A Tree Segmentation Algorithm for Airborne Light Detection and Ranging Data Based on Graph Theory and Clustering

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Using different clustering and graph-based algorithms, the authors proposed a tree segmentation algorithm for point cloud data. Despite having a practical topic, the manuscript is yet ready for publication. In the following, I provided a list of critical issues that should be addressed:

The authors should clearly state the novelty of their study. Why should this algorithm be used and what are the advantages over other tree segmentation algorithms?

Citing the obtained accuracies when reviewing the literature is unnecessary and sometimes misleading. I suggest removing this information from the literature review section.

The proposed method has several steps, and providing a pseudo-code can help its understanding. Besides, the purpose of each step should be mentioned. 

The article must compare the results of the proposed method with state-of-the-art algorithms using the same data set and under the same setting. Citing the obtained results of other articles can be considered as a comparison.

Author Response

Response to Reviewer 1

Manuscript ID: forests-3015974

Title: A Tree Segmentation Algorithm for Airborne LiDAR Data Based on Graph Theory and Clustering

Firstly, the authors would like to sincerely thank the reviewer for his/her valuable comments and recommendations and all the time devoted to our manuscript. For each comment, we provide our response and give an information how the manuscript was updated in this regard. While reviewer comments are written in standard font and black colour, our responses are given in italics and green colour.

Using different clustering and graph-based algorithms, the authors proposed a tree segmentation algorithm for point cloud data. Despite having a practical topic, the manuscript is yet ready for publication. In the following, I provided a list of critical issues that should be addressed:

The authors should clearly state the novelty of their study. Why should this algorithm be used and what are the advantages over other tree segmentation algorithms?

We have described that more according to your suggestion in the Introduction section.

Citing the obtained accuracies when reviewing the literature is unnecessary and sometimes misleading. I suggest removing this information from the literature review section.

We selected articles using the same metrics which are used in our study, therefore they should be generally comparable. We kept the accuracies in article to provide baseline about state-of-the-art and overall scope of detection possibilities.

The proposed method has several steps, and providing a pseudo-code can help its understanding. Besides, the purpose of each step should be mentioned.

For a better understanding of the method and its processing chain and individual steps, we presented Fig. 4 with schematic diagram of the proposed method. We agree that pseudocode could be useful, but given the number of steps in the method, it would be too long for an article if it were to contain a meaningful description of them. Conversely, in our opinion, a concise pseudocode would not provide information beyond the current Fig. 4.

The article must compare the results of the proposed method with state-of-the-art algorithms using the same data set and under the same setting. Citing the obtained results of other articles can be considered as a comparison.

Based on your suggestion, we have included a comparison with commercial Lis Pro 3D software for complex point cloud data processing which provides a dedicated forestry module. Please, see more in the Results section.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The submitted work is interesting and useful for various applications. However, it requires major revision. It should be reworked, rewritten, and resubmitted.

The Manuscript requires English Language corrections. Improper use of ‘itself’ in multiple places.

The Introduction is short and not well written. It requires improvement in describing the use of LiDAR technology or data in various forestry applications, including height and biomass estimation, structure mapping, species mapping, etc. Introduce the different satellite LiDAR data available with their characteristics, and how the UAV LiDAR data are different from them. Mention the wavelength and other characteristics. Add more background or details on the existing and major methods for 3D point cloud data processing, for example, two or three lines on the Watershed algorithm.

The Methodology section is ill-structured and requires major revision. The use of various parameters and functions is not well described. I couldn’t find a tree segmentation map for the entire study (two sites) as expected from the title of the manuscript.

Figure 1: one field-measured circular plot seems to be outside of the study area. Remove if it is outside of the study area. One another circular plot is un-labelled.

L130: what refers to the ‘FL all; FL only measurement satisfying required criteria’

L136: I couldn’t find ‘AOI M2 (visible in Figure 1 and marked by X)’

It is unclear how the orthophotos were geo-rectified.

L171: How the outliers were decided? What radius filtering value was used here and how was it decided?

L178: what is the source of the decimation value?

L179-182: It is unclear how the division of forests into smaller homogeneous areas is useful to avoid the mixed forest structure.

L184: What is the flood fill algorithm?

L184-185: How the minimum area was decided? Did the authors try with other area thresholds?

L186: unclear: 4-pixel connectivity.

L189: 2 m radius: how was it decided and did the author try other radius values? As the authors themselves mentioned the window size is not optimal for all forests.

L201: ‘Centroids for each group are created, and those centroids are then used as nodes.’ – was it created or identified?

L272-274: Did the authors assess the accuracy of point cloud and orthophoto? How 4 m is decided here? 

