Composite Detection of Continental Shelf Fronts from Sea Surface Temperature and Altimetry: Assessing the Added Value of SWOT KaRIn

Juhl, Marie-Christin; Passaro, Marcello; Dettmering, Denise; Hart-Davis, Michael G.; Saraceno, Martin

doi:10.5194/egusphere-2026-4751

Preprints

https://doi.org/10.5194/egusphere-2026-4751

Preprints

13 Aug 2026

| 13 Aug 2026

Composite Detection of Continental Shelf Fronts from Sea Surface Temperature and Altimetry: Assessing the Added Value of SWOT KaRIn

Marie-Christin Juhl, Marcello Passaro, Denise Dettmering, Michael G. Hart-Davis, and Martin Saraceno

Abstract. We present a composite front detection approach that uses horizontal gradients in Absolute Dynamic Topography (ADT) and Sea Surface Temperature (SST) to identify oceanic frontal structures on the Southwestern Continental Atlantic Shelf, a region characterized by strong mesoscale variability and multiple front types. SST fields are derived from the OSTIA satellite product, while four ADT datasets are compared: SWOT MIOST, CMEMS, OK-STv2, and GLORYS12v1. A joint front probability metric is introduced to quantify the co-occurrence of SST- and ADT-derived frontal signatures and to evaluate consistency and differences across products. The method is applied to assess the spatial and seasonal variability of major regional fronts, including the Shelf-Break Front, the San Matías Front, and the Magellan Plume Front. The ADT datasets showed significant differences in their ability to co-detect continental shelf fronts. The Shelf-Break Front was consistently represented across all datasets, with maximum joint front probabilities occurring in austral summer (DJF) and exceeding 96 % between 35 and 45° S. In contrast, the seasonal and coastal San Matías Front exhibited stronger dataset dependence, with the highest joint probabilities obtained using SWOT MIOST ADT (73.63 %), followed by CMEMS (54.95 %), OK-STv2 (41.86 %), and GLORYS12v1 (35.16 %). In the mid-shelf region (38–41° S), elevated joint frontal probabilities indicate that ADT and SST products, particularly altimetry-based datasets, capture a persistent Mid-Shelf Front. The Magellan Plume Front showed strong joint signatures, most pronounced in SWOT MIOST (97.85 %), followed by CMEMS (80.65 %), OK-STv2 (72.83 %), and GLORYS12v1 (51.09 %), with a distinct coastal plume structure during austral winter (JJA). The approach could be further evaluated using ADT from SWOT KaRIn L3 data during SWOT's Cal/val phase, potentially enabling improved detectability through the sharper ADT gradients provided. Overall, the results show that combining SST- and ADT-based gradient detection enhances the characterization of frontal dynamics. The intercomparison further demonstrates that SWOT KaRIn–enhanced gridded altimetry (SWOT MIOST) substantially improves the detection of coastal and seasonal fronts compared with conventional ADT products.

Received: 06 Aug 2026 – Discussion started: 13 Aug 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Marie-Christin Juhl, Marcello Passaro, Denise Dettmering, Michael G. Hart-Davis, and Martin Saraceno

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-4751', Anonymous Referee #1, 07 Sep 2026
Review of “Composite Detection of Continental Shelf Fronts from Sea Surface Temperature and Altimetry: Assessing the Added Value of SWOT KaRIn” by Marie-Christin Juhl et al.
The authors present a composite front-detection approach based on horizontal gradients of sea surface temperature (SST) and absolute dynamic topography (ADT). The methodology is applied to altimetry-based and assimilated model products distributed by the Copernicus Marine Service, as well as to Level-3 altimetry observations from the SWOT mission and an experimental Level-4 product incorporating SWOT observations distributed by AVISO+. The study aims to characterize the surface dynamics of oceanic fronts over the southwestern Atlantic continental shelf (SWACS) region, with particular emphasis on assessing the potential added value of SWOT KaRIn observations for the detection and characterization of frontal structures. The results indicate that the integration of SWOT KaRIn observations into Level-4 gridded sea-level products may contribute to improved detection of coastal and seasonal fronts, likely benefiting from their enhanced spatial resolution and the resulting sharper ADT gradients. The SWOT KaRIn Level-3 product also appears to provide a more detailed representation of sharper and more spatially confined frontal structures compared with conventional altimetry products, particularly in complex coastal regions. Overall, the study provides interesting evidence that combining SST and ADT gradients, together with the complementary information provided by SWOT observations, can contribute to a more detailed characterization of oceanic fronts over the southwestern Atlantic continental shelf.
In my view, the manuscript is well structured. The objectives are clearly stated, and the datasets and methodology are clearly described. The study addresses relevant scientific questions that are well aligned with the scope of Ocean Science. The results provide a solid basis for the interpretations and conclusions presented by the authors. Overall, I consider the manuscript to be a valuable contribution to the field and potentially suitable for publication in Ocean Science, provided that the authors address the minor comments and points for clarification outlined below.
Minor comments:
The mean front probabilities obtained from ADT-only and SST-only approaches are currently presented in the Appendix. It might be helpful to consider moving these figures to the main text, as they would provide useful context for the reader and help illustrate the motivation for, and added value of, the composite approach.

Maybe add DOI to the datasets description in section 2.1

In the methodology section, the computation of horizontal gradients is described for gridded products. It would be useful to provide some additional details on how the horizontal gradients are computed for the SWOT KaRIn observations. The authors might consider referring to Section 4.6 in the methodology to clarify how the approach is adapted to the specific characteristics of the KaRIn dataset.

The legend of Figure 4 could perhaps be made slightly more explicit regarding the temporal period represented (e.g., “in different seasons” or “DJF season”); If the figure specifically corresponds to the DJF season ?

Line 220: The latitude value appears to be missing before °S. Could the authors please confirm whether the intended value is 40°S?

Based on the results presented, the discussion could perhaps be further developed to address several aspects that would help contextualize the results and their broader implications. In particular, it could be interesting to further discuss some of the potential limitations of the proposed composite front-detection methodology, with a view to providing insight into its robustness and general applicability. Some relevant aspects are already mentioned in the manuscript and could potentially be developed further. For instance, gridded altimetry products, including SWOT-based MIOST, may have an effective spatial resolution that is substantially coarser than their nominal grid resolution. This may limit their ability to resolve some small-scale features, such as submesoscale structures, coastal fronts, etc... It could therefore be useful to discuss further the extent to which current operational Level-4 gridded products can capture these features, as well as whether dedicated gridding approaches could potentially make better use of the high-resolution spatial information provided by SWOT KaRIn observations. As a complementary perspective, the authors may also wish to mention recent or ongoing developments in altimetry–SST mapping approaches aimed at improving the reconstruction of upper-ocean dynamics. For example, approaches such as NEUROST (Martin et al., 2024), VarDyn (Le Guillou et al., 2025), and Ciani et al. (2025) could provide relevant examples in this context. Discussing the application of the proposed front-detection methodology to such experimental mapping approaches could also be an interesting avenue for future work. Finally, the validation of the Level-3 SWOT observations appears to rely on a single pass over a relatively limited time period. This may constrain the robustness of the assessment and the extent to which the reported SWOT impact can be generalized across different dynamical regimes. The authors could perhaps discuss whether similar results might be expected in regions characterized by ocean dynamics different from those of the SWACS region, or when considering multiple KaRIn swaths acquired during the Science Phase. It could also be valuable to discuss the potential contribution of additional in situ observations to further validate and strengthen the findings, particularly with respect to the representation of small-scale frontal structures.

References:
Ciani, D., Fanelli, C., and Buongiorno Nardelli, B.: Estimating ocean currents from the joint reconstruction of absolute dynamic topography and sea surface temperature through deep learning algorithms, Ocean Sci., 21, 199–216, https://doi.org/10.5194/os-21-199-2025, 2025.
Le Guillou, F., Chapron, B., & Rio, M.-H. (2025). VarDyn: Dynamical joint-reconstructions of sea surface height and temperature from multi-sensor satellite observations. Journal of Advances in Modeling Earth System, 17, e2024MS004689. https://doi.org/10.1029/2024MS004689
Martin, S. A., Manucharyan, G. E., & Klein, P. (2024). Deep learning improves global satellite observations of ocean eddy dynamics. Geophysical Research Letters, 51, e2024GL110059. https://doi.org/10.1029/2024GL110059
Citation: https://doi.org/10.5194/egusphere-2026-4751-RC1
RC2:
'Comment on egusphere-2026-4751', Anonymous Referee #2, 20 Oct 2026
Review of the manuscript
“Composite Detection of Continental Shelf Fronts from Sea Surface Temperature and Altimetry: Assessing the Added Value of SWOT KaRIn”
This manuscript examines shelf fronts in the Southwestern Atlantic using a composite SST-ADT approach, with particular attention to the contribution of SWOT. I found the comparison across different frontal environments interesting, especially the contrasting results for the shelf-break and coastal fronts. The study addresses a relevant question for Ocean Science, and the workflow is easy to follow. My main concern is how the differences in joint front probability are interpreted. The results show that some products more frequently produce ADT gradients near SST fronts, but I am less convinced that this alone demonstrates better front detection, or that the differences can be attributed specifically to KaRIn. Some additional analysis would help here, together with a more careful description of what the metric measures. Several methodological details also need clarification. I think the paper has merit, but I would recommend major revisions before publication. My main comments are below.

