Benchmarking Geometry-Based Leaf-Filtering Algorithms for Tree Volume Estimation Using Terrestrial LiDAR Scanners
Round 1
Reviewer 1 Report (New Reviewer)
Comments and Suggestions for AuthorsThis paper compared four method of leaf-filtering with clear method introduction and reasonable results. It was well written and might be very interested by the LIDAR users. The paper can be further improved in its figures and some parts of the method section.
1. More introduction and review is necessary for leaf-filtering to give a better background.
2. Lines 191-196 is not clear enough.
3. Section 2.2.4 might be better to give details about how the random forest used to leaf-filtering in previous studies just as the structure of other used methods.
4. Section 2.2 and section 4.2 could be better integrated. Section 2.2 is the background of leaf-filtering method and can be more concise, and looks not necessary to introduce one by one as you introduced much specific information in Section 4.2. Section 4.2 could give more details about their algorithm mechanism and the processes.
5. 4.1 AdQSM should be 4.3? Is it better to put in section 4.2.2 Volume Comparison?
6. Line 425 the referred paper should be added.
7. 4.2 Performance Assessment should be 4.4?
8. The letter in equation 5-6 should be clearly described.
9. Lines 466-468 could you explanation about why those variables played important roles?
10. Figure 3 is not very readable, it can put the data (different point cloud densities) in one figure together. Or can be deleted as you discussed some related information in figure11. By the way, the figures are lack of appreciation and academic.
11. It will be better to show and compare some classified point cloud in the form of 3D information like Figure A1 with more details, instead of all showed statistics in the text.
Author Response
Comment 1: More introduction and review is necessary for leaf-filtering to give a better background.
Response 1: We have provided a sufficiently detailed introduction and review concerning leaf-filtering and the algorithms used. If there are specific areas within the introduction and review that you believe need further attention or expansion, please let us know.
Comment 2: Lines 191-196 is not clear enough.
Response 2: We appreciate your attention to detail. After a thorough review, we believe the description accurately conveys the semi-supervised learning techniques employed by the CANUPO algorithm. If you have any further suggestions or specific points to address, please feel free to let us know.
Comment 3: Section 2.2.4 might be better to give details about how the random forest used to leaf-filtering in previous studies just as the structure of other used methods.
Response 3: We have addressed this concern by providing details on how random forest was used for leaf-filtering. As not many studies reported RF classification for wood-leaf classification, we included relevant literature that has utilized it.
Comment 4: Section 2.2 and section 4.2 could be better integrated. Section 2.2 is the background of leaf-filtering method and can be more concise, and looks not necessary to introduce one by one as you introduced much specific information in Section 4.2. Section 4.2 could give more details about their algorithm mechanism and the processes.
Response 4: Thank you for your valuable suggestion, but as different leaf-filtering algorithms have distinct concepts, we introduce the background in Section 2.2 and provide detailed information on how we utilized and optimized parameters in Section 4.2. In response to your feedback, we have added a brief paragraph in Section 2.2 to synthesize the explanations and provide readers with a clear picture. This enhancement is highlighted in the updated manuscript. We hope this adjustment improves the flow and understanding of our methodology. If you have any further recommendations, please feel free to let us know.
Comment 5: 4.1 AdQSM should be 4.3? Is it better to put in section 4.2.2 Volume Comparison?
Response 5: Thank you for pointing this out. We've corrected this in the updated manuscript and shifted the relevant paragraph to Section 4.2.2, as suggested.
Comment 6: Line 425 the referred paper should be added.
Response 6: We appreciate your observation and have included the reference in the updated manuscript.
Comment 7: 4.2 Performance Assessment should be 4.4?
Response 7: Thank you for catching that. We've made the correction in the updated manuscript.
Comment 8: The letter in equation 5-6 should be clearly described.
Response 8: While we found the letters used to be self-explanatory, we have included a paragraph explaining them in the updated manuscript.
Comment 9: Lines 466-468 could you explanation about why those variables played important roles?
Response 9: We've added a paragraph (lines 478-486) explaining why those variables are crucial, as suggested.
Comment 10: Figure 3 is not very readable, it can put the data (different point cloud densities) in one figure together. Or can be deleted as you discussed some related information in figure11. By the way, the figures are lack of appreciation and academic.
