Forest Vertical Structure Mapping Using Multi-Seasonal UAV Images and Lidar Data via Modified U-Net Approaches

Round 1

Reviewer 1 Report

Review of Jin-Woo Yu et al. Manuscript ID: remotesensing-2322794. Title: “Forest Vertical Structure Mapping Using Multi-Seasonal UAV Images and Lidar Data via Modified U-Net Approaches”

 

General comments

The paper proposes a methodology to map the forest structure using machine learning techniques applied to multi-seasonal unmanned aerial vehicle (UAV) optic and LiDAR data. This new methodology improves on those used previously. I consider these results will be of interest to the readers of the Remote Sensing journal.

In general, the manuscript is well organised. However, the whole text could be improved after revising the English writing. In particular, the Methods and Results sections must be improved to make them clearer and easier to read.

I see two major shortcomings in the manuscript. The first concerns the analyses used to decide which model performed better. The authors used one visual and one quantitative analysis. They chose as the best model the one that performed best in the quantitative analyses. Although this may seem reasonable, it needs to be discussed and explained in the text. Furthermore, the visual analysis needs to be explained in detail. The other major issue is novelty in relation to previous work. The authors mention this briefly at the end of the manuscript (lines 392-398). They mention that the models presented here improve the performance obtained in a previous work. It is not possible to verify this as they do not give references to the previous work mentioned. However, if this is the case, this improvement should be highlighted in the abstract, and these previous results should be mentioned as part of the literature review presented in the Introduction.

 

Specific comments

Line 23: It is unclear what "This" refers to.

Line 26: The description of Model 1 repeats what was described in lines 23-24.

Introduction section: In this section, it would be desirable to cite the previous study mentioned, against which the results obtained in this study are compared in lines 392-398.

Lines 34-36: Both statements are correct, but the way it is worded it sounds contradictory to say that the advantages of forests are that they store carbon and, at the same time, they are good fuels, which implies burning them and releasing the stored carbon. The authors should review how they state both benefits.

Lines 38-40: Any method requires time and money. The authors should explain why the proposed method requires less time and money than the traditionally used methodology.

Line 64: Can you add a sentence or two explaining what "semantic information" is?

Line 76-77: Please rewrite this sentence. For example, it should commence as "In the present study, we trained....." as "This study" cannot train a model.

Lines 89-90: This sentence is very general. Here it would be better to describe the study area in more detail. For example, describing the main tree species present, whether they are deciduous or perennial, among other vegetation characteristics relevant to the present study.

Lines 90-91: If the forest in the study area “consists of three-layered structures”, why later in this manuscript did the authors mention the presence of a fourth layer in the forest? For example, see lines 97, 309, and 317, among others.

Lines 127-146: What these lines describe seems to be more detailed in the following paragraphs. I suggest removing this paragraph.

Lines 154-156: Awkward writing, hard to follow. Please try rewording these sentences.

Lines 172-173 and 176-177: Consider making one sentence from these two.

Lines 213-214: Numbers under ten should be written in letters.

Lines 220-222: It is unclear what the difference between “input data” and “label data” is. It would be desirable to have more detail on how the data was cut.

Lines 223-226: It is unclear which classes are being referred to.

Lines 226-227: These figures do not respect the 80-20 ratio mentioned in lines 204-205.

Line 229: Where was the preprocessing described?

Lines 234-235: It is unclear which was “the previous U-Net structure”? And which one is “the original U-Net decoder”?

Lines 265-266: Is it possible to cite a work describing this statement?

Lines 271-276: This should be described in the Figure caption rather than in the text.

Lines 276-277: Additionally to what? In which images is it difficult to make the above distinction? Please, be more specific.

Line 290: “between” should be “among”.

Lines 299-302: Here and in other references to the Figures throughout the text. This detailed description should be in the Figure caption, not in the main text.

Line 308: Avoid using Fall / Winter and October / November interchangeably. It can be confusing. It is better to choose one at the beginning of the manuscript and always use the same one. In this case, I think it is better to use the name of the month because when a season is referred to, the readers might think of the whole three months of the season.

Lines 317-318: It is unclear how the authors arrived at this result solely from Figure 5.h. Please provide more details.

Line 331, Table 1: The table repeats what is expressed in lines 326-330. Consider removing the Table. 

Lines 333-337: This description should be in the Figure caption rather than in the text. Please check this here and in every mention of the Figures.

Lines 344-351: It is not explained how the authors did this analysis. Besides, why they used these two boxes A and B? Why in those locations and with those particular shapes? This analysis should be better described.

Lines 356-358: Should be in Figure caption.

Lines 366-368: This is unclear. This seems to be the opposite of what is stated in lines 344-351. Please clarify and discuss in more detail.

Lines 371-373: How did the authors arrive at this conclusion? I need clarification on why the results presented in Figure 6 differ from those of the quantitative analysis shown in Figure 7. Although it is not explained in the text, the authors decided to prioritize the quantitative analysis. Why? That should be explained and discussed. Please clarify.

Line 377: Even though it can be guessed, it would be desirable to explain how the “total precision” is calculated.

Line 384: Is this model 1? Say it explicitly.

Lines 387-389: These sentences could be more precise. It seems like without data augmentation, models 2 and 3 could have a better performance, even better than model 1. If that is the case, it should be discussed and explained.

Line 390: Which previous study? Please cite it.

Lines 392-398: I consider this result as one of the main outcomes of the present study. It should be included in the abstract, and the previous study must be referenced here and in the Introduction section.

Author Response

Thank you very much for giving us a chance to resubmit the manuscript. We would like to express our gratitude towards the editor and the reviewers for their valuable comments and suggestions that helped improving the manuscript.

We agree with almost all their comments, and we have carefully considered and addressed all the comments from the reviewers, point by point. The following paragraphs include a response to reviewers’ comments and suggestions. To carry out the revisions, we received the help from other co-authors which were added on the manuscript. We hope that this improved manuscript is acceptable for publication in Remote Sensing.

The files were merged in the order of comments, revised articles, and authorship modifications. Please verify. Thank you.

Author Response File: Author Response.pdf

Reviewer 2 Report

This work trained the model and analyzed its effects by minimizing location errors using multi-season data and a modified U-Net structure based two periods of UAV optic and Lidar data, the authors generated spectral indices and a filtered canopy height maps before model input. The aim of this study is to reduce the tree location errors owing to changes in the timing and location of acquisition between images by using the Modified U-Net structure, and enhance the accuracy of vertical forest structures map. The paper is a pure application, and lacked sufficient innovation, I consider it is important to explore a new method to reduce the error derived from tree location shift in different UAV images rather than modify the U-Net structure. I confirm that this study verified the modified U-Net structure can improve the map accuracy, but as a scientific paper, there are some issue should be considered.

(1) What are the specific problems and research implications addressed by the article?

(2) What are the related research works? What are their shortcomings?

(3) What is the connection between the contribution of the paper and other approaches? None of the above questions are mentioned in the paper.

(4) Line 289: replace “(f) DTM provided by NGII” with “(g) DTM provided by NGII.”

(5) Line 413: replace “Modified” with “modified”

(6) Please discuss the limitations and strength of your approach in the manuscript.

(7) Please increase the resolution of Figure 6.

(8) In the introduction part, the background and the state of the art about the forest vertical structure map is lacked in the manuscript.

(9) Authors and the reference [2] considered that the global warming is attributed to forest absorb and store carbon dioxide?  

(10) In addition, the paper needs an in-depth English proof reading by a native speaker.

Author Response

Thank you very much for giving us a chance to resubmit the manuscript. We would like to express our gratitude towards the editor and the reviewers for their valuable comments and suggestions that helped improving the manuscript.

We agree with almost all their comments, and we have carefully considered and addressed all the comments from the reviewers, point by point. The following paragraphs include a response to reviewers’ comments and suggestions. To carry out the revisions, we received the help from other co-authors which were added on the manuscript. We hope that this improved manuscript is acceptable for publication in Remote Sensing.

The files were merged in the order of comments, revised articles, and authorship modifications. Please verify. Thank you.

Author Response File: Author Response.pdf

Reviewer 3 Report

An excellent manuscript. I could only find one apparent error. The flight height of the drone for photography was 200m but the LiDAR instrument had a range of 100m.  Were there two flights?  This could be clarified. 

Author Response

Thank you very much for giving us a chance to resubmit the manuscript. We would like to express our gratitude towards the editor and the reviewers for their valuable comments and suggestions that helped improving the manuscript.

We agree with the comments, and we have considered and addressed the comments from the reviewers. The following paragraphs include a response to reviewers’ comments. To carry out the revisions, we received the help from other co-authors which were added on the manuscript. We hope that this improved manuscript is acceptable for publication in Remote Sensing.

The files were merged in the order of comments, revised articles, and authorship modifications. Please verify. Thank you.

Author Response File: Author Response.pdf