Major comments
Sections 2.2.2 and 4; Fig. 12: As I understand Eq. (6), P_joint measures how often the two gradient criteria are satisfied within the chosen spatial tolerance. It does not directly measure detection accuracy. For example, a broader or more persistent area of elevated ADT gradients could increase P_joint without necessarily improving the location of a front. The agreement with previously documented frontal regions is encouraging, although I do not think it is sufficient to infer that false detections are uncommon, as suggested in lines 200-208.

It would help to see the SST-only, ADT-only, and joint occurrence fields together, so that the reader can assess what the composite criterion adds and which features it excludes. For a few representative fronts, a comparison with independent observations, where available, would make the interpretation more convincing. A controlled test with known front locations could also be informative. I am not suggesting that an extensive new validation dataset is essential, but some assessment beyond maximum P_joint seems important if improved detection is to remain a central claim. If independent validation is not feasible, a more limited interpretation in terms of frontal co-occurrence would still be useful.

Section 2.1.1; lines 393-402: I was not entirely clear whether CMEMS and SWOT MIOST differ only in the inclusion of KaRIn, or whether there are also relevant differences in their mapping and upstream processing. This matters for the paper's main conclusion. A compact table giving the exact product versions, contributing missions, DT/NRT status, mapping method, MDT, and corrections would help. It would also be useful to confirm explicitly that the CMEMS release used here contains no KaRIn observations.

Ideally, the comparison would use otherwise matched reconstructions with and without KaRIn. I appreciate that these may not be available to the authors. In that case, I would suggest acknowledging the other processing differences and being more cautious about assigning the full difference to SWOT. Ballarotta et al. (2025) “Integrating wide-swath altimetry data into Level-4 multi-mission maps” seems particularly relevant to this discussion.

lines 145-153; Appendices A and B: The percentile approach is straightforward, but I had some difficulty determining exactly which samples enter the threshold calculation. Are the percentiles calculated over the entire space-time domain, and are they then held fixed for the seasonal analyses? Is the same spatial mask used for all products? Giving the actual thresholds in physical units, along with their calculation domain, would make the comparison easier to assess.

I am not fully persuaded by the narrower dynamic range of ADT gradients as the explanation for using a lower percentile, since rescaling gradient magnitudes alone would not change their percentile rank. The choice may well be reasonable in practice, but its justification would be stronger if linked to the detection results. Fig. B1 is helpful in this respect. However, it concerns SWOT MIOST in DJF, whereas the broader conclusion depends on differences among all four products. I would suggest extending the sensitivity analysis to check whether the main product comparisons remain similar over a reasonable range of thresholds and matching radii.

Eqs. 3-6; Fig. 12: A little more detail on the transition from the 0.05-degree OSTIA grid to the 0.125-degree analysis grid would be helpful. In particular, are SST gradients computed before or after remapping? How are coastal boundaries and missing values treated? The physical extent of the matching neighbourhood also varies with latitude, which is worth making explicit.

I noticed a difference between Eq. (6), which uses a fixed T, and the public example notebook, where seasonal probabilities are divided by valid-sample counts while the annual plot uses a direct mean of the binary mask. This may not affect the results where coverage is complete, but it would be helpful to clarify which denominator was used in the manuscript, especially for the swath analysis, and how validity is handled when an SST match lies in a neighbouring cell.

Section 4.6; Fig. 11: There appears to be a date inconsistency: Section 2.1.2 gives March-July 2024 for Cal/Val, whereas Section 4.6 gives March-July 2023. I assume this is a typographical error, but the actual dates and cycles should be clear. The phrase “filtered SSH + MDT” would also benefit from the exact variable name, since SSH and sea-level anomaly are not equivalent here.

The L3 results are promising, but I found the processing description too brief to judge the comparison fully. My concern is that larger L3 gradients could reflect both finer ocean signals and residual errors, particularly near the coast in this macrotidal region. A representative matched transect, perhaps accompanied by a check of sensitivity to filtering or quality control, would help support the interpretation.

Sections 4.2-4.5 and Conclusion: The discussion of regional mechanisms is useful, and I agree with the caution expressed in lines 240-244 about inferring causality. Co-located SST and ADT gradients are consistent with a frontal circulation, but do not uniquely identify a baroclinic density front or its forcing. Salinity compensation and the contribution of background or barotropic flow may also matter. Finally, Fig. 9 marks a transect, but I could not find the corresponding profiles. These would be useful for assessing the peak around 100 km offshore and the claim of improved frontal sharpness (lines 259-265 and 408-410).

Specific comments
The abstract, lines 15-17, describes the Cal/Val application as future work, although it is already included in Section 4.6. The wording could be brought into line with the analysis presented.

Fig. 4 appears to compare products in DJF, rather than different seasons as stated in the caption. The description of the shelf mask as “continental shelf (outside the 200-m isobath)” is also confusing. The reference to Fig. 4 as an annual mean in lines 228-229 seems to need correction.

A latitude value is missing in line 220, after “South of approximately”.

The statement that GLORYS12v1 does not capture the San Matías front (lines 248-250) seems too categorical given the reported maximum of 35.16%. Would “a weaker or less complete representation” describe the result more accurately?

The third row of Fig. B1 is not explained in the caption. Identifying all three regions and the product shown would help.

For Fig. 10 and the summary statistics, it would be useful to indicate where the quoted maximum occurs and whether it belongs to the Magellan plume front or the Tierra del Fuego tidal front. I was not always sure how the two features were separated in the comparison.

Some labels and annotations are difficult to read, particularly in Figs. 11 and B1. Slightly larger text and consistent labelling of probabilities versus percentages would improve readability.

I suggest removing the title “Bathymetry of the Study Area” from Figure 2.
Citation: https://doi.org/10.5194/egusphere-2026-4751-RC2

Marie-Christin Juhl, Marcello Passaro, Denise Dettmering, Michael G. Hart-Davis, and Martin Saraceno

Data sets

Daily gridded Sea Level Anomalies from satellite altimetry using Spatio Temporal Window Kriging Marie-Christin Juhl, Marcello Passaro, Denise Dettmering https://doi.org/10.17882/103947

Interactive computing environment

Composite Front Detection (Jupyter Notebook) Marie-Christin Juhl et al. https://github.com/mariejuhl/Composite_Front_Detection

Marie-Christin Juhl, Marcello Passaro, Denise Dettmering, Michael G. Hart-Davis, and Martin Saraceno

Viewed

Total article views: 388 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
217	130	41	388	54	40

HTML: 217
PDF: 130
XML: 41
Total: 388
BibTeX: 54
EndNote: 40

Views and downloads (calculated since 13 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	32	14	4	50
Sep 2026	85	60	18	163
Oct 2026	93	52	17	162
Nov 2026	7	4	2	13

Cumulative views and downloads (calculated since 13 Aug 2026)

Month	HTML	PDF	XML	Total
Aug 2026	32	14	4	50
Sep 2026	85	60	18	163
Oct 2026	93	52	17	162
Nov 2026	7	4	2	13

Viewed (geographical distribution)

Total article views: 373 (including HTML, PDF, and XML) Thereof 373 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 03 Nov 2026

Short summary

Ocean fronts are boundaries where water masses meet, influencing how heat, nutrients, and marine life move across continental shelves. Satellites have long tracked these fronts using sea temperature and sea level, but in coastal waters, they remain difficult to resolve. We combine both signals to map fronts along the Patagonian continental shelf, showing that a new high-resolution satellite mission improves detection of coastal and seasonal fronts compared with conventional satellite data.


Total:	0
HTML:	0
PDF:	0
XML:	0