Response 10: We've addressed this issue in the updated manuscript to ensure better readability of Figure 3. If there are specific areas needing improvement, please let us know.
Comment 11: It will be better to show and compare some classified point cloud in the form of 3D information like Figure A1 with more details, instead of all showed statistics in the text.
Response 11: We've added figures and tables for each resulting statistic defined or computed. If there are specific areas for improvement, please let us know.
Reviewer 2 Report (New Reviewer)
Comments and Suggestions for AuthorsThe manuscript entitled “Benchmarking Geometry-Based Leaf-Filtering Algorithms for Tree Volume Estimation using Terrestrial LiDAR Scanners” examined the applicability of terrestrial LiDAR data for the discrimination of wood and leaf, and the estimation of the tree volume. Authors used four different filtering algorithms, and they assessed and compared the accuracy of these algorithms. The results indicated the highest accuracy when RF forest algorithm was employed and high accuracy of tree volume estimation. The paper highlighted the importance of the choice of algorithm to achieve highly accurate classification results.
The methods are clearly described and the results are also clearly presented. The conclusion is supported by the results indicating the future research directions.
The manuscript is well-written and also interesting. It provides valuable information for researchers in the similar field and forest management. The manuscript is publishable in the present form and I have no comments.
Thank you.
Author Response
Thank you for your positive feedback.
Reviewer 3 Report (New Reviewer)
Comments and Suggestions for AuthorsThis article is written and well-organized. The introduction and analysis are reasonable, given the premise of the article.
Firstly, I would suggest including links to access the data on the location of the research area or by countries. As mentioned, this study relies on open-source data from three sources or five references, which would be better in Table 2 to include the original data link and the type of data source.
Secondly, to be clearer on the study area, I suggest including a map of the study area or process flow that can informatively explain the study area, and the type of data, data processing, data validation, and data datasets for each country.
Then on page 18, line 562, the effect of point cloud density, what are the limitations of previous studies, to answer the effectiveness of cloud density by 30%, 50%, 70%, and 90% in confirming these findings. This includes answering in the conclusion how the scale and efficiency of using the assessment through this algorithm can be applied in aspects of forestry monitoring, especially in areas that have challenges with cloud cover.
Author Response
Comment 1: This article is written and well-organized. The introduction and analysis are reasonable, given the premise of the article.
Firstly, I would suggest including links to access the data on the location of the research area or by countries. As mentioned, this study relies on open-source data from three sources or five references, which would be better in Table 2 to include the original data link and the type of data source.
Response 1: Thank you for your positive feedback. However, the links for the data access were provided in the Data Availability Statement. We have also included link in the Table 2 as per your suggestion.
Comment 2: Secondly, to be clearer on the study area, I suggest including a map of the study area or process flow that can informatively explain the study area, and the type of data, data processing, data validation, and data datasets for each country.
Response 2: We have included the map describing the geographical locations across different regions as Figure 1 highlighted in updated manuscript.
Comment 3: Then on page 18, line 562, the effect of point cloud density, what are the limitations of previous studies, to answer the effectiveness of cloud density by 30%, 50%, 70%, and 90% in confirming these findings. This includes answering in the conclusion how the scale and efficiency of using the assessment through this algorithm can be applied in aspects of forestry monitoring, especially in areas that have challenges with cloud cover.
Response 3: As far as we know, none of the studies reported the impact on leaf-filtering algorithms in terms of different point cloud densities. Regarding the challenges with cloud cover, terrestrial lidar does not possess problems in such scenarios.
Reviewer 4 Report (New Reviewer)
Comments and Suggestions for AuthorsThe article presents an evaluation of different algorithms to leaf-filtering algorithms for Tree volume estimation using a terrestrial LiDAR scanner.
The article is well-written and well-structured. The references used are adequate and updated. The state of the art is good, the goals are well presented and defined. The methodology is quite extensive but well explained. Good discussion and conclusions
Minor comments.
- There are duplicated sections, for example: 2.2.3 CANUPO and 4.2.3 CANUPO. The first one is state-of-the-art, and the second presents the methodology. Maybe is better to present the methods (or algorithms) once (in methodology). It will reduce the extension of the article and the repetition of sections.
- I miss a map representing the study areas.
- Table 3 needs to be described better in its title and 5.
- The section results start with a chart, Is strongly recommended to introduce the chart narratively before the chart.
