UAV-Based Multispectral Imagery for Area-Wide Sustainable Tree Risk Management
, Tomasz Jelonek 2, Adam Juźwiak 3 and Marcin Kubus 4Round 1
Reviewer 1 Report (Previous Reviewer 2)
Comments and Suggestions for Authors- The authors should further validate effectiveness by adding experiments of more than two kinds of heights and more than two kinds of speeds for UAV/UAVs. A major issue of this manuscript is that the dimensions of data, experiments, or parameters may be too limited. It is recommended that the authors spend more time to supplement the dimensions of the research contents of the manuscript by rearranging corresponding experiments.
- The authors should add major contributions for the research work of the manuscript to the section of 1. Introduction.
- The authors should open a new section which should be 2.6. Evaluation Metrics.
- The authors should mention the combination of ISA/BMP and TRAQ methods in section of 3. Results.
- The authors should conduct more quantitative experiments and analysis in section of 3. Results. For example, the authors should conduct more experiments for deep analysis of differentiation and effectiveness for each of all three categories of dead trees, low vegetation, trees in good condition, and bare soil in 3. Results.
- The authors should adjust heights of Figure 1(a) and Figure 1(b) to make them consistent in heights. Besides, an overall title should be added to the title of Figure 1.
- The authors should adjust heights of Figure 2(a), Figure 2(b) and Figure 2(c) to make them consistent in heights. Besides, an overall title should be added to the title of Figure 2.
- The authors should make Figure 1(b), Figure 2(c) and Figure 12 more distinguishable, as those three Figures currently appear to have a lot of repetition in their contents.
- The authors should add more images to Figure 2. For example, the authors may need to add images under different application scenes such as UAVs of different heights, UAVs under different backgrounds, etc.
- The authors should use a three line table to rearrange Table 2.
- The authors should reformat the references among line 730 and line 837 according to format requirements of this journal. For example, all lines after the first line of each reference should be indented from the left margin.
- The authors should revise the title of the manuscript. The revised title should be similar with Sustainable Area-Wide Risk Management of Trees Using UAV-Based Multispectral Aerial Imagery.
- The authors should remove revision marks as much as possible, since it may be very difficult for reviewers to review a manuscript having too much revision marks. The authors could highlight revisions by a bright background color.
- The authors should use vector Figures as much as possible to represent Figure 1, Figure 2, Figure 3, Figure 4, Figure 6, Figure 7, Figure 8, Figure 9, Figure 10, Figure 11, Figure 12, and Figure 13, and ensure texts involved in those Figures could be easily identified.
- The authors should make following Figures more distinguishable and combine them as much as possible, i.e., Figure 5, Figure 6, Figure 7, Figure 8, Figure 9, Figure 10, and Figure 11.
Comments on the Quality of English LanguageThe English of the manuscript could be further polished.
Author Response
|
1. Summary |
|
|
|
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding corrections in track changes in the re-submitted files.
|
||
|
2. Point-by-point response to Comments and Suggestions for Authors |
||
Comments and Suggestions for Authors
Comments: The authors should further validate effectiveness by adding experiments of more than two kinds of heights and more than two kinds of speeds for UAV/UAVs. A major issue of this manuscript is that the dimensions of data, experiments, or parameters may be too limited. It is recommended that the authors spend more time to supplement the dimensions of the research contents of the manuscript by rearranging corresponding experiments.
Response: We sincerely thank the Reviewer for this valuable suggestion. We acknowledge that in the current version of the manuscript, the experiments were conducted using a limited number of UAV heights and speeds. While we did not include extended analyses with more varied flight parameters in this study, we agree that expanding the range of experimental conditions would enhance the generalizability and robustness of the results. We will certainly consider incorporating a broader set of heights and speeds in our future work to further validate and strengthen our findings.
Comments: The authors should add major contributions for the research work of the manuscript to the section of 1. Introduction.
Response: We thank the Reviewer for this valuable comment. In the revised manuscript, the Introduction was expanded to highlight the main contributions of our work: (I) providing the first operational test of UAV-based multispectral imagery for tree risk detection with 78% accuracy, and (II) formulating a second hypothesis that CIR imagery is more reliable than RGB for identifying declining trees, with clear operational implications for large-scale risk management. We believe these additions strengthen the focus and novelty of the study, while keeping the Introduction concise and methodologically relevant.
Comments: The authors should open a new section which should be 2.6. Evaluation Metrics.
Response: We appreciate the Reviewer’s suggestion. In the current version of the manuscript, the evaluation metrics were described within the relevant subsections of the methodology section, where they are directly related to the specific analyses performed (e.g. subsection 2.2 with description of classification of declining trees as trees with NDVI values in the range 0.20–0.28 and exhibiting partial crown dieback or subsection 2.5. standardized a fall risk zone of 1.5 tree heights determined as the hazard zone). For clarity and contextual relevance, we chose to present the metrics alongside the procedures they pertain to. However, if required, we are open to reorganizing the content and adding a separate Section 2.6 dedicated to Evaluation Metrics.
