Development of an Algorithm for Assessing the Scope of Large Forest Fire Using VIIRS-Based Data and Machine Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this study, ST-MASK was developed using VIIRS data and ST-DBSCAN algorithm to distinguish forest fires from other industrial hotspots to support global forest fire monitoring. This research is important for forest fire prevention and control, helping to detect fires in a timely manner and improving the efficiency and effectiveness of fire response. However, there are the following problems that need to be corrected by the authors:

1, I think it would be better to put Status of Study Areas before Research Method.

2, I recommend that the authors reorganize the logical structure of the introduction section to ensure that there is a natural transition between sections so that the reader can clearly understand the authors' motivation and direction of the study.

3, Please embellish this sentence from lines 123-125 in the paper. (‘’Therefore, to fully automate forest fire detection, the aim is to develop an algorithm that extracts spatiotemporal patterns for the active removal of false positives and other active fires and the detection of forest fires.”)

4, Please note the use of punctuation in lines 310-311. “Equation (2) shows that dNBR is calculated as the difference between “PrefireNBR” and “PostfireNBR.””

5, Please check the language of the paper to make sure it is correct.

 

Comments on the Quality of English Language

Please check the language of the paper to make sure it is correct.

Author Response

Point 1: I think it would be better to put Status of Study Areas before Research Method.

• Response 1: (Page 3 line 134~158): As you suggested, we have moved the status of the study area to the first section of the Methods section to better explain the content of the study.

 

Point 2: I recommend that the authors reorganize the logical structure of the introduction section to ensure that there is a natural transition between sections so that the reader can clearly understand the authors' motivation and direction of the study.

• Response 2: (Page 3 line 121~132): We found that the transition between the sections was not seamless, with two sections referring to past research trends and one referring to a specific part of this research. Therefore, we have reordered the sections to describe remote sensing-based wildfire detection/monitoring techniques and then VIIRS-based forest fire monitoring to enhance the transition.

 

Point 3: Please embellish this sentence from lines 123-125 in the paper. (‘’Therefore, to fully automate forest fire detection, the aim is to develop an algorithm that extracts spatiotemporal patterns for the active removal of false positives and other active fires and the detection of forest fires.”)

• Response 3: (Page 3 line 124~128): Thank you for pointing out that the sentence didn't flow well, and we've revised it as you suggested.

 

Point 4: Please note the use of punctuation in lines 310-311. “Equation (2) shows that dNBR is calculated as the difference between “PrefireNBR” and “PostfireNBR.””

• Response 4: (Page 11 line 374~385): We have replaced the terms used in the equations with TeX formatting rