The Use of an Unmanned Aerial Vehicle for Tree Phenotyping Studies

Round 1

Reviewer 1 Report

GENERAL COMMENTS

The authors present an interesting problem with the use of unmanned aerial vehicle (UAV commonly known as drones), equipped with sensors to classify tree species, however, I have doubts whether the manuscript is sufficiently related to the scope of the journal Separations (MDPI).

 

SPECIFIC COMMENTS AND SUGGESTIONS

1. The title does not exactly reflect the content of the manuscript, I suggest changing it, for example: The use of unmanned aerial vehicle for tree phenotyping studies.
2. Lines 185-188: there is no need to repeat in the description how many square meters a hectare is, in my opinion, unnecessary explanations may be removed.
3. I think it is necessary to add a location map of the study area to the text of the section Methodology.
4. I propose to extend the “Methodology” section to “Mathodology and Apparatus” and to describe in more detail the apparatus used in the research.
5. Figure 2 is poorly, unclearly described and should be improved.
6. Conclusions are too general, they are not convincing.

Author Response

1. The title does not exactly reflect the content of the manuscript, I suggest changing it, for example: The use of unmanned aerial vehicle for tree phenotyping studies.

Thank you for the suggestion we have revised the title.

1. Lines 185-188: there is no need to repeat in the description how many square meters a hectare is, in my opinion, unnecessary explanations may be removed.

agreed, so we have reverted to m2 only.

1. I think it is necessary to add a location map of the study area to the text of the section Methodology.

agreed. we have added a new figure 1(a) showing the location of the site in the UK. Figure 1 now has the original (labelled figure 1(b) and the new part (a).

1. I propose to extend the “Methodology” section to “Mathodology and Apparatus” and to describe in more detail the apparatus used in the research.

agreed, this was an omission on our part not to have included the apparatus. A new section has been added as section 2.2.

1. Figure 2 is poorly, unclearly described and should be improved.

agreed, a fuller description has been added. new lines 326-336.

1. Conclusions are too general, they are not convincing.

ok, the conclusion has been revised and amended. see new lines 818-822 and 829-841.

Reviewer 2 Report

Dear authors,

I have finished reviewing your manuscript and I would suggest major revisions. Due to reasons that I will list below, I feel that this manuscript should be rejected but I would like to see a makeover of this work and a convincing presentation of a thorough data analysis and a clear presentation of implications of this research.

At this stage, I would like to list a number of issues that I encountered and ask you to address them carefully.

First of all, the method is somewhat new in the way that you present it, but all steps that you are incorporating are in principle already part of a standard and quite conservative repertoire of methods in  remote sensing data analysis. If you want to stress its novelty and original character, I would appreciate if you could show what is really new.

2. Correct me if I am wrong, but the setup of your data analysis is made in a way that it is only applicable to that very specific case that you are investigating as you introduce some index calculations that are dependent on the actual inventory of tree species. This also means, it requires some manual interaction and visual inspection. Is that correct? If so, how useful is this approach for investigations outside this specific species domain?

2b. What is a principal component of a single band? Are you sure you wanted to say that? Also, L246: how do you obtain %-variance for tree species? What are the percentages referring to? They are surely the first 3 PCs but how does that refer to any tree signatures?

2c. How did you produce orthoimages? Where is the surface model, which resolution does it have? What is your reference system. The reader certainly also appreciates if you do not just list some meaningless parameters that you used, but if you provided detail on their meaning and importance. 

2d. Please explain figure 5 properly. What are x and y exactly showing? What is an Eigenvector of 0.5 of PC1 for Band 1 in figure? Are you really sure this is what you want to say? Also check spelling of labels. 

3. I somewhat liked reading the introduction, but the manuscript seems to fall apart when reaching the methodology. Would you mind explaining which MSI sensor and drone system you have used? What happens to the MSI signal during quantisation? Did you operate with DN values that were quantised onboard, or did you use reflectance values? What are the center wavelengths and the FWHMs or bandwidths? 

4. The section on k-means segmentation is not useful as it stands right now. Please provide proper explanations and some more concrete insights into your approach.

4b. Can you provide an accuracy assessment? This is all missing. You have some ground-truth validation, but that is it, and that is not enough, in particular when we talk about classification.

5. What are the conclusions? Where do you discuss the relevance of this research? What are the implications? How is all that applicable in real life and what are the implications for monitoring strategies (also costs)?

Kind regards.

Author Response

First of all, the method is somewhat new in the way that you present it, but all steps that you are incorporating are in principle already part of a standard and quite conservative repertoire of methods in  remote sensing data analysis. If you want to stress its novelty and original character, I would appreciate if you could show what is really new.

thank you for the comments. We have extensively revised the manuscript in light of your feedback.

2. Correct me if I am wrong, but the setup of your data analysis is made in a way that it is only applicable to that very specific case that you are investigating as you introduce some index calculations that are dependent on the actual inventory of tree species. This also means, it requires some manual interaction and visual inspection. Is that correct? If so, how useful is this approach for investigations outside this specific species domain?

This approach is useful at classifying any type of tree species and is not only applicable for this case. For any type of tree species when PCA is performed the spectral bands would behave differently and the proportion of eigenvectors values for each spectral band in the PC would change according to the tree species. Therefore, the information from the PC can be used to derive the most suitable spectral index in each case for the tree species to be classified. 

 

2b. What is a principal component of a single band? Are you sure you wanted to say that? Also, L246: how do you obtain %-variance for tree species? What are the percentages referring to? They are surely the first 3 PCs but how does that refer to any tree signatures?

We have revised the text, for clarification, on new lines 388-395.

2c. How did you produce orthoimages? Where is the surface model, which resolution does it have? What is your reference system. The reader certainly also appreciates if you do not just list some meaningless parameters that you used, but if you provided detail on their meaning and importance. 

we have revised the text on new lines 326-336.

2d. Please explain figure 5 properly. What are x and y exactly showing? What is an Eigenvector of 0.5 of PC1 for Band 1 in figure? Are you really sure this is what you want to say? Also check spelling of labels. 

the text has been revised, new lines 544-552.

3. I somewhat liked reading the introduction, but the manuscript seems to fall apart when reaching the methodology. Would you mind explaining which MSI sensor and drone system you have used? What happens to the MSI signal during quantisation? Did you operate with DN values that were quantised onboard, or did you use reflectance values? What are the center wavelengths and the FWHMs or bandwidths? 

sorry, somehow we omitted this. new lines 238-254, and 260-305. 

4. The section on k-means segmentation is not useful as it stands right now. Please provide proper explanations and some more concrete insights into your approach.

new text has been added, new lines 687-701.

4b. Can you provide an accuracy assessment? This is all missing. You have some ground-truth validation, but that is it, and that is not enough, in particular when we talk about classification.

The approach used in the current study does not build any classification models. It uses spectral indices which helps is classification, and k-means segmentation which segments the classified tree clusters by the spectral index. The clusters formed are pixel based clustered which has been quantified and shown in Table 2 and the accuracy is confirmed with data obtained from a ground level field study.

5. What are the conclusions? Where do you discuss the relevance of this research? What are the implications? How is all that applicable in real life and what are the implications for monitoring strategies (also costs)?

new text has been added around the conclusions, new lines 818-822 and 829-841.

Round 2

Reviewer 2 Report

Dear authors,

thank you for addressing my comments and for taking a closer look at my concerns.

I would not ask for an additional revision as all issues have been addressed. I might have missed it but I still do not see any mentioning about the quantization method of the MSI -- are you operating on 8 bit DNs or on radiance values? The reasons why I am mentioning that is because your classification relies on individual ratios. While the PCA should be insensitive to how DN values are spread over the 8 bit range, it still deserves some mentioning I feel. Anyway, you might want to take a closer look at that for your next publications perhaps, I feel it's worth investigating.

Thank you and kind regards.

Author Response

thank for the additional comment. We are measuring in MSI using radiance values. We have amended the manuscript for clarification on new lines: 236-237.

Author Response File: Author Response.pdf