Next Article in Journal
Multimodal Data-Driven Visual Sensitivity Assessment and Planning Response Strategies for Streetscapes in Historic Districts: A Case Study of Anshandao, Tianjin
Previous Article in Journal
Fluctuating Development Traits of Industrial Land Mismatch and Its Influence on Urban Ecological Modernization
 
 
Article
Peer-Review Record

Application of Remote Sensing and GIS in Monitoring Forest Cover Changes in Vietnam Based on Natural Zoning

Land 2025, 14(5), 1037; https://doi.org/10.3390/land14051037
by An Nguyen 1,*, Vasily Kovyazin 1 and Cong Pham 2
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3:
Land 2025, 14(5), 1037; https://doi.org/10.3390/land14051037
Submission received: 6 April 2025 / Revised: 30 April 2025 / Accepted: 7 May 2025 / Published: 9 May 2025
(This article belongs to the Section Land – Observation and Monitoring)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In lines 32-33, the phrase "High-tech methods are actively employed..." is clichéd and somewhat vague. It would be better for the author to specify more clearly from the beginning what the exact focus of the paper is.

In lines 43-45, the sentence "Remote sensing (RS) remains the only effective method..." requires a credible source. Additionally, the phrase "appropriate methodologies have not yet been fully developed" is too general. It would be better to specify which aspect of the methods has not yet been fully developed (e.g., classification? temporal analysis?).

In lines 69-71, the list of advantages reads like marketing content from software companies. It can be shortened and made more precise, focusing specifically on the advantages for forest cover monitoring.

It is suggested that the literature review be explained in more detail. What gaps exist in previous studies, and how does this research specifically and innovatively address them? Clearly stating the novelty of the study will strengthen the introduction and provide better justification for the importance of the research.

In lines 118–121, it is necessary to explain how the use of topographic data (DEM and slope) played a role in the land cover change analysis process (e.g., as input for the classification model or for restricting areas). Additionally, it would be helpful to specify the resolution of the Digital Elevation Model used (e.g., SRTM with a 30-meter resolution?).

In lines 138–142, it is suggested to focus on the practical role of NDVI in the final classification instead of repeating general information. Was it used as an input feature in the Random Forest algorithm, or was it only used to create a visual map? Additionally, it is unclear how the NDVI maps for each year were integrated with other data.

In lines 143–148, the explanation about band combinations is useful, but it seems more educational than a description of the research method. It would be better if the authors specify whether these combinations were used in the visual preprocessing stage or if they also played a role as input features for the classification algorithm.

In lines 149–156, it would be better to clearly explain how the description of mixed and broadleaf forest characteristics influenced the classification process. Additionally, the mention of "increasing reference points with band combinations and NDVI calculation" is ambiguous—does this refer to increasing training samples? If so, the number and method of selecting these samples should be explained.

In lines 157–161, more details are needed regarding the Random Forest algorithm, such as: The number of trees in the model. The feature selection criteria. The sampling rate. How the training and validation data were generated (training/validation split), the number of points, and the accuracy evaluation method (confusion matrix and Kappa) should be explained.

In lines 169–178, more explanation is needed regarding how the Transition Matrix was generated. Specifically, it is unclear what the base data for producing the matrix is and from which years the data were taken, which software was used, and how the prediction accuracy was evaluated.

In lines 187–191, mentioning the advantages and limitations of the Ca-Markov model is good, but the statement "this study aims to address some of these limitations" is vague. It is necessary to specify exactly which limitations this study attempts to address and how it does so. Was auxiliary data used? Was there a combination with another algorithm?

In lines 207–216, it is unclear which exact data were used for the layers and inputs in the ANN in this study. For example, were climate and demographic data obtained from local or global sources? Additionally, the spatial resolution of these data and their sources are not specified.

In lines 217–223, the formula Y = ANN(X) is simple and understandable, but it is recommended that the authors explain how the model's accuracy was evaluated and whether model validation (e.g., using real data for future years) was conducted.

