Unsupervised Methodology for Large-Scale Tree Seedling Mapping in Diverse Forestry Settings Using UAV-Based RGB Imagery

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper addresses the challenge of mapping and monitoring tree seedlings in reforested areas, which is a current topic and aligns with the scope of the journal. The authors propose an unsupervised approach using UAV-based RGB imagery to detect and map young conifer seedlings. The integration of different methods into the workflow, such as the use of 3D-based photogrammetric information, RGB-based indices, or a row segment detection algorithm to reduce false positives, is, in my opinion a significant contribution. The methodology has been validated in various plots and across diverse settings, including varying weed infestations, harvest residue, and terrain conditions, demonstrating the method's robustness. The results are statistically sound, and present indicators to validate the methodology, including precision, sensitivity, and F1 score.

The manuscript is well-structured and written in a clear and plain English. In my opinion, it could be considered for publication in the journal after addressing some minor comments:

-        I think an interesting aspect of this method is its reliance on conventional (RGB) images, eliminating the need for complex sensors (multispectral, LiDAR, etc). This advantage should receive more prominent emphasis in the introduction since it may go unnoticed until one reaches the methodology section.

-        The authors might even consider revising the title to "UAV-based RGB imagery" instead of "UAV imagery."

-        2.1 Study site. Geolocate the study plots using global coordinates.

-        Line 135: I understand that "DJI P4P system" refers again to the Phantom 4 Pro, so it might be best to eliminate duplicated information, such as "(DJI Ltd., Shenzhen, China)."

-        Lines 140-146: Please provide a more comprehensive explanation of the procedure used for annotating seedlings and how the field data was verified.

-        Figure 2: Considering that only RGB information is used in the workflow, it's advisable not to use the term "Multispectral UAV imagery" as it might be misleading.

-        I recommend adding a scale bar to the aerial images in Figures 4, 5, 6d, 8, and 9 to provide readers with a sense of scale.

Comments on the Quality of English Language

The manuscript is written in a clear and plain English

Author Response

The authors extend their gratitude to the reviewer for their invaluable feedback. Majority of the reviewer’s comments/suggestions have been incorporated into the manuscript, and a summary of these inclusions is provided in the table below. In instances where suggestions could not be integrated for valid reasons, explanations have been provided detailing why they could not be included in the manuscript. Please refer to the attached document for the authors' responses to specific comments.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The study proposes a seedling UAV image segmentation method combining spatial, spectral and structural parameters, which utilizes spectral and spatial information filtering to obtain a purer point cloud of seedling vegetation; by constructing a multicriteria evaluation system and utilizing the structural parameters to further filter and determine the detected points, the precision of seedling detection is effectively improved. It is a relatively novel method. Specific comments are as follows:

1.The introduction needs to add a comparison between this study and the direct use of LiDAR point cloud data

2.The study chose 8 sample plots of different types to test the method; however, the performance of the method in different plots lacks systematic analysis;

3.P1, L33: keywords are written throughout to better facilitate understanding;

4.How to determine whether a seeding was missing detectation or flagged as missing?

5.P14, L424: a sensitivity analysis was conducted on the "cut-off score"; while the "Maximum allowable distance for crown delineation", which also needs to be customized by the user, was not analyzed.

6.P13, L385: it is suggested to add a reference to the detection accuracy evaluation method.

7.What is the basis for evaluating the impact of multiple criteria on the accuracy of seedling detection?

 

Author Response

The authors extend their gratitude to the reviewer for their invaluable feedback. Majority of the reviewer’s comments/suggestions have been incorporated into the manuscript, and a summary of these inclusions is provided below. In instances where suggestions could not be integrated for valid reasons, explanations have been provided detailing why they could not be included in the manuscript. Please refer to the attached document for the authors' responses to specific comments.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript "Unsupervised segmentation of tree seedlings in UAV imagery" is a study about mapping and watching over young trees, which is very important for reforestation and restoration. The study uses a new, unsupervised method that makes use of several types of information collected from drones to map young seedlings. The manuscript is detailed, but the authors must address several concerns before its publication.

The title should be more descriptive and provide information about the method and what is expected to be read in the manuscript. For example, something like "Unsupervised UAV-Based Methodology for Large-Scale Tree Seedling Mapping and Monitoring in Diverse Forestry Settings". Moreover, in the introduction, the manuscript gives a lot of background information and describes the processes, but, the presentation of this information could be simplified and made more direct.

