Individual Tree Detection and Qualitative Inventory of a Eucalyptus sp. Stand Using UAV Photogrammetry Data

Round 1

Reviewer 1 Report

The article presents a method to identify trees automatically, estimate their total heights and monitor the initial silvicultural quality of a Eucalyptus sp plot using UAV photogrammetry.

The topic of the paper is suitable for the Remote Sensing journal. The article and methods are well-explained. However, it was not clear to this reviewer what the gap in the knowledge or methodology is that this paper is addressing. As the authors explain, other papers already similarly addressed this issue, so what are the differences between this work and others such as the one developed by Hentz et al.? (Estimating forest uniformity in Eucalyptus spp. and Pinus taeda L. stands using field measurements and structure from motion point clouds generated from unmanned aerial vehicle (UAV) data collection).

Moreover, there are pretty similar articles that do not appear in this work. E.g.

Juan Guerra-Hernández, Diogo N. Cosenza, Luiz Carlos Estraviz Rodriguez,Margarida Silva, Margarida Tomé, Ramón A. Díaz-Varela & Eduardo González-Ferreiro(2018): Comparison of ALS- and UAV(SfM)-derived high-density point clouds for individualtree detection in Eucalyptus plantations, International Journal of Remote Sensing, DOI:10.1080/01431161.2018.1486519

The writing style of the paper needs to be made more scientific and formal, and a thorough review and edit by a native English speaker would greatly help the quality of the paper. In addition, verbs are missing in sentences throughout the article so makes it difficult to understand.

The authors can reduce the extension of the paper by deleting unnecessary things such as well-known equations (RMSD, SD…), which are not specific to this paper.

 

Specific comments:

Line 20 and 22

Do not cite the same reference twice in the same paragraph

Line 44

Correct English usage: “data has [been]”

Line 79

“Nevertheless, few are the studies that…” please rewrite

Line 82-86

what is the novelty of this paper? This is a critical point that needs to be further identified and described in this section.

Line 88 “2.1. Study area”

Please specify the coordinates of the centre of the plot.

reference for weather station where this data was obtained

line 108

Please specify better how the authors measured the tree heights using the rule. For example, did they see the top of the rule?

line 109

“and had their ht remeasured” Please explain this better.

Line 111

Is it necessary to define common statistical expressions such as RMSD, SD…? The same for CV (line 123) and Gini (line 129)

Line 147

The name of the paragraph is “2.4. Digital Aerial Photogrammetry” but there is no information about sensors/cameras. Please specify it.

Line 157

Did the authors obtain the mean ground sample distance (GSD)? or it was an output of the software? Please specify.

Line 265

“One may notice the adequate terrain…” Please rewrite in a more scientific/formal way.

Line 306

“The regression model eliminated Bias”

Please explain this fact better. Do you mean the sample mean is considered an unbiased estimate of the population mean μ?

Line 332

The authors should specify what these limitations are.

Reference 37: where is the year? Write the references correctly.

Author Response

Review 1 

The article presents a method to identify trees automatically, estimate their total heights and monitor the initial silvicultural quality of a Eucalyptus sp plot using UAV photogrammetry.

The topic of the paper is suitable for the Remote Sensing journal. The article and methods are well-explained. However, it was not clear to this reviewer what the gap in the knowledge or methodology is that this paper is addressing. As the authors explain, other papers already similarly addressed this issue, so what are the differences between this work and others such as the one developed by Hentz et al.? (Estimating forest uniformity in Eucalyptus spp. and Pinus taeda L. stands using field measurements and structure from motion point clouds generated from unmanned aerial vehicle (UAV) data collection).

Moreover, there are pretty similar articles that do not appear in this work. E.g.

