Review Reports - Developing a Quality Flag for SAR Ocean Wave Spectrum Partitioning with Machine Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study proposes a Machine Learning approach that assigns a multi-level Quality Flag with SAR-derived swell partition using collocated WaveWatch III data, improving partition accuracy and reliability.

Line 64
"These discrepancies are largely attributed to the motion of ocean waves with radial velocity components..." What is the effect of correcting radial velocity in existing methods? Can ML method better replace existing methods in terms of data correction accuracy and data applicability for address such non-linear effects?

Line 134
"The accuracy of the model's wave parameter has been validated against buoy and altimeter measurements..." The citations also mention that four sets of parameterizations available in the WAVEWATCH III modeling framework are compared with the wave partition exported from SAR. Has the relative accuracy of the WW3 model and SAR data been considered in ML methods? Will ML methods introduce errors from the WW3 model itself into SAR?

Line 174
"This combined error, denoted as Cerror and defined in Eq. 6, is calculated by multiplying the three primary error estimates..." Can the combined error in product form reasonably reflect the combined effect of the individual errors of three variables? For example, some extremely small values close to 0 will completely eliminate the errors of other variables in the product, so that even if there are large values of other errors, they cannot be reflected in the combined error.

Line 377
"Poor: This class is more frequently observed at high latitudes in both hemispheres..." The strong winds and variable wind speeds in high latitude regions (prevailing westerlies), as well as complex atmospheric phenomena, also have an impact on the data quality of the WW3 model. Is it complete that the source of Poor QF is only attributed to SAR? How to consider the relationship between the real situation, WW3 model, and SAR data.

LIne 404
The citation and Fig numbers in the text are displayed incorrectly in "?".

Author Response

Dear Reviewer,

We would like to sincerely thank you for your careful review and constructive feedback on our manuscript. Your comments have been very helpful in improving the clarity and quality of the paper.

Please find attached a detailed response document, where we address each of your remarks point by point.

We remain at your disposal for any further clarification and hope the revised version meets your expectations.

With best regards,
Amine Benchaabane
on behalf of all co-authors

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Recommendation: major revision

General

This paper proposes a novel method for developing a Quality Flag (QF) for Sentinel-1 SAR ocean wave spectrum partition data using a machine learning algorithm (XGBoost). It combines this with SHAP for model interpretability to gain deeper insights into the physical features affecting SAR wave retrieval errors, aiming to solve the problem of how to effectively assess and label the quality of partitioned wave data.

In general, the analysis presented in this paper is significant and has promising applications, but in some cases further specification of details would add value to the manuscript; please see below for details. The English can be improved and is confusing at times.

Detailed suggestions

There are too many keywords; it is recommended to have no more than five.

In line 44, the term "WM" is used. Its function appears identical to the "WV" mode described earlier in the text. Furthermore, "WM" is not defined in the list of abbreviations on page 21. This appears to be an error and needs to be corrected.

The term "partitioned Significant Wave Height (Hs)" is used on line 57, while "partition-effective Hs" is used on line 309. Please confirm whether these two terms refer to the same concept and ensure consistency throughout the manuscript.

In section 3.1, why was multiplication chosen as the method for calculating the combined error metric (Cerror) instead of other forms? The product form has a potential drawback: if any single error term is close to zero, the combined error will be small, even if the other two terms are very large. This could lead to a misjudgment of the partition's quality. The rationale for this choice needs to be explained.

The entire methodology is based on defining errors with respect to the WW3 model data as the "ground truth." However, relying solely on model data for validation may introduce bias. Although it is mentioned in section 3.2 that "we place greater trust in the observational data as a more reliable reference," a more in-depth analysis is required. The inherent limitations of this approach should be explicitly stated. Could other independent observational data (such as buoys or altimeters) be introduced to further validate or optimize the QF definition?

There are several placeholder citations in the manuscript, such as [?] and Fig. ?? in lines 404-406. It is essential to carefully proofread the manuscript and correct these placeholders.

Author Response

Dear Reviewer,

We would like to sincerely thank you for your careful review and constructive feedback on our manuscript. Your comments have been very helpful in improving the clarity and quality of the paper.

Please find attached a detailed response document, where we address each of your remarks point by point.

We remain at your disposal for any further clarification and hope the revised version meets your expectations.