Monitoring of Nitrogen Indices in Wheat Leaves Based on the Integration of Spectral and Canopy Structure Information

Round 1

Reviewer 1 Report

The article deals with the monitoring of nitrogen content in wheat leaves using combined canopy spectral and structural information. Authors compare the effectiveness of three modelling algorithms: linear regression, PLSR and random forest. The study was carried at a field level.

Remote monitoring of nitrogen status of wheat is important for timely estimation of crop state and development. The idea to use spectral reflectance of crop canopy and structural features to determine nitrogen indices is promising and can have some useful applications. However, there are some inconsistencies in the manuscript that need to be corrected or clarified.

Methodology section needs more detailed description, especially the part on field experiments. Discussion sections needs to be corrected. It should not contain the results.

Please, also see specific comments below.

Specific comments:

Lines 23-24 you meant based on true colour image?

Lines 57-60 – the impact of nitrogen on spectral reflectance of crops is indirect. It influences the state of the crop canopy, which, in turn, affects crop canopy spectral reflectance.

Line 68 – red envelope?? You meant red edge??

Line 70-71 – This is oversimplification. Crop canopy reflectance can contain information on a number of parameters, which can have direct or indirect influence on it. Not only nitrogen and structural information. In some cases, crop canopy reflectance can contain information on soil background. Especially at early stages of crop development.

Line 119 – why linear regression is not mentioned?

Line 122 – informative canopy structural features?

Lines 128-129 – when nitrogen fertilizers were applied (please provide dates)? which fertilizers were applied on each plot?

Lines 125 – 138 – Please, provide a scheme of test plots and experiment.

Line 130-132 – 3 row widths for each nitrogen application rate?

Lines 141-147 – Please, specify the dates of spectral data acquisition. How many spectral curves were acquired? Spectral data was collected for each phase of crop development? How long after the nitrogen fertilizer application?

Lines 145-146 When you measured spectral reflectance of wheat canopy, how you considered the influence of soil background on it? Especially at early stages of wheat development? Because it can impact the results of modelling nitrogen indices based on spectral reflectance of crop canopy.

Line 148 – How many images? When they were taken?

Lines 161-167 – How many plant samples were taken? When they were taken? How long it takes for the nitrogen after it is applied to accumulate in leaves or to impact leaf nitrogen content?

Subsec. 2.3.3 It might be better to rename it,  because otherwise it seems that you are proposing your own new indices

Line 210 – please specify the indices here. Do not refer to them as first and second before naming them. It is confusing.

Equation 2 – why these particular bands were chosen for RVI calculation?

Lines 230-231 – It is not clear why you included only vegetation indices in linear regression model and did not include canopy structural information. If the problem is multicollinearity, you need to check the correlation and exclude highly correlated parameters.

As later you compare the effectiveness of three modelling methods, you need to use the same information in all of the methods, otherwise the comparison is not fair.

Line 397 – why vegetation indices? you used only AIVI here

Line 408 – you meant AIVI?

Please, correct figure captions for Figure 5 (what are S1, S2, S3??), Figure 6, 7, 8 (D1, D2, D3???).

Line 498 – phenotypes studies?

Line 530 – what do you mean by systematically?

Line 534 How are they affected by multicollinearity? The meaning is not clear. Multicollinearity does not impact the variables or their importance. It means that some variables are strongly correlated.

Line 536 – how are they integrated? They used separately in models.

Line 539 – which coverage?

 

Author Response

Dear Silvia Liu,

First, we really appreciate you and the reviewers’ helpful comments and valuable suggestions. According to the comments of the reviewers, we have made corrections and modifications as follows:

Reviewer #1

Point 1. Lines 23-24 you meant based on true colour image?.

Response 1: Thanks for the reminder, we have found the error in this sentence.

 

Point 2. Lines 57-60 – the impact of nitrogen on spectral reflectance of crops is indirect. It influences the state of the crop canopy, which, in turn, affects crop canopy spectral reflectance.

Response 2: Thanks for pointing out the problem, we read some literature and revised the sentence.

 

Point 3. Line 68 – red envelope?? You meant red edge??

Response 3: Thanks for the reminder. This is a low-level problem caused by translation. We meant red edge here.