Comments: The authors should mention the combination of ISA/BMP and TRAQ methods in section of 3. Results.
Response: We sincerely thank the Reviewer for this valuable suggestion. The widely used tree stability assessment methods of recent decades, namely the ISA/BMP and TRAQ methods — whose combination was applied in this manuscript — consider dead and declining trees as hazardous. For the purposes of this study, a common element from these approaches was adapted: according to the standards, as mentioned above, a fall risk zone equal to 1.5 times the tree height was used to define the hazard zone. According to the Reviewer suggestion we have added in Result section (subsection 3.2. Hazard Zone Delineation) the name of methods determining the risk zone caused by trees (lines 471-472).
Comments: The authors should conduct more quantitative experiments and analysis in section of 3. Results. For example, the authors should conduct more experiments for deep analysis of differentiation and effectiveness for each of all three categories of dead trees, low vegetation, trees in good condition, and bare soil in 3. Results.
Response: We thank the Reviewer for this suggestion. However, the scope of this study is primarily methodological, and it is not feasible within a single field experiment to control and quantitatively analyze all possible categories (dead trees, low vegetation, healthy trees, bare soil). In natural sciences dealing with living organisms, such comprehensive control is practically impossible. We therefore focused on demonstrating the workflow and providing representative results, while clearly stating the limitations. We also emphasize in the Discussion and Future Work sections that broader experiments including additional categories are needed to further validate and generalize the approach.
Comments: The authors should adjust heights of Figure 1(a) and Figure 1(b) to make them consistent in heights. Besides, an overall title should be added to the title of Figure 1.
Response: The changes in the text have been made.
Comments: The authors should adjust heights of Figure 2(a), Figure 2(b) and Figure 2(c) to make them consistent in heights. Besides, an overall title should be added to the title of Figure 2.
Response: The changes in the text have been made.
Comments: The authors should make Figure 1(b), Figure 2(c) and Figure 12 more distinguishable, as those three Figures currently appear to have a lot of repetition in their contents.
Response: Given that each of the figures presents a different aspect of the research — Fig. 1(b) (border in context), Fig. 2(c) (the UAV mission plan, photo from the DJI RC Pro controller), and Fig. 12 (tree risk buffer zone) — and due to concerns that merging these elements might reduce clarity of provided information, we propose keeping them in their current, separate form.
Comments : The authors should add more images to Figure 2. For example, the authors may need to add images under different application scenes such as UAVs of different heights, UAVs under different backgrounds, etc.
Response: Since Fig. 2 illustrates the equipment used in the study, the UAV in flight, and the mission display on the DJI RC Pro controller and given that the study was conducted using a single flight altitude within a specific spatial context we believe that adding other UAV use scenarios would not reflect the actual experiment and could obscure the key information. Therefore, we propose keeping Fig. 2 in its current form.
Comments: The authors should use a three line table to rearrange Table 2.
Response: The changes in the text have been made.
Comments: The authors should reformat the references among line 730 and line 837 according to
format requirements of this journal. For example, all lines after the first line of each reference should be indented from the left margin.
Response: The changes in the text have been made.
Comments: The authors should revise the title of the manuscript. The revised title should be similar with Sustainable Area-Wide Risk Management of Trees Using UAV-Based Multispectral Aerial Imagery.
Response: In accordance with the reviewer's suggestion, we have revised the title to: “UAV-Based Multispectral Imagery for Area-Wide Sustainable Tree Risk Management”.
Comments: The authors should remove revision marks as much as possible, since it may be very difficult for reviewers to review a manuscript having too much revision marks. The authors could highlight revisions by a bright background color.
Response: We thank the Reviewer for catching this. All revision marks have been removed. The document now contains only tracked corrections to allow for better control and transparency of the modifications made.
Comments: The authors should use vector Figures as much as possible to represent Figure 1, Figure 2, Figure 3, Figure 4, Figure 6, Figure 7, Figure 8, Figure 9, Figure 10, Figure 11, Figure 12, and Figure 13, and ensure texts involved in those Figures could be easily identified.
Response: We thank the Reviewer for this suggestion. Most figures in our manuscript (e.g., UAV orthomosaics, classification outputs, mission screenshots) are raster-based products generated directly by UAV sensors and photogrammetric software, which do not allow for vector representation without loss of detail. Where feasible (e.g., schematic diagrams or maps), we ensured that outputs were exported in high resolution with clearly legible labels to maximize readability. We believe this approach preserves both clarity and fidelity of the original data.