Juan Guerra-Hernández, Diogo N. Cosenza, Luiz Carlos Estraviz Rodriguez,Margarida Silva, Margarida Tomé, Ramón A. Díaz-Varela & Eduardo González-Ferreiro(2018): Comparison of ALS- and UAV(SfM)-derived high-density point clouds for individualtree detection in Eucalyptus plantations, International Journal of Remote Sensing, DOI:10.1080/01431161.2018.1486519

The writing style of the paper needs to be made more scientific and formal, and a thorough review and edit by a native English speaker would greatly help the quality of the paper. In addition, verbs are missing in sentences throughout the article so makes it difficult to understand.

The authors can reduce the extension of the paper by deleting unnecessary things such as well-known equations (RMSD, SD…), which are not specific to this paper.

The authors thank the reviewer for the careful review and valuable comments. All specific suggestions have been addressed, as described below.

 

The description of the knowledge/methodology gap that the manuscript addresses was included in the introduction (lines 81-101), in addition to the citation of Juan Guerra-Hernández et al. (2018) (lines 52 and 87). A review of the entire writing of the work was also carried out, along with a complete review/editing by a native English speaker, making the text more scientific and formal. The well-known equations in the literature have been removed throughout the text, as suggested. The specific comments are addressed below.

Specific comments:

Line 20 and 22

Do not cite the same reference twice in the same paragraph

Thanks for catching this error. The duplicated citation has been removed.

Line 44

Correct English usage: “data has [been]”

Done.

Line 79

“Nevertheless, few are the studies that…” please rewrite

This sentence has been removed.

Line 82-86

what is the novelty of this paper? This is a critical point that needs to be further identified and described in this section.

We thank the reviewer for this suggestion. We addressed this point in the two last paragraphs of the introduction (lines 81-101).

Line 88 “2.1. Study area”

Please specify the coordinates of the centre of the plot.

reference for weather station where this data was obtained

The coordinates of the centre of the area and the reference for the weather data have been added to the text (lines 105).

line 108

Please specify better how the authors measured the tree heights using the rule. For example, did they see the top of the rule?

Details of tree height measurement have been provided (lines 125-130)

line 109

“and had their ht remeasured” Please explain this better.

Additional details on the height measurements have been provided (lines 130-132). 

Line 111

Is it necessary to define common statistical expressions such as RMSD, SD…? The same for CV (line 123) and Gini (line 129)

We thank the reviewer for this suggestion. The equations for RMSD and SD have been removed. For clarity, we opted for maintaining the equations describing stand uniformity indices.

Line 147

The name of the paragraph is “2.4. Digital Aerial Photogrammetry” but there is no information about sensors/cameras. Please specify it.

Technical details about the camera have been added to the text (lines 169-164)

Line 157

Did the authors obtain the mean ground sample distance (GSD)? or it was an output of the software? Please specify.

The GSD was an output of the software. This information has been added to the text (lines 194-197).

Line 265

“One may notice the adequate terrain…” Please rewrite in a more scientific/formal way.

This paragraph has been rewritten in a more formal manner as suggested (lines 288-292).

Line 306

“The regression model eliminated Bias”

Please explain this fact better. Do you mean the sample mean is considered an unbiased estimate of the population mean μ?

Yes. This paragraph has been rewritten to make this point clearer.

Line 332

The authors should specify what these limitations are.

Additional details concerning the limitations have been added to the text (lines 328-333).

Reference 37: where is the year? Write the references correctly.

 Done.

 

Author Response File: Author Response.docx

Reviewer 2 Report

Dear authors,

Thank you for giving me the opportunity to review your research work. The topic of the study is interesting as it attempts to answer a problem in the identification and designation of tree attributes from UAV photogrammetry data. Both the DTM quality analysis of the classified and supervised point cloud and the subsequent automatic detection of tree canopies are correctly identified, statistically analyzed, and represented.

I congratulate them for the structure of the research and the easy reading of the article.

Please consider some comments and suggestions:

Line 20: Check if "Eucalyptus spp." or "Eucalyptus sp.". Check lines 26 and 48 and more

Line 112. The height the tree "ht" must be italic.

