Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Comments

Abstract:

-       There is a lack of consistency between the title and the content of the author's abstract. The abstract indicates that the paper focuses on model competition and data-driven comparison to produce canopy cover information on a regional scale. However, the title suggests that the paper will discuss and analyze the main barriers to producing wall-to-wall canopy height information. This could be achieved by conducting a systematic or extensive literature review about model limitations, the effect of data quality in terms of spatial and temporal coverage, and a comprehensive analysis of the limitations of already published methodologies or canopy height products. From my perspective, there is a discrepancy between what the title suggests and what the abstract actually covers. Therefore, I recommend changing the title to align with the document's goals.

 

 Introduction:

-       Lines 88-92. Please provide more details about the objectives of the paper. The goal of "determining the challenges" is too broad and does not provide a clear understanding of the problem being addressed. I suggest that the introduction should clearly state the problem and the questions that the researchers aim to answer. By setting specific objectives, you can limit the literature review to relevant material that supports your methods and demonstrates the scientific or applied contribution of your research.

 

Methods:

-       Lines 101-112. Please make sure to include an explanation in either this section or the introduction as to why conducting the study in this specific region is sufficient to fully explore the challenges involved in generating a canopy height estimation. It would be helpful to clarify why an alternative location with contrasting topography and forest compositions was not included in order to verify the accuracy and performance of the model.

-       Lines 121-129, 198-213, 229-237, 252-256, 288-301. I want to suggest moving these sections to the introduction. Adding a section in the introduction that describes the types of RS data/sources used for Canopy copy will be beneficial. It would be helpful to include a table that compares the relevant data attributes, their classification, and potential uses/limitations. This way, it will be easier to justify why you chose these RS data sources and focus on preprocessing details in the method section.

-       Lines 341-359, 372-391, 399-407. Similar to the idea above, This section could be relocated to the introduction. Additionally, a section only for ML in the context of Canopy cover prediction would be beneficial.

-       Line 440. I am confused. Are you also working on modeling the AGB?

-       Lines 502-507. It appears that this paragraph is repeating the definition of PRMSE. Are the RMSE and PRMSE computed using cross-validation? Usually, the cross-validation RMSE is a better measure of the model's predictivity ability.

-       Line 511. Do you have any thoughts on the maximum canopy height of the GEDI and ICESat-2? It looks quite high for your area of study.

-       Fig 6. Please revise a line off the position.

-       Lines 552-554. Although the two R2 are different, they are technically not comparable. Note that the total sums of squares for both regressions are different.

-       Line 557. Why use a graph to summarize statistics? Can't you present results in a table instead? What additional insights can be gained from the graph?

-       Section 3.5. I would like to suggest an improvement to your results interpretation. Please take into consideration the following reference: https://amstat.tandfonline.com/doi/full/10.1080/00031305.2016.1154108#.Vt2XIOaE2MN

-       Section 4. It appears that you have included more results in the discussion section. Please revise this section and focus on discussing the implications and significance of the results.

-       Section 4.5. What is the purpose of this section? It seems it’s not supported by your data or modeling results. 

-       Section 4.6. I appreciate the insightful discussion and analysis you provided. However, I noticed that your methods and results seem to be focused on developing a prediction model for a particular area. I didn't come across any findings that help to measure or assess the effects of the challenges mentioned on the accuracy of the canopy cover information.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This is a well-written and informative review of the manuscript "Challenges associated with generating wall-to-wall canopy height information from discrete data provided by spaceborne-LiDAR system". It accurately summarizes the key points of the material and methods section, including the data sources, machine learning models, and evaluation metrics used in the study. It is a clear and concise study, with a balanced discussion of the results and limitations. Nevertheless, some minor points should be addressed. In particular,

1. Place c and d images under a and b in order to enlarge Figure 1.

2. If possible, provide a brief analysis of the strengths and weaknesses (especially the limitations of your study) of the manuscript, going beyond just summarizing the content within the discussion.

3. Remove colors from the tables if are not necessary.

4. Provide a new image for Figure 15. At this point, it makes manuscript not readable.

 

Comments on the Quality of English Language

A minor editing of the English language is required.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Brief summary:

This study investigates challenges in generating wall-to-wall canopy height using machine learning models from discrete data provided by spaceborne LiDARs. The research compares models like gradient boosting, k-nearest neighbor, and random forest, using ICESat-2 and GEDI-based canopy height data for training. Results indicate gradient boosting as the most accurate model, followed by k-nearest neighbor and random forest, with respective optimal RMSE/PRMSE values. Integration of Sentinel-1 with Sentinel-2 data enhances canopy height modeling accuracy, emphasizing the importance of accurate canopy height estimation for sustainable forest management and mitigating potential environmental impacts.

General concept comments:

The abstract provides a succinct overview of the study's objectives, methods, and key findings, emphasizing the importance of accurate canopy height estimation for forest management. The conclusion effectively summarizes the main results and implications of the study, highlighting the significance of factors such as satellite data, regression models, and training data in predicting canopy height accurately and its crucial role in various ecological and environmental applications.

Integrating data from diverse remote sensing sources, each with varying resolutions, poses challenges in achieving a homogeneous dataset for analysis. This study adeptly addresses these challenges by rescaling data from sources like Sentinel-1, Sentinel-2, topographic data, and LiDAR to ensure consistency. However, the process of rescaling may introduce potential impacts on the original data quality, necessitating careful consideration during analysis. Despite these challenges, the integration of heterogeneous data sources offers valuable insights, underscoring the importance of methodological rigor in addressing data heterogeneity.

Understanding the complexity of vegetation structure is crucial for effective application of machine learning models in canopy height prediction. However, this study doesn’t highlight the challenges in capturing the diverse characteristics of vegetation, such as variations in tree species, forest types, and growth stages.

Specific comments:

1.     Subheading 1, lines 78 and 79:

The statement implies that TLS, UAV-LiDAR, and ALS can provide continuous data, but it suggests a limitation in terms of cost.  The illogical aspect in the text is the implication that the coverage is inherently limited to small areas solely due to cost. In reality, the coverage area depends on various factors beyond just cost, including the methodology employed and the operational capabilities of the equipment used.

2.     Subheading 1, lines 109 and 110:

The statement regarding the average annual destruction of 526 hectares by fire within and around the area from 1963 to 2019 is pertinent to the discussion. However, it is crucial to note the absence of a specific source or reference to substantiate this claim. Including a proper citation or reference is essential to ensure the credibility and reliability of the information presented. Providing the source not only enhances the transparency of the article but also enables readers to verify the accuracy of the data and evaluate its relevance to the topic at hand.

3.     Subheading 1, figure 1

The representation of the "Study site boundary" in Figure 1 appears to be flawed, possibly due to layers being superimposed incorrectly. Correcting this discrepancy is essential to ensure the accuracy and clarity of the visual depiction of the study area's boundaries.

4.     Subheading 1, figure 1

The image provided as Figure 1 outlines the geographical location of the study site and training/validation data; however, it lacks essential spatial information such as the north direction or a coordinate grid. Including these elements is crucial to enhance the map's utility and aid in spatial orientation for readers analyzing the study area.

5.     Subheading 2․2․2, lines 208-209 and 214-215:

Both statements highlight the importance of utilizing various methods, such as multiple spectral bands, vegetation indices, and ancillary data, to mitigate saturation issues in canopy height estimation. While the phrasing differs slightly, they essentially convey the same concept․

6.     Subheading 2:6, lines 418 and 419:

The phrase "satellite data, regression models, and training data" suggests that these are factors, but in reality, they are tools or components used in the analysis. It would be more accurate to say something like: "The factors that were identified to be analyzed in this study include characteristics of satellite data, types of regression models, and composition of training data." This way, it clarifies that the factors being analyzed are related to these components rather than being the components themselves.

7.     Subheading 3․3

The explanation of the results, particularly regarding the identification of overfitting in the GB and RF models and the analysis of feature importance, could be more detailed. Providing specific examples or insights into why certain configurations or predictors were deemed more effective would enrich the interpretation of the findings.

8.     Figures 10, 11, 14

The absence of spatial components such as a north sign or coordinate grid in Figures 10, 11 and 14 detracts from its utility as a spatial distribution visualization. Including these elements would enhance the clarity and orientation for readers analyzing the spatial features depicted in the figure.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Additional Comments:

 

Abstract: Please include a brief mention of the study site in the abstract. Your findings are linked to a particular area, and they may not be applicable to other regions.

 

Introduction: You have made improvements to the introduction section, but the purpose/questions of this work are still unclear. Please be more specific. "Generating wall-to-wall canopy..." does not clearly explain the paper's contribution. Instead of summarizing what you are doing in the paper, make a list of specific objectives you have accomplished in the paper.

 

Regarding the maximum canopy height of GEDI and ICESat-2, I am not recommending removing data from your analysis, but please ensure that you provide a discussion of the possible causes of these generated extreme values in your data sources and the implications these few observations could have for your model evaluation/performance.

Thanks

Author Response

Please see the attachment

Author Response File: Author Response.pdf