Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study used a convolutional neural network (CNN)-based approach to enhancing SWOT-derived SSHA and geostrophic velocity fields. The results show that the proposed CNN-based filtering strategy can be used in filtering SWOT-derived SSHA observations. However, the results only give one example cycle 7 to present SSHA and geostrophic current, which is not persuasive. Also, the RMSE comparison between SWOT_CNN and others shows little difference. The authors are suggested to do more detailed and in-depth analysis.

Comments on the Quality of English Language

Overall, the manuscript was written well and show clear data/methods descriptions. But the content needs to be improved.

 

Author Response

The manuscript has been revised 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Attached.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript shows a CNN-based approach to filter the SWOT SSHA to get small scale features and achieve a substantial noise reduction in the spectral domain. Although there are other filter methods, here the authors argue that their method works specifically to SWOT measurements.

 

The innovation of the manuscript is not so prominent. There are so many methods for filtering [ref 10-14], how can you judge that the methods in your manuscript are not suitable for the SWOT measurement, or why you use CNN? So, the motivation is too weak.

There are some specific questions.

1. Can the locations of the drifter velocity observations be provided? If the CNN method be operationally applied, the in-situ drifter velocity observations should be used.
2. Is it possible to describe the CNN filtering more specifically? The long paragraph in 3.1seems not logically enough. When the CNN is trained, which is the target data or reference data? DUACS?
3. 6 Reconstruction of the SWOT track should be expressed more specifically. The process of the data seems complicated, so, will it affect the evaluation of the geostrophic velocity, rather than the CNN filtering?
4. How is the “Spectral Noise Reduction” in table 1 calculated? Such as 77.9%, it is ambiguity. Besides, 77.9% is smaller than 81.3%, so we can say SWOT_Gómez is better?
5. There is a mismatch between the numbers of several images and their explanations in Figure 3. Please revise it. Also Figure 4.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The authors argue that "the CNN is trained to minimize the difference between surface geostrophic velocities (computed from the filtered SSHA) and in-situ drifter velocities." It appears that in-situ drifter velocities can measure accurate geostrophic velocities, and that these geostrophic velocities are treated as the reference or target. If this is the case, an explanation should be added in the paper.

Author Response

Comment 1: The authors argue that "the CNN is trained to minimize the difference between surface geostrophic velocities (computed from the filtered SSHA) and in-situ drifter velocities." It appears that in-situ drifter velocities can measure accurate geostrophic velocities, and that these geostrophic velocities are treated as the reference or target. If this is the case, an explanation should be added in the paper.

Response 1. We thank the reviewer for this comment. Actually, we clarified in the manuscript (line 189-193) that "Although drifter velocities obviously reflect the full dynamical components (including Ekman transport and other ageostrophic motions), using this loss function in combination with ancillary wind data (provided as input to the network) proved effective in filtering unbalanced motions."