnostics for studies of freshwater pathways, tracer transport, and Arctic connectivity.
- Preprint
(25932 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-3629', Anonymous Referee #1, 21 Aug 2026
-
AC1: 'Reply on RC1', Júlia Crespin Esteve, 17 Sep 2026
We thank the reviewer for their thorough reading of our manuscript, the positive assessment of our work, and their constructive suggestions. We have detailed our point-by-point responses to all comments and a description of the changes made to the manuscript in the attached PDF.
-
AC1: 'Reply on RC1', Júlia Crespin Esteve, 17 Sep 2026
-
RC2: 'Comment on egusphere-2026-3629', Anonymous Referee #2, 21 Oct 2026
The paper compares surface currents from six different products in the Arctic against drifter data. While this exercise idea is nice, the manuscript seems somewhat overcomplicated relative to its central message that surface current products in the Arctic should not only improve the current estimates, but also the transport geometry – shapes and positions of the relevant features – eddies, currents, etc. Improved Lagrangian behavior ultimately requires improvements in the velocity product itself – the quality of the underlying velocity fields, improved observations/mapping, resolution, forcing, assimilation, or model physics. Moreover, the products provide different physical quantities/components and effective depths; the drifters’ data used are spatially and temporally limited; the dynamical regimes compared are very different, and the analysis evaluates near-surface pathways over 5-day integrations. While the authors mention the limitations in Section 5.5 and the Conclusion, I suggest clarifying the study's goal and perhaps focusing more on the methodological framework and application in the Arctic, rather than on the comparisons themselves. Nevertheless, I still consider the paper a worthwhile study, after a revision.
My main points are:
1) Eulerian comparison. Could the Eulerian comparisons be performed against the same set of drifter observations? Part of the differences in RMSE, correlation, bias, etc., may reflect different spatial, temporal/seasonal sampling, or different dynamic regions, rather than product performance. Also, how exactly is the collocation procedure performed, given the different product resolutions? And could it be worth weighting the observations? Densely sampled regions may otherwise have a disproportionate influence on the aggregated statistics.
2) Lagrangian Validation Framework. I am not fully convinced by the interpretation of the spatially averaged LUQ. For the Arctic, a 10 km box is not necessarily dynamically small; it can incorporate a front (for example, in the Bering Strait). So, in the end, the authors are averaging together the particles' pathways, which may belong to different dynamical regimes and therefore are not expected to follow the same path.
Another question I have to this: since the product resolutions differ, part of the LUQ results might reflect the interpolation process. The neighborhood is sometimes smaller than the product grid spacing, let alone the effective resolution. Also, I didn’t understand how many particles are simulated within the 10 km box?
Similarly, the temporal resolution of each product may also affect the results: how do you simulate within a 5-day window, if, e.g., AVISO is provided every 3 days? Could the results of the sensitivity analysis of the chosen scales be provided?
Another point – the whole Lagrangian analysis is based on only nine drifter case studies, which seems rather limited for drawing broader conclusions about Arctic transport regimes. Each regime is sampled by only one selected drifter trajectory during a specific season, year, and particular forcing during those 5 days. Since the Eulerian analysis shows a pronounced seasonal dependence in product skill, it is difficult to separate the influence of circulation regime from seasonal/event-specific conditions in the LUQ comparison.
3) While authors do state the limitations, I would still advise caution in statements that one product “outperforms” another, particularly in the Abstract and Conclusions. The products compared represent different physical quantities. E.g., while the AVISO product “underestimates” measured by the drifters current speed, this should not necessarily be interpreted as a product deficiency, since AVISO is intended to represent only the geostrophic component of the flow.
Minor comments:
Lines 27, 35, 90, 100, etc.: for the geostrophic currents, “Absolute Dynamic Topography (ADT) gradients” would be more precise than “sea surface height gradients,” as correctly stated at line 105. The description of the geostrophic products is also repeated several times throughout these sections and could, in my view, be shortened.
Section 2.1.3. Please add the reference to Fig. 1d for OSCAR. You may also consider swapping the OSCAR and NeurOST panels so that the figure panel order follows the subsection order.
Figure 1: I would suggest replacing “current intensity” in the colorbar title with “current speed” for consistency.