Line 166: The procedure to be followed for the geolocation of the GCPs is not random, but must meet a series of characteristics, among others: GCPs should be evenly distributed across the entire map area. Ensure that the GCPs are at least far enough away from the boundary of your study area to be visible later in the image

Line 188: The cell size was 50 meters?. The study area has a length of 30 meters?

Line 272: The largest differences are from 0.558 to -1.35 m. as shown in Figure 2e.

Line 281: Do not you think that the nonidentification (OE) of individuals below a ht of 1.15 m. is a problem?. It is mentioned in line 408.

Line 284: In Figure 3, the circle, diamond, and triangle symbology is not displayed correctly.

Figure 5 and 6: In the description of the figure, the text "...in northeastern Brazil" is not necessary.

Figure 1 and 7: There is an error in the graphical scale representation.

Line 330: The limitation of terrain estimation from DAP-UAV products in stands with high tree canopy density is a major problem that is reduced with the use of LiDAR techniques. The low canopy cover index (CCR = 0.56) of the stand minimizes such errors, but I believe that, to perform a canopy identification and height estimation of the trees, and thus be able to perform a complete monitoring of the stand with different ages, it would not be advisable.

Up to what stage of growth in the eucalyptus stand could the UAV be used for inventory data?

Line 361: I agree with the sentence: "The supervised classification of terrain points can reduce the overestimation of elevation, increasing the quality of products and highlighting the potential of UAV photogrammetry data for areas with sparse vegetation", but such a process would increase the processing time and therefore the response time to stand monitoring.

Can angular and rotational variation in UAV imaging influence better tree identification or improvements in DTM estimation?

Kind regards,

Author Response

Review 2

Dear authors,

Thank you for giving me the opportunity to review your research work. The topic of the study is interesting as it attempts to answer a problem in the identification and designation of tree attributes from UAV photogrammetry data. Both the DTM quality analysis of the classified and supervised point cloud and the subsequent automatic detection of tree canopies are correctly identified, statistically analyzed, and represented.

I congratulate them for the structure of the research and the easy reading of the article.

Please consider some comments and suggestions:

The authors thank the reviewer for the careful review and valuable comments. All specific suggestions have been addressed, as described below.

Line 20: Check if "Eucalyptus spp." or "Eucalyptus sp.". Check lines 26 and 48 and more

We have checked the text and corrected the abbreviation where needed. Both “sp.” and “spp.” can be used, depending on the context (i.e., if referring to a single or several species within the genus).

Line 112. The height the tree "ht" must be italic.

Done.

Line 166: The procedure to be followed for the geolocation of the GCPs is not random, but must meet a series of characteristics, among others: GCPs should be evenly distributed across the entire map area. Ensure that the GCPs are at least far enough away from the boundary of your study area to be visible later in the image

We agree with these recommendations and tried to follow them in our study as much as possible.

Line 188: The cell size was 50 meters?. The study area has a length of 30 meters?

We thank the reviewer for catching this error. The cell size is 5 m and the text has been corrected (line 209).

Line 272: The largest differences are from 0.558 to -1.35 m. as shown in Figure 2e.

We thank the reviewer for catching this error.  These values are correct and have been revised in the text (line 295).

Line 281: Do not you think that the non identification (OE) of individuals below a ht of 1.15 m. is a problem?. It is mentioned in line 408.

We thank the reviewer for bringing this point to our attention. 1.15 m is the height of the tallest tree that was not detected by the algorithm. However, the algorithm did detect trees smaller than 1.15 m (as seen in Figures 4 and 5). The sentence has been removed from the text and a new sentence has been added to provide information on the minimum detected tree height (lines 304-305).

Line 284: In Figure 3, the circle, diamond, and triangle symbology is not displayed correctly.

The tree locations derived from the different NPCs were very close, so the symbols tend to overlap. We tried using different symbols and colors in a new version of the figure in an attempt to make it clearer.

Figure 5 and 6: In the description of the figure, the text "...in northeastern Brazil" is not necessary.

We thank the suggestion. The location of the study area has been removed from the descriptions.

Figure 1 and 7: There is an error in the graphical scale representation.