In lines 224–228 (Advantages and limitations): Mentioning the advantages and limitations of the ANN is helpful. However, it would be better to specify how these limitations were addressed in this study. For example, was the lack of social data compensated by using averaged data? Or was the selection of the number of layers and neurons based on trial and error, or was there a specific optimization method used?

In lines 291–294, it is unclear whether the predictions regarding the land cover distribution for 2020 were compared with actual data for 2020. Including comparative results with real data could strengthen the credibility of the predictions. Additionally, providing more details about the model evaluation criteria (such as prediction accuracy, evaluation indices) would be helpful.

In lines 299–302, the comparison of the prediction results between the CA-Markov and MOLUSCE models is very important. However, the details of this comparison are not fully explained. It is suggested to mention the key differences between the two models in terms of prediction accuracy and the possible reasons for these differences (such as simpler modeling in CA-Markov or higher accuracy in the MOLUSCE model due to the use of more data). Additionally, Table 6, which shows the deviations for each method, could be analyzed further to help the reader better understand the differences between the two models.

The section on predicting land cover changes for 2025 and 2030 is generally well-organized. Providing more precise information on data analysis and management policies in the provinces would be valuable. However, adding more details on comparing predictions with actual data, evaluating prediction accuracy, and providing a more precise assessment of the impacts of forestry policies would strengthen this section and enhance its credibility.

Lines 363–383 should more clearly address the impact of environmental and human factors on forest cover changes and better clarify how the results relate to the research objectives. This will provide a stronger connection between the findings and the study's overall goals.

Lines 384–394 should more specifically address the reasons for the superiority of the MOLUSCE module over the Ca-Markov model. This would provide a clearer justification for the choice of MOLUSCE and highlight its advantages in the context of the study.

In the discussion section, clearly address the limitations of the research, including data limitations, methodological constraints, and the limitations of the models used. Based on the study's findings, provide specific recommendations for future research. This will help contextualize the study's results and guide subsequent investigations in related areas.

 

 

In lines 32-33, the phrase "High-tech methods are actively employed..." is clichéd and somewhat vague. It would be better for the author to specify more clearly from the beginning what the exact focus of the paper is.

In lines 43-45, the sentence "Remote sensing (RS) remains the only effective method..." requires a credible source. Additionally, the phrase "appropriate methodologies have not yet been fully developed" is too general. It would be better to specify which aspect of the methods has not yet been fully developed (e.g., classification? temporal analysis?).

In lines 69-71, the list of advantages reads like marketing content from software companies. It can be shortened and made more precise, focusing specifically on the advantages for forest cover monitoring.

It is suggested that the literature review be explained in more detail. What gaps exist in previous studies, and how does this research specifically and innovatively address them? Clearly stating the novelty of the study will strengthen the introduction and provide better justification for the importance of the research.

In lines 118–121, it is necessary to explain how the use of topographic data (DEM and slope) played a role in the land cover change analysis process (e.g., as input for the classification model or for restricting areas). Additionally, it would be helpful to specify the resolution of the Digital Elevation Model used (e.g., SRTM with a 30-meter resolution?).

In lines 138–142, it is suggested to focus on the practical role of NDVI in the final classification instead of repeating general information. Was it used as an input feature in the Random Forest algorithm, or was it only used to create a visual map? Additionally, it is unclear how the NDVI maps for each year were integrated with other data.

In lines 143–148, the explanation about band combinations is useful, but it seems more educational than a description of the research method. It would be better if the authors specify whether these combinations were used in the visual preprocessing stage or if they also played a role as input features for the classification algorithm.

In lines 149–156, it would be better to clearly explain how the description of mixed and broadleaf forest characteristics influenced the classification process. Additionally, the mention of "increasing reference points with band combinations and NDVI calculation" is ambiguous—does this refer to increasing training samples? If so, the number and method of selecting these samples should be explained.

