Review Reports

Round 1

Reviewer 1 Report

Minor comments;

Line 444-445 "The F2F-NN is extremely accurate in 445 describing the regional structure" 

Line 460-464

"which can roughly describe the true Lorenzo wind field structure.  The wind field of F2F-NN (Figure 7b) is very similar to the true value (Figure 7a), the hurricane eye structure is accurate, the RMW is in perfect agreement with the true value,  and the structure of the high wind area around the hurricane is close to the true wind field. "

Please change this everywhere (bold) in this manuscript and replace it with actual numbers.

 

Major Comments:

 1- Section 2.3 Buoy Data (NDBC and TAO)

You validation data set is very small, you need to include the SFMR data for Tropical cyclone.

 

 2- Subsection  2.5.3. . Height Match of Wind Field

It is important to make sure that the time averaging of your data set is the same. Please include a subsection that convers the time averaging and height match.

 

3- Need to include the time in Fig 7 and 8, also you need to add a zoomed in version of these figures. 

 

4- Remove figures 8 c, d, and f, they do not help. Replace them with  a high resolution regional model surface wind analysis.

 

5- The whole discussion in section 4.3 needs to be redone. The conclusion is wrong. The reasoning is incorrect (e.g Fig7 and 8).

 

6- For tropical cyclones, you need to have high temporal and resolution validation data set.

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

see attached.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have satisfactorily addressed most of my concerns.

Thank you.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

I acknowledge the revisions the authors have made. All of them have been addressed adequately except a couple:

1. Regarding Figure 6, this figure is fine and can be kept in the paper. But I still believe that a global map of average retrieved wind, including a global map of the wind speed error (your retrieved wind - collocated model wind such as from ECMWF ) would complement the paper well. The authors have included global statistics and a case study to validate their product. A global geographical distribution of the error (for the whole test period) is what is missing.

2. You did address comment 2 in the text. It is noted. However, I think you should include more explicit information about the actual computational cost. For example. how long does it actually take to process a whole day, or a full orbit from fy-3d data? That's not really clear to me.

Once these two comments are addressed, I believe the paper can be published.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

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

 

The discussion of the training of the neural network is adequate, and I have no significant issues except that the English grammar (throughout the paper) is not particularly good.

Concerning the evaluation results (retrievals vs ERA5 or buoys), I have more reservations than agreement. The issue of rain impact on the passive microwave wind speed retrieval is not adequately addressed. The history of passive remote sensing has been that wind speed cannot be adequately retrieved in the presence of rain. Not only is rain the dominant signal in linearly polarized Tb's (compared to WS, SST, WV and CLW), but also there is the issue of the spatial distribution of rainfall within the sensor IFOV (beam-fill fraction). This is the fatal flaw in trying to make WS retrievals within the coarse resolution of the passive microwave antenna IFOV's. Further, ERA5 is a mesoscale numerical model that cannot model rain; so using this as surface truth (especially in tropical cyclones) is not creditable. Now the argument is that rain globally is only about 5%; but in the tropics (ITCZ) rain can exceed 25% spatial/temporal coverage. I cannot see how rain can be neglected in the NN training and retrieval estimation. I suggest that the global average statistics presented are misleading because they neglect rain. If the authors could present WS retrieval statistics as a global image, then it would allow me to assess the impact of rain in the tropics.

 

Reviewer 2 Report

Major comments;

a- Line 440-441 "Compared with the traditional wind speed retrieval method, this method does not need to rely on complex geophysical models". This is a problem and it shows in Figure 8. You need to use real examples of cyclones and make adjustments to your method.

b-Line 362 "Although the amount of training data in this wind speed area is relatively small, the accuracy of the retrieval results is excellent". This is not correct, please replace it with numbers. 6(g) and 6(h) are different in location.

c- Line 418 " and the description of the cyclone structure is closer to reality structure Fig 7". Do you have real examples? 

d- Line 472-473 "F2F has a clearer and more complete description of the high-speed wind field structure (including the typhoon structure)." This is wrong. Please see the above comments.

You need to read on real examples of typhoon structure, then make the necessary adjustments in your manuscript.

 

Minor revision:

a- "Figure 1. Locations of TAO (blue)...". It is not showing blue.

b- Line 509-510 "(bright temperature value) and the sea surface wind speed more accurately". Please correct.