Generation of Non-Linear Technique Based 6 Hourly Wind Reanalysis Products Using SCATSAT-1 and Numerical Weather Prediction Model Outputs

Round 1

Reviewer 1 Report

This paper describes what appears to be promising method to construct a wind analysis by combining measurements from SCATSAT-1 with NWP model output.  Unfortunately, the paper contains several major shortcomings that make it impossible for me to recommend it for publication without major revision. 

1.        The descriptions of the particle filter technique are confusing and incomplete.  After reading the description several times, I am left with the following questions:

a.       Do the final particles used include both overall biases *and* spatially dependent biases??

b.       How was the number of particles chosen?  Did the authors do any sensitivity tests to determine this number?

c.       What is the spatial structure of the chosen particles?  A figure could be helpful here?

d.       Were any other sources of particles considered?  Why chose the monthly mean differences??

e.       Does any information from the previous time step carry over to the analysis step in question, or does the process start over each time step?

f.        I don’t understand how the multiple copies of the high-likelihood particles impacts the analysis.  Is there a dynamical step that was not described?

2.       Figures 7 and 8 have strong signatures of precipitation influence.  Does the SCATSAT-1 data have a rain flag or rain correction?  Was it used in this analysis? There is also a lot of vertical striping in these figures.  If this due to residual cross-scan biases in the satellite data or ??

3.       Figure 11.  I would not say that the PF winds match well to the other winds.  The PF winds show much, much fewer winds in the 0 - ~4 m/s range.  I think this would be more obvious is the histograms we normalized so that the area under the curves is equal.

4.       Figure 12.  This figure is hard to evaluate.  I recommend plotting binned means of the wind mean and standard deviation of the wind difference.  The data should be binned by the average of the two winds speeds to avoid statistical effects at low and high wind.

Some other issues:

Sometimes it appeared that words disappeared between pages.

Line numbers sometimes were not there, and then resumed on a later page.

The version of CCMP used is not described, but it appears to be version 2.0.  Versions 3.0 is now available.

Author Response

The authors would like to thank Reviewer and the Editor for a prompt response. The comments provided by the esteemed reviewer and the editor improved the quality of work carried out in this paper. We incorporated all the suggestions provided in the revised manuscript.  Responses to all the comments and suggestions are given below.

Reviewer-1: Authors are grateful to reviewer-1, whose suggestions have strengthened the validation part of the study. All the suggestions provided are incorporated as given below. 

This paper describes what appears to be promising method to construct a wind analysis by combining measurements from SCATSAT-1 with NWP model output. Unfortunately, the paper contains several major shortcomings that make it impossible for me to recommend it for publication without major revision. 

1. The descriptions of the particle filter technique are confusing and incomplete. After reading the description several times, I am left with the following questions:

Ans: Thank you for pointing out the mistakes. We have revised parts of the description in the revised manuscript to make it clearer for the reader. We highlighted the revision in yellow.

1. Do the final particles used include both overall biases *and* spatially dependent biases??

Ans: Initially, as part of the sensitivity study, we applied a constant bias to all grid points and generated 128 particles. These constant biases are 128 random numbers generated between the bounds of standard deviation of the wind field. At the end of this exercise, the RMSE between observation (SCATSAT-1) and Background (NCMRWF winds, in this case) is governed by the bounds of these constant biases. The result of applying particle filter on these particles yielded wind field that were more skewed to the background field. Henceforth, we tried using spatially varying biases, which we constructed from the monthly mean differences between the winds from SCATSAT-1 and Background. After getting the encouraging result from the second experiment, we continued with only spatially varying bias. This point is now more clearly written in the revised manuscript

1. How was the number of particles chosen? Did the authors do any sensitivity ests to determine this number?

Ans: We considered 128 particles in the first experiment as an arbitrary guess. We generated these particles by adding up a small constant bias between -1m/s to +1 m/s (which corresponds to standard deviation of the background field) to the original background field. The minimum RMSE between SCATSAT-1 and background was for a bias of around 0.4 m/s, which is somewhere between the minima and maxima of the additive bias. The maximum RMSE was however found at -1m/s. (figure- 3). We then increased and decreased the number of particles and found the same outcome every time. Thus, the distribution of the bias between the upper and lower bound is more crucial than considering many particles.  So, when we started using monthly mean difference as the additive bias, we took 20 particles with an increment of 5% of the monthly bias so that we could accommodate 100% of the bias. This point is now mentioned in the revised paper.  

What is the spatial structure of the chosen particles?  A figure could be helpful here?

Ans: The spatial structure of the mean perturbation for the initial population of the generated particles is 52.5% of the monthly mean difference field i.e. Mean of the 5% to 100% of the bias. However, after the application of the filter procedure, the set of particles changes. Some weak particles whose weights are less are rejected and some high value particles are repeated, such that the total number of particles are same, in this case 20. Thus, the mean perturbation of the final set of population vary largely. For example, on May 30, 2018, the mean bias of the final population was 69.75% and for 31st May 2018 it was 76.25% of the monthly mean difference. We have now incorporated figure 5b and tried to explain the same in the revised text.

1. Were any other sources of particles considered?  Why chose the monthly mean differences??

Ans: The particles from other source i.e.  by putting constant biases were used initially but the generated fields are closer to the background. So, the global difference between observation and background are considered as additive bias, so that the background field can absorb these biases and particles similar to SCATSAT-1 field can be generated.

Further, these global differences between observation and background biases are taken on a monthly scale because on daily scale we do not get global coverage of SCATSAT-1 as it has 2-day repetivity. A few days of bias can be influenced by high wind over several locations across the globe facing tropical/subtropical storms. Even winds have great seasonal variability, so taking bias at more than a month scale would not be suitable for the study. Thus, we preferred mean differences on the monthly scale. We also added this point in the data and method section.

 Does any information from the previous time step carry over to the analysis step in question, or does the process start over each time step?

Ans: This is an excellent suggestion and authors are thankful for this. As of now this process starts at each time step. This point is now mentioned in the methods section. However, in the future, we would definitely attempt to look for the impact of temporal carry over and assess an analysis window for the process.

1. I don’t understand how the multiple copies of the high-likelihood particles impacts the analysis.  Is there a dynamical step that was not described?

Ans: The multiple copies of a high-likelihood particle would eventually replace a low-weight particle so that the total population of 20 particles is constant. This impacts the mean percentage of bias that is to be introduced in the background field. More details of how we selected the particles are given in the particle filter and its implementation section.

2. Figures 7 and 8 have strong signatures of precipitation influence.  Does the SCATSAT-1 data have a rain flag or rain correction?  Was it used in this analysis? There is also a lot of vertical striping in these figures.  If this due to residual cross-scan biases in the satellite data or??

Ans: The reviewer has rightly pointed out that there are influences of precipitation and cross-scan bias in both figures. SCATSAT-1 L2B data have rain-corrected wind speed, which we used in our analysis besides using the “good quality” data flag. We mentioned this in the result and discussion section.

However, the cross-scan bias is still present in the SCATSAT-1 data and its sources are yet to be known. This caused stripping in L2B data of SCATSAT-1.  Even L4AW daily gridded analysed field have embedded features of this bias. One of the positive aspect of PF based technique for generation of wind is that it removes this scan dependent biases from the final wind product.

3. Figure 11. I would not say that the PF winds match well to the other winds.  The PF winds show much, much fewer winds in the 0 - ~4 m/s range.  I think this would be more obvious in the histograms we normalized so that the area under the curves is equal.

Ans. The reviewer is absolutely correct, in lower wind range the PF wind has a fewer number of points. We have now changed the figure -11 in revised manuscript following the suggestion. We have normalized the histogram. Further we have also added L2B wind speed normalized histogram with the previously plotted L4AW, BUOY and PF-WIND.   The revised plot shows that L4AW matches the normalized count of buoy observations in lower wind range, however if we consider L2B data of SCATSAT-1 the PF wind match is better.

4. Figure 12.  This figure is hard to evaluate.  I recommend plotting binned means of the wind mean and standard deviation of the wind difference.  The data should be binned by the average of the two wind speeds to avoid statistical effects at low and high wind.

Ans. This figure has now been changed based on the suggestion of the reviewer.

 Some other issues:

5. Sometimes it appeared that words disappeared between pages.
6. Line numbers sometimes were not there, and then resumed on a later page.

Ans: We rectified all the above issues.

7. The version of CCMP used is not described, but it appears to be version 2.0.  Versions 3.0 is now available.

Ans. The reviewer is absolutely right. This is indeed Version -2 of CCMP data and is mentioned in the revised manuscript. We would surely attempt to validate the PF winds using Version 3 CCMP data in our next analysis. Reanalysis using version-3 cannot be done in the limited time given.

Reviewer 2 Report

Authors have made an attempt to combine the ocean surface wind observations acquired from the scatterometer satellite (SCATSAT-1) with wind field from a numerical weather prediction (NWP) model at a global level for the year 2018. The authors have utilized the particle filer technique for combining both data. As far as validation is concerned, SCATSAT-1 L4_analyzed wind products were utilized. The outcome of the research work is important in terms of the many ocean applications. However, some of the observations are as follows:

1.     In the title of the manuscript, the word SCATSAT-1 must be there instead of using satellite.

2.     Mention the full name of SCATSAT-1 in the abstract and each abbreviation in the keywords must be specified.

3.     Authors have missed the international status of the scatterometer in the introduction section. One dedicated paragraph is required on the same. Pls check it out this one: https://doi.org/10.1109/MGRS.2022.3145500 or Liu, W.T., 2002. Progress in scatterometer application. Journal of Oceanography, 58(1), pp.121-136.

4.     P2L53, NWP? Make sure that each abbreviation must be specified (full form) as it arrived first in the manuscript (exclude the abstract or key words). Check the same error throughout the manuscript.

5.     Under Section “Data & methods”, specify the date and time of data acquisition.

6.     P4, Fig. 1, Unit of the legend is missing.

7.     P5L132, Equation number error.

8.     P6, I think figure 2 could be more elaborated in details especially green block.

9.     P7, paragraph must be split into two, one is based on particle generation and another one is on resampling.

10.  P7, last few lines are not on suitable places such as there is not need to mention the methodology part as you already explained.

11.  P7, Recheck the last line, something is missing.

12.  P8, Figure 3 needs to be revised with small dot on the graph.

13.  Results section is ok but unit of legend is missing in all figures.

14.  Error in Line 260, I,n…

15.  Reference part is too much weak. It is due to international status of the scatterometers and their contribution with advanced algorithms in oceanography is missing.

Author Response

Authors have made an attempt to combine the ocean surface wind observations acquired from the scatterometer satellite (SCATSAT-1) with wind field from a numerical weather prediction (NWP) model at a global level for the year 2018. The authors have utilized the particle filer technique for combining both data. As far as validation is concerned, SCATSAT-1 L4_analyzed wind products were utilized. The outcome of the research work is important in terms of the many ocean applications. However, some of the observations are as follows:

Authors would thank to reviewer 2 for his motivating words and creative suggestions. All the suggestions provided are incorporated. Kindly find the response to each point. 

1. In the title of the manuscript, the word SCATSAT-1 must be there instead of using satellite.

Ans: Following the suggestion, the title of the  manuscript is changed to ‘’Generation of non-linear techniques based 6 hourly wind reanalysis product using SCATSAT-1 and numerical weather prediction model outputs’’

2. Mention the full name of SCATSAT-1 in the abstract and each abbreviation in the keywords must be specified.

Ans: The Scatterometer Satellite -1 is popularly called as SCATSAT-1. It is now mentioned in abstract.  We now elaborated the abbreviations in the keywords.

3. Authors have missed the international status of the scatterometer in the introduction section. One dedicated paragraph is required on the same. Pls check it out this one: https://doi.org/10.1109/MGRS.2022.3145500or Liu, W.T., 2002. Progress in scatterometer application. Journal of Oceanography, 58(1), pp.121-136.

Ans: Following this suggestion, we have now incorporated a dedicated paragraph in introduction section   on international status of scatterometer of revised manuscript. We also added the suggested references.

4. P2L53, NWP? Make sure that each abbreviation must be specified (full form) as it arrived first in the manuscript (exclude the abstract or key words). Check the same error throughout the manuscript.

Ans: Thanks for pointing to this mistake. We provided full form of   each abbreviation in the revised manuscript. 

5. Under Section “Data & methods”, specify the date and time of data acquisition.

Ans: Done  

6. P4, Fig. 1, Unit of the legend is missing.

Ans: Legend is added in figure -1 in the revised manuscript.

7. P5L132, Equation number error.

Ans: This is corrected now.

8. P6, I think figure 2 could be more elaborated in details especially green block.

Ans: We have now elaborated  figure -2 in revised manuscript  

9. P7, paragraph must be split into two, one is based on particle generation and another one is on resampling.

Ans: This is done.

10. P7, last few lines are not on suitable places such as there is not need to mention the methodology part as you already explained.

Ans: These lines are removed in revised manuscript.

11. P7, Recheck the last line, something is missing.

Ans: We modified these sentences in revised manuscript. Sorry for the mistake.

12. P8, Figure 3 needs to be revised with small dot on the graph.

Ans: We replaced Figure -3 following the suggestion. Thank you.

13. Results section is ok but unit of legend is missing in all figures.

Ans: We added units of legend to all the figures in the revised manuscript. Sorry for missing the legend.  

14. Error in Line 260, I,n…

Ans : Corrected.

15. Reference part is too much weak. It is due to international status of the scatterometers and their contribution with advanced algorithms in oceanography is missing.

Ans: A paragraph on international status and their contributions in oceanography is now added in revised manuscript. We also added suggested references. 

Round 2

Reviewer 2 Report

The authors have addressed all the queries.