In lines 157–161, more details are needed regarding the Random Forest algorithm, such as: The number of trees in the model. The feature selection criteria. The sampling rate. How the training and validation data were generated (training/validation split), the number of points, and the accuracy evaluation method (confusion matrix and Kappa) should be explained.

In lines 169–178, more explanation is needed regarding how the Transition Matrix was generated. Specifically, it is unclear what the base data for producing the matrix is and from which years the data were taken, which software was used, and how the prediction accuracy was evaluated.

In lines 187–191, mentioning the advantages and limitations of the Ca-Markov model is good, but the statement "this study aims to address some of these limitations" is vague. It is necessary to specify exactly which limitations this study attempts to address and how it does so. Was auxiliary data used? Was there a combination with another algorithm?

In lines 207–216, it is unclear which exact data were used for the layers and inputs in the ANN in this study. For example, were climate and demographic data obtained from local or global sources? Additionally, the spatial resolution of these data and their sources are not specified.

In lines 217–223, the formula Y = ANN(X) is simple and understandable, but it is recommended that the authors explain how the model's accuracy was evaluated and whether model validation (e.g., using real data for future years) was conducted.

In lines 224–228 (Advantages and limitations): Mentioning the advantages and limitations of the ANN is helpful. However, it would be better to specify how these limitations were addressed in this study. For example, was the lack of social data compensated by using averaged data? Or was the selection of the number of layers and neurons based on trial and error, or was there a specific optimization method used?

In lines 291–294, it is unclear whether the predictions regarding the land cover distribution for 2020 were compared with actual data for 2020. Including comparative results with real data could strengthen the credibility of the predictions. Additionally, providing more details about the model evaluation criteria (such as prediction accuracy, evaluation indices) would be helpful.

In lines 299–302, the comparison of the prediction results between the CA-Markov and MOLUSCE models is very important. However, the details of this comparison are not fully explained. It is suggested to mention the key differences between the two models in terms of prediction accuracy and the possible reasons for these differences (such as simpler modeling in CA-Markov or higher accuracy in the MOLUSCE model due to the use of more data). Additionally, Table 6, which shows the deviations for each method, could be analyzed further to help the reader better understand the differences between the two models.

The section on predicting land cover changes for 2025 and 2030 is generally well-organized. Providing more precise information on data analysis and management policies in the provinces would be valuable. However, adding more details on comparing predictions with actual data, evaluating prediction accuracy, and providing a more precise assessment of the impacts of forestry policies would strengthen this section and enhance its credibility.

Lines 363–383 should more clearly address the impact of environmental and human factors on forest cover changes and better clarify how the results relate to the research objectives. This will provide a stronger connection between the findings and the study's overall goals.

Lines 384–394 should more specifically address the reasons for the superiority of the MOLUSCE module over the Ca-Markov model. This would provide a clearer justification for the choice of MOLUSCE and highlight its advantages in the context of the study.

In the discussion section, clearly address the limitations of the research, including data limitations, methodological constraints, and the limitations of the models used. Based on the study's findings, provide specific recommendations for future research. This will help contextualize the study's results and guide subsequent investigations in related areas.

 

 

 

Author Response

Comment 1: In lines 32-33, the phrase "High-tech methods are actively employed..." is clichéd and somewhat vague. It would be better for the author to specify more clearly from the beginning what the exact focus of the paper is.

Response 1: We removed this sentence. Instead, we opened by directly stating the importance of RS and GIS for forest monitoring in Vietnam, emphasizing the context and technological advances (in lines 32-34).

Comment 2: In lines 43-45, the sentence "Remote sensing (RS) remains the only effective method..." requires a credible source. Additionally, the phrase "appropriate methodologies have not yet been fully developed" is too general. It would be better to specify which aspect of the methods has not yet been fully developed (e.g., classification? temporal analysis?).

