Modeling Canopy Height of Forest–Savanna Mosaics in Togo Using ICESat-2 and GEDI Spaceborne LiDAR and Multisource Satellite Data

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

This revised manuscript discusses the estimation of forest canopy height in the forest-savannah mosaics of Togo using ICESat-2 and GEDI LiDAR data combined with multisource satellite data. The integration of diverse data sources and machine learning models remains a commendable effort. The authors have addressed some previous feedback, including better explanations of data usage and reporting some validation results. However, several issues must still be resolved before the manuscript can be considered for publication.

 

The primary contribution of the study lies in using multisource data for canopy height prediction. However, similar studies have already been conducted extensively. The authors need to clearly position their findings within the context of existing research and emphasize whether this study offers unique innovations in data processing, variable selection, model development, or application of results.

 

The manuscript employs four machine learning algorithms: Random Forest, SVM, XGBoost, and Deep Neural Network (DNN). However, the DNN model, despite its higher complexity, does not show significantly better performance than the others. This could indicate that the dataset lacks the feature dimensionality required for effective training of DNN.

 

When applying the model across the landscape, the manuscript continues to assume spatial homogeneity. Given the variability in canopy structure, spatial autocorrelation analysis or stratified validation is essential to ensure robust predictions.

 

Author Response

Please find attached the responses to the Reviewer1's comments.

Author Response File: Author Response.pdf

Reviewer 2 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

Regarding the manuscript entitled “Modelling Canopy Height of Forest-Savanna Mosaics in Togo Using ICESat-2 and GEDI Spaceborne LiDAR and Multisource Satellite Data”, I appreciate the authors' efforts to address my suggestions. However, I noticed that Question 3 has not been adequately tackled. If the authors address this issue, I believe the manuscript will be ready for publication.Even though I am not a native English speaker, I can read the manuscript comfortably and find it clear overall.

Additionally, I kindly request the authors to thoroughly proofread the manuscript, as I have identified several grammatical errors. Below are my specific comments:

Line 49–50: What is the difference between forest biomass and above-ground biomass? Please clarify this sentence.

Lines 99–100: The term “Random Forest (RF)” is mentioned twice. Please revise for clarity.

Line 99: “(r =) 0.78” – This requires clarification. Please check the text.

Line 132: The section title "2.1. Study aera" contains a grammatical error. Please correct it.

 

Line 159: Can this statement be supported by literature? If so, please include references. If it is not relevant can be removed.

Author Response

Please find attached the responses to the Reviewer 2's comments.

Author Response File: Author Response.pdf

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

Review Comments for the Paper: "Modelling Canopy Height of Forest-Savannah Mosaics in Togo Using ICESat-2 and GEDI Spaceborne LiDAR and Multisource Satellite Data "

 

This manuscript provides valuable insights into estimating forest canopy height using ICESat-2 and GEDI data combined with other remote sensing data. The integration of multiple models and data sources is commendable, and the results have the potential to contribute significantly to forest structure mapping. However, several issues need to be addressed, both in terms of content and the organization of the manuscript. I recommend a major revision before this work can be considered for publication.

 

Major Comments:

Content-related Issues

1. The analysis of feature importance, particularly the greater importance of optical and topographic data over radar data, is not adequately discussed. This finding contradicts some existing literature, where radar data have shown strong contributions to forest structure modeling. The authors need to provide a more detailed explanation or hypothesis for why this might be the case in the study area.
2. There is a lack of clear discussion on model validation outside the training dataset. The manuscript should include an evaluation of how the model might generalize to areas not included in the training process, especially given the heterogeneity of forest types in the study region.
3. The Random Forest (RF) model is said to outperform other models, but it remains unclear whether all models were optimized equally. The authors should provide more details on the hyperparameter optimization for each model and clarify whether fair optimization was conducted across all models to avoid biased performance comparisons.
4. The study implicitly assumes spatial homogeneity across the landscape when applying machine learning models, but the complex forest structures in Togo may not support this assumption. The authors should discuss the implications of spatial heterogeneity in the canopy structure and consider incorporating spatial autocorrelation measures or spatial cross-validation to account for these variations.
5. There is a potential issue of overfitting, especially considering the relatively large number of features and machine learning models used. The authors should discuss steps taken to prevent overfitting and report metrics such as training vs. testing error to provide evidence that overfitting has been avoided.