The manuscript does an excellent job explaining the methods used in the study and what the outcomes mean, and each step of identifying and locating the young trees is explained in a lot of detail. Nevertheless, in several sections, breaking down this information would improve readability.

Moreover, the figure captions are very descriptive. However, the quality of the figures should be improved, as the letters and numbers in the images are barely legible (e.g. in Figure 5, the red numbers are barely legible).

Finally, the "references" section of the manuscript critically requires revision and improvement. Currently, the references are inaccurate, incomplete, and improperly formatted, which undermines the credibility of the manuscript. For instance, the reference “Finn, A., et al., Unsupervised spectral-spatial processing of drone imagery for identification of pine seedlings. ISPRS Journal of 607 Photogrammetry and Remote Sensing, 2022. 183: p. 363-388.” is incorrectly cited and formatted.

Furthermore, there is a concerning issue of misrepresentation in the manuscript. The authors seem to have included references that do not accurately support the statements made. For example, in line 480, the authors reference studies that supposedly discuss the correlation of NDVI with leaf area index (LAI), biomass, and plant vigour. However, upon review, these references do not substantiate the authors' claims, and some even contradict them. Specifically, Reference 49 (“Segmentation of unhealthy leaves in cruciferous crops for early disease detection using vegetative indices and Otsu thresholding of aerial images.”) focuses on vegetation indices based on RGB imaging and does not employ NDVI. Similarly, Reference 48 (“New research methods for vegetation information extraction based on visible light remote sensing images from an unmanned aerial vehicle (UAV)”) discusses various vegetation indices, to be more accurate: NGBDI, GRVI, GLI, EXG, RGBVI, MGRVI, and NGRVI, but NDVI is not among them, and on the contrary, the study suggests that most consumer UAVs operate without calculating NDVI.

Another example is in line 504: “Not only are the predictions useful …. related to growth and vigour [50-53].” This assertion, again, is inadequately substantiated by the provided references. The cited materials do not explain the application of masks or convex hulls, with only reference 52 affirming the statement associated with convex hulls. In the specific comments section, some suggestions are offered that should be considered carefully by the authors.

There are more "references" with the same problem, and these are only a few examples. Therefore, it is imperative that the authors meticulously review and revise the references throughout the manuscript to ensure that each citation accurately supports the claims made. References should not be included merely as a formality but should genuinely corroborate the statements and findings presented.

  

Specific comments:

Lines 8-32:

"Mapping and monitoring tree seedlings... supplementing training datasets."

The abstract is well-written but slightly lengthy. The authors should focus on the main findings and contributions of the study.

Lines 49-65:

"Remote sensing (RS) offers rapid... broadened the horizons of UAV-based remote sensing in forestry."

Please, clarify how each method improves or fails to improve seedling detection.

Line 56

“airborne LiDAR”

This statement is not accurate. UAVs can be equipped with cheap LiDAR sensors, such as Zenmuse L1.

Lines 81-89:

"An alternative possibility is to use objective... a broad spectrum of forest settings."

The explanation about explaining the advantages of unsupervised techniques is good. However, the authors should explain why these techniques are particularly suited for this study or what makes them stand out in comparison to other methods.

Lines 105:

"Study site"

Please, add the coordinates and the crs of each plot in the text and in Figure 1.

Lines 132-142:

"High resolution imagery for Kaingaroa... were verified on the ground.”

The data collection section is detailed but could be made clearer. A summarized tabulation of the different sites, the equipment used, and the data collected would be better.

Line 291-308:

“Within the multicriteria evaluation … their nearest neighbours.”

This section is quite dense and could be better structured for clarity. Consider breaking down the steps and processes further.

Line 329-354:

“Ideally, the seedlings are expected to … iteratively for all the confirmed seedling locations.”

The methodology in these lines is described in a rather complex manner. Please, simplify the language and consider breaking down the steps and processes further.

Line 330

“blobs”

What is a “blob”?

Line 355-364:

“At this stage, as the locations of most … information of points in each segment”

What specific seedling metrics are being referred to? (authors should relate this information to the information given before in the other sections), and how they contribute to the overall objectives of the study?

Line 377-397:

“Detected seedling locations … the proposed seedling detection method.”

