A Neural Network Method for Retrieving Sea Surface Wind Speed for C-Band SAR

Round 1

Reviewer 1 Report

A novel method based on the NN technique and Bayesian regularization algorithm is proposed to retrieve sea surface wind speed for C-band SAR sensors. In general, the manuscript is well prepared. However, there are some issues need to be considered more carefully. 1.How the radiometeric calibration of SAR data, including long-term consistency, is performed?  2.Please show also the std of wind differences. Biases may caused by calibration which is not performed for ASCAT wind used for model training. If the bias is consistent, it can be removed. 3.What are the definition of inshore and offshore?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors addressed ally comments.

Author Response

We would like to thank the effort from the reviewer, which improved the manuscript to a large extent. 

Reviewer 3 Report

The manuscript presents a method for SAR wind retrievals based on Neural Networks. The conclusion is that this new method outperforms the GMF approached, which is widely used at present.

Overall, the manuscript is interesting and it is relevant to apply artificial intelligence in connection with wind retrievals from satellite observations. 

General comments:

1. Effects of atmospheric stability are not properly considered. They could influence the reference wind speeds from ocean buoys as well as the SAR wind retrievals. The findings are therefore less solid.
   
   
2. I think the idea of testing the method over an independent area of interest in China is good. I miss, however, some reasoning over what we might expect to see and how differences between the Gulf of Mecico and the Chinese seas might affect the results.

Please see the attached file for my specific comments.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Overall comments

• While situ observations are more accurate than model/reanalysis – what is the true innovation in this work? There doesn’t appear to be a meaningful change in method or impact by switching data source, particularly because the method appears to require buoy data as input. If you need directional buoy data as an input – where can this method be used and/or extended to? What is the fundamental difference in the method that advances it over previous work?

 

• Line 255: there appears to be a clear outlier causing the really high bias for CMOD5.N – given your small sample size, it may not be appropriate to consider the outlier in the overall performance.

 

• With such a small sample size tested, are the differences statistically significant?

 

 

Line by line comments:

 

Line 21: “unlike most previous”

Line 36: fronts (plural)

Line 71: “and relative”

Line 74: buoy (no s) (and other lines)

Line 89: fix sentence beginning with “And, ..”

Line 103: totally -> exclusively

 

Match verb tense throughout (were vs are)

Figure 2: red markers (?) and labels unintelligible

Line 171 – fix grammar

 

Line 175: accessing -> assessing

 

Line 179: what do you mean by “matchup dataset”? does that mean collocated?

Line 188: rephrase line 188

Line 206: better description of “p is a parameter” – tuning parameter? How is it chosen? Is there a fixed value?

 

Line 214: “labelled”

Figure 3 (the second) :

• mislabelled – scatter plot should be figure 4
• Improve plot: all the points overlap, one cannot see where most of the data points are – replot as a 2D PDF or 2D histogram
• the overall bias might be close to zero, but it appears that high winspeeds are biased low, and low windspeeds appear biased high? This should be assessed more clearly.

 

Line 222: the wind direction from NDBC buoy is used as an input? Therefore, is this method only feasible proximate to a buoy with directional information?

 

Figure 6: why does RMSE appear lower for 12.5 km and 20 mins away?

Line 244: accessed -> assessed

Line 253: OPEM -> OPEN

 

Line 270: prevailing?

Line 274: fix grammar

 

Line 275 – what do you mean “matches the SAR image well” – what features are you comparing?

Line 287: fix grammar

Line 292: fix grammar

Line 293: fix grammar

Line 297-8: fix grammar

 

Line 316: [m] NOT [m/s]

 

Line 319: assess based on relative error then if the two datasets have different magnitudes

Line 326: grammar

 

Comments for author File: Comments.pdf

Reviewer 2 Report

This paper used the neural network method for estimating wind speed using satellite scaterometry and SAR data. The authors used 5 years of wind data from NDBC buoys in the Gulf of Mexico to train the satellite data and then evaluated the results using 2% of the independent data as well as buoy wind data on China coasts.