 

Structure-related Issues

1. The section numbering is inconsistent, and some sections lack clarity in their titles. For example, the "2.2. Methodology" section could be better organized to reflect the step-by-step process of data collection, preprocessing, and model development. I suggest to better organize the flow of the paper and separate major sections like "Data Processing," "Feature Extraction," and "Model Optimization."
2. There are several instances of repeated information, especially in the discussion section. The authors frequently repeat model performance statistics that were already presented in the results section. A more concise presentation, focusing on interpretation rather than reiterating figures, would improve the flow and readability.
3. The manuscript would benefit from clearer transitions between sections, particularly between the methods and results sections. A brief summary at the end of each major section could help the reader understand how each part of the study connects to the next.
4. The equations presented in the paper are not consistently formatted, and variables are not always clearly defined when introduced. For example, in Equation (1), some variables are explained much later in the text, which can confuse readers. All variables should be clearly defined directly alongside the equations.
5. The resolution, font, and font size of the figures are inconsistent throughout the manuscript. Several figures appear blurry and difficult to interpret, making it challenging for readers to extract meaningful information. It is crucial to ensure that all figures have a high enough resolution for clear visibility and that the font style and size are uniform across all visual elements for consistency. This will significantly improve the overall presentation of the results.

Minor Points

• Line 56-58: Specify whether the satellite data were atmospherically corrected.
• Line 195: Explain how field data were spatially matched with satellite data, and whether there were challenges due to differing resolutions.
• Line 249: Use consistent terminology when describing the relative heights extracted from GEDI and ICESat-2 (i.e., RH98). It’s unclear if the same height metric was used for both datasets.
• Line 339: Avoid overreliance on acronyms without proper introduction. Some readers may not be familiar with all terms, such as SHAP.
• Figure 6: Enhance label readability, as some text is too small and difficult to interpret.

Reviewer 2 Report

Comments and Suggestions for Authors

The research entitled “Modelling canopy height of forest-savannah mosaics in Togo using ICESat-2 and GEDI Spaceborne LiDAR and Multisource” is interesting and fits well with the aim of the journal. This study evaluates different powerful algorithms using a combination of S1/S2 and topographic metrics to predict canopy heights in savannah sites. Although the study is intersting, I encourage the authors to be more concise in the Materials and Methods and Results sections. Some parts of the Results read like Methodology sections, and certain sections of the Materials and Methods lack proper support from previous studies. The study also presents numerous abbreviations that are quite similar, which can hinder the readability of the manuscript. I recommend major revisions. I have no specific comments for the editors.

Title:

The title effectively informs readers about the study.

Abstract:

The abstract is concise, clear, and readable, providing an easy understanding of the study's main aim.

Introduction:

Line 42: I believe that forest clearing refers to the permanent removal of trees and bushes by humans. Were you referring to forest disturbance or cut-to-length harvesting activities?

Lines 47-50: Please rephrase and provide proper citations. I think the term “forest dynamics” is not properly linked to forest biomass, but their variability over time can provide key insights into forest dynamics. Canopy height and tree diameter are the most important variables for calculating volume, and therefore, biomass. Perhaps,”Canopy height is a key parameter for monitoring forest biomass” would be more accurate.

Lines 97-112: It would be better to include a table indicating the method, key details, achieved accuracy, and citations.

Line 124: Did you refer to the performance of models predicting canopy height using ICESat-2 and GEDI, or did you study the performance of each covariate and then separately ICESat-2 and GEDI? I suggest rephrasing for clarity.

Line 126: What do you mean by "establish"? Do you mean monitor, assess, or generate?

Materials and Methods:

Line 150: Can you roughly mention the names and percentages of tree species in the Togo Mountains? How is the forest cover in the Togo Mountains?

2.5.2: Data filtering is part of pre-processing and should be properly supported by scientific studies.

2.5.3: The zonal statistic is applied to a 10m pixel. What does the generated pixel measure?

I find the algorithm section too long. I suggest keeping only crucial information and moving the rest to supplementary material.

Results:

Lines 539-540: This reads like a Methodology section.

Figures 8 and 10: These figures seem more suited to the Methodology section, not the Results.

The results are very interesting and well explained, but I would prefer a more concise Results section.

The scenarios (S1-S7) and Config1-Config9 need to be clearly explained in the Methodology section.

 

Lines 124-125: This was already mentioned in the Materials and Methods. It would be advisable to move it there.

Reviewer 3 Report

Comments and Suggestions for Authors

line 59:

"Regardless of whether the data are optical or obtained from radar, signal saturation (especially in dense forests) can constitute a substantial limitation [32–35]"

This sentence is oversimplifying the obtained knowledge, especially in the filed of SAR.

Indeed, for shorter wavelengths such as X-band, the saturation exists in very dense forests if underlying topography is not known or selected baseline is not appropriate. However, the situation for L- and especially P-band is different.

The authors use "radar" to describe any SAR data which is also an oversimplification.  Depending on the polarimetric configuration (single-, dual-quad-pol) and (across-track) baseline, the saturation may or may not be reached. In this study, only Sentinel 1 C-band dual-pol SAR data with zero (across-track) baseline data are used.

Therefore, I recommend not using generalised term "radar" across the manuscript. The potential outllook, in my view, consists in adding other bands in InSAR/Pol-InSAR configurations and potentially TomoSAR (from BIOMASS mission) data.