Response 2: We explained why RS is the primary method (difficult terrain, diverse climate, limitations of field methods), thus providing a logical justification (in lines 39-47). We also clarified that the limitations are in classification accuracy, temporal change detection, and environmental data integration (in lines 47-51).

Comment 3: In lines 69-71, the list of advantages reads like marketing content from software companies. It can be shortened and made more precise, focusing specifically on the advantages for forest cover monitoring.

Response 3: We wrote a paragraph explaining how Arcgis, GEE and QGIS are valuable for forest monitoring: comprehensive analytical toolset, massive data processing, cloud-based corrections, integrating multiple data types, and scalable modeling (in lines 68-77). 

Comment 4: It is suggested that the literature review be explained in more detail. What gaps exist in previous studies, and how does this research specifically and innovatively address them? Clearly stating the novelty of the study will strengthen the introduction and provide better justification for the importance of the research.

Response 4: We stated the gaps: reliance only on spectral data, lack of supplementary variables, environmental factors (in lines 52-56), simplistic predictive models (in lines 92-94). We then stated how this study addresses those by integrating NDVI, DEM, slope, and using advanced machine learning algorithm RF. The last paragraph of the Introduction now clearly states the novelty: integrated use of spectral and environmental data + comparison of two different predictive modeling approaches in three ecologically distinct provinces (in lines 103-113). 

Comment 5: In lines 118–121, it is necessary to explain how the use of topographic data (DEM and slope) played a role in the land cover change analysis process (e.g., as input for the classification model or for restricting areas). Additionally, it would be helpful to specify the resolution of the Digital Elevation Model used (e.g., SRTM with a 30-meter resolution?).

Response 5: We stated DEM and slope were input features in Random Forest classification and also included in prediction modeling. We also specified 30-meter resolution of the Digital Elevation Model used (in lines 160-163).

Comment 6: In lines 138–142, it is suggested to focus on the practical role of NDVI in the final classification instead of repeating general information. Was it used as an input feature in the Random Forest algorithm, or was it only used to create a visual map? Additionally, it is unclear how the NDVI maps for each year were integrated with other data.

Response 6: We wrote that NDVI was not only for visualization but was also an input feature for the classifier. We explained how the NDVI maps for each year were integrated with other data (in lines 184-190).

Comment 7: In lines 143–148, the explanation about band combinations is useful, but it seems more educational than a description of the research method. It would be better if the authors specify whether these combinations were used in the visual preprocessing stage or if they also played a role as input features for the classification algorithm.

Response 7: We clarified that spectral composites were used to assist visual selection of training samples. These combinations allowed for more effective identification of land cover categories (in lines 191-195).

Comment 8: In lines 149–156, it would be better to clearly explain how the description of mixed and broadleaf forest characteristics influenced the classification process. Additionally, the mention of "increasing reference points with band combinations and NDVI calculation" is ambiguous—does this refer to increasing training samples? If so, the number and method of selecting these samples should be explained.

Response 8: We added a paragraph that explains that broadleaf forests have higher NDVI and more uniform canopy. Mixed forests have more variability and slightly lower NDVI. We explained how high-NDVI (>0.7) and stable spectral areas were chosen for broadleaf forests, and moderate-NDVI (0.5–0.7) heterogeneous areas for mixed forests. We stated that this informed sampling strategy helped Random Forest better learn to separate forest types and thus improve classification performance (in lines 196-207). 

Comment 9: In lines 157–161, more details are needed regarding the Random Forest algorithm, such as: The number of trees in the model. The feature selection criteria. The sampling rate. How the training and validation data were generated (training/validation split), the number of points, and the accuracy evaluation method (confusion matrix and Kappa) should be explained.

Response 9: We added that random sampling was used, with sample sizes (~200–250 points/category) and 70/30 training-validation split. We specified 500 trees, Gini index, sampling strategy, and validation method (in lines 210-222). 

Comment 10: In lines 169–178, more explanation is needed regarding how the Transition Matrix was generated. Specifically, it is unclear what the base data for producing the matrix is and from which years the data were taken, which software was used, and how the prediction accuracy was evaluated.