Thanks for catching this error. The scale has been corrected.

Line 330: The limitation of terrain estimation from DAP-UAV products in stands with high tree canopy density is a major problem that is reduced with the use of LiDAR techniques. The low canopy cover index (CCR = 0.56) of the stand minimizes such errors, but I believe that, to perform a canopy identification and height estimation of the trees, and thus be able to perform a complete monitoring of the stand with different ages, it would not be advisable.

Up to what stage of growth in the eucalyptus stand could the UAV be used for inventory data?

This is an important question. Unfortunately, older Eucalyptus stands were not assessed in the present study, and the authors are not aware of studies in the specialized literature that assessed the performance of DAP-UAV in eucalyptus plantations as a function of the stand age.

 

However, studies carried out by Guerra et al. (2018) and (2019) on a 7-year-old Eucalyptus plantation in Portugal showed good results, suggesting that this method can be useful for monitoring stand density and tree height throughout the entire rotation of plantations used for pulp and paper or the charcoal-based steel industry (the primary uses in Brazil).

 

We have cited these studies in the introduction and discussion sections and recommended in the conclusions that the methods presented in our study should be tested in plantations with different ages, tree spacing, management regimes, etc.

Line 361: I agree with the sentence: "The supervised classification of terrain points can reduce the overestimation of elevation, increasing the quality of products and highlighting the potential of UAV photogrammetry data for areas with sparse vegetation", but such a process would increase the processing time and therefore the response time to stand monitoring.

We agree with this comment. The supervised classification can improve the results, but requires an extra step that cannot be easily automated and will unavoidably increase the processing time. Although the added time may not be significant for occasional or small area inventories, it will add up in continuous, large-scale inventories. However, for the operational inventory of eucalyptus stands, alternatives can be used to reduce the total processing time. Once the terrain is mapped, for instance when the soil is visible during pre-planting operations, the DTM generated by either supervised or unsupervised classification can be used as the reference for normalization of point clouds generated in future acquisitions, when the vegetation is present. In addition, if a good Lidar, InSAR or topography-based DTM is available, it can also be used as the reference surface for normalization in future UAV acquisitions. We have added this discussion to the text (lines 394-402.

Can angular and rotational variation in UAV imaging influence better tree identification or improvements in DTM estimation?

This is a good question. We believe that angular and rotational variation in UAV imaging has the potential to improve tree identification and DTM quality, as different viewing angles will increase the chance of seeing the ground and facilitate the identification of key and tie points. Unfortunately, this method was not tested in this study. A recommendation that it be tested in future studies has been added to the conclusion (lines 487-490).

Author Response File: Author Response.docx

Reviewer 3 Report

Generally, the article is okay.  It uses many terms and abbreviations to represent the data so it is sometimes difficult to read. The English style can be improved to make reading easier. Below are some specific suggestions.

• some references miss critical information (e.g. year in [37])
• L79-81: There is research such as Krause, S.; Sanders, T.G.M.; Mund, J.-P.; Greve, K. UAV-Based Photogrammetric Tree Height Measurement for Intensive Forest Monitoring. Remote Sensing 2019, 11, 758, doi:10.3390/rs11070758. did this. I noticed you compared your results to theirs in the discussion but maybe try to highlight your significance compared to previous studies to wow the readers.
• L92: Please describe the characteristics of 'As' climate class.
• L169-171: Is the error report based on the checkpoints? I noticed from Figure 1, there is one checkpoint outside the extent of GCPs. From previous studies, the accuracy outside the extent of GCPs degrades quickly so the error report outside the extent of GCP may not be reliable.
• L175-176: As far as I know, Agisoft metashape can only produce orthomosaic based on either mesh or DEM. Do you use other software to produce the orthomosaic based on point cloud?
• L190-196: This point classification criterion based on colour may opt-out shaded pixels. Could you please justify this idea?
• Figure 2: Please use the same range of colour scheme for the same product comparisons (i.e., DEM, the difference between DEMs, etc)
• L346-363: I suggest discussing the influence of supervised classification methods on the accuracy of DTM. You used the colour as the threshold, but in some sites, the colour of the ground may not be homogeneous enough to use this method.
• I feel the discussion part can be more organised. For instance, put the discussion of similar topics (e.g. supervised classification) in one cluster and so on.

