EddyDet: A Deep Framework for Oceanic Eddy Detection in Synthetic Aperture Radar Images

Round 1

Reviewer 1 Report

This paper presents a deep learning network EddyDet for eddy detection in SAR images. This work is of practical significance for operational SAR-based eddy detection. However, the following comments have to be addressed in a minor revision.

1. How to set alpha and beta in equation (1)?

2. In Figure 6, please present visualization results of other models.

3. The EddyDet adopts SAR sub images range from approximately 600 × 600 to 1200 × 1200 pixels as input. Why not use a uniform size? Why not use the whole SAR image as input? This is more convenient for practical application. Although this might cause memory issue, you can reduce memory usage by downsampling following Khachatrian et al. (2023).

Ref: Khachatrian, E., Sandalyuk, N., & Lozou, P. (2023). Eddy Detection in the Marginal Ice Zone with Sentinel-1 Data Using YOLOv5. Remote Sensing, 15(9), 2244.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

-       Manuscript will benefit from additional proofreading and editing for clarity; the manuscript is frequently wordy to an unnecessary degree.

-       Authors need to ensure that all acronyms are defined

-       All eddy detection methods and datasets have limitations and error. Authors should temper expectations in the introduction and make claims such as their method being able to detect every single eddy in the ocean perfectly. This is improved in the results and discussion.

o   This paper varies wildly in its categorization of success. The first half makes it seem as if this eddy detection system is near-perfect, but the discussion seems to only include negative results.

-       Figures 4/5/6: while the colored shapes are useful for indicating where the eddies are, would it not be more useful to show the actual eddy contours as detected by the algorithm? Or are these the contours?

-       Description of eddy detection (around Figure 5) could be clearer.

-       Immediately before Section 3.1.3, the authors say that they adopted Gaussian filtering and then immediately say that they abandoned it. Is Gaussian filtering utilized in the final product? If so, what are the details (i.e., half-widths, etc.)? If not, then why was it abandoned? Does this mean that absolutely no planetary wave filtering is included?

-       What exactly is produced by this eddy tracking system? Eddies are “detected”, but what exactly is the output? Contours? Are eddy trajectories tracked?

-       Would there be a way to combine your results with another satellite dataset to better improve identification? What about HF radar? Wind speed is mentioned, but is this incorporated? Or is that future work?

Manuscript would benefit from additional proofreading for grammar and clarity. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

In general, I found this manuscript an interesting and new contribution to the SAR eddy detection field.

However, the manuscript has certain shortcomings and could be significantly improved by the authors.

I would recommend major revision before publication in MDPI Remote Sensing.

General comments:

Major

1.     You have not provided any information about the study region. Is there any reason why did you choose the Western Mediterranean Sea region? What is so special about it in terms of eddy variability? What are the mechanisms of eddy generations in this region?  If you are developing an eddy detection algorithm that can be potentially used for any area of the World Ocean why not choose the more dynamically active region such as Gulfstream or Agulhas Current?  These are very important points that you completely missed in the paper.

2.     The SOED dataset consists of subsets of SAR images which has eddy instances. It seems that such an approach for the construction of the dataset oversimplifies the process of model training and eddy detection. What happens if I feed to the algorithm the whole SAR scene for the study area? Could you please elaborate on this matter?

Minor

Note to all figures with the SAR images. The SAR images you have provided in the text do not have any georeferenced information. Please provide this information either for every image in the manuscript or add another illustration that has information on the total geospatial coverage of all the images used in this study.

Ln. 40.  If SAR is such an ideal data for eddy detection why till this day the main source on the global mesoscale eddy variability are altimetry datasets merged from the multiple missions?  Is it only because of the absence of robust eddy detection methods from SAR images? Or maybe altimetry still has some advantages over SAR data? Please elaborate more on this subject.

Ln. 41-53. This paragraph is supposed to present a brief literature review on the topic of automatic eddy detection. It definitely lacks citations. I feel that authors should do a better work. There are dozens of automated eddy detection and tracking algorithms that were developed over the last 15 years and authors should mention at least such fundamental works as Chelton et al. (2011) (http://doi.org/10.1016/j.pocean.2011.01.002), Faghmous et al. (2015) (http://dx.doi.org/10.1038/sdata.2015.28). It should be also noted that the absolute majority of these algorithms were developed for using with altimetry data.

Ln. 54-65. The same for this paragraph. There are a number of papers on the eddy detection from SAR data have been published in recent years (for example: https://dx.doi.org/10.3389/fmars.2022.1023624, https://dx.doi.org/10.1109/ACCESS.2019.2946852, https://www.mdpi.com/2072-4292/15/9/2244)  but you mentioned only couple of them.

Ln. 111-112. How do you define the outer boundary of the eddy? based on which criteria?

Ln. 132. Again, how does this correspond to eddy sizes in the Western Mediterranean Sea? You have to connect these results to the well-known eddy characteristics in your study region.

Figure 3. Please add units for width and height. Also, the histogram would benefit if you add borders to the bars.

Ln. 154-155. Again, what is your definition of eddy boundary?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The authors have addressed all my questions.