Response 10: We explained that transition matrices were generated from 2010 and 2015 classified maps using IDRISI TerrSet software. We stated that 2020 predictions were compared to actual 2020 maps, calculating overall deviation percentages of predicted land cover areas (in lines 248-258).

Comment 11: In lines 187–191, mentioning the advantages and limitations of the Ca-Markov model is good, but the statement "this study aims to address some of these limitations" is vague. It is necessary to specify exactly which limitations this study attempts to address and how it does so. Was auxiliary data used? Was there a combination with another algorithm?

Response 11: We explained that CA-Markov model relies on historical transition probabilities, can’t incorporate external drivers, and this study aimed to address these limitations by supplementing the prediction process with an ANN-based modeling approach that integrates a broader range of input variables influencing forest dynamics (in lines 261-265).

Comment 12: In lines 207–216, it is unclear which exact data were used for the layers and inputs in the ANN in this study. For example, were climate and demographic data obtained from local or global sources? Additionally, the spatial resolution of these data and their sources are not specified.

Response 12: We listed the exact input layers: land cover maps, DEM, slope, distance to roads, distance to urban centers, and WorldClim climate data, including their spatial resolution (in lines 282-290).

Comment 13: In lines 217–223, the formula Y = ANN(X) is simple and understandable, but it is recommended that the authors explain how the model's accuracy was evaluated and whether model validation (e.g., using real data for future years) was conducted.

Response 13: We described that ANN-predicted 2020 maps were compared to classified 2020 maps, calculating overall deviation percentages of predicted land cover areas and Kappa coefficients (in lines 303-306).

Comment 14: In lines 224–228 (Advantages and limitations): Mentioning the advantages and limitations of the ANN is helpful. However, it would be better to specify how these limitations were addressed in this study. For example, was the lack of social data compensated by using averaged data? Or was the selection of the number of layers and neurons based on trial and error, or was there a specific optimization method used?

Response 14: We added that distance-based proxies were used to approximate human activity and that ANN architecture (hidden layers, neurons) was optimized by trial-and-error process to achieve a balance between model complexity and overfitting (in lines 311-315).

Comment 15: In lines 291–294, it is unclear whether the predictions regarding the land cover distribution for 2020 were compared with actual data for 2020. Including comparative results with real data could strengthen the credibility of the predictions. Additionally, providing more details about the model evaluation criteria (such as prediction accuracy, evaluation indices) would be helpful.

Response 15: We stated that the MOLUSCE 2020 predictions were compared with the real classified 2020 maps. We added that accuracy was assessed by (1)-overall deviation percentages (1.61%, 1.14%, 1.80%), and (2)-confusion matrices, Overall Accuracy (>83%), and Kappa coefficients (range 0.78–0.82) (in lines 376-387).

Comment 16: In lines 299–302, the comparison of the prediction results between the CA-Markov and MOLUSCE models is very important. However, the details of this comparison are not fully explained. It is suggested to mention the key differences between the two models in terms of prediction accuracy and the possible reasons for these differences (such as simpler modeling in CA-Markov or higher accuracy in the MOLUSCE model due to the use of more data). Additionally, Table 6, which shows the deviations for each method, could be analyzed further to help the reader better understand the differences between the two models.

Response 16: We wrote a full comparative analysis explaining performance and why each model performed the way it did. We reported deviation rates clearly for both models, side-by-side, and discussed the magnitude of improvement. We explained that CA-Markov assumes stationary processes and does not use external drivers, while MOLUSCE with ANNs uses multiple environmental, anthropogenic inputs and models nonlinear relationships. For table 6, We specifically discussed the size of deviation reductions (5–7% improvement) and what it means for model selection (in lines 393-414).