Line 182. The authors state that, to ensure methodological consistency across all datasets, they use the surface velocities from GLORYS. However, for GlobCurrent, 15 m was used instead of 0 m. These choices do not seem fully consistent. Have the authors compared both levels? It would be nice to know how this choice affects the results, if it does.
I would suggest providing a compact summary table of the used products. It could considerably improve clarity. It could include product type, spatial resolution, temporal resolution, effective sampling depths, spatial coverage, principal observational/model inputs, different dynamical components represented. Please also check the independence of the products from drifter data: are any of the products using or calibrated against drifter observations? (GlobCurrent or OSCAR)?
Figure 3 and in the text: The Beaufort Gyre is more commonly abbreviated as BG in Arctic oceanography. This is a minor point, but using the conventional abbreviation would make the terminology more familiar to the community.
Figure 5: while the shapes of the markers help understand the diagram, the AVISO circle doesn’t appear orange, as it is stated in the caption, particularly in comparison with Figs. 6 and 7, where it is indeed orange. Also, the legend is cut off by formatting.
Once the abbreviations are defined (e.g., RMSE, U, V, etc.), there is no need to write out the full term again.
Line 544: comma at the beginning of the line
Figures 12 and 13. Please make the color scale ranges the same across the right-panel figures.
Lines 561-562 and lines 680-682: as I understood, the authors still simulate the trajectories over 5-day windows for all LUQ calculations, while the 30-, 60-, and 150-day periods refer to the duration of the selected drifter records over which the 5-day LUQ calculations are repeatedly evaluated at different initial times. If so, the statements that long cases demonstrate the accumulation of errors over time are misleading here. Otherwise, please clarify and revise the interpretation. Is this an experiment of integration-time dependence?
Figure 14, 15, 16 – the LUQ is normalized, but the y-axis ranges are not the same
Figure 17: The titles “Standard scale cases” and “High Scale” should be at the top of the figure.
Figures 14–17: The Lagrangian results are interpreted mainly in terms of circulation regime, but the selected case studies occur in different seasons. Since the Eulerian analysis shows strong seasonal dependence, please clarify whether seasonality may also influence the LUQ differences between cases.
Lines 215 – 220 and elsewhere: The manuscript contains a fair amount of repetition that could be reduced. For example, the next subsection repeats the statement of these lines. Lines 251 – 254 also repeat what has already been said. Another example – the sentence that small velocity differences may accumulate over time and lead to large deviations in trajectories is repeated on line 267, then on line 274, and then again on line 314. Another example – the LUQ is already explained in the Methodology section; there is no need to explain it again on page 25. I feel the paper can be shortened, which will improve it.
Citation: https://doi.org/10.5194/egusphere-2026-3629-RC2
Model code and software
Software and Computational Notebooks for "Eulerian and Lagrangian Assessment of Arctic Surface Current Products Using Drifter Observations" A. Rosquete-Estevez et al. https://doi.org/10.5281/zenodo.20624431
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 214 | 169 | 47 | 430 | 36 | 41 |
- HTML: 214
- PDF: 169
- XML: 47
- Total: 430
- BibTeX: 36
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of the manuscript EGUSPHERE-2026-3629
In this manuscript, the authors compare the observational, quality-controlled data gathered by the ocean surface drifters from the well-known ‘Global Drifter Program’ with the output from various ocean current gridded products: AVISO altimetry-derived geostrophic currents, NeurOST machine-learning reconstructions, GlobCurrent and OSCAR multi-observational/mixed-layer total currents, and the GLORYS and TOPAZ reanalyses. The data comparison spans over a decade: years 2011-2021. In addition to the usual Eulerian evaluation of point-to-point similarity of current magnitude and direction, the authors use a Lagrangian method to show both discrepancy and coherence of the ocean surface’s circulation patterns. The identification of Lagrangian Coherent Structures (LCSs) as well as the use of the Lagrangian Uncertainty Quantification (LUQ) framework not only allows for trajectory-to-trajectory comparison between the observations and the chosen product but also permits transport assessment over a wider area (in the vicinity of the trajectory), organised by transport barriers and coherent structures.
The authors present and thoroughly analyse the statistical diagnostics, providing regional and seasonal differences between observations and the products as well. The presented case studies cover various dynamical regimes: from more energetic ones, where mesoscale processes have a major impact on the flow structure, to more coherent, large-scale, more stable patterns in the gateways to the Arctic Ocean. This leads the authors to two major conclusions: 1) Products incorporating observational constraints and wind effects outperform altimetry-only estimates; 2) good local velocity agreement does not guarantee realistic transport; and 3) no one product would provide the best results for all types of circulation patterns.
General assessment:
The manuscript's topic aligns well with the scope of Ocean Science and addresses the importance of validating publicly available gridded ocean current products. Despite the somewhat arbitrary choice of regional areas, the authors provide a logical explanation of the large number of metrics. Thanks to the application of two different methods, the authors successfully convinced the reader of their complementarity.
My bigger concern is the authors’ choice of the analysed case studies: why add these non-Arctic ones? How do they relate to the title of the article itself? I have a feeling that some of the cases were added not necessarily because they relate to the topic but rather because they are good to illustrate the differences between the products analysed. While the Lofoten Gyre and the Labrador Current are more closely connected to the Arctic (outflow and inflow areas), I would consider the Gulf Stream almost entirely out of scope of the paper. Perhaps adding ‘and sub-Arctic’ to the title would solve this issue?
Among others, the conclusion about the usefulness of the TOPAZ product for the ice-influenced cases, despite its weaker agreement in the global speed-based Eulerian statistics, is promising, especially due to the lack of observational data there, as illustrated by sparse GDP drifter coverage.
In my opinion, the manuscript is well written and illustrated and makes a proper research statement on the covered scientific topic. However, I also found it a bit too long and too repetitive in some parts. As mentioned above, some case choices seem slightly misleading. Therefore, I suggest that it can be published after some minor corrections.
Minor comments and remarks:
Title: ‘Eulerian and Lagrangian Assessment of Arctic Surface Current Products Using Drifter Observations’ – consider adding ‘and sub-Arctic’.
This section clearly describes the research background, motivation, data and methods and briefly introduces the next sections. The bibliography used is up to date and demonstrates a knowledge of the subject matter.
In this section, more information about the observations and gridded products is provided.
Line 182: ’This choice is made to ensure methodological consistency across all datasets’- Are all the other products similar? From their description, one may understand that they differ in the layers analysed.
In this section, the processing of the GDP drifters, Eulerian metrics, Langrangian validation framework and sensitivity analysis are described in detail. Furthermore, reasoning behind the chosen case studies is given, as well.
Line 272: ‘In ocean transport problems, relatively small velocity differences can accumulate over time and generate large deviations in particle trajectories.’ - This sentence is like the sentence in the previous section. Please consider rewriting/shortening it.
Line 334: ‘This analysis supported the use of τ = 5 days and R = 10 km’. – Which other values were considered in your sensitivity analysis?
This section provides a number of good illustrations of the differences in the seasonal, temporal and regional variabilities emerging from all products after the methodology has been used. However, the time series of the normalised LUQ values do not cover the same periods for all cases but are scattered among several years (in subplots of one figure). Is there a reason for this?
Line 439: ‘narrow Arctic gateways’ – I think you may add a sentence or two about the worst results that were obtained for S9 (the Canadian Archipelago); it really stands out in Figure 9.
I would also advise the authors to consider removing some repetitions that protract the text in this section as well: for example, these two sentences below are very similar:
Line 477: ‘Rather than evolving as a random diffusive cloud around the ground-truth trajectory, transport deviations are organized by the transport geometry of the underlying velocity field and by the Lagrangian coherent structures governing particle evolution.’
Line 506: ‘Accordingly, transport deviations are not uniformly distributed around the observed drifter position, but are organized by the underlying recirculation and its associated transport barriers.’
Line 574: ‘that products with with similar local velocity’ – remove repetition
In this part of the manuscript, the results are thoroughly discussed, together with the added value and limitations of the performed analysis. There are some repetitions from the previous parts, but this is probably justified and aims to highlight the importance of proposals.
This section lacks a broader context and a comparison of the results obtained with similar studies and some implications for future work. This might be helpful for the readers conducting similar data experiments or using similar methods.
Line 668: Section title - make it a verbless sentence to be consistent with previous subsection titles
This part is a summary of the obtained results; here the repetitions are unavoidable.
The manuscript contains 17 figures and 2 tables, and they are correctly prepared for publication. The fonts are large enough, the lines are not too thin, and the colours might probably be better in some of them (Figures 6, 7, 14-16). Please check the best colour palettes for scientific figures and data; this may help in your future submissions.