Unveiling Forest Density Dynamics in Saihanba Forest Farm by Integrating Airborne LiDAR and Landsat Satellites

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article “Unveiling Forest Density Dynamics in Saihanba Mechanical Forest Farm in Hebei Province, China by Integrating Airborne LiDAR and Landsat Satellites”. Authors developed a long-term (1988–2023), high-resolution forest density dataset for Saihanba by combining multi-seasonal Landsat time-series with airborne LiDAR individual tree segmentation. They employed airborne LiDAR-derived tree density as reference and extracted in 96 seasonal features from Landsat SR products, including commonly used vegetation indices NDVI and EVI, to train a regression model for estimating tree density. Forest density references were derived from LiDAR and linked to Landsat surface reflectance and vegetation indices, including NDVI (Normalized Difference Vegetation Index) and EVI (Enhanced Vegetation Index). Three machine learning models—Random Forest, Support Vector Regression, and XGBoost—were trained and compared.

The study provides a valuable tool for the scientific community interested in forests development and degradation around the world. I recommend accepting the article with the following minor comments:

Study Area and Data

Line 108 mention the area in km2

Data and Preprocessing

Figure 2 quality needs to be improved. Please fix the Y-axis scale. Please use scientific number scale on the x-axis, please include units. The whole 3D figure is not clear.

 

Figure 3, what is the x – axis, what is the y axis, include units.

Figure 4, same questions as in figure 3, legend box is hiding the bottom part of the figure.

Figure 5, what is the x – axis, what is the y axis, include units.

Lines 165 – 170. Authors should explain how they created the vector grid shown in figure 5 and why they chose the 30 m resolution i.e., what are the steps to go from Figure 3 to Figure 5?

 

2.2.2. Satellite Imagery

Table 1, authors are recommended to list the band frequency range for each band in columns 2 and 3

3. Method

Authors should comment on why using the statistical quantity median in their statistical analysis. Why not the average? Are there many statistical outliers?  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This paper takes Saihanba Forest Farm as the research area, integrates airborne LiDAR data and Landsat data from 1988 to 2023, and constructs a high-resolution forest density dataset. It extracts multi-seasonal spectral and vegetation index features, trains Random Forest (RF), Support Vector Regression (SVR), and XGBoost models, and determines that the XGBoost model has the optimal performance through comparison. The XGBoost model is then selected to invert forest density at 5-year intervals. The paper analyzes the impacts of tree species, ecological function zones, and slope gradients on forest density, verifies the effectiveness of afforestation, and proposes a large-scale ecological monitoring framework. The overall quality of the paper is quite good, but there are still some details that need to be further improved and perfected.

1. The location of the study area in Figure 1 is not prominent enough and the figure is not visually appealing, with excessive blank space. It is suggested that the authors highlight the location of the study area on the map and further improve the figure's aesthetics.
2. It is recommended that the authors increase the font size of the axis titles on the right side of Figure 5.
3. The technical flowchart (Figure 6) is overly simplistic and not visually attractive. It is suggested that the authors condense the research content and further refine the aesthetics of the technical flowchart.
4. The font sizes of Formula (1) and Formula (2) in the text are inconsistent; the authors are advised to check this carefully.
5. The method of constructing a long-term forest density dataset by combining LiDAR and Landsat data in this paper has innovation, but the text fails to elaborate on the differences and unique features compared with other existing studies. It is suggested that the authors add relevant explanations.
6. Although the paper compares two machine learning models, it does not explain why the XGBoost model was chosen over other more advanced deep learning models (such as CNN or Transformer).
7. In the feature extraction section, the paper mentions calculating seasonal statistics of NDVI and EVI, but it does not explain why the maximum value, minimum value, and median are selected. The authors are recommended to supplement the reasons for this selection.
8. The paper emphasizes "seasonality", but the main text does not quantitatively analyze the independent impact of images from different seasons on model performance. It is suggested to add a comparison of model accuracy across different seasons.
9. The paper mentions "182,000 valid samples" but does not clarify whether the distribution of these samples across different slope gradients is balanced.
10. The paper states that "tree species classification is conducted based on 50 features derived from CCD images and point cloud data", but it does not specify the specific types and attribute information of these 50 features.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This study addresses an important topic by producing a long-term, high-resolution forest density dataset for Saihanba. While the work is valuable, the manuscript requires substantial revision for clarity, structure, and methodological transparency. I recommend a major revision.

Abstract

• The current abstract reads more like a list of results than a connected narrative. I recommend restructuring it into four logical components:
  (1) Background – why forest density matters and why Saihanba is a relevant case study;
  (2) Methods – how the dataset was developed (Landsat + LiDAR + ML);
  (3) Results – highlight the main findings (validations, overall trends, species differences, slope effect);
  (4) Implications – what these results mean for forest restoration and ecological monitoring.
• Lines 18–19: The abstract reports RMSE = 174 trees ha⁻¹ for the XGBoost model, but it is unclear whether this refers to self-validation with a testing dataset or validation against independent field measurements. Please clarify. It would also strengthen the abstract to report the coefficient of determination (R²) alongside RMSE.
• Lines 20–21: The phrase “species-specific analysis” is vague. How many species were included in the analysis? And what variables were considered in “crown structure”? For instance, does this refer to canopy height, crown width, canopy cover, or other structural metrics? Please specify.

 

Main Text

• Line 57: Please use a proper and professional citation format for the referenced study.
• Line 60: Clearly explain the differences between this study and Zhang et al. beyond just the spatial scale (large vs. small area).
• Line 69: The term “ground truth data” is awkward. Please replace with “field measurements” or “ground-based measurements.”
• Line 96: The method should not be described as an “inversion model.” It is more accurately characterized as a framework that integrates LiDAR and satellite imagery.
• Line 114: The study area map is unclear. Please provide more geographic context, such as location details or an overview map that situates the study site within a broader region.
• Lines 131–175: This section only describes data sources, airborne instruments, and preprocessing. The results of individual tree segmentation, classification, and crown density estimation should be moved to the Results section, not Methods.
• Line 151: I could not find where this figure is cited in the main text. Please check and correct.
• Line 199: Please describe in more detail how tree crown segmentation was performed and how forest density was calculated. This is a central part of the study and should be explained clearly. By contrast, the machine learning methods require less detailed explanation, as they are not novel.
• Line 294: Since airborne LiDAR data were used to derive tree segmentation and crown density, which serve as key inputs to the machine learning models, it is essential to validate the LiDAR-derived crown density. Please report whether such validation was conducted.
• Line 294 (continued): I could not find evidence of independent validation of the forest density product using field measurements or intercomparison with other existing products. Such validation is critical to establish the reliability of your dataset.
• Line 355: The meaning of “true values” is unclear. Does this refer to field-measured data? Please clarify.
• Sections 5.2 and 5.3: These sections read more like Results rather than Discussion. Please revise or relocate content accordingly.

 

Comments on the Quality of English Language

must be improved.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Authors have addressed all comments. I have no further comments. 

Author Response

We are truly grateful for your careful review and kind confirmation. Thank you for your valuable time and support.

Reviewer 3 Report

Comments and Suggestions for Authors

The revised manuscript has significantly improved and is suitable for publication in its current form.

Author Response

Thank you very much for your kind assessment and positive recommendation. We are truly grateful for your time and thoughtful review.