Evaluation of Three Algorithms and Forest Fire Risk Prediction in Zhejiang Province of China

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study focused on the evaluation of three algorithms and forest fire risk prediction in Zhejiang province of China. Overall, this is a decent and well-structure study with adequate scientific contribution. I think that issues on visualization and discussion parts of the study must be improved before publication.

Figure 1 has a very poor resolution. Moreover, DEM color scale should be improved, as elevations about 2000 m are typically not visualized using white color.

Fourth-rank subsections in 2.3.2. Machine learning algorithms are redundant (e.g. 1. Data processing). Please remove them from the text.

It is mandatory to add a figure representing all training and test samples after SMOTE in a figure with high resolution.

Figure 2A is crucial for this manuscript. Please provide it as a separate figure and provide it enlarged in high resolution. I suggest removing Figure 2D as you did not insert a figure representing these administrative units anywhere. They are also not relevant for international readers.

Figure 5, 6, 7, 8, 9 and 10 have legends with very low visibility. All parts of all figures must be clearly visible.

Figure 11: These are not “changes” (?!) Please remove the entire 3.2.3. subsection.

The Discussion should be improved with a special focus on methodological approach presented in this study according to previous studies, including advantages and limitations of the proposed approach relative to state of the art.

Author Response

 

 

 

Review 1

• This study focused on the evaluation of three algorithms and forest fire risk prediction in Zhejiang province of China. Overall, this is a decent and well-structure study with adequate scientific contribution. I think that issues on visualization and discussion parts of the study must be improved before publication.

Response: Thank you for your thoughtful review and for taking the time to evaluate our manuscript. We appreciate your positive comments regarding our study. We fully agree with your suggestion that the visualization and discussion sections could benefit from improvement. In response, we have revised the visualizations to enhance clarity and better highlight the key findings of our analysis. Additionally, we have refined the discussion to provide a more comprehensive interpretation of our results, emphasizing their implications and addressing any potential limitations in greater detail.

• Figure 1 has a very poor resolution. Moreover, DEM color scale should be improved, as elevations about 2000 m are typically not visualized using white color.

Response: We appreciate your valuable suggestion. Figure 1 has been replaced with a higher-resolution version in the revised manuscript. The DEM color scale has also been updated to use a red gradient for elevations, which is more consistent with standard visualization practices and improves readability.

• Fourth-rank subsections in 2.3.2. Machine learning algorithms are redundant (e.g. 1. Data processing). Please remove them from the text.

Response: Thank you for identifying the redundancy. The fourth-level subsections in 2.3.2 Machine Learning Algorithms have been removed to streamline the presentation. This section has been rewritten for improved clarity and cohesiveness in the revised manuscript.

• It is mandatory to add a figure representing all training and test samples after SMOTE in a figure with high resolution.

Response: Thank you for emphasizing this point. We have added a high-resolution figure (Figure 2) representing all training and test samples after applying the SMOTE technique. This figure visually demonstrates how SMOTE balances the dataset by creating synthetic samples, ensuring better representation of minority classes in the model. This addition aids in the clear understanding of our methodology.

Figure 2. The training and test samples after SMOTE balancing.

• Figure 2A is crucial for this manuscript. Please provide it as a separate figure and provide it enlarged in high resolution. I suggest removing Figure 2D as you did not insert a figure representing these administrative units anywhere. They are also not relevant for international readers.

Response: We appreciate your suggestion. (1) A new, enlarged, high-resolution version of Figure 2A has been provided as Figure 3 in the revised manuscript for better visualization. (2) Figure 2D has been removed as it does not directly contribute to the manuscript's objectives, particularly for international readers. A new figure (now Figure 4) has been included for improved relevance and clarity.

• Figure 5, 6, 7, 8, 9 and 10 have legends with very low visibility. All parts of all figures must be clearly visible.

Response: Thank you for pointing out the issue with figure legends. The legends in Figures 5, 6, 7, 8, 9, and 10 have been adjusted for better visibility and improved aesthetics in the revised manuscript to ensure clarity and readability.

• Figure 11: These are not “changes” (?!) Please remove the entire 3.2.3. subsection.

Response: We agree with your observation that Figure 11 does not accurately represent “changes.” Therefore, the entire 3.2.3 subsection has been removed, and the section has been rewritten in the revised manuscript for consistency and clarity.

• The Discussion should be improved with a special focus on methodological approach presented in this study according to previous studies, including advantages and limitations of the proposed approach relative to state of the art.

Response: We appreciate your suggestion to enhance the discussion by focusing on the methodological approach in comparison to prior studies. In the revised manuscript, we have emphasized the advantages of our integrated machine learning approach, including its ability to handle complex, non-linear relationships between variables and its superior performance relative to traditional statistical methods. We also discussed the limitations, such as potential biases introduced by SMOTE and the absence of certain socioeconomic variables. The comparison with state-of-the-art methods has been elaborated to highlight the unique contributions of this study to forest fire risk modeling.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study presents an innovative approach based on remote sensing data, geospatial data analysis, and machine learning to construct a map demonstrating forest fire risk. Zhejiang province is selected as the study area. The authors relied on MODIS dataset to obtain information regarding fire occurrences. A set of variables are used as fire risk’s influencing factor. The proposed framework has high application potential due to good predictive performance (F1 score up to 0.92). The research goal is good and the research method is appropriate. In general, the reviewers appreciate the author’s effort in data collection, model construction and validation, especially the high application potential of the study. The authors are encouraged to revise the paper according to the following comments:

1. Forest fire detection and susceptibility mapping are highly active research theme. Therefore, the current literature review section is definitely not sufficient to take into account recent advancements in these fields. Please review more related papers especially the ones cover:

-Machine learning and geospatial approaches for forest fire detection and susceptibility

-Deep learning, convolutional neural networks for forest fire detection and susceptibility

-Unsupervised learning approaches

-Hybridizations of machine learning and metaheuristic approaches

2. In section 1, please add at least 1 paragraph to summarize the research gaps.

3. In section 1, add at least 1 paragraph to clearly point out the innovative contributions of the paper and how the current work can help fill the existing research gaps in the literature.

4. Elaborate on the process used to derive the fire radiative power (FRP).

5. Related to the fact “from 2013 to 2023, a total of 2,713 forest fire events were recorded in Zhejiang…”, please state the source or reference to the forest fire inventory.

6. Provide more details on the climatic data, including precipitation, temperature (TEM), land surface temperature (LST), sunshine duration (SD), and relative humidity (RH). The data source, information about the stations, and the methods used to obtain the measurements should be clearly stated.

7. The explanatory variables used in previous works should be reviewed with more details. The authors should provide comments or discussions about the explanatory variables that were used in previous work but were not employed in the current work.

8. Consider the uses of demographic factors (e.g., population density), distance to roads, distance to residential areas, and distance to tourist hubs because they have been proven to be relevant in forest fire modeling.

9. Why were factors such as waterbody density, distance to major waterbody, or distance to rivers not considered in the current work? Please provide some comments.

10. When analyzing the effect of the influencing factors, are there any differences in the governing factors of forest fire for different zones in the study area, such as northern vs. southern regions or coastal vs. inland?

11. SMOTE is used to balance the dataset. Although SMOTE is an effective method, it actually creates artificial data samples in certain classes. Please test the model’s predictive capability when all testing samples are not generated by SMOTE.

12. Please elaborate on how the findings of the current work helps support the land use planning and forest management policy.

Author Response

Review 2

• This study presents an innovative approach based on remote sensing data, geospatial data analysis, and machine learning to construct a map demonstrating forest fire risk. Zhejiang province is selected as the study area. The authors relied on MODIS dataset to obtain information regarding fire occurrences. A set of variables are used as fire risk’s influencing factor. The proposed framework has high application potential due to good predictive performance (F1 score up to 0.92). The research goal is good and the research method is appropriate. In general, the reviewers appreciate the author’s effort in data collection, model construction and validation, especially the high application potential of the study. The authors are encouraged to revise the paper according to the following comments:

Response: Thank you for your positive and constructive feedback on our manuscript. We are pleased to hear that you find the research goal and methodology to be appropriate and valuable. Your acknowledgment of the study’s high application potential, particularly the predictive performance of the framework, is very encouraging. We have worked hard on data collection, model construction, and validation, and it is gratifying to know that these efforts are appreciated. We value your insights and will continue to refine the manuscript to ensure that it meets the highest standards for clarity and impact.

• Forest fire detection and susceptibility mapping are highly active research theme. Therefore, the current literature review section is definitely not sufficient to take into account recent advancements in these fields. Please review more related papers especially the ones cover:

-Machine learning and geospatial approaches for forest fire detection and susceptibility

-Deep learning, convolutional neural networks for forest fire detection and susceptibility

-Unsupervised learning approaches

-Hybridizations of machine learning and metaheuristic approaches

Response: Thank you for your kind suggestion. According to your advices, we have modified this section (line 58 to 101) in blue font in the revised text.

• In section 1, please add at least 1 paragraph to summarize the research gaps.

Response: Thank you for your kind suggestion. In the line 99 to 101, and 108 to 111, we have added the research gaps.

• In section 1, add at least 1 paragraph to clearly point out the innovative contributions of the paper and how the current work can help fill the existing research gaps in the literature.

Response: Thank you for your kind suggestion. At present, there is little research on machine learning for fire prediction in Zhejiang Province. Our results provide important reference for forest fire prevention in Zhejiang Province. We have added the related content in the line111 to 118.

• Elaborate on the process used to derive the fire radiative power (FRP).

Response: Thank you for your kind suggestion. Fire Radiative Power (FRP) represents the rate at which fire radiative energy is released. We have added the formula in lines 142 to 157 of the revised manuscript.

• Related to the fact “from 2013 to 2023, a total of 2,713 forest fire events were recorded in Zhejiang…”, please state the source or reference to the forest fire inventory.

Response: We have added the forest fire inventory in the line 195 to 207 of revised manuscript. And we also modified the describe of this paragraph in line 266.

• Provide more details on the climatic data, including precipitation, temperature (TEM), land surface temperature (LST), sunshine duration (SD), and relative humidity (RH). The data source, information about the stations, and the methods used to obtain the measurements should be clearly stated.

Response: Thank you for your kind suggestion. The climatic data used in this study include precipitation (PRE), temperature (TEM), land surface temperature (LST), sunshine duration (SD), and relative humidity (RH). These data were obtained from the Resource and Environmental Science Data Platform (https://www.resdc.cn/) and were collected from more than 70 meteorological stations across Zhejiang Province. The data have a spatial resolution of 1 km and were interpolated using ArcGIS 10.8 to ensure consistency across all covariates. The measurements were acquired using standard ground-based meteorological instruments and comply with national measurement standards. We have provided the detail information about the climatic data in the line 170 to 176 in the revised manuscript.

• The explanatory variables used in previous works should be reviewed with more details. The authors should provide comments or discussions about the explanatory variables that were used in previous work but were not employed in the current work.

Response: Thank you for your kind suggestion. Previous studies have commonly employed explanatory variables such as NDVI, slope, wind speed, and fuel moisture for forest fire prediction. However, this study did not include wind speed or fuel moisture due to the lack of long-term and complete observational data in the study area. Additionally, the model tests demonstrated that the selected variables, such as elevation and precipitation, were sufficient to achieve high prediction accuracy, minimizing the necessity of including additional variables with incomplete datasets.

• Consider the uses of demographic factors (e.g., population density), distance to roads, distance to residential areas, and distance to tourist hubs because they have been proven to be relevant in forest fire modeling.

Response: Thank you for your kind suggestion. Variables like population density, distance to roads, and distance to residential areas have proven relevant in some studies for forest fire modeling. However, this study focuses primarily on natural environmental factors, aiming to assess the influence of climate and topography on forest fires. Incorporating socioeconomic factors could enhance the comprehensiveness of future models and will be considered in subsequent studies.

• Why were factors such as waterbody density, distance to major waterbody, or distance to rivers not considered in the current work? Please provide some comments.

Response: Factors such as waterbody density and distance to major rivers were not included because forest fires in Zhejiang Province predominantly occur in mountainous and dry areas rather than near water sources. Initial model tests showed that these variables had limited significance in this region. However, they may be explored in future research to account for potential localized effects.

• When analyzing the effect of the influencing factors, are there any differences in the governing factors of forest fire for different zones in the study area, such as northern vs. southern regions or coastal vs. inland?

Response: Thank you for your kind suggestion. We have rewritten the section of discussion in the revised manuscript. Our study reveals that the factors influencing forest fires vary across regions in Zhejiang Province. For instance, southern regions exhibit higher fire risks due to greater topographic variation and lower precipitation, whereas northern areas, with relatively flat terrain and higher precipitation, have lower fire risks. In addition, coastal areas also have lower fire risks, especially the Ningbo, and Zhoushan areas.

• SMOTE is used to balance the dataset. Although SMOTE is an effective method, it actually creates artificial data samples in certain classes. Please test the model’s predictive capability when all testing samples are not generated by SMOTE.

Response: The SMOTE method generates synthetic instances from the minority classes that have fewer data points, thereby augmenting the dataset to achieve balance. This process helps to equalize the representation of each class in the dataset, making it more reflective of a balanced scenario. It's important to clarify that this technique does not alter the predictive accuracy of the model; rather, it supplies the model with additional instances that are variations of the existing samples within the same category. Most researches have used this method to resample data in their works.

Reference:

Zhang L, Shi C, Zhang F. Predicting Forest Fire Area Growth Rate Using an Ensemble Algorithm. Forests, 2024, 15(9): 1493.

Tavakoli F, Naik K, Zaman M, et al. Big Data Synthesis and Class Imbalance Rectification for Enhanced Forest Fire Classification Modeling//ICAART (2). 2024: 264-275.

Tang J, Weeramongkolkul M, Suwankesawong S, et al. Toward a more resilient Thailand: Developing a machine learning-powered forest fire warning system. Heliyon, 2024, 10(13).

• Please elaborate on how the findings of the current work helps support the land use planning and forest management policy.

Response: Thank you for your kind suggestion. We have rewritten the section of discussion in the revised manuscript. We provide valuable insights for land use planning and forest management by identifying high-risk areas for forest fires. For instance, regions like Lishui and Wenzhou, identified as high-risk zones, can prioritize resource allocation for fire prevention and restrict high-risk activities. Moreover, by identifying key influencing factors such as precipitation and elevation, the study offers actionable data to support targeted fire prevention measures, enhancing policy formulation for sustainable forest management. In a future study, we plan to focus on conducting on-site data collection to make the model more suitable for the area where it will be deployed.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

The manuscript is interesting and has the potential to be published and become a reference in the scientific literature. However, the manuscript needs adjustments, corrections and more in-depth discussions of the study results. Suggestions, corrections and considerations regarding the manuscript are included in the attached file.

Comments for author File: Comments.pdf

Author Response

The manuscript is interesting and has the potential to be published and become a reference in the scientific literature. However, the manuscript needs adjustments, corrections and more in-depth discussions of the study results. Suggestions, corrections and considerations regarding the manuscript are included in the attached file.

Response: Thank you for your positive and constructive feedback on our manuscript. We are pleased to hear that you find the research goal and methodology to be appropriate and valuable. According to your suggestions, we have modified the manuscript.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have revised the paper well.

The reviewer recommends the publication of the manuscript.