The paper is relevant to the remote sensing and Oceanography community and can be published in the JMSE after major revisions. Followings are the issues that need to be addressed in the revision:

-Ln 35-36: Reference [2] is not appropriate to address the application of wind data in wave and circulation modeling. Please add more related/relevant references.  You can use the following references one for wave modeling and the other for wave modeling:

 

Allahdadi, M.N., Gunawan, B., Lai, J., He, R., Neary, V.S., 2019. Development and validation of a regional-scale high-resolution unstructured model for wave energy resource characterization along the US East Coast. Renewable Energy 136, 500–511. https://doi.org/10.1016/j.renene.2019.01.020

-Ln 57-59: Although C-band and Ku-band are technical terms in remote sensing since they have been frequently used in this paper, it is better to add a couple of sentences to define and differentiate them.

- Ln 92-94 Present example references for the rare studies that included buoy data for model training.

- Ln 142-148: Based on what rationale the authors selected the 11 NDBC buoys shown in figure 1 for model training? In the GoM there are many other buoys with more complete data history 9like 42002, 42003, 42036, and 42039). Wind and wave data from these stations have been extensively used for wave modeling (see Allahdadi et al., 2021 as cited below):

 

-Ln 142-148: Please add a table in the paper that includes the coordinate and depth of the buoy data in both GoM and China coasts.

- Figure 1: the station numbers ae small and not clear.

-Figure 2: No station number/name is specified for the buoy data along the China coast.

- Section 2.3: More details about the implementation of the NN method (OPEN)are required. Specifically, the authors need to add a schematic showing the NN layers and Neuron connections as well as input and output data. See the following pape:

 

James, S.C., Zhang, Y., O’Donncha, F., 2018. A machine learning framework to forecast wave conditions. Coastal Engineering 137, 1–10. https://doi.org/10.1016/j.coastaleng.2018.03.004

 

-Section 2.4, equation (3):  the relationship needs more details. What is the final equation for estimation wind speed? What about wind direction?

-Section 3, Results: What about wind direction? Did you implement the data training to get estimations for the wind direction?

 

Comments for author File: Comments.pdf

Reviewer 3 Report

My assessment is that the authors did not present their research work in a clear and easy-to-follow manner. While the text might be written with sufficient grammatical correctness, the manuscript is filled with vague and what seems to be unsubstantiated claims and statements.

The section 2.3 “The OPEN method for C-band radars” lacks sound description and substantiated rational on author’s methodology of how to set up and configure the NN OPEN learning. The one 10-line paragraph of this section contains only a fair amount of hand-waving statements which should not be acceptable in a scientific paper.  

In my opinion the (poorly) presented research does not advance the science and understanding of wind scatterometery. The two most important graphs in this manuscript (Figure 5 and 7 in section 3 “Results”) show comparisons between winds retrieved using NN OPEN and CMOD functions against buoy measured winds. Visual inspection of the graphs in Figure 5 and 7 suggests that the claimed improvement in the wind retrievals by NN OPEN for the entire range of winds (2m/s to 18 m/s) for the sample data set is a result that for a small subset of data points [buoy winds above 14 m/s and up to 18 m/s] the NN OPEN estimated retrieved winds have a saturated value of 14 m/s;  while for the same data points (buoy winds between 14 m/s and 18 m/s ) CMOD retrieved winds are in a range between 14 m/s and 20 m/s. It is my assessment that this claimed improvement does not represent any valuable advancement in interpretation or accuracy of ocean wind retrievals from scatterometer data.  

The phrase “for C-band Radars” that is being used throughout the manuscript and in the title, suggest that this method is limited only to C-band radars, while the authors failed to explain that limitation. Also, the more appropriate designation for this type of instrument is a scatterometer rather than radar.

Based on the lack of scientific merit I recommend that this manuscript is rejected for publishing. 

Reviewer 4 Report

Dear Authors,

I enjoyed reading the draft. It's well written and easy to understand. I'd recommend to check the language one more time. Overall I'm more than satisfied with this study.

Your Reviewer