Comments: The authors should make following Figures more distinguishable and combine them as much as possible, i.e., Figure 5, Figure 6, Figure 7, Figure 8, Figure 9, Figure 10, and Figure 11.
Response: The figures mentioned in the comment illustrate different aspects of the conducted research. Therefore, we propose keeping them separate, as combining them may result in the loss of important information.
|
4. Response to Comments on the Quality of English Language |
|
Point 1: The English of the manuscript could be further polished. |
|
Response: We thank the Reviewer for the remark. We have taken great care to ensure the correctness and clarity of the English language throughout the manuscript. |
|
|
Author Response File: 
Author Response.docx
Reviewer 2 Report (New Reviewer)
Comments and Suggestions for Authors1. Summary and General Recommendation
This manuscript is dealing with the quite relevant and also practically important issue of using UAV-based multispectral imagery for detecting hazardous or declining trees in a suburban pine stand, and although the general framework is clear enough and the results like 78% agreement with ground inspection are presented, still I found that there are inconsistencys, methodological gaps, and formatting issues which together reduce the reliability of the study, and therefore my judgement is that the paper should be considered only after major revision and not in the current form, because without these clarifications the contribution is not strong enough.
2. Section-by-Section Comments
(A) Title and Abstract
-
The abstract at present is rather vague, since it does not mention key detail such as the exact size of the study area (6.69 ha), the number of ground reference trees which was only 51, nor does it explain that the red-edge band was collected but not used, and therefore the reader cannot reproduce the experiment; I suggest that the authors expand the abstract by including these information and also to quantify the relative saving in labor and costs, since only saying “significant” is not scientific (L34–L51; L52–L60).
(B) Introduction
-
The introduction is very much descriptive about general problems of institutions lacking capacity, but it does not focus on what exactly is the research gap in UAV-based tree risk detection, and there is no clear hypothesis formulated (like CIR imagery being more reliable than RGB imagery), which makes the study objectives look blurred; also some of the references are wrong formatted, e.g. “[31, 3227,28]” which makes it confusing and must be corrected (L65–L81; L28–L33).
(C) Study Area
-
The description of the site is detailed and informative, however the assumption that the hazardous zone is defined simply as 1.5 tree heights in radius is given without any reference or justification, which is problematic because reviewers and practitioners would ask where this rule coming from (L3–L10).
(D) Methods
-
There is contradictory information: in one place the flight is said to have been conducted in December 2024 while in another part the results section writes that the survey was in “mid-summer,” so which is true, because this strongly affects leaf phenology and NDVI, and therefore the authors must resolve this inconsistency (L29–L35; L60–L63).
-
The description of RTK is confusing: the text says D-RTK 2 was used, but the figure clearly shows the module not mounted; please clarify if the UAV actually recieved correction signals or not, and what was the achieved positional accuracy (L18–L21; L13–L28).
-
The flight parameters reported (AGL=100 m, 70% overlap, GSD=4.15 cm) are insufficient, since crucial aspects like shutter interval, exposure settings, solar angle, and also processing parameters in WebODM are missing, and without them the reproducibility is very low.
-
The red-edge (RE) band was available but not used, while in the discussion the authors admit that NDVI-RE is often more sensitive for pine stress, so this omission is strange and needs justification (L55–L57; L41–L44).
-
The NDVI thresholds (0.20 and 0.28) are based only on 51 trees, with no cross-validation, ROC analysis or sensitivity test, which makes them arbitrary and not robust, thus stronger statistical treatment is required (L9–L14; L34–L40).
-
The ground survey reliability is not described: who did it, what expertise they had, if there was more than one observer, and whether agreement was measured (like Cohen’s kappa), which is important for validation (L73–L81).
(E) Results
-
The results only provide an overall agreement of 78% but not a full confusion matrix, so we cannot know about false positive or false negative rates, therefore the authors should report precision, recall, specificity etc (L19–L22; L42–L45; L67–L68).
-
Misclassification is briefly mentioned, that dead trees were confused with hard surfaces, but this is not analyzed in detail, although adding a training class “bare soil or pavement” could solve it (L11–L17).
-
Figures numbering is inconsistent, with strange labels like “Fig. 125” and “Fig. 136,” which looks like editing errors and must be cleaned (L21–L24; L29–L30; L7–L13).
(F) Discussion
-
The discussion mentions that UAV data may be more reliable for simple structures like avenues, but it does not really give actionable guidance for managers, e.g. which zones (playgrounds, paths) should be prioritized, and such practical implications would strengthen the contribution (L32–L41; L48–L51).
-
When comparing with other studies, the authors only say “similar accuracy,” but they do not explain why their study reached 78% and not higher, nor do they control for differences in stand type or season (L1–L7).
-
Future work is only superficially mentioned, but I believe it would be important to propose concrete improvements, like conducting repeat flights in different seasons, using oblique imagery to reduce canopy occlusion, and integrating LiDAR (L12–L19).
(G) Conclusion
-
The conclusions restate feasibility but are too generic; they should include more operational guidance, like minimum data quality standards, workflow steps, and thresholds for acceptable accuracy (L17–L29).
-
Data availability statement is weak (“on request”); for transparency and reproducibility, it is better to deposit UAV images and shapefiles in an open repository (L7–L10).
3. Formatting and Editorial Issues
-
Problem 1 (L73–L74): Section numbering errors (e.g., “2.32.1”). Please fix.
-
Problem 2 (L1–L5): Residual Polish formatting marks (“sformatowano”) appear throughout. Remove.
-
Problem 3 (L18–L21; L13–L28): Inconsistent RTK statements between text and figure captions. Standardize.
-
Problem 4 (L42–L49; L16–L23): Reference formatting inconsistent; missing DOIs. Correct per journal style.
4. Language Issues
-
Example 1 (L65–L71): “The study was conducted assuming flight altitude …” → revise to “The flights were conducted at an AGL of 100 m …”.
-
Example 2 (L36–L39): “In Addition the drone …” → should be “In addition, the drone …”.
-
Example 3 (L69–L70): “ground Sampling Distance” → “ground sampling distance.”
-
Example 4 (L30–L37): Long technical descriptions of battery specs are excessive; move to appendix.
Author Response
|
Response to Reviewer 2 Comments
Comments and Suggestions for Authors Comments 1: (A) Title and Abstract The abstract at present is rather vague, since it does not mention key detail such as the exact size of the study area (6.69 ha), the number of ground reference trees which was only 51, nor does it explain that the red-edge band was collected but not used, and therefore the reader cannot reproduce the experiment; I suggest that the authors expand the abstract by including these information and also to quantify the relative saving in labor and costs, since only saying “significant” is not scientific (L34–L51; L52–L60). Response: We thank the Reviewer for this valuable comment. We agree that the abstract should provide more specific details for transparency and reproducibility. We have therefore included the exact study area size (6.69 ha), the number of ground reference trees (51), and explicitly noted that although the red-edge band was collected, it was not used in the current analysis. Furthermore, we quantified the relative savings in labor and cost compared to traditional surveys, rather than using only qualitative wording (lines 23-27).
|
|||||||||
|
Comments 1: (B) Introduction
Response: We acknowledge this comment. In the revised manuscript, we clarified the research gap in UAV-based tree risk detection by emphasizing the lack of operational validation studies in public safety contexts. We also reformulated the hypothesis, stating explicitly that CIR multispectral imagery is expected to be more reliable than RGB imagery in detecting declining trees (lines 118-124). Furthermore, we corrected the reference formatting errors.
Comments 1: (C) Study Area
Response: Thank you for pointing this out. We respectfully note that this information is already included in the manuscript (section 2.5, Tree risk assessment, lines 385-395), where we explicitly state that the fall risk zone was defined as 1.5 times tree height according to ISA/BMP and TRAQ standards. To make it even clearer, we highlighted the reference to ISA/BMP (2012) and ISA TRAQ (2021) in the revised text. Comments 1: (D) Methods
Response: We thank the Reviewer for catching this. The UAV flight took place in December 2024, while the ground survey was conducted in November 2024; the isolated “mid-summer” wording in Results was a residual error and has been corrected in Sections 3.3 (and Limitations where applicable). We also clarify in Section 2.3.1 that a DJI D-RTK 2 GNSS base station was used externally to provide real-time corrections; it was not mounted on the UAV, which explains the appearance of the aircraft in the photo, and it ensured sub-meter positional accuracy.
Response: For reproducibility, we have added in the Manuscript that imagery was acquired with automatic exposure and 0.7-s shutter-interval; processing used WebODM v3.3.2 with default settings unless stated otherwise (Sections 2.3.1–2.3.2). These additions complement the reported AGL, overlap and GSD and allow the workflow to be reproduced with the same hardware/software stack.
Response: We thank the Reviewer for spotting this. We added a short justification in Section 2.3.2: due to full overcast and low solar elevation during acquisition, the RE mosaics were inconsistent, so we focused on G, R, NIR and NDVI composites for stable interpretation. This decision is now explicitly stated (lines: 315-317).
Response: We agree that cross-validation would be desirable; however, the reference set (n = 51) limits the stability of ROC-derived thresholds. We therefore transparently report local, context-dependent thresholds (0.20; 0.28) calibrated against field vitality classes and explicitly caution against treating them as universal. We note this as a limitation and a priority for future, larger-n studies.
Response: Thank to the Reviewer for this valuable comment. We now state in the Manuscript that the ground survey was performed twice by an experienced ISA-TRAQ-certified arborist, two weeks apart, focusing on dead and R3 trees per Roloff (Section 2.4.1). While we did not compute inter-observer statistics, the repeated survey by a certified specialist improves reliability and is acknowledged as a limitation.