Consider elaborating on what the ‘annotated seedling dataset’ involves and how it’s used as a benchmark to validate the performance of the proposed method.

Lines 480-483

“Any other VIs (e.g., normalised difference vegetation index – NDVI) that are reported to have strong correlations with the leaf area index (LAI), biomass and vigour of plants could also be potential candidates for vegetation point isolation [30, 41, 48, 49].”

This is not true. Some of the cited references do not discuss or support the relationship between NDVI and LAI, biomass, or plant vigour as stated. The authors need to ensure the accuracy of their claims by citing references that directly corroborate their statements. The authors can find several articles related to images, NDVI and LAI in this review:

“Remote Sensing Vegetation Indices in Viticulture: A Critical Review” https://doi.org/10.3390/agriculture11050457

Lines 502:

“5.5. Downstream applications”

This paragraph presents the advantages of the approach, linking the methodology and results to wider applications. It is a good point and should be developed further.

Line 504-507

“Not only are the predictions useful …. related to growth and vigour [50-53].”

Again, this statement seems to be unsupported by the cited references. The references provided do not discuss the use of masks, or convex hulls. Only reference 52 (“Tree species classification and estimation of stem volume and DBH based on single tree extraction by exploiting airborne full-waveform LiDAR data”) supports the statement related to convex hulls. Some articles that the authors should consider:

“Towards Tree Green Crown Volume: A Methodological Approach Using Terrestrial Laser Scanning” https://doi.org/10.3390/rs12111841

"Innovation in Olive-Growing by Proximal Sensing LiDAR for Tree Volume Estimation” https://doi.org/10.1109/MetroAgriFor55389.2022.9965016

And specifically, for 2D convex hull:

“A Novel Technique Using Planar Area and Ground Shadows Calculated from UAV RGB Imagery to Estimate Pistachio Tree (Pistacia Vera L.) Canopy Volume” https://doi.org/10.3390/rs14236006

“Measurement and Calculation of Crown Projection Area and Crown Volume of Individual Trees Based on 3D Laser-Scanned Point-Cloud Data” https://doi.org/10.1080/01431161.2016.1265690

Comments on the Quality of English Language

Authors should revise the manuscript to correct minor mistakes. For example, in line 505: "2D snd 3D crown".

Author Response

The authors express their sincere gratitude to the reviewer for providing invaluable and detailed feedback, which has significantly contributed to improving the manuscript. The majority of the reviewer’s comments and suggestions have been incorporated into the manuscript, and a summary of these inclusions is provided below. In cases where suggestions could not be integrated for valid reasons, explanations have been provided to clarify why they were not included in the manuscript. Please refer to the attached document for the authors' responses to specific comments.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript has certainly undergone significant improvements, with the authors taking into account majority of the previous comments. However, there are still crucial corrections that need to be performed to make the manuscript align more with the journal’s requisites and to augment the clarity of the presented work.

Firstly, while addressing previous comments, the authors seemed to have opted for removing certain information and references, rather than improving and enhancing the discussion and references. Therefore, authors should address previous comments with higher precision. For instance, as stated in the previous review, the utilization of convex hulls is a key aspect of this study, yet the discussion lacks a comprehensive exploration of the implications of employing this specific geometry as opposed to others. Further clarification is required on the methodological choices made in the study, such as the decision to use a 0.6 threshold for the circularity of the convex hull value (line 400). Additionally, the discussion should be improved by including a more detailed comparison with results from other researchers who have also worked with convex hulls. This is crucial as convex hulls play a significant role in tree crown measurement and characterization. Similarly, the discussion regarding correlations between vegetation indices and leaf area index needs more depth and elaboration (line 547), rather than merely changing some information.

 

Finally, the authors have not followed the journal guidelines available at https://www.mdpi.com/journal/remotesensing/instructions. Particularly, the "Materials and Methods" section should be appropriately named, and a clear distinction should be made between the "Results" and "Discussion" sections. Such separation is indispensable to distinctly present objective results and the authors’ interpretations and implications of these results, as clearly advised by the journal.

Author Response

The authors extend their appreciation to the reviewer for their valuable feedback. All of the reviewer's comments have been thoroughly addressed, and their suggestions have been incorporated into the discussion section of the manuscript. Detailed responses to specific comments are provided below.

Author Response File: Author Response.pdf