Author Response
Comment 1: - There are duplicated sections, for example: 2.2.3 CANUPO and 4.2.3 CANUPO. The first one is state-of-the-art, and the second presents the methodology. Maybe is better to present the methods (or algorithms) once (in methodology). It will reduce the extension of the article and the repetition of sections.
Response1: We appreciate your suggestion, but as the different leaf-filtering algorithms have different concepts, we first define the concept behind their algorithms in the introduction section (Section 2.2), and then explain how we used and optimized parameters for different algorithms in Section 4.2.
Comment 2: - I miss a map representing the study areas.
Response 2: We've included the map describing the geographical locations across different regions as Figure 1, highlighted in the updated manuscript.
Comment 3: - Table 3 needs to be described better in its title and 5.
Response 3: We've revised the title of Table 3 as per your suggestion, and highlighted in the updated manuscript.
Comment 4: - The section results start with a chart, Is strongly recommended to introduce the chart narratively before the chart.
Response 4: Thanks for pointing this out. We've updated this as per your suggestion in the updated manuscript.
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 is a very useful comparison, as people who use TLSs and lack the requisite programming experience (particularly with python) must navigate a bevy of papers and/or methods that are being proposed, most of which do not work very well for data from different tropical forest types or dense forest stands that do not have evenly spaced trees. Moreover, the inclusion of an assessment of the impact of subsampling is great as many papers that tout different methods do not include the effects of subsampling leaving the reader to decide how best to approach this issue.
Specific comments
To make this comparison even more useful, especially for people who are not seasoned users of TLSs, the authors should possibly include the information the software they used in the methods (and not the results) and add more information such as did they use CANUPO in CloudCompare? This was not made sufficiently clear. Additionally, can I assume that the geometric features were first generated in CloudCompare and then exported to Python? If so, please state this as this will help immensely for anyone not familiar with this approach.
Additionally, and this is important, the authors state that accuracy decreases as a function of trees size. For some users of TLS like me who use TLS in a range of forest types, some of which tend to be dense and have much more smaller trees than large trees, having some idea of how much accuracy will decrease with tree size is important. I might be asking for too much, but is it possible for the authors to include a figure with tree size versus accuracy for each method? The x-axis could just include DBH classes such as 2 – 10, 10 – 20, 20 – 30, etc. and accuracy on the y-axis? This would be very useful. Also, food for thought, but perhaps the inclusion of relative height resulted in decreased accuracy with smaller trees for the RF? I have found that not including height as a predictor in a RF increases the classification accuracy for features that are closer to the ground, i.e., smaller trees. This may or may not be the case for the authors assessment though.
The authors seem to have a missing reference as there was an Error! Reference source not found message found through out the manuscript.
This is a bit minor, but there is no need to constantly use the phrase “It is observed that”. Unfortunately, it is commonly used phrase in undergraduate reports.
Author Response
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.
Specific comments:
Comment 1: To make this comparison even more useful, especially for people who are not seasoned users of TLSs, the authors should possibly include the information the software they used in the methods (and not the results) and add more information such as did they use CANUPO in CloudCompare? This was not made sufficiently clear. Additionally, can I assume that the geometric features were first generated in CloudCompare and then exported to Python? If so, please state this as this will help immensely for anyone not familiar with this approach.
Response: The information about the software has been incorporated in the introduction section. CANUPO is employed as a plugin in CloudCompare, as specified in Section 3.2.3. Additionally, the generation of geometric features in CloudCompare and their subsequent computation in Python using the jakteristics library is clarified in Section 3.2.4.1, highlighted in yellow.
Comment2: Additionally, and this is important, the authors state that accuracy decreases as a function of trees size. For some users of TLS like me who use TLS in a range of forest types, some of which tend to be dense and have much more smaller trees than large trees, having some idea of how much accuracy will decrease with tree size is important. I might be asking for too much, but is it possible for the authors to include a figure with tree size versus accuracy for each method? The x-axis could just include DBH classes such as 2 – 10, 10 – 20, 20 – 30, etc. and accuracy on the y-axis? This would be very useful. Also, food for thought, but perhaps the inclusion of relative height resulted in decreased accuracy with smaller trees for the RF? I have found that not including height as a predictor in a RF increases the classification accuracy for features that are closer to the ground, i.e., smaller trees. This may or may not be the case for the authors assessment though.
Response 2: Figure 10 now presents the accuracy versus DBH scatter plot, addressing this valuable suggestion.
Comment 3: The authors seem to have a missing reference as there was an Error! Reference source not found message found through out the manuscript.
Response 3: It was likely a result of an inaccessible error during the conversion from .docx to .pdf. The necessary corrections have been made to ensure the accurate and complete presentation of references in the document. If you encounter any lingering concerns or notice further discrepancies, please bring them to our attention for thorough resolution.
Comment 4: This is a bit minor, but there is no need to constantly use the phrase “It is observed that”. Unfortunately, it is commonly used phrase in undergraduate reports.
Response 4: We rephrased several sentences in the manuscript to avoid “It is observed that…..” These changes highlighted with yellow.
Reviewer 2 Report
Comments and Suggestions for AuthorsDear author,
First of all, thank you for this interesting paper. Although I was not able to read an flawless version of the paper, since there seems some errors occurred during the creation of the pdf, I would like to state the importance of error coefficients (such as RMSE) of the validation results.
Author Response
Thank you for your positive feedback. I would like to inform you that, in our study, we opted for Mean Absolute Deviation (MAD) and Mean Absolute Percentage Deviation (MAPD) as our error coefficients instead of RMSE. The decision to use MAD and MAPD was driven by their advantages in our specific context. Unlike RMSE, MAD directly uses the absolute values of the errors, does not magnify large deviations i.e. not contaminated by outliers, not influenced by the magnitude or direction, making it a more suitable choice for our study. We believe that this approach provides a more accurate representation of the error metrics in the context of our research. If you have any specific questions or concerns about this choice, please feel free to let us know.
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper evaluated various algorithms designed for the crucial task of wood-leaf 3D TLS point classification. Overall, the structure of this paper is well organized, and the presentation is relatively clear. However, there are still some crucial problems that need to be carefully addressed before a possible publication.
1. The motivations or remaining challenges are not so clear as or what kinds of issues or difficulties this task is facing. Please give more details and discussion about the key problems solved in this paper, which is largely different from existing works.
2. A deep literature review should be given, particularly regarding state-of-the-art deep learning and AI methods in remote sensing and point clouds, e.g., 10.1109/LGRS.2017.2764938, 10.3390/rs15092380, 10.1016/j.isprsjprs.2023.03.022, and 10.1109/LGRS.2023.3294748.
3. Please clarify the contributions further. For example, which are your existing ones and which are your own ones?
4. It is well-known that the point data tend to suffer from various occlusions, and noise effects in the point cloud processing. Please give the discussion and analysis by referring to the paper titled e.g., ASFM-Net: Asymmetrical Siamese Feature Matching Network for Point Completion. The reviewer is wondering what will happen if the proposed method meets the various occlusions.
Author Response
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.
Specific comments:
Comment 1: The motivations or remaining challenges are not so clear as or what kinds of issues or difficulties this task is facing. Please give more details and discussion about the key problems solved in this paper, which is largely different from existing works.
Response 1: While numerous wood-leaf separation algorithms have been developed, there remains a lack of clarity regarding their comparative effectiveness, inherent strengths and weaknesses, and the scope of their applicability across diverse forest types. Questions about the achievable accuracy of different algorithms also persist. This research addresses these critical gaps by systematically testing various algorithms and introducing a robust RF algorithm. Our comprehensive evaluation includes a thorough comparison of their performances in terms of pointwise classification accuracy, volume extraction through AdQSM, impact under decreased point cloud density through subsampling, and computational efficiency. By addressing these aspects, our research aims to provide valuable insights into the selection and optimization of wood-leaf separation algorithms, contributing to advancements in forest inventory methodologies.
Comment 2: A deep literature review should be given, particularly regarding state-of-the-art deep learning and AI methods in remote sensing and point clouds, e.g., 10.1109/LGRS.2017.2764938, 10.3390/rs15092380, 10.1016/j.isprsjprs.2023.03.022, and 10.1109/LGRS.2023.3294748.