Comments 1: (E) Results
Response: Thank you for this comment. We agree that a full confusion matrix is a common way of reporting classification accuracy. However, in our case the reference dataset was limited to 51 trees, which makes the resulting values of precision, recall and specificity statistically unstable and of limited interpretative value. For this reason, we reported the overall agreement (78%) and additionally described the nature of false negatives (22%) in detail, which we believe provides the necessary transparency. We also emphasized this limitation in the Discussion and suggested that future studies with larger datasets should include full confusion matrices.
Response: Thank you, we agree. We clarified in Section 3.5 that future deployments should add a dedicated “asphalt/concrete” class and consider simple texture/morphological filters to reduce confusion between dead wood and hard surfaces.
Response: Thank you for spotting this. We systematically cleaned figure numbering and cross-references; all figures now follow a continuous sequence and the legends were de-duplicated (see Fig. 13).
Comments 1: (F) Discussion
Response: We thank the Reviewer for comments. We added actionable guidance in Section 4.1: after UAV triage, managers should prioritize high-occupancy micro-sites (playgrounds, main pedestrian routes, bus stops, benches/picnic areas, parking-lot edges) within ≥1.5× tree-height hazard buffers for first-pass ground verification of all dead/dying trees.
Response: We sincerely thank the Reviewer for this valuable suggestion. We have expanded the comparison in Section 4.1 to explain why our accuracy was 78% and not higher. Specifically, we now note that the lower agreement likely reflects under-canopy occlusion in a discontinuous canopy, the timing of a single late-autumn acquisition under overcast conditions, and nadir-only imaging at 4.15 cm GSD. These factors increase false negatives compared to multi-date or leaf-on datasets reported in other studies.
Response: We thank the Reviewer for this suggestion. We have now expanded Section 4.4 to include concrete directions for future work, highlighting multi-season acquisitions, oblique imagery, LiDAR integration, and advanced classification algorithms as ways to improve reliability.
Comments 1: (G) Conclusion
Response: We thank the Reviewer for these valuable remarks. We have strengthened the Conclusions by adding concrete operational guidance, including minimum data quality standards (e.g., GSD < 5 cm, CIR composites, ≥75% agreement with field inspections), recommended UAV flight parameters, and guidance on local NDVI threshold calibration and ground verification in high-use zones. In addition, we revised the Data Availability Statement: while raw UAV images cannot be openly shared due to municipal ownership restrictions, all processed outputs (NDVI rasters, shapefiles, classification results, and workflows) are now explicitly offered upon request to ensure transparency and reproducibility. Comments 1: Formatting and Editorial Issues
Response: The changes in the text have been made.
Response: The UAV was equipped with the optional D-RTK 2 High Precision GNSS mobile station, which ensured enhanced positioning accuracy, RTK as optional equipment, is not visible on the Figure, but was used during data collection. We provided the explanation in the text.
Response: The references have been formatted according to the journal's style.
4. Response to Comments on the Quality of English Language |
|||||||||
Response: The changes in the text have been made.
Response: The changes in the text have been made.
Response: The changes in the text have been made.
|
|||||||||
|
Response: The changes in the text have been made, we shortened the text and changed it to: “The UAV is powered by LiPo batteries with a total capacity of 5000 mAh, allowing for a maximum flight time of 43 minutes and coverage of up to 200 ha per charge.” |
|||||||||
|
|
|||||||||
Author Response File: Author Response.docx
Reviewer 3 Report (New Reviewer)
Comments and Suggestions for AuthorsGeneral Comments
The main research addresses UAV-based multispectral aerial imagery as an effective and efficient tool for sustainable tree risk management over a wide area, particularly in identifying dead or declining trees, and how its accuracy compares to traditional ground-based surveys. The study specifically tests the hypothesis that remote sensing methods can detect high-risk trees at a level comparable to ground-based surveys, offering significant advantages in cost, labor, and scalability.
The topic is relevant and within the scope of the Sustainability Journal. Tree risk management in urban and peri-urban forests is a growing concern due to climate change, pest outbreaks, and public safety responsibilities. Using consumer-grade UAVs for this purpose is a fast-track approach. The application of multispectral sensors and vegetation indices for risk assessment adds a layer of practical applicability that is very timely.
The study focuses on a clear gap between technological capability and practical, validated application in forest management. While many studies utilize UAVs for forest health monitoring, few explicitly focus on their operational integration into public safety risk management protocols, directly comparing the results with standardized field verification methods. This bridges the gap between remote sensing science and applied arboriculture/forestry.
Specifically, this study adds several key elements compared to other published materials, including:
a) It provides a concrete and quantifiable accuracy rate (78%) for a specific, commercially available UAV system (DJI Mavic 3 Multispectral) in a real-world scenario, which is highly valuable for managers considering adopting this technology.
b) It frames the results not only in terms of tree health, but explicitly in the context of risk management (e.g., by defining a "fall risk zone" of 1.5x the tree height), making the results directly applicable to managers with legal safety obligations. c) It consciously uses equipment and software (WebODM, QGIS, SAGA GIS) accessible to public institutions, such as municipalities, making the study's conclusions more actionable for its target audience. d) The article honestly discusses false negatives (trees under cover, broken trees) and the challenges of absolute NDVI values, providing a realistic picture of the technology's capabilities.
The main results corroborate the central thesis of a 78% detection rate, corroborating the conclusion that UAVs are a valuable tool for large-area assessments but not a complete replacement for ground-based surveys, especially in high-traffic areas. The arguments regarding cost and time efficiency are logically derived from the methodology.
The limitations discussed are directly reflected in the results (a false negative rate of 22%).
The conclusion that CIR composites are more useful than RGB composites is corroborated by the experience reported by the authors and is well established in the remote sensing literature.
The conclusions accurately address the main issue, affirming the potential of UAVs for risk management over large areas while clearly describing their current limitations.
The reference list is generally appropriate and comprehensive, covering key areas: ecosystem services, tree risk assessment methods, UAV technology, and remote sensing applications in forestry.
Recommendations and Suggestions
The ground survey was conducted in November 2024, while the UAV flight was conducted in December 2024. For deciduous trees, this is a critical issue (leaves would have fallen, affecting the NDVI). Although the stand is predominantly pine, this should be explicitly addressed as a potential confounding factor, even if the impact is minimal. Future studies should strive for simultaneous data collection.
It is stated that the NDVI thresholds for dead (<0.20) and declining (0.20–0.28) trees were calibrated from 51 field trees. The calibration process requires further detail. Was this a statistical analysis or based on visual correspondence? A scatterplot of field-assessed vitality versus the NDVI value for these 51 trees would significantly strengthen this fundamental methodological step.
To reduce bias, the analysis of UAV imagery should ideally be performed by an analyst blinded to the ground survey results. It is unclear whether this was done. Stating that this protocol was followed would increase the objectivity of the results.
The study was conducted in a pure Scots pine stand. The authors should more explicitly state that the developed NDVI thresholds and overall accuracy are likely specific to conifer species and may not be directly transferable to mixed or broadleaf forests without recalibration.
The description of the supervised classification in SAGA GIS is brief. More information about the algorithm used, the number and type of training samples per class, and the validation method would be beneficial for reproducibility.
Author Response
|
1. Summary |
|
|
The main research addresses UAV-based multispectral aerial imagery as an effective and efficient tool for sustainable tree risk management over a wide area, particularly in identifying dead or declining trees, and how its accuracy compares to traditional ground-based surveys. The study specifically tests the hypothesis that remote sensing methods can detect high-risk trees at a level comparable to ground-based surveys, offering significant advantages in cost, labor, and scalability.
The topic is relevant and within the scope of the Sustainability Journal. Tree risk management in urban and peri-urban forests is a growing concern due to climate change, pest outbreaks, and public safety responsibilities. Using consumer-grade UAVs for this purpose is a fast-track approach. The application of multispectral sensors and vegetation indices for risk assessment adds a layer of practical applicability that is very timely.
The study focuses on a clear gap between technological capability and practical, validated application in forest management. While many studies utilize UAVs for forest health monitoring, few explicitly focus on their operational integration into public safety risk management protocols, directly comparing the results with standardized field verification methods. This bridges the gap between remote sensing science and applied arboriculture/forestry.
We thank very much the Reviewer for taking the time to review this manuscript. Please find the detailed responses below and the corresponding corrections in track changes in the re-submitted files.
|
2. Point-by-point response to Comments and Suggestions for Authors |
Comments and Suggestions for Authors
Comments 1: General Comments
Specifically, this study adds several key elements compared to other published materials, including:
- a) It provides a concrete and quantifiable accuracy rate (78%) for a specific, commercially available UAV system (DJI Mavic 3 Multispectral) in a real-world scenario, which is highly valuable for managers considering adopting this technology.
b) It frames the results not only in terms of tree health, but explicitly in the context of risk management (e.g., by defining a "fall risk zone" of 1.5x the tree height), making the results directly applicable to managers with legal safety obligations. c) It consciously uses equipment and software (WebODM, QGIS, SAGA GIS) accessible to public institutions, such as municipalities, making the study's conclusions more actionable for its target audience. d) The article honestly discusses false negatives (trees under cover, broken trees) and the challenges of absolute NDVI values, providing a realistic picture of the technology's capabilities.