Comment 17: The section on predicting land cover changes for 2025 and 2030 is generally well-organized. Providing more precise information on data analysis and management policies in the provinces would be valuable. However, adding more details on comparing predictions with actual data, evaluating prediction accuracy, and providing a more precise assessment of the impacts of forestry policies would strengthen this section and enhance its credibility.

Response 17: We stated that projections were based on 2010–2020 historical trends, with model validation against 2020 real data (in lines 418-425). In original version of the article, we detailed the expected increases/decreases and tied them logically to economic development, policy changes, and conservation actions, so we didn’t add information about management policies. We explained that although 2025/2030 data do not exist yet, the high 2020 prediction accuracy (deviations <2%) lends confidence to the projections.  We also analyzed how policy, reforestation programs, and conservation efforts are expected to influence predicted changes in each province (in lines 466-472).

Comment 18: Lines 363–383 should more clearly address the impact of environmental and human factors on forest cover changes and better clarify how the results relate to the research objectives. This will provide a stronger connection between the findings and the study's overall goals.

Response 18: We explained how elevation, slope, proximity to roads and urban centers, and policies influenced forest loss and recovery in each province. We added that the RS/GIS approach supported the goals of better classification and prediction by modeling environmental and human drivers. The last two paragraphs of section 4.2.  connected the findings with methodological innovations and the objectives set in the Introduction (in lines 527-542).

Comment 19: Lines 384–394 should more specifically address the reasons for the superiority of the MOLUSCE module over the Ca-Markov model. This would provide a clearer justification for the choice of MOLUSCE and highlight its advantages in the context of the study.

Response 19: We explained that MOLUSCE integrates multiple variables, captures nonlinear relationships, and better models spatial trends. We linked the model choice directly to study's goals: improving reliability and accuracy in predicting forest cover changes. We emphasized that in dynamic, human-affected environments like Vietnam, MOLUSCE's ability to account for environmental and human factors was crucial (in lines 544-565).

Comment 20: In the discussion section, clearly address the limitations of the research, including data limitations, methodological constraints, and the limitations of the models used. Based on the study's findings, provide specific recommendations for future research. This will help contextualize the study's results and guide subsequent investigations in related areas.

Response 20: We discussed missing socio-economic data, coarse climate data, and the use of proxies. We explained that low temporal resolution (five-year intervals) was a limitation. We stated that ANN is a black-box model and that future conditions might diverge from historical trends. We gave specific recommendations (in lines 576-610).

 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript (land-3600697)monitors forest cover changes in three provinces of Vietnam (Thanh Hoa, Quang Son and Dong Nai) between 2010 and 2020 using remote sensing (RS) and geographic information system (GIS) techniques combined with natural zoning methods. The authors used Ca-Markov model and MOLUSCE module to predict the forest cover change in 2030. I think the article is logical and clear, but it still needs to be revised before publication.
1. In the second paragraph, it is mentioned that “Vietnam, due to the complex terrain and climatic conditions, forest land monitoring requires the use of geoinformation technologies”. technologies”, but does not explain how these technologies overcome the complexity of terrain and climate. It is suggested that a brief paragraph be added to explain the advantages of the specific application of these technologies in Viet Nam.
2. The introduction presents the background of the application of remote sensing and GIS technologies in forest resource monitoring, but it should also emphasize the background of Ca-Markov modeling and MOLUSCE technologies in the study of forest resources under the complex terrain and climate conditions in Vietnam. The following references are suggested to enhance the synthesis. - Multi-Scenario Simulation of Land Use Change and Ecosystem Service Value Based on the Markov-FLUS Model in Ezhou City, China; Land cover changes and carbon dynamics in Central India's dry tropical forests: A 25-year assessment and nature-based eco-restoration approaches; Neotropical urban forest allergenicity and ecosystem services. urban forest allergenicity and ecosystem disservices can affect vulnerable neighborhoods in Bogota, Colombia
3. “3.2. Forest Cover Changes Prediction Using the CA-Markov Model and the MOLUSCE Module”, the authors mentioned the results of the two prediction methods but did not analyze the advantages and disadvantages of the two methods in detail. The author mentions the results of the two prediction methods but does not analyze the advantages and disadvantages of the two methods in detail. In my opinion, it is necessary to explain why the MOLUSCE Module is more accurate and the application scenarios of the two methods.
4. “4.1 Advantages of Remote Sensing and GIS Technology Over Traditional Methods”, the article mentions the limitations of traditional methods, but does not explain in detail how these limitations affect the accuracy of forest monitoring. The article mentions the limitations of traditional methods, but does not elaborate on how these limitations affect the accuracy of forest monitoring.
5. In the section “4.2. Application of Remote Sensing and GIS in Assessing Forest Cover Changes”, the article mentions the trends of forest cover changes in three provinces, but does not analyze in detail the reasons for these changes. It is suggested to explain the specific factors leading to these changes, such as policy changes, economic activities, etc.
6. In the section “4.3. Using the MOLUSCE Module to Predict Forest Cover Changes to 2030”, the results of the MOLUSCE Module's predictions are mentioned, but the specific significance of these results for forest management and conservation is not explained in detail.
7. In the conclusion section, it is mentioned that “this study employs a remote sensing approach integrated with GIS technology to monitor forest cover changes in Vietnam. “but does not elaborate on how these technologies can support the sustainable management of forest resources.

Author Response

Comment 1: In the second paragraph, it is mentioned that “Vietnam, due to the complex terrain and climatic conditions, forest land monitoring requires the use of geoinformation technologies”. technologies”, but does not explain how these technologies overcome the complexity of terrain and climate. It is suggested that a brief paragraph be added to explain the advantages of the specific application of these technologies in Viet Nam.

Response 1: 

To increase logic and coherence, we removed this sentence. Instead, we explained why Remote Sensing is the primary method (difficult terrain, diverse climate, limitations of field methods), thus providing a logical justification (in lines 39-47).

We wrote a paragraph explaining how Arcgis, GEE and QGIS are valuable for forest monitoring: comprehensive analytical toolset, massive data processing, cloud-based corrections, integrating multiple data types, and scalable modeling (in lines 68-77).

Comment 2: The introduction presents the background of the application of remote sensing and GIS technologies in forest resource monitoring, but it should also emphasize the background of Ca-Markov modeling and MOLUSCE technologies in the study of forest resources under the complex terrain and climate conditions in Vietnam. The following references are suggested to enhance the synthesis. - Multi-Scenario Simulation of Land Use Change and Ecosystem Service Value Based on the Markov-FLUS Model in Ezhou City, China; Land cover changes and carbon dynamics in Central India's dry tropical forests: A 25-year assessment and nature-based eco-restoration approaches; Neotropical urban forest allergenicity and ecosystem services. urban forest allergenicity and ecosystem disservices can affect vulnerable neighborhoods in Bogota, Colombia.

Response 2: We added two new paragraphs explaining how CA-Markov and MOLUSCE methods are used for land cover modeling, their strengths, and their relevance. We integrated first reference into the discussion, showing examples of where the Markov model has been applied. We did not see any clear relevance of the remaining two references to Ca-Markov model and MOLUSCE module, so we used different references to enhance the synthesis following your suggestions (in lines 82-102).

Comment 3: “3.2. Forest Cover Changes Prediction Using the CA-Markov Model and the MOLUSCE Module”, the authors mentioned the results of the two prediction methods but did not analyze the advantages and disadvantages of the two methods in detail. In my opinion, it is necessary to explain why the MOLUSCE Module is more accurate and the application scenarios of the two methods.

Response 3: We added a comparative analysis explaining performance and why each model performed the way it did. We reported deviation rates clearly for both CA-Markov Model and the MOLUSCE Module, side-by-side, and discussed the magnitude of improvement. We explained that CA-Markov assumes stationary processes and does not use external drivers, while MOLUSCE with ANNs uses multiple environmental, anthropogenic inputs and models nonlinear relationships. We specifically discussed the size of deviation reductions (5–7% improvement) and what it means for model selection (in lines 393-414).