Response 2: While the concept is intriguing, our current study has not delved into the realm of deep learning technologies. This area holds potential for future exploration, offering an opportunity to compare the accuracy enhancements brought by deep learning methods over machine learning approaches in the specific context of wood-leaf separation. This investigation would involve scrutinizing the extent of improvement, if any, achieved by deep learning models compared to the RF model elucidated in this research paper. Moreover, it requires an examination of the additional computational power demands and overall efficiency associated with deploying deep learning methods. Deciding whether to adopt deep learning over machine learning becomes a pivotal question, particularly given the notable accuracy achieved by the RF model presented in our research. These questions remain part of the future work in this domain.
Comment 3: Please clarify the contributions further. For example, which are your existing ones and which are your own ones?
Response 3: LeWoS and TLSeparation were initially designed for leaf-wood separation, while the CANUPO algorithm was primarily employed for rock, soil, and vegetation classification. In our study, we repurposed CANUPO for wood-leaf classification, evaluating and comparing its performance in this context. Additionally, although previous studies have employed the RF algorithm, their models exhibited certain limitations. In contrast, our research introduces a robust RF model that not only surpasses the performance of other algorithms but also demonstrates versatility. Notably, we trained the RF model on a singular dataset (Cameroon) but validated its effectiveness across a diverse range of datasets, including Guyana, Indonesia, Peru, and Germany. This emphasizes the universal applicability of our RF model.
The text about same is added in manuscript and highlighted with blue.
Comment 4: It is well-known that the point data tend to suffer from various occlusions, and noise effects in the point cloud processing. Please give the discussion and analysis by referring to the paper titled e.g., ASFM-Net: Asymmetrical Siamese Feature Matching Network for Point Completion. The reviewer is wondering what will happen if the proposed method meets the various occlusions.
Response 4: The study exclusively utilized real-world data rather than simulated datasets. The training data, being real and inherently subject to occlusions, ensures that the developed RF model adapts to such conditions. It is expected that the model's performance might decline with increased occlusion. In the specific context of leaf-wood separation, the RF model exhibits stability, leveraging its pointwise classification approach on real-world data. In contrast, the performance of other algorithms, like TLSeparation, which involves voxelization in its intermediate steps, may be more susceptible to influences and variations.
Reviewer 4 Report
Comments and Suggestions for AuthorsComment to Authors:
While the authors have employed a valuable dataset, the manuscript suffers from significant shortcomings in its presentation. The writing lacks coherence and flow, making it challenging to grasp the content. I kindly request you to revises the manuscript to ensure it is well-organized and concise, thereby enhancing its readability for the intended audience. Please see some other points below.
Abstract:
1. L10-12: You may wish to revise as “Terrestrial LiDAR scanners (TLS) have the potential to revolutionize forestry by enabling precise estimation of forest inventory parameter at plot level such aboveground biomass, vital for forest carbon management.”
2. L12-15: This study addresses ……….. diverse global locations. Merged these two sentences.
3. L17: Delete (QSM), as you did not use it later in abstract.
4. L18-19: Just simply say RF model outperformed in place of these: “groundbreaking innovation”, “remarkable robustness”. I suggested using other models before saying that, for example Stochastic Gradient Boosting.
Introduction:
Please follow a well structure. Current introduction has no structure and lack of flow, in appropriate referencing. Please simply follow the structure below: 1st paragraph- answer the question “Why Forest inventory parameters need to be precisely measured?” here. It is a general background with problem statement in current forest inventory and how TLS and your potential modeling approach could help to assess forest inventory parameters more precisely than conventional measure. Write these paragraph 4-8 sentences.
2nd paragraph:
Answer the question: How traditionally forest inventory parameters estimated?
Write here about conventional forest inventory and its limitations. End with a transitional sentence to move TLS based approached. Write this paragraph 5-10 sentences with appropriate referencing.
3rd and 4th paragraphs:
Literature on TLS application in forest inventory parameter assessment. What are the different methodologies applied? What are their limitations? What are the possible ways to improve it?
Which approach do you want to focus on and what the current literature on it?
5th paragraph
What are your specific objectives and where do you want do this study briefly mention. Do you have any hypotheses related to your objectives? Mention all these issues here.
5. L34-57: Please add references appropriately.
6. L57 -62: Please keep author name in the text and in reference dropped author name.
7. L92 Other models also perform better than RF. Please check literature intensively...
8. L102-115, Your paper is an article not a review paper. Finish your introduction here and what ever you want to say about your methodology integrate it into your method and materials part concisely with appropriate citations.