Author Response

Review 3

Generally, the article is okay.  It uses many terms and abbreviations to represent the data so it is sometimes difficult to read. The English style can be improved to make reading easier. Below are some specific suggestions.

The authors thank the reviewer for the careful review and valuable comments. All specific suggestions have been addressed, as described below.

 

some references miss critical information (e.g. year in [37])

L79-81: There is research such as Krause, S.; Sanders, T.G.M.; Mund, J.-P.; Greve, K. UAV-Based Photogrammetric Tree Height Measurement for Intensive Forest Monitoring. Remote Sensing 2019, 11, 758, doi:10.3390/rs11070758. did this. I noticed you compared your results to theirs in the discussion but maybe try to highlight your significance compared to previous studies to wow the readers.

 

We thank the reviewer for this suggestion. A better description of the significance of our work compared to previous studies that addressed the same topic, including Krause et al. (2019) and Juan Guerra-Hernández et al. (2018) , has been included in the introduction and discussion (lines 87, 454, and 491).

 

L92: Please describe the characteristics of 'As' climate class.

 

The characteristics of the As climate class has been added to the text (lines 107-110).

 

L169-171: Is the error report based on the checkpoints? I noticed from Figure 1, there is one checkpoint outside the extent of GCPs. From previous studies, the accuracy outside the extent of GCPs degrades quickly so the error report outside the extent of GCP may not be reliable.

 

Yes, the error report is based on the checkpoints. We thank the reviewer for this suggestion. We discarded the checkpoint outside the extent of the GCPs and recalculated the error. This point has been removed from Figure 1 and the new results have been added to the text (lines 192-193).

 

L175-176: As far as I know, Agisoft metashape can only produce orthomosaic based on either mesh or DEM. Do you use other software to produce the orthomosaic based on point cloud?

 

We used Agisoft Metashape and the orthomosaic was produced from the DEM. This information has been added to the text (lines 232-233).

 

L190-196: This point classification criterion based on colour may opt-out shaded pixels. Could you please justify this idea?

 

This is an interesting point. We agree that the colour-based ground classification is limited, in the sense that shaded ground points may be left out of the interpolation. However, our results suggest that the benefit of avoiding vegetation points in the more conservative colour-based classification method outweigh the harm of leaving out shaded pixels that could be used in the interpolation. This is illustrated by comparing the results obtained in Figures 2c and 2d.

 

Figure 2: Please use the same range of colour scheme for the same product comparisons (i.e., DEM, the difference between DEMs, etc)

 

We thank the reviewer for this suggestion. The same colour scheme/range is now used for these figures.

 

L346-363: I suggest discussing the influence of supervised classification methods on the accuracy of DTM. You used the colour as the threshold, but in some sites, the colour of the ground may not be homogeneous enough to use this method.

 

We thank the reviewer for this suggestion.  In sites with different types/colours of soils, this method may not represent the variation well if a small number of points is used for training. However, although not tested in this study, we believe that a larger sample size, covering all the colour variation in the area, may still produce good results. In these cases, setting a higher tolerance limit (> 2) may also help with the ground classification. We have added this discussion to the text (lines 387-392).

 

I feel the discussion part can be more organised. For instance, put the discussion of similar topics (e.g. supervised classification) in one cluster and so on.

We thank the reviewer for this comment. The discussion has been reorganized into topics as suggested.  

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The new version is acceptable, and the changes are enough to consider the publication of the manuscript.

Line 331
RMS or RMSE?

Author Response

The authors thank the reviewer for the careful review and valuable comments. All specific suggestions have been addressed, as described below.

Line 331 - Done. Inserted RMSE.

A specialized company performed the English revisions of the manuscript.

Author Response File: Author Response.pdf