Geographic Object-Oriented Analysis of UAV Multispectral Images for Tree Distribution Mapping in Mangroves

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

The paper utilizes Unmanned Aerial Vehicle (UAV) high-resolution remote sensing technology, combined with Geographic Object-Based Image Analysis (GEOBIA) and Machine Learning (ML) methods, to classify mangrove species at two ecologically distinct sites in Brazil: Cardoso Island and Suape. This demonstrates the powerful potential of this technology in detailed mapping of mangrove species and structure.

1 Research Significance and Innovation: The topic of the paper has high practical significance and scientific value. The protection and management of mangroves require precise species distribution information, and UAV remote sensing technology provides an efficient and economical solution. The combination of GEOBIA and ML for mangrove species classification is an innovative methodological exploration that helps overcome the limitations of traditional remote sensing classification methods, improving classification accuracy and practicality.

2. Results and Discussion: The research results clearly showcase the superiority of this method in species classification, providing higher-precision mangrove species distribution maps compared to traditional methods. The paper provides an in-depth discussion of the results, not only verifying the effectiveness of the method but also exploring its potential applications in ecological monitoring and protection, enhancing the practicality and impact of the research.

3. Writing and Structure: The paper has a clear structure and logical flow, with natural transitions between sections covering the research background, methods, results, and discussion, making it easy to understand.

4. Suggestions and Future Directions: Although the research has achieved significant results, further exploration is needed to assess the stability and applicability of this method under different environmental conditions (such as seasonal changes and lighting conditions). Consider expanding the research scope to more mangrove types or regions to verify the method's universality and provide broader data support for global mangrove protection. Additionally, further discussion on how to integrate this technology into existing mangrove management systems to enhance its practical application value is warranted.

Author Response

We are grateful for the thoughtful feedback and recommendations provided. Your insights have allowed us to refine our study and improve its clarity and impact. Below, we outline our responses to each point and describe the corresponding changes made to the manuscript.

 

1. Stability and Applicability under Different Environmental Conditions

Response: We recognize the significance of evaluating the robustness of our methodology under different environmental conditions, including seasonal variations and lighting differences. To address this, we have expanded the discussion at line 612 to examine how these factors could affect classification accuracy and to propose future research directions that explore seasonal and illumination impacts on UAV-based multispectral analysis.

 

2. Expanding the Study to Other Mangrove Types or Regions

Response: Following your suggestion, we have elaborated on the applicability of our method to diverse mangrove ecosystems. At line 637, we emphasize the necessity of validating our approach across multiple latitudinal gradients, geomorphological conditions, and species distributions. This revision strengthens our discussion on the adaptability of the methodology and the importance of broader ecological assessments.

 

3. Integration into Mangrove Management Systems

Response: To enhance the practical relevance of our research, we have included a new paragraph at line 601 discussing the integration of UAV-based classification techniques into existing mangrove monitoring and conservation programs. This section highlights the role of our approach in ecosystem management, restoration planning, and carbon stock estimation, demonstrating its potential for real-world conservation applications.

 

We deeply appreciate the time and effort dedicated to reviewing our work. Your constructive input has contributed significantly to strengthening our manuscript, and we thank you for your valuable contributions.

Reviewer 2 Report (Previous Reviewer 3)

Comments and Suggestions for Authors

The manuscript addresses mangrove species classification using an OBIA-based approach with high-resolution UAV remote sensing data. However, the framework presented lacks novelty in both its methodology and result interpretations. To enhance the manuscript, the following suggestions are provided:

Comments:

1. Clarify the novelty of the approach, particularly in terms of its contribution to mangrove mapping and classification.

2. Provide additional details on the UAV data pre-processing steps and the derivation of the products used in the analysis.

3. Clarify the objectives and contributions of the study, highlighting its significance within the context of existing research.

4. Include details on the machine learning models, specifically the hyper-parameters used and the training process employed for the classification.

5. Provide an analysis of feature importance in relation to the classification process, which would strengthen the understanding of the model's performance and interpretability.

Author Response

We sincerely appreciate the comments and suggestions provided. Below, we address each point and outline the modifications made to the manuscript accordingly.

 

1. Clarify the novelty of the approach, particularly in terms of its contribution to mangrove mapping and classification.

Response: We acknowledge the need to better highlight the novelty of our approach. To address this, we have revised the introduction (Line 87) to explicitly emphasize the unique contributions of our study. A new paragraph has been included in the discussion section (Line 558), incorporating a recently published reference (2024) to reinforce the advancements presented in our study. These revisions clarify how our work builds upon existing methodologies and contributes to improved species-level classification in mangrove ecosystems using UAV-based GEOBIA and ML approaches.

 

2. Provide additional details on the UAV data pre-processing steps and the derivation of the products used in the analysis.

Response: We have included a new paragraph in the methodology section (Line 222) detailing the pre-processing steps undertaken before classification. This section now explicitly describes the pre-processing workflow, including radiometric and geometric corrections, mosaicking, and noise reduction techniques. We also clarify that shadow masking was applied during post-processing, as highlighted in Line 370, to improve classification accuracy and ensure consistency in data interpretation.

 

3. Clarify the objectives and contributions of the study, highlighting its significance within the context of existing research.

Response: To strengthen the articulation of the study's objectives and contributions, we have rewritten and expanded the corresponding paragraph in the introduction (Line 89). This revision provides a clearer delineation of the research gaps being addressed, particularly in improving the efficiency and accuracy of mangrove species classification through UAV-based OBIA and ML techniques. Furthermore, we explicitly discuss the implications of our findings for conservation planning and ecosystem monitoring.

 

4. Include details on the machine learning models, specifically the hyperparameters used and the training process employed for the classification.

Response: We have now included a detailed explanation of the machine learning hyperparameters and training process in the methods section (Line 471). The specific tools and libraries used (Python code) for model training and optimization are now described (the code with the hyperparameters tunning funcrions and libaries is avaiable in supplement material). This addition enhances transparency and reproducibility, providing insight into how parameters such as tree depth (for Random Forest), learning rates (for Neural Networks), and prior probability distributions (for Naïve Bayes) were selected and validated.

5. Provide an analysis of feature importance in relation to the classification process, which would strengthen the understanding of the model's performance and interpretability.

Response: A new paragraph has been added to the results section (Line 610), discussing feature importance in the classification process. We now explicitly present a statistical analysis of feature contributions, demonstrating how spectral, textural, and height-based parameters influenced model performance. Additionally, we refer to supplementary material that includes detailed statistical outputs reinforcing our interpretations. This inclusion enhances the interpretability of our findings and provides a stronger basis for the classification approach adopted.

We greatly appreciate the constructive feedback, which has helped refine and strengthen our manuscript. Thank you for your valuable insights.

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

This study applies UAV remote sensing technology combined with
geographic object image analysis (GEOBIA) and machine learning (ML)
methods to classify species in different mangrove areas. With
multispectral data captured by UAV, this study successfully achieved an
accurate classification of major mangrove species. This study helps to
improve the accuracy of UAV-based mangrove mapping, provides detailed
insight into the spatial distribution and structure of mangrove forests,
and can effectively contribute to the development of mangrove
conservation and management strategies.

1. Although the importance of mangroves and the advantages of UAV
technology are mentioned in the introduction, the specific
innovativeness and uniqueness of the study is not sufficiently
emphasised. In addition, an overview of the sample size and resolution
of mangrove classification studies is missing from the introduction. It
is recommended that the authors add an overview of the sample size and
resolution of mangrove classification studies to summarise the results
and shortcomings of previous studies in this field and to provide a more
solid theoretical foundation and background for this study. In addition,
the description of the problems facing the current study is vague. It is
suggested that the authors clearly list the specific objectives of this
study and closely relate the shortcomings of the existing studies to
highlight the necessity and relevance of this study, so that the readers
can quickly understand the core values and key issues of this study.

2. The layout of Figure 3 is slightly crowded, and it is suggested that
the authors redesign the layout to highlight the core content of each
step and add more text notes to explain the specific operation of each
step. For example, for key steps such as data collection, preprocessing,
feature extraction, model training and validation, the inputs, outputs,
and main operation processes of each step should be clearly labeled to
help readers better understand the overall research process and enhance
the logic and readability of the article.

3. This study lacks an in-depth discussion of the reasons for choosing
Random Forest, Plain Bayes, and Deep Neural Networks as machine learning
algorithms. It is recommended that the authors provide a detailed
explanation for the selection of these algorithms, including their
strengths and applicability in processing mangrove multispectral data
and species classification tasks, as well as a comparison with other
algorithms.