The main results corroborate the central thesis of a 78% detection rate, corroborating the conclusion that UAVs are a valuable tool for large-area assessments but not a complete replacement for ground-based surveys, especially in high-traffic areas. The arguments regarding cost and time efficiency are logically derived from the methodology.
The limitations discussed are directly reflected in the results (a false negative rate of 22%).
The conclusion that CIR composites are more useful than RGB composites is corroborated by the experience reported by the authors and is well established in the remote sensing literature.
The conclusions accurately address the main issue, affirming the potential of UAVs for risk management over large areas while clearly describing their current limitations.
The reference list is generally appropriate and comprehensive, covering key areas: ecosystem services, tree risk assessment methods, UAV technology, and remote sensing applications in forestry.
Recommendations and Suggestions
The ground survey was conducted in November 2024, while the UAV flight was conducted in December 2024. For deciduous trees, this is a critical issue (leaves would have fallen, affecting the NDVI). Although the stand is predominantly pine, this should be explicitly addressed as a potential confounding factor, even if the impact is minimal. Future studies should strive for simultaneous data collection.
It is stated that the NDVI thresholds for dead (<0.20) and declining (0.20–0.28) trees were calibrated from 51 field trees. The calibration process requires further detail. Was this a statistical analysis or based on visual correspondence? A scatterplot of field-assessed vitality versus the NDVI value for these 51 trees would significantly strengthen this fundamental methodological step.
To reduce bias, the analysis of UAV imagery should ideally be performed by an analyst blinded to the ground survey results. It is unclear whether this was done. Stating that this protocol was followed would increase the objectivity of the results.
The study was conducted in a pure Scots pine stand. The authors should more explicitly state that the developed NDVI thresholds and overall accuracy are likely specific to conifer species and may not be directly transferable to mixed or broadleaf forests without recalibration.
The description of the supervised classification in SAGA GIS is brief. More information about the algorithm used, the number and type of training samples per class, and the validation method would be beneficial for reproducibility.
Response: We thank the Reviewer for the constructive and balanced feedback. We agree that the points raised are important for transparency and reproducibility. Below we provide clarifications and indicate how we addressed them:
Timing of surveys (November ground vs. December UAV).
The study stand is composed of >90% Scots pine, with only a minor admixture of deciduous trees. Therefore, the leaf-off timing had negligible impact on NDVI interpretation. We now explicitly acknowledge this in the Limitations section.
NDVI threshold calibration (0.20 and 0.28)
Thresholds were derived from visual vitality classes (Roloff) for 51 reference trees, and then aligned with corresponding NDVI values. Given the limited sample size, we refrained from formal ROC analysis, but we note in the Methods that thresholds were calibrated against field vitality assessment, and in the Discussion that future work with larger datasets should include scatterplots and statistical cross-validation.
Blinding of UAV analyst
The UAV classification was performed independently of the ground survey data to reduce bias; this protocol is now stated in the Methods (Section 2.4.1).
Species-specific thresholds
We explicitly note in the Discussion that thresholds and accuracy values are stand-specific and primarily applicable to Scots pine; transferability to broadleaved or mixed stands would require recalibration.
SAGA GIS supervised classification details
To improve reproducibility, we expanded the Methods (Section 2.3.2) with details on the algorithm used (Maximum Likelihood), training samples per class, and validation procedure.
We appreciate the Reviewer’s perspective and believe that these clarifications have strengthened the methodological transparency of the paper, while keeping the scope consistent with the current dataset.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report (New Reviewer)
Comments and Suggestions for AuthorsEverything is fine; the authors have addressed all my previous concerns about the manuscript. Therefore, I recommend accepting and publishing the manuscript immediately.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis study deals with an analytical method to efficiently identify dead and declining trees in forested areas by combining UAV image analysis with GIS analysis. Although similar methodologies have already been reported in several research papers, the results of tree health assessment are influenced by factors such as the spectral bands used for image analysis and the condition of the forest. Therefore, accumulation of case studies is important.
In the present paper, RGB and CIR orthomosaics are used to detect dead and declining trees. However, from the manuscript, the following points are not clearly described:
(1) What classification criteria were used in the image analysis to determine “dead” and “declining” trees.
(2) Whether the criteria were based on numerical thresholds (e.g., NDVI < 0.2), or on visual judgment of color tone and morphology.
(3) Whether the classification was done entirely by visual inspection, or by using automated or semi-automated classification methods.
(4) If thresholds were set, whether they were adopted from previous studies or calibrated using original field measurement data in this study.
Without a clear explanation of the analytical methods and classification criteria, reproducibility of this study cannot be ensured, and it is also impossible to evaluate the accuracy of the classification. Furthermore, the validity of the analytical method used in this study cannot be judged. In the current manuscript, there remains the possibility that the results are only a case where an analysis happened to be conducted. The paper only presents the detection rate (78%) without describing the analytical procedures and the basis for setting thresholds, which leaves problems from the viewpoint of research transparency and reproducibility. In particular, without any validity assessment, it is not possible to judge whether the method can be applied to other regions or other tree species.