9. L115-416 Write a concise methodology with proper citations.
Result:
10. Organized your result accordingly with your objectives.
11. 608-616 In result just mention what you have found. In discussion, evaluate your result.
Discussion:
Write a short introductory paragraph on your key findings in line with your objectives and write a one sentence implication. Then go ahead with your deep dive on the discussion evaluation: three point you mention here.
12. L651-653 For making a paragraph you need at least three sentences: a topic sentence, body (you may extend based on the context) and a concluding sentence. Follow this across the whole manuscript.
Conclusions: Is it a separate discussion! Please make it smaller than your abstract.
Comments on the Quality of English Language
The quality of the writing in the manuscript is subpar, and I recommend that it undergoes review by both an expert in the field of study and a professional editor. This collaboration can significantly improve the overall quality of the document.
Author Response
Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.
Specific comments:
Comment 1: L10-12: You may wish to revise as “Terrestrial LiDAR scanners (TLS) have the potential to revolutionize forestry by enabling precise estimation of forest inventory parameter at plot level such aboveground biomass, vital for forest carbon management.”
Response 1: We have revised the text as per your recommendation and highlighted with green.
Comment 2: L12-15: This study addresses ……….. diverse global locations. Merged these two sentences.
Response 2: We have revised the text as per your recommendation and highlighted with green.
Comment 3: L17: Delete (QSM), as you did not use it later in abstract.
Response 3: We have revised the text as per your recommendation.
Comment 4: L18-19: Just simply say RF model outperformed in place of these: “groundbreaking innovation”, “remarkable robustness”. I suggested using other models before saying that, for example Stochastic Gradient Boosting.
Response 4: We have revised the text as per your recommendation and highlighted with green.
Comment:
Introduction:
Please follow a well structure. Current introduction has no structure and lack of flow, in appropriate referencing. Please simply follow the structure below: 1st paragraph- answer the question “Why Forest inventory parameters need to be precisely measured?” here. It is a general background with problem statement in current forest inventory and how TLS and your potential modeling approach could help to assess forest inventory parameters more precisely than conventional measure. Write these paragraph 4-8 sentences.
2nd paragraph:
Answer the question: How traditionally forest inventory parameters estimated?
Write here about conventional forest inventory and its limitations. End with a transitional sentence to move TLS based approached. Write this paragraph 5-10 sentences with appropriate referencing.
3rd and 4th paragraphs:
Literature on TLS application in forest inventory parameter assessment. What are the different methodologies applied? What are their limitations? What are the possible ways to improve it?
Which approach do you want to focus on and what the current literature on it?
5th paragraph
What are your specific objectives and where do you want do this study briefly mention. Do you have any hypotheses related to your objectives? Mention all these issues here.
Response: Thank you for providing the suggested structure. We've already implemented a structure similar to your recommendation. Our arrangement comprises sections on Motivation and Challenges, the Role of RS Technologies, Issues with Allometric Equations, Integration of TLS and ALS, the Importance of Leaf-filtering Algorithms, the Objectives of the Study, and an Introduction to the Algorithms.
Comment 5: L34-57: Please add references appropriately.
Response 5: We have revised the text as per your recommendation and highlighted with green.
Comment 6: L57 -62: Please keep author name in the text and in reference dropped author name.
Response 6: We have revised the text as per your recommendation and highlighted with green.
Comment 7: L92 Other models also perform better than RF. Please check literature intensively...
Response 7: Your observation is correct but we chose RF method, aligning with numerous studies that have highlighted its robustness across various domains. One key advantage is its ability to mitigate overfitting issues, which enhances the model's generalization performance. Additionally, RF offers valuable insights into feature importance, proving beneficial for our analysis. Its robustness to outliers, parallelization capabilities, and user-friendly nature further contribute to its suitability for our study.
Comment 8: L102-115, Your paper is an article not a review paper. Finish your introduction here and what ever you want to say about your methodology integrate it into your method and materials part concisely with appropriate citations.
Response 8: We have revised the text as per your recommendation.
Comment 9: L115-416 Write a concise methodology with proper citations.
Response 9: Following the content on L115, the manuscript proceeds with the continuation of the Introduction. This section introduces various algorithms for wood-leaf separation and provides insights into the algorithms adopted for this study. Subsequently, the document transitions to a dedicated section on Materials and Study Area to provide readers with essential background information. The narrative then seamlessly transitions into the Methodology section, which, while concise, effectively outlines the key steps and approaches employed in our study. This section is structured to ensure clarity and coherence, guiding readers through the methodology in a streamlined manner.
If you have any specific areas within these sections that you would like to emphasize or if there are particular details you'd like to discuss further, please feel free to provide additional guidance or questions.
Comment 10: Result: Organized your result accordingly with your objectives.
Response 10: We structured the manuscript by first delving into individual algorithm results. Initially, we meticulously outlined the features selected for developing the RF model. Following this, we dedicated sections to elucidate the effects of parameter optimization in both LeWoS and TLSeparation algorithms. These results were considered prerequisites as they laid essential groundwork before delving into our main objectives.
By addressing the specifics of each algorithm and their nuances, we aimed to provide readers with a comprehensive understanding of the intricacies involved. These detailed insights not only contribute to the transparency of our methodology but also set the stage for the subsequent sections that focus on our main results.
Should you have any particular aspects you'd like to emphasize further or if there are specific points you wish to discuss, please feel free to provide additional guidance. We remain committed to ensuring the clarity and coherence of the manuscript
Comment 11: 608-616 In result just mention what you have found. In discussion, evaluate your result.
Response 11: We have revised the text as per your recommendation.
Comment: Discussion:
Write a short introductory paragraph on your key findings in line with your objectives and write a one sentence implication. Then go ahead with your deep dive on the discussion evaluation: three point you mention here.
Response: We have revised the text as per your recommendation and highlighted with green.
Comment 12: L651-653 For making a paragraph you need at least three sentences: a topic sentence, body (you may extend based on the context) and a concluding sentence. Follow this across the whole manuscript.
Conclusions: Is it a separate discussion! Please make it smaller than your abstract.
Response 12: We have revised the text as per your recommendation and highlighted with green. While we've endeavoured to streamline the content, achieving a length shorter than the abstract proved challenging. This is primarily due to the necessity of presenting detailed outcomes, algorithmic behaviors, pros and cons, potential improvements, identified gaps, and future directions. We've carefully structured the information to ensure clarity and coherence. If there are specific areas you suggest for further trimming or if you have additional preferences, please provide guidance, and we'll make the necessary adjustments accordingly.
Round 2
Reviewer 3 Report
Comments and Suggestions for AuthorsMy concerns have not been addressed, and the novelty of this paper is limited.
Reviewer 4 Report
Comments and Suggestions for Authors
After a careful review of the revised manuscript, I regret to inform you that I am not satisfied with the current state of the paper. However, I firmly believe that with certain improvements, the paper has the potential to meet the desired standards.
Feedback for Improvement:
Introduction:
The introduction currently lacks structure, flow, and proper referencing. It should serve as a strong foundation for the rest of the paper.
The specific objectives of the study should be explicitly stated. Additionally, hypotheses related to these objectives should be clearly mentioned.
The paper should focus on its content rather than attempting to be a review. The introduction should be concise, and any method-related content should be integrated into the method and materials section.
Consider referring to a well-structured paper (e.g., Cao et al., 2023;Benchmarking airborne laser scanning tree segmentation algorithms in broadleaf forests shows high accuracy only for canopy trees) as a guide for how to improve the introduction and overall paper structure.
Methodology:
It is advisable to use a 10-fold cross-validation approach and perform an ANOVA to assess potential differences between the algorithms used in the modeling validation and performance.
Ensure that the methodology section is clear, concise, and includes proper citations.
Citations and References:
Proper citation and referencing are essential throughout the manuscript. Address any issues with citations (e.g., L264-268).
Provide clarity in references (e.g., L307-308, "For example???").
Results:
The figures presented in the manuscript need improvement. They should be defined more clearly and organized to enhance readability and comprehension.
The results should be organized according to the study's objectives to provide a clear and logical flow.
Discussion:
The discussion should be closely aligned with the study's objectives and results. It should serve to interpret the findings and their implications.
Begin the discussion with an introductory paragraph summarizing key findings and their broader implications.
Ensure that the discussion logically flows from the results and directly addresses the study's objectives. Provide insights and interpretations based on the study's goals and outcomes.
Conclusions:
The conclusion section should be concise. It should summarize the key takeaways based on the study's objectives and results.
Comments on the Quality of English Language
Manuscript need to improve the writing for maintating flow and coherence.
