Mapping Tropical Forested Wetlands Biomass with LiDAR: A Machine Learning Comparison

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper "Mapping Tropical Wetlands Biomass with LiDAR: A Machine Learning Comparison" represents a relevant contribution to the field of remote sensing biomass in tropical wetland forests using LiDAR technology and machine learning. The methodological approaches used are ok. The authors have clearly formulated the problem, set objectives, and precisely presented the results. The paper is well-structured, with a clear introduction, a detailed "Materials and Methods" section, and logically and comprehensibly presented results with adequate graphical representations.

However, due to the following issues, I suggest that the paper can be accepted only after revision and addressing the following points:

Section 2
If the authors agree, it might be beneficial to include a workflow diagram in this section.

Section 2.1
Figure 1: Indicate the north direction on the map.

Section 2.2
Comment: If I understood correctly, the authors collected the necessary field data to calculate AGB for specific trees, recorded the coordinates, and then used these calculated values for training, tracking how the LiDAR metrics affect the results, and then used these for prediction where direct measurements were not taken. If this understanding is correct, the authors should clarify this explanation somewhere.

Section 2.2.1
Sentence: "These plots were separated between 150 and 350 m, following two gradients: distance to the lagoon and distance to the widest secondary channel (Fig. 1)."
Comment: Make this explanation clearer.
Sentence: "Once the AGB for each tree was obtained, the plot’s AGB was calculated as the sum of all the trees’ AGB and extrapolated to 1 ha."
Question: Why are these extrapolations not shown anywhere?

Section 2.2.2
Comment: "DBH stands for DBH" - clarify or correct.

Section 2.2.3
Sentence: "An airborne LiDAR point cloud acquired in March 2014 was used to extract several metrics characterizing the vertical structure of the forest. The point cloud showed up to seven returns per pulse, a point density of 8 points / m2, and encompassed approximately 9.5 km2. The point cloud was classified by the provider into two classes: ground and unclassified."
Comment: This description of the LiDAR data acquisition is quite brief. The authors should specify the data source, the total number of points, which sensor was used, etc. Additionally, they should include an image of the scanned area to enhance clarity.

Sentence: "These metrics included: vertical height (z) and intensity (i) statistical metrics such as mean, maximum (max), standard deviation (sd), skewness (skew), kurtosis (kurt), and percentiles in 5% steps (q5-95). Additionally, the cumulative percentage of return in the ith layer according to [62] (pcum1-9), as well as height percentage of returns above the mean and 2 m (pzabovezmean and pzabove2), percentage of xth returns (p1-5th), number of points (n), the approximate area covered by each plot (area), and average absolute scan angle (angle)."
Comment: How did the authors determine which metrics to use? If this selection was arbitrary, they can use for example a feature importance analysis to show which metrics/indicators contribute the most to the results.

Comment: LiDAR data were collected in 2014, while field data were collected between 2016 and 2017. Couldn’t significant changes in forest structure have occurred during this period? Further justification is needed to explain how this time difference might affect the results.

Section 2.3.2
Sentence: "From the complete dataset, 20 randomly selected plots were selected as the training dataset (80%) and the remaining five as a test set (20%)."
Question: A sample of 25 plots may be insufficient for get generalized conclusions. It would be useful to discuss the potential effects of small sample size on model stability. Is it possible to refine the model with a larger sample size?

Section 3.2
Sentence: "The model that achieved the lowest error on the test set was the random forest (RMSE = 20.25, rRMSE = 12.25 %, R2 = 0.74), followed by the linear regression (RMSE = 31.8, rRMSE = 19.24 %, R2 = 0.36) and XGBoost (RMSE = 36.22, rRMSE = 21.91 %, R2 = 0.16; Table 1, Fig. 3). All the models achieved a smaller error on the training while they differed mostly in their performance on the test set."
Question 1: Model parameters should be listed somewhere.
Question 2: In addition to RMSE and R^2, additional metrics such as Mean Squared Error (MSE), Mean Average Error (MAE), and Mean Absolute Percentage Error (MAPE) could be included for a more comprehensive assessment of model performance.

Section 3.4
Figure 5: Indicate the north direction on the map.

Section 4
Although this Discussion section might be overly detailed, it still seems that the authors should more clearly highlight how the obtained results align with or differ from previous studies.

References
Most references come from the period 2014–2021, indicating that the authors use relatively recent literature, but they should include more sources from 2022, 2023, and 2024.

 

Comments for author File: Comments.txt

Comments on the Quality of English Language

I am not competent enough to comment on the quality of the English language.

Author Response

Please see the attachment

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Comments to the text:

1. Page 1, keywords:

• Add AGB shortcut in brackets, as well as LiDAR.

2. Page 1, first paragraph of the Introducton:

• The geographic extent of mangroves and tropical swamp forests should also be mentioned.

3. Page 4, LiDAR Data:

• What is the source of the Airborne Laser Scanning data? Is it proprietary or public domain?
• How accurate is it?
• Only two classes? Was any further classification done for the purposes of the article (separation of vegetation, artificial surfaces etc.).
• There are many kriging methods, please state which one you used to generate the elevation data - also, it is a Digital Elevation Model (DEM), not a Digital Terrain Model (DTM).

4. Page 8, AGB prediction on the complete study area:

• Please correct the description under figure 5, without emphasizing again the 1000 randomizations. Figre 5. Left - Mean AGB prediction of the study area, right - CoV of AGB prediction of the study area.

5. Page 11, last paragraph of the Discussion chapter:

So far, there has been no explanation of the GEDI abbreviation, nor has it been stated whether you used or not this dataset in the article.

6. Page 12, Abbrevations section:

You also use abbreviations: GEDI, DBH, Sar or inSAR.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

 

The manuscript “Mapping Tropical Wetlands Biomass with LiDAR: A Machine

Learning Comparison” has been reviewed, and my comments are as follows.

 

1 Why choose these three models for modeling needs to be explained, and the differences in model results need to be discussed.

 

2 In figure 3, I think it is necessary to add precision indicators such as R2 and RMSE.

 

3 The data source for this study is lidar data, and I believe that in the future, the research scope can be further expanded by combining information such as optical satellite images, such as Advancing the mapping of mangrove forests at national-scale using Sentinel-1 and Sentinel-2 time-series data with Google Earth Engine: A case study in China.

 

4 How to extract the spatial range of mangrove forests, which is a prerequisite for estimating biomass, seems not mentioned in the article. It is recommended to include a detailed explanation.

 

5 The sampling time and lidar data seem to be inconsistent, what impact does this have on biomass inversion.

 

6 What impact will seasonal variations in biomass have on the results of this study.

 

7 This article only uses one biomass inversion model, but in fact, there are many biomass inversion methods. Why choose this model needs to be explained or compared with the results of other models.

Author Response

Please see the attachment

Author Response File: Author Response.pdf