| 08 May 2026
Intercomparison of Three SWOT-Derived Level-4 Products: From Mapping Accuracy to Multi-Scale Dynamical Representation
Abstract. Oceanic submesoscale dynamics play a critical role in energy cascades and vertical tracer transport. The Surface Water and Ocean Topography (SWOT) mission, through its high-resolution wide-swath sea surface height (SSH) observations, provides an unprecedented capability for resolving these processes. While this enhanced spatial coverage represents a major advance over conventional nadir altimetry, it also introduces new challenges for constructing dynamically consistent gridded Level-4 products. To address these challenges, a range of data fusion and reconstruction approaches have been developed to incorporate SWOT observations into next-generation SSH mapping systems. This study presents a comparative evaluation of three SWOT-derived Level-4 products (MIOST, 4DvarQG, and 4DvarNet) over the North Atlantic (25° N–50° N, 80° W–10° W). The assessment combines Eulerian metrics of mapping accuracy, Lagrangian trajectory mprediction skill based on surface drifter observations, and diagnostics of dynamical structures using Rossby number (Ro) and finite-size Lyapunov exponent (FSLE) fields, with SWOT Level-3 data as a reference. The results reveal a pronounced scale dependence in product performance. In mesoscale-dominated regimes such as the Gulf Stream, 4DvarQG achieves the highest velocity reconstruction accuracy and improves short-term (0–4 days) Lagrangian prediction skill, reflecting the benefits of quasi-geostrophic dynamical constraints. In contrast, 4DvarNet shows greater sensitivity to smaller-scale variability, characterized by sharper SSH gradients, elevated Ro, and more filamentary strain structures, indicating an enhanced representation of fine-scale features. However, the physical realism of these intensified small-scale signals requires further validation against higher-resolution or less-filtered observations. MIOST demonstrates stable and consistent performance across a wide range of spatial scales for global ocean mapping. These results highlight inherent trade-offs between dynamical consistency and small-scale variability representation among current SWOT-based Level-4 products. Future developments may therefore benefit from hybrid approaches that integrate data-driven flexibility with explicit physical constraints.
This manuscript intercompares three different level-4 SWOT-derived SSH products within their common domain in the North Atlantic, a region which includes the western-boundary current and lower kinetic energy sub-regions. The "level-4" products are those which blend information from multiple satellite altimeters, forming a sea level estimate on a gap-free spatially uniform grid, which typically requires some form of interpolation to estimate SSH in the gaps between observations, and it also involves smoothing the observations to reduce the impact of measurement noise. Level-4 products are widely used in applications for identifying eddies or other sea level features, and for estimating geostrophic currents. Therefore, I think the topic of this manuscript will be of interest to many readers.
The authors use a nice methodology to intercompare the products. They examine Eulerian currents (velocity at fixed locations) as well as Lagrangian flow structures (following particle trajectories) through comparison with drifter observations. They also examined dynamically-relevant statistics of the products, such as the Rossby number, and qualitative aspects of the flow field around a small set of eddies.
Overall, I found the paper does a nice job of illustrating the different aspects of the level-4 products. The presentation style and quality of writing is excellent. I think the article could be published as is, but in a few places I disagree with their characterization of results from the 4DvarNET product. While a lot of good work has gone in to producing the 4DVarNET product, I think the results here demonstrate rather conclusively that the approach based on universal approximators with neural nets fails to produce meaningfully useful results. The fact that such approaches can produce qualitatively realistic results---which are quantitatively less accurate than much simpler linear estimators---should be held up as a warning or a cautionary note, rather than being lauded as "particularly promising."
Detailed comments:
l41-l42: Provide references for "recent studies" or rephrase.
Aha: The next 3 sentences provide these references.
l68-l71: Good summary of introduction.
l100-l109: Aha! Two of the products are not available on a global grid!
Well stated in l110-l112.
l119: Not sure what an "event-based perspective" could refer to.
l214-l229: Nice summary of Eulerian velocity comparisons.
l273: States that DUACS was used as a benchmark for normalizing
the Lagrangian comparisons. Good idea. Fig 6 is nice.
l406: The word "captures" implies that these strain structures are
realistic. I think a better word is "contains" since the realism of the
structures is unknown.
l421: 4DvarNET "better preserves fine-scale spatial organization"
is totally speculative. A more justified or balanced statement
could be, "4DvarNET generates spurious small-scale flow features which look
realistic, but which are not supported by any validation data. Hence, dynamical
inferences from this product should be avoided."
l438: How do you know that 4DvarNET exhibits "enhanced sensitivity to
fine-scale variability"? Could this not simply be due to spurious
structures in the training data or mis-tuning of the assumed noise level?
l495-l496: omit "as it better preserves ..."
l502: Why does inclusion of the neural networks "appear particularly
promising"? Your results appear to show the opposite is true.
Please proofread the references list more carefully. I note some problems
in Le Guillou 2025 with a quick spot-check, for example.