 

Point 4. Line 70-71 – This is oversimplification. Crop canopy reflectance can contain information on a number of parameters, which can have direct or indirect influence on it. Not only nitrogen and structural information. In some cases, crop canopy reflectance can contain information on soil background. Especially at early stages of crop development.

Response 4: We agree with you that crop canopy reflectance can contain information on a number of parameters. What we wanted to mean here was that the canopy structure has a large influence on the spectral reflectance. We have revised this sentence

 

Point 5. Line 119 – why linear regression is not mentioned?

Response 5: Thanks for the reminder, linear regression should indeed be mentioned as well. The reason we didn't mention it before was that we only see linear regression as an algorithm for comparison.

 

Point 6. Line 122 – informative canopy structural features?

Response 6: We have rephrased the sentence to make it clearer.

 

Point 7. Lines 128-129 – when nitrogen fertilizers were applied (please provide dates)? which fertilizers were applied on each plot?

Response 7: The base fertilizer for experiment 1 was applied on October 27, 2019, and the jointing fertilizer was applied on March 6, 2020. The base fertilizer for experiment 2 was applied on October 30, 2020, and the jointing fertilizer was applied on March 7, 2020. The scheme of fertilizers applied on each plot has been added (figure 1).

 

Point 8. Lines 125 – 138 – Please, provide a scheme of test plots and experiment.

Response 8: The scheme of test plots and experiment has been added.

 

Point 9. Line 130-132 – 3 row widths for each nitrogen application rate?

Response 9: We did set 3 row widths for each nitrogen application rate. The specific content is shown in figure 1.

 

Point 10. Lines 141-147 – Please, specify the dates of spectral data acquisition. How many spectral curves were acquired? Spectral data was collected for each phase of crop development? How long after the nitrogen fertilizer application?

Response 10: Spectral data were collected at jointing, booting and heading stages. Data acquisition dates for Experiment 1 were March 13, March 23, and April 7, and data acquisition dates for Experiment 2 were March 14, March 20, and April 2. The first spectral data collection time is about a week after jointing and fertilization every year. The first spectral data collection time was about a week after the application of jointing fertilizer every year. The total number of spectral curves finally obtained in this study was 135.

 

Point 11. Lines 145-146 When you measured spectral reflectance of wheat canopy, how you considered the influence of soil background on it? Especially at early stages of wheat development? Because it can impact the results of modelling nitrogen indices based on spectral reflectance of crop canopy.

Response 11: We agree with you that soil background has a big impact spectral reflectance of crop canopy, but we did not control the effect of soil background because coverage was included in the canopy structure indices of our study. Coverage is an important factor in determining the influence of soil background. Therefore, we believe that adding coverage as an input feature to build a model is one way to reduce the influence of soil background.

 

Point 12. Line 148 – How many images? When they were taken?

Response 12: Three locations in each plot were randomly selected to acquire images, and each location acquired an RGB image and a depth image. Images were taken at 4:00 p.m. on the same date as the spectral data acquisition. The number of RGB and depth images was 405 in total, and the average value of the images of three sample points were taken as the canopy structural features of each plot.

 

Point 13. Lines 161-167 – How many plant samples were taken? When they were taken? How long it takes for the nitrogen after it is applied to accumulate in leaves or to impact leaf nitrogen content?

Response 13: The plant sample collection time was consistent with the spectral data collection time. The number of plant samples was also 135. The first plant samples collection time was also about a week after the application of jointing fertilizer every year. This date has been more than 4 months from the base fertilizer, and the LNC and LNA under different treatments have been very different.

 

Point 14. Subsec. 2.3.3 It might be better to rename it, because otherwise it seems that you are proposing your own new indices

Response 14: Thanks for your advice. We have changed the name to the Selection of spectral indices.

 

Point 15. Line 210 – please specify the indices here. Do not refer to them as first and second before naming them. It is confusing.

Response 15: Thanks for your suggestion. We have reorganized this section to avoid ambiguity.

 

Point 16. Equation 2 – why these particular bands were chosen for RVI calculation?

Response 16: The reason why these bands were chosen is because we have done research on the sensitive bands of the LNA before. We have added references to this part of the study.

 

Point 17. Lines 230-231 – It is not clear why you included only vegetation indices in linear regression model and did not include canopy structural information. If the problem is multicollinearity, you need to check the correlation and exclude highly correlated parameters. As later you compare the effectiveness of three modelling methods, you need to use the same information in all of the methods, otherwise the comparison is not fair.

Response 17: The reason why we did not include canopy structure indices as input features for linear regression was not because of multicollinearity. The linear regression model based on vegetation index is the most common empirical model. We constructed linear regression models based on vegetation index to represent the inversion accuracy of LNA/LNC in the most common cases. However, this type of model generally has the problems of insufficient information and few invertible indicators. The purpose of this study is to add canopy structure information to compensate for the defects of vegetation index in LNC and LNA monitoring. That is to say, instead of comparing the performance of the three algorithms, we want to explore the effect of canopy structure information on improving the accuracy of nitrogen index inversion. As to why RF and PLS were chosen over other algorithms such as multiple linear regression, we explained separately in the introduction section.

 

Point 18. Line 397 – why vegetation indices? you used only AIVI here

Response 18: The vegetation index here is actually AIVI. We have realized that writing this way created ambiguity and this issue has been revised.

 

Point 19. Line 408 – you meant AIVI?

Response 19: Thanks for the reminder, here is AIVI.

 

Point 20. Please, correct figure captions for Figure 5 (what are S1, S2, S3??), Figure 6, 7, 8 (D1, D2, D3???).

Response 20: Figure captions have been corrected. We have explained these abbreviations in figures.

 

Point 21. Line 498 – phenotypes studies?

Response 21: We mentioned phenotypic studies here was to illustrate the importance of canopy structure feature extraction.

 

Point 22. Line 530 – what do you mean by systematically?

Response 22: The systematically here actually referred to the correlation analysis between different features. We realize that the meaning of this word here is unclear and prone to misunderstanding and has been revised.

 

Point 23. Line 534 How are they affected by multicollinearity? The meaning is not clear. Multicollinearity does not impact the variables or their importance. It means that some variables are strongly correlated.

Response 23: Thank you for pointing out our problem. We have recognized that multicollinearity did not actually affect our results. We have reorganized the Conclusions section.

 

Point 24. Line 536 – how are they integrated? They used separately in models.

Response 24: Your comment is correct. These indices were used separately in models. We have revised this sentence

 

Point 25. Line 539 – which coverage?

Response 25: The coverage here is actually FVC, and we've corrected that.

 

Point 26. Discussion sections needs to be corrected. It should not contain the results.

Response 26: We agree that Discussion sections should not contain the results, and we noticed that we listed some of the data in the Discussion section. However, we do not think that the data presented here can be considered as the results of this study, as these are of little relevance to the main purpose of this paper. We put these figures here to better explain why canopy structure information could improve the accuracy of spectral monitoring of nitrogen indicators. Now, we've added some theoretical explanations and references so that this section is not just a data showing, but a deeper discussion of the problem. We hope that the discussion section can more satisfy to your requirements.

 

We hope that the revised manuscript could satisfy you and the requirements for publication in the journal. Thank you and the reviewers again for your help.

Yours sincerely,

Huaimin Li and Jun Ni

Nanjing Agricultural University

No.1 Weigang Road

Nanjing, Jiangsu 210095

P.R.China

Author Response File: Author Response.doc

Reviewer 2 Report

1. Refer to some comments from the reviewed pdf
2. The latest research on - "Leaf nitrogen content estimation using hyperspectral data" must be added. Do a literature review of the latest papers and include them in the introduction story properly. 
3. Add an overall framework/flowchart of the research work
4. The logic of selecting two indices is not convincing. Either refer to good published papers in order to support the index selected by you OR create a correlation heatmap between all possible indices and the LNC (and LNA) values. This can be generated using the collected ground truth data. You may get this approach in multiple papers. 
5. The latest papers related to canopy height estimation using optical (RGB) images also need to be cited and results should be compared against your results. 
6. No mention of BRDF effect on data and results! Plese, discuss its effect on the indices and results preferably in the discussion section.  