Additional comments:
In Fig. 15, an area more than 1.5 times the tree height is indicated as a hazard zone. How was the tree height determined? Was it based on actual measurements or on image analysis? Also, in Fig. 15, there are elongated dark red areas—what do they represent?
In Fig. 16, classification results are presented. However, it seems that the individual trees judged as “dead” or “declining” in the field survey are not clearly indicated. In addition, from this figure alone, readers cannot understand what classification criteria were applied.
Line 425–428: Although tree species identification is mentioned, no related description can be found in the results section.
Line 438–441: The logic appears to make a sudden leap.
Line 452–453: Whether the values given in raster data are absolute values is a separate issue.
Line 460–461: I do not find any description in the text regarding the temporal change of NDVI.
Line 475–496: The content described here appears to be an introduction of previous studies and seems more appropriate to be placed in the Introduction section.
Reviewer 2 Report
Comments and Suggestions for Authors- The authors should cut texts among line 92 and line 94, and paste those texts to the end of texts in line 53. Besides, the authors should check whether there is a issue for "source: [links in original]" in line 79.
- The authors should merge texts among line 55 and line 84 into one paragraph, and then delete "Visual ground-based inspection" in line 46, "Current Remote Sensing Capabilities Using UAVs" in line 54, and "Land Cover Classification" in line 85.
- The authors should append major contributions of manuscript to the end of section of 1. Introduction.
- The authors should merge the Fig. 1 and Fig. 2 into one figure, where Fig. 1. should be located at the left, and Fig. 2. should be located at the right. Texts of Fig. 1. and Fig. 2. in the subtitles should be replaced by (a) and (b) respectively.
- The authors should merge the Fig. 3, Fig. 4 and Fig. 5 into one figure, where Fig. 3. should be located at the left, Fig. 4 should be located in the middle, and Fig. 5 should be located at the right. Texts of Fig. 3., Fig. 4. and Fig. 5. in the subtitles should be replaced by (a), (b) and (c) respectively.
- The authors should cut texts among line 147 and line 157, and paste those texts to the beginning of texts in line 271. The authors should delete "Tree Assessment" in line 146, and "UAV-Based Data Processing" in line 218.
- The authors should merge texts among line 175 and 178, and texts among line 185 and line 188, since texts in these two paragraphs are almost the same.
- The authors should delete "Assumed Flight Parameters" in line 202. Besides, the authors should merge texts among line 203 and line 207 into one paragraph, and cut the merged paragraph and paste the merged paragraph to the end of texts in line 195.
- The authors should reorganize the structure as follows:
- Line 105: Study Area -> 2.1. Study Area
- Line 158: Aerial Data Collection -> 2.2.1. Aerial Data Collection
- Line 209: Processing of Data Collected Using UAVs -> 2.2.2. Processing of Data Collected Using UAVs
- Line 231: Ground-Based Data Collection -> 2.3.1. Ground-Based Data Collection
- Line 263: Processing of Ground-Based Survey Data -> 2.3.2. Processing of Ground-Based Survey Data
- Line 270: Tree risk assessment -> 2.4. Tree risk assessment
- Insert a blank line and add text of "2.2. UAVs-Based Data Collection and Processing" on the previous line of the line for the title of 2.2.1.
- Insert a blank line and add text of "2.3. Ground-Based Data Collection and Processing" on the previous line of the line for the title of 2.3.1.
- The authors should merge the two categories of "trees in good condition" and "trees in very good condition" into one category of "trees in good condition".
- The authors present only one Fragment for most results. This way of presenting results may have locality. The authors should better present results for all four categories of dead trees, low vegetation, trees in good condition, and bare soil, in 3. Results.
- The authors should present key data results by tables, such as 78%, 7.62, 22%. Otherwise, it may difficult to understand relationship among those results.
- The authors should further validate effectiveness by adding experiments of more than two kinds of height and more than two kinds of speeds for the UAV.
- The authors should open a new section, namely 2.5. Evaluation Metrics.
- The authors should may mention the combination of ISA/BMP and TRAQ methods in 3. Results. Besides, the authors should conduct more quantitative experiments and analysis.
- The authors should divide 3. Results into several subsections, such as 3.1. XXX, 3.2. XXX, 3.3. XXX ...... Besides, the authors should better divide 4. Discussion into several subsections, such as 4.1. XXX, 4.2. XXX ......
- The authors should add the four sections of Author Contributions, Funding, Data Availability Statement, and Conflicts of Interest, before the References.
- The authors should reformat the references among line 602 and line 693 according to format requirements of this journal.
Comments on the Quality of English LanguageThe English of the manuscript could be further polished.