Comments on the Quality of English Language

Requires revision

Author Response

Response to Reviewer 2

Manuscript ID: forests-3015974

Title: A Tree Segmentation Algorithm for Airborne LiDAR Data Based on Graph Theory and Clustering

Firstly, the authors would like to sincerely thank the reviewer for his/her valuable comments and recommendations and all the time devoted to our manuscript. For each comment, we provide our response and give an information how the manuscript was updated in this regard. While reviewer comments are written in standard font and black colour, our responses are given in italics and green colour.

The submitted work is interesting and useful for various applications. However, it requires major revision. It should be reworked, rewritten, and resubmitted.

The Manuscript requires English Language corrections. Improper use of ‘itself’ in multiple places.

The Manuscript grammar was checked and corrected by English native speaker.

The Introduction is short and not well written. It requires improvement in describing the use of LiDAR technology or data in various forestry applications, including height and biomass estimation, structure mapping, species mapping, etc. Introduce the different satellite LiDAR data available with their characteristics, and how the UAV LiDAR data are different from them. Mention the wavelength and other characteristics. Add more background or details on the existing and major methods for 3D point cloud data processing, for example, two or three lines on the Watershed algorithm.

The introduction has been rewritten to include more information about LiDAR, its use and different platforms based on your suggestion.

The Methodology section is ill-structured and requires major revision. The use of various parameters and functions is not well described. I couldn’t find a tree segmentation map for the entire study (two sites) as expected from the title of the manuscript.

Methodology section was partly rewritten. The segmentation map is now included in the manuscript. As the segmentation map of the whole area was not very clear at this resolution, a sample of a smaller area has been included instead.

Figure 1: one field-measured circular plot seems to be outside of the study area. Remove if it is outside of the study area. One another circular plot is un-labelled.

We have modified Figure 1 according to your suggestion, thank you for this correction.

L130: what refers to the ‘FL all; FL only measurement satisfying required criteria’.

We have explained the term more specifically.

L136: I couldn’t find ‘AOI M2 (visible in Figure 1 and marked by X)’

We have updated the figure according to your suggestion.

It is unclear how the orthophotos were geo-rectified.

Information is now included in the Section 2.2.2 Natural ground control points were used to georeference the orthophotos.

L171: How the outliers were decided? What radius filtering value was used here and how was it decided?

We have included that information according to your suggestion.

L178: what is the source of the decimation value?

We have included that information according to your suggestion.

L179-182: It is unclear how the division of forests into smaller homogeneous areas is useful to avoid the mixed forest structure.

We have rewritten the appropriate text and explain the approach. The division into smaller homogeneous areas is based on the tree heights, not tree species. We admit writing about mixed tree species and tree heights was a bit confusing in the previous version of the manuscript.

L184: What is the flood fill algorithm?

We have included a short description of the algorithm.

L184-185: How the minimum area was decided? Did the authors try with other area thresholds?

The min size of the area is used to limit the number of areas passed into processing. Different sizes were not tested.

L186: unclear: 4-pixel connectivity.

We have revised the sentence to make the term clearer.

L189: 2 m radius: how was it decided and did the author try other radius values? As the authors themselves mentioned the window size is not optimal for all forests.

Two meter size is just a starting point to ensure, that even relatively small particles will be identified but also does not divide grown trees too much. It was selected after testing of different values as optimal starting point.

L201: ‘Centroids for each group are created, and those centroids are then used as nodes.’ – was it created or identified?

Probably some misunderstanding, it is stated in text. They are created.

L272-274: Did the authors assess the accuracy of point cloud and orthophoto? How 4 m is decided here? 

In the revised manuscript we now provide an explanation of how the value "up to 4 m" was determined. The reason for this difference is the tilt of the trees from the vertical axes on the orthophoto, the inaccuracy of the orthophoto georeferencing and inaccuracy of tree-top locations derived from the point cloud (detected tree-top does not have to exactly represent the real one).

In the section 2.1, information about UAV trajectory accuracy is provided. The georeferencing accuracy of the point cloud is directly related to that, further influenced by the accuracy of laser measurements which is expected at the few centimeters level for the used device (according to the official documentation of Riegl). Expected orthophoto georeferencing accuracy is at the level of decimeters. Based on the applied approach for the realized comparison, the described situation does not influence the obtained results

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have addressed the given comments. The current version can be accepted with minor revisions of the Introduction section and overall English language editing. 

Comments on the Quality of English Language

Minor English language editing is required. 

Author Response

Dear reviewer, thank you for your feedback. We have revised the Introduction part and realized English language editing of the whole manuscript.