Comments for author File: Comments.pdf

Author Response

Dear Silvia Liu,

First, we really appreciate you and the reviewers’ helpful comments and valuable suggestions. According to the comments of the reviewers, we have made corrections and modifications as follows:

Reviewer #2

Point 1. Refer to some comments from the reviewed pdf.

Response 1: Thanks for your comments. The issues mentioned in the reviewed pdf have been resolved.

 

Point 2. The latest research on - "Leaf nitrogen content estimation using hyperspectral data" must be added. Do a literature review of the latest papers and include them in the introduction story properly.

Response 2: We've added newly published articles on "Leaf nitrogen content estimation using hyperspectral data".

 

Point 3. Add an overall framework/flowchart of the research work.

Response 3: An overall flowchart has been added.

 

Point 4. The logic of selecting two indices is not convincing. Either refer to good published papers in order to support the index selected by you OR create a correlation heatmap between all possible indices and the LNC (and LNA) values. This can be generated using the collected ground truth data. You may get this approach in multiple papers.

Response 4: Thanks for your suggestion. We have reorganized this section and added some references. We have understood the method you mentioned and we admit that it is indeed an excellent method for screening vegetation index. However, we think that this approach is inconsistent with the purpose of our article. Canopy structure has an effect on the spectral index, which is likely to be reflected in any vegetation index. The purpose of this study is to add canopy structure information to compensate for the defects of vegetation index in LNC and LNA monitoring. Therefore, we chose to select two vegetation indices with large differences in spectral properties from the literature to show that our research results can be applied to different types of vegetation indices.

 

Point 5. The latest papers related to canopy height estimation using optical (RGB) images also need to be cited and results should be compared against your results.

Response 5: Thank you for reminding us of the missing literature on canopy height estimation using optical (RGB) images. We have added this to the Discussion section.

 

Point 6. No mention of BRDF effect on data and results! Plese, discuss its effect on the indices and results preferably in the discussion section.

Response 6: Thanks for your suggestion. We have read some literature on BDRF and added the effect of canopy structure on the directionality of spectral reflectance. In addition, we also focused on the analysis of the difference in effect on the two vegetation indices in the Discussion section.

 

We hope that the revised manuscript could satisfy you and the requirements for publication in the journal. Thank you and the reviewers again for your help.

Yours sincerely,

Huaimin Li and Jun Ni

Nanjing Agricultural University

No.1 Weigang Road

Nanjing, Jiangsu 210095

P.R.China

Author Response File: Author Response.doc

Reviewer 3 Report

The article is well written and timely submission needs minor revision. The following points need to be addressed.

1. The abstract ok.
2. The introduction should be reframed with the clear-cut hypothesis.
3. Methodology is ok
4. Results well presented
5. Discussion must be supported with newly published reports
6. The conclusion must be reorganized it should not be in points. It should be a recommendation of the study. 
7. The language of the article is fairly ok but critical rechecking is suggested.
8. Fram the references as per the style and pattern of the journal. cross-checking suggested.

Author Response

Dear Silvia Liu,

First, we really appreciate you and the reviewers’ helpful comments and valuable suggestions. According to the comments of the reviewers, we have made corrections and modifications as follows:

Reviewer #3

 

Point 1. The introduction should be reframed with the clear-cut hypothesis.

Response 1: The introduction has been reframed. The topic and conclusion of each paragraph are now clearer

 

Point 2. Discussion must be supported with newly published reports.

Response 2: Thanks for your advice. Some recent literature has been added to the Discussion section

 

Point 3. The conclusion must be reorganized it should not be in points. It should be a recommendation of the study.

Response 3: The conclusion has been reorganized.

 

Point 4. The language of the article is fairly ok but critical rechecking is suggested.

Response 4: The language of the article has been rechecked.

 

Point 5. Fram the references as per the style and pattern of the journal. cross-checking suggested.

Response 5: We have checked the format of the references.

 

We hope that the revised manuscript could satisfy you and the requirements for publication in the journal. Thank you and the reviewers again for your help.

Yours sincerely,

Huaimin Li and Jun Ni

Nanjing Agricultural University

No.1 Weigang Road

Nanjing, Jiangsu 210095

P.R.China

Author Response File: Author Response.doc

Round 2

Reviewer 2 Report

Figure 5 does not add any value to manuscript. You may consider removing it.