Comment 4: “4.1 Advantages of Remote Sensing and GIS Technology Over Traditional Methods”, the article mentions the limitations of traditional methods, but does not explain in detail how these limitations affect the accuracy of forest monitoring. The article mentions the limitations of traditional methods, but does not elaborate on how these limitations affect the accuracy of forest monitoring.

Response 4: We explained that physical access limitations of traditional methods result in incomplete spatial coverage, temporal infrequency of ground-based surveys reduce the relevance and responsiveness of forest management decisions, and human subjectivity causes classification errors (in lines 476-499). 

Comment 5: In the section “4.2. Application of Remote Sensing and GIS in Assessing Forest Cover Changes”, the article mentions the trends of forest cover changes in three provinces, but does not analyze in detail the reasons for these changes. It is suggested to explain the specific factors leading to these changes, such as policy changes, economic activities, etc.

Response 5: We explained how different factors influenced forest loss and recovery in each province. We added that the RS/GIS approach supported the goals of better classification and prediction by modeling environmental and human drivers. The last two paragraphs connected the findings with methodological innovations and the research objectives (in lines 527-542).

Comment 6: In the section “4.3. Using the MOLUSCE Module to Predict Forest Cover Changes to 2030”, the results of the MOLUSCE Module's predictions are mentioned, but the specific significance of these results for forest management and conservation is not explained in detail.

Response 6: We explained that MOLUSCE module integrates multiple variables, captures nonlinear relationships, and better models spatial trends. We linked the model choice directly to study's goals: improving reliability and accuracy in predicting forest cover changes. We emphasized that in dynamic, human-affected environments like Vietnam, MOLUSCE's ability to account for environmental and anthropogenic factors was crucial (in lines 544-565).

Comment 7: In the conclusion section, it is mentioned that “this study employs a remote sensing approach integrated with GIS technology to monitor forest cover changes in Vietnam. “but does not elaborate on how these technologies can support the sustainable management of forest resources.

Response 7: We added information providing advantages of RS and GIS technologies for forest monitoring. We explained that RS/GIS enable threat detection, prioritization for conservation, policy evaluation, and proactive management planning (in lines 631-639).

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

1) Better explain the process of selecting landmarks and how they affect land cover classification.

2) Add a more thorough explanation of the differences between the Ca-Markov model and the MOLUSCE module, for example, a table comparing the models (e.g.: Principle of operation, accuracy, prediction deviation, applications, limitations, advantages, etc.).

Author Response

Comment 1: Better explain the process of selecting landmarks and how they affect land cover classification.

Response 1: We described that landmarks were selected via random sampling, guided by visual interpretation of satellite imagery, Google Earth data and field maps (in lines 210-212). We clarified that NDVI and spectral composites were used to assist visual selection of training samples (in lines 184-195). We added that approximately 200–250 sample points were collected for each land cover category, and the dataset was divided into 70% for training and 30% for validation. We explained that the quality and representativeness of the samples directly influenced classification performance. Well-distributed, accurately labeled samples enabled the Random Forest model to learn the unique spectral, vegetative, and topographic characteristics of each category more effectively. This was particularly important for distinguishing spectrally similar categories, such as mixed forests and broadleaf forests (in lines 212-218).

Comment 2: Add a more thorough explanation of the differences between the Ca-Markov model and the MOLUSCE module, for example, a table comparing the models (e.g.: Principle of operation, accuracy, prediction deviation, applications, limitations, advantages, etc.).

Response 2: In this revised version, we have added more information in different sections to explain the differences of the Ca-Markov model and the MOLUSCE module-in lines 82-102 (introduction), 261-265 (method), 393-414 (results).

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for addressing the feedback on the methodology. The added details have improved clarity and transparency.

 

Back to TopTop