Skip to Content
MDPIMDPI

Advanced

by
  • Maxime Beauchamp 1,*,
  • Ronan Fablet 1 and
  • Bertrand Chapron 4
  • et al.

Reviewer 1: Anonymous Reviewer 2: Anonymous Reviewer 3: Anonymous

Round 1

Reviewer 1 Report

The paper “Intercomparison of data-driven and learning-based interpolations of along-track Nadir and wide-swath Swot altimetry observations” by Maxime Beauchamp et al presents comparison of some interpolation and a learning-based methodologies applied to pseudo observational altimetric along-track nadir and wide-swath SWOT datasets for the reconstruction of sea surface height (SSH) spatio-temporal fields. Performance of an analog data assimilation (AnDA), an up-to-date version of the DINEOF algorithm, and a new neural networks-based end-to-end learning framework has been described on two types of altimetric pseudo observational datasets generated for the two small regions of the North-Atlantic basin: a part of GULFSTREAM and OSMOSIS area. Energetic mesoscale dynamics is peculiar for the GULFSTREAM area while OSMOSIS area is less energetic but with noticeable spatial patterns. Therefore, both are well suited for performance evaluation of relatively small scales features ranging up to 30 … 40 km. From analysis of data some improvements connected with the newly introduced neural networks method regarding spatio-temporal resolution of SSH data is demonstrated.

There are some issues with the paper that the authors needs to address before it can be published in Remote Sensing. My major concern is the clarity of Figures. Font sizes of text and/or numbers are often too small and difficult to read. Not all quantities and units defining axes in Figures are given and/or sufficiently explained in text. Some points in text also need clarification. Maybe it will be helpful to collect into a special table and redefine a large number of rather complicated long abbreviations presently used in text and figures for marking different calculation algorithms. Shorter and simpler abbreviations could improve the figures simplifying legends and texts. This will increase the available space for enlarged charts.

Suggestions and questions for authors.

L78: Suggest: …variance  σ2 = (4 ... 9) cm2 

82-83: …time window tk ± d needs explanation: Is the time window 1 day, if d = 0? Then, if d = 5, the time window likely should be 11 days, see also Figure 3.

Figure 2: I suggest: One and 11 days accumulated along-track nadir and wide-swath pseudo-observations on August 4, 2013. (a,b) and August 5, 2013 (c,d).

Are the observations of Figure 2a performed during a day?

Why different dates for d = 0 and 5, August 4 and 5?

The numbers at axe’s scales are too small.

What is shown in Figure 2 with color scale? Relevant quantity/unit must be given, and explanation in text if relevant.

I suggest in Figures 2 placing 2a and 2b side by side in a first row and 2c an 2d just below. The same color scale is valid for the whole group, thus a single color scale would be sufficient. Joint scales are possible also for geographic coordinates. The same comments apply to Fig. 10.

Figure 3: Suggest: Variance of the Observation error ɛk as a function of the hourly lag between the observations and the day to estimation time. Correct position of zero time for the estimation lag?

Axe’s titles and units are not given. Referring to Figure 2, not only date but also the time of observations should be accounted for to make it agree with Figure 3. The font size of numbers at axe's scales are too small. Two different error levels in Figure 3 need to be explained.

Figure 5 and 6, 11 and 12: Suggest using larger font sizes for text and numbers for both axes, and reformatting date values. Time is shown with dates and not in days. Legend should be simplified, charts enlarged. RMSE unit should be added.

Figure 8 and 9, 14 and 15: Suggest placing enlarged sub-figures more side by side, with the single color scale for all, and adding quantity/unit for the presented color scale.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Most interesting result of the paper is related to evaluation efficiency of a new neural networks-based learning framework approach that was applied successfully earlier to the interpolation of SST data. It is shown that the proposed neural networks-based learning framework approach provides advantage against the standard AVISO OI technique and other data-driven interpolation technologies in filling data free regions. An ability of the neural networks-based learning framework to extend resolved scales up to 50 – 60 km is manifested in case of joint use of nadir and SWOT altimeters which is better than in case of OI application.

Thus, paper results show prospects of a new interpolation methodology to improve quality of interpolated altimetry products after the launch of SWOT mission.

I have some general comments, which may be helpful for better presentation of the paper significance.

 

  1. I think that it may be good to include some simple explanation of advantage of the neural networks-based learning framework application to make the paper reader-friendly for broader range of readers.
  2. It is not clear from the text of the paper how OI technique was applied for interpolation of nadir altimeters and SWOT data. Does OI interpolated an anomaly field with properly evaluated covariance matrix?
  3. It is shown that different kind of the learning strategy should be applied depending on the energy of natural variability with respect to the observational noise. Is it possible to discuss how this fact should be taken into account in a practical algorithm?
  4. Authors have a hope that the use of longer learning base will increase efficiency of the proposed methodology. However, do you need some kind assumption about stationarity of natural processes during the learning period?
  5. I guess that it is possible to omit Appendix A and pictures from Appendix B because there is no not discussion of their differences

 

Some editorial comments

  1. It will be easier to compare graphs if to make the same marks on the left coordinate axe of Fig.5 a,b.. Also good to have the same marks on Fig. 6.
  2. 5 right axe – does it percent or fraction?
  3. It is not clear why the level of barplots for nadir data is different on Fig. 5 and 6?
  4. Color bars does not explained on Fig. 2 and 10.
  5. Is it possible make clearly the difference between 20 –days and 80 -days validation periods that appears in the text?
  6. Is it possible to explain terms “supervised” and “unsupervised” on page 8 should?
  7. Page 11, discussion of Fig. 8,9. Where are “unnecessary artefacts”? What part of pictures reader needs to look at?
  8. Page 14, 2-nd paragraph “and” repeated twice
  9. Fig’s 14, 15 not discussed in the text of the paper

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

The Swot of the title should be capitalized.

Line 7: Please spell out the full name of OSMOSIS here.

Line 13: Please spell out the full name of SWOT here.

Line 56: It is recommended to explain why this area (OSMOSIS) was selected, and what is the significant difference between this area and the ocean physical background of GULFSTREAM?

Figure 1: Please add latitude and longitude coordinates.

Author Response

Please see the attachment

Author Response File: Author Response.pdf