4. The diagrams and graphs in the article are clear but some of them are
not labeled in sufficient detail, which may cause difficulties for
readers to understand them, for example, the lack of clarity in the
color gradient part of the hotspot distribution graph in Figure 10.

5. The discussion section lacks in-depth analyses of possible sources of
error in species classification and their effects on the results. For
example, the study mentions the confusion between Rhizophora mangle and
Avicennia schaueriana, but does not analyze the reasons for these
errors. It is recommended that the authors provide an in-depth analysis
of the possible causes of these errors and discuss how classification
accuracy can be further improved in the future by optimizing the
algorithm, enhancing the accuracy of data collection, or increasing the
sample size.

6. It is recommended that the authors strengthen the transition between
different sections of the paper to help readers better understand the
background of the study, the purpose of the study, and the expected
results after implementing the methodology.

Comments on the Quality of English Language

Pay attention to the fluency and coherence of the language, avoiding the
hard splicing of sentences. 

Author Response

We appreciate your thorough review and constructive feedback. Your comments have helped us refine our manuscript, and we have made the following revisions in response to each point.

 

1. Emphasizing the Innovativeness and Uniqueness of the Study

Response: We have revised the introduction to better highlight the novelty of our approach. Additional considerations have been incorporated at lines 54, 94, and 102, explicitly addressing the relevance, originality, and objectives of the study. These modifications provide a clearer distinction between our work and previous research, reinforcing its contribution to mangrove mapping and classification.

 

2. Improvements to Figures

Response: Based on your feedback, we have remade the relevant figure(s) to enhance clarity and ensure a more effective visual representation of the methodology and findings. We hope the revised version better illustrates the core concepts of the study.

 

3. Methodology Clarifications

Response: We have clarified specific methodological aspects in line 366, ensuring that key details regarding data processing and classification steps are explicitly stated. This should provide a clearer understanding of our workflow and analytical approach.

 

4. Figure Contrast and Quality

Response: We have enhanced the contrast in figures and maps to improve readability. However, we acknowledge that the quality of the proof version may not fully reflect the resolution of the original figures. We trust that the editorial team will assess the final quality to ensure optimal visual clarity in the published version.

 

5. Addressing Possible Sources of Classification Errors

Response: We have expanded the discussion section by adding two new paragraphs at line 263, specifically analyzing potential sources of classification errors. These additions provide a more in-depth examination of challenges such as species misclassification, shadow effects, and variations in spectral responses, strengthening the discussion on model accuracy and reliability.

 

6. Paper Structure and Flow

Response: While the structure of the paper follows a conventional format (Introduction, Methods, Results, Discussion, and Conclusion), we have made efforts to improve the coherence and connectivity between paragraphs. These refinements should enhance the logical flow of the manuscript and address any concerns regarding readability and transitions.

We sincerely appreciate your valuable feedback, which has helped us refine and improve our manuscript. Thank you for your insightful comments and suggestions.

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 Authors

1. 1. Abstract: Describe briefly how the model improves upon traditional methods.

   2. Line 33-58: Start with a more focused statement on the significance of forests. For example: "Forests are critical to Earth's ecological balance, climate regulation, biodiversity conservation, and human resource provision."

   3.line 60-82: Enhance Explanation of Prediction Models, Explain why parameters are difficult to ascertain and how this impacts prediction accuracy. 

   4. line 89-92: Refine the objective to directly state the benefits and expected outcomes.

   5. Please provide a data table that clearly specifies the source, type, time, and access time for all data.

   6. Providing a Sample Reference Code in the Appendix to Enhance Reader Engagement

   7. Please correct the inconsistent font sizes/styles in the text of Figure 2 - the flowchart.

   8. The data used in this paper exhibits inconsistencies in temporal and spatial scales. How do you propose to correct or reduce these errors?

   9. When validating the model, in addition to comparing it with satellite data and previous research results, it is also advisable to consider adding some additional validation metrics.

   10. I suggest separating the conclusion and discussion sections..

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this study, the authors classified mangrove tree species using multispectral data collected by a DJI drone at 30, 50, and 120 m flight altitudes in two representative mangrove growth areas. Individual trees were segmented using e-Cognition software, and classification methods such as Random Forest, Naïve Bayes, and Deep Neural Networks were employed to classify the mangrove forests, and then the Getis-Ord statistic was used to identify hotspot areas for each species, leading to a more in-depth analysis of their spatial intensity and distribution patterns. The main problems with the thesis are listed below:

1. Introduction. Some research gap was mentioned in lines 77-85, however, there is on response in research purposes or research content. Paper presented “some methodological aspects remain still open (especially, point-3, ascertaining the ideal quantity of training samples required for accurate classifications)” in line 81, but I didn't see any aspect of sample optimization in the article.

2. Materials and Methods. Section 2.3 of the paper is missing. In section 2.4, were the segmentation parameters "ESP = 120, S = 0.9, and C = 0.5" appropriate for both typical zones simultaneously?

3. In section 2.5, five forest inventory sample plots were established. It is recommended that the spatial locations of the five forest inventory sample plots be plotted in Figure 1 and Figure 2. At the same time, if the data from the sample plots are applied to the classifier, it is necessary to add statistical information, such as the “maximum value, minimum value, mean value, etc.” of the samples, in order to enhance the credibility of the results.

4. In section 2.6, “Training samples for the classifiers were sourced from images captured during lowaltitude flights (with a resolution of 30m to 1.4cm per pixel)." and "For each class, a minimum of 130 samples were designated, with 100 allocated for training purposes and the remaining used to test the models' accuracy". It is recommended to include a map showing the spatial distribution of the training and validation samples, as well as descriptive statistics for these samples. For instance, what is the sample size in each of the two typical zones? What constitutes a training sample versus a validation sample? How many individual trees of each species were sampled?

5. In Figures 6 and 7, does the confusion matrix represent the classification accuracy of the classifier or the test accuracy? Please clarify. If only the classification accuracy is given, it is recommended to include the test accuracy as well.

6. The study frequently mentions multiple flight altitudes, but their description is somewhat unclear in the paper. Additionally, in lines 482-490, the role of different flight altitudes is reiterated, yet only some general observations about UAV performance are provided. The paper does not reach a clear conclusion regarding the optimal flight altitude or observation scale.

7. In the full text, it is not clear how the model is built, what features are used, and how many samples are used. And the sentence, "In contrast, the shape indices had minimal impact on the performance of the classifiers algorithms " in lines 480-481, which is a lack of corroboration. I really didn't see any details about any importance ratings for the modeling features (independent variables), please show more details on the modeling.

8. The Latin scientific names of species appearing in the text should be considered in italics. (e.g., lines 18-19).

9.In the discussion section (see lines 491-502), the random forest model shows slightly better performance than deep neural networks and plain Bayes in terms of object classifier performance . The conclusion is not  credible. It is suggested that the authors further should analyze whether it is because the training samples for deep learning are not large enough resulting in inferior classifier performance to random forests or whether it is a problem with the model itself.

Comments on the Quality of English Language

The quality of English is good in this paper.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Overall, the manuscript does not contribute any novel framework or methods for mangrove detection/identification or mapping. The author may choose to include the following comments to improve the manuscript. 

Comments

-            The objective of the study is not mentioned in a more specific way. For instance, contributing significantly to conservation and monitoring, how?

-            The use of UAVs has escalated in recent years for many applications such as precision agriculture [1], crop disease detection [2] and image segmentation  [3]. Better to include a standing paragraph highlighting the use case of UAV.

-            Fix the Typos throughout the paper. For instance,  pixel ~6,2 cm (line), dense dense point cloud (line-195) etc.

-            Is this needed “Further insights into these methodologies can be referenced in review works by [43, 44, 45] and references within”?

-            Provide some analysis of different band combinations in an appendix. (lines 223-224).

-            Enhance the Equations in the tables that are blurred.

-            Fig-3 is not readable. So methodological flow of data is hard to understand/read.

-            This text does not make any sense “This section may be divided by subheadings. It

-            should provide a concise and precise description of the experimental results, their interpretation, as well as the experimental conclusions that can be drawn. (lines 388- 390”)

-            This line is repeated twice “After various model parameterization tests, the best results (e.g. models that best 'fit' the data distribution) were incorporated into the previously described OBIA data models.”

-            Figure 7, only caption, no figure.

Overall maps need to be enhanced. They are not quite good in resolutions. Very blur.

 

-            The work is just the experimentation of existing methods (OBIA and ML) into a new dataset. No Novelty at all.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf