Study on Individual Tree Segmentation of Different Tree Species Using Different Segmentation Algorithms Based on 3D UAV Data

Round 1

Reviewer 1 Report

The paper is an interesting study about tree crown segmentation, comparing three crown segmentation algorithms. Tree crown segmentation is an issue specially for natural forests and although the paper is about only four species in a planted forest, it gives to the paper importance. However, the writing is confuse and I suggest it to be rewrite and resubmitted. As English is not my first language, I perfectly understand the challenging to write in a foreign language, and, in this case, it seems also in a different alphabet. I stop correcting it in the summary because to review the whole paper I would have to rewrite it myself.

Comments

Summary

Line 11-12. What forest resource and dynamic change?

Line 12. 3D points data. I suggest to use only 3D data in all the text

Line 13-15. I can understand the point, but it is really confuse

Line 16. LiDAR points data. I suggest to use only “LiDAR data” in all manuscript”

Line 17-18. “points segmentation algorithms”. I suggest to use “tree crown segmentation algorithms” in all manuscript

18. remove “of”

Line 18. I suggest “used” instead of “utilized”

Line 18. Replace “segment individual trees” by “segment individual trees crowns”

Line 19-20. Confuse, consider to rewrite it

Line 20. What do you mean by “two point data”?

Lines 22-24. Again I can understand that you are saying that PointNett++ presented the best result and LSS the worst, but it is really confuse.

Line 24-30. Instead of “segmentation results” use “crown segmentation of”

Line 28. Use past tense when writing about your results

Line 31. What do you mean by “points data types”?

Line 32-36. It is an interesting conclusion I was expecting that LiDAR data would be significantly better than the stereo images derived point clouds.

The paper is an interesting study about tree crown segmentation, comparing three crown segmentation algorithms. Tree crown segmentation is an issue specially for natural forests and although the paper is about only four species in a planted forest, it gives to the paper importance. However, the writing is confuse and I suggest it to be rewrite and resubmitted. As English is not my first language, I perfectly understand the challenging to write in a foreign language, and, in this case, it seems also in a different alphabet. I stop correcting it in the summary because to review the whole paper I would have to rewrite it myself.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

In the article entitled “Study on Individual Tree Segmentation of Different Tree Species using Different Segmentation Algorithms based on UAV 3D Points Data”, lidar data and high-resolution stereo images were collected with UAVs, and three different segmentation algorithms were used to segment individual trees for four tree species to explore the impact of points data types, individual tree segmentation algorithms, and tree species on the segmentation results of individual trees.

The tree species segmentation results of LiDAR points data are generally better than that of images derived points data. Among the segmentation algorithms tested, the deep learning PointNet++ approach provided the best result. While regarding the tree species, the ones with a larger crown length and a significant height difference between the top and edge of the crown such as Liriodendron, provided the best result.

In my opinion the article is well written, all sections are correctly presented, and the English appears accurate. Nevertheless, I suggest minor changes in order to improve it before publication. Authors are requested to solve and responde the following suggestions, with special attention to the final considerations.

Lines 12-13. Please consider to add i. the points data types, ii. segmentation algorithms, and iii. tree species. So as to focus reader attention on these aspects.

Lines 25-29. Please, use italic with scientific name of tree species.

Lines 90-108. Thanks to the authors for the comprehensive review of the algorithms and their performance. If possible, try to use the same accuracy indicators for different studies.

Lines 107-108. Please make a further investigation to check for the absence of other studies on this topic (improve references)

Line 114. What was tree species percentages?

Line 118. Tang et al., please add full stop and reference number

Lines 119-130. Please, use italic with scientific name of tree species.

Line 146. Consider remove “From Figure 1”.

Line 149. Check the Precipitation measurement units, maybe mm

Line 149. Consider to remove "adequate". It appears out of context without further considerations.

Line 150. The same as above. For whom are the weather conditions excellent?

Figure 1. If possible, use km instead miles in the image scale.

Table 1. Please use the SI notation. m^-1

Figure 2-3 caption. Consider to rephrase as: lidar point front view of (images poins front view of) (b)…, (c)…

Line 214. What are the proportions for training and test?

Line 230. How is the dataset sample divided? Is it maintained the proportion between tree species?

Paragraph 2.4. Please, provide a brief explanation of the choice of algorithms used.

Figure 11. If possible, find a different way to show this result. Moreover it could be useful show the proportion between species or number of each tree per specie.

Line 467 add space after PointNet++

Line 501. Julia et al. Add reference number. Moreover, where the study was conducted and what were the tree species?

 Overall considerations

Despite the lower point density, the ability to penetrate the canopy of lidar data is crucial to the results. Regarding the raw data, some aspects remain to be clarified for readers:

What were the survey times with the two sensors? Or at least it is asked to highlight whether the findings are comparable or not.

On the other hand, regarding computational times, are there relevant differences between lidar and images points and for different algorithms?

 The last consideration was regarding the application of segmentation in forest conditions that are more complex in terms of morphology, number of species ecc. I appreciate the final comment in the conclusion (lines 553-558). However, I would ask that the homogeneity situation of trees be highlighted in the study area paragraph.

This is a very good article, I have enjoyed to read it. I suggest minor revisions, but I am pretty sure you may easily solve my requests without so much effort.

English is accurate to me, no needs to further checks.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

That was a good surprise that the author could rewrite the paper in so short time. It is now much better organized, and I think it is now sufficiently improved to publication

Author Response

Thank you for your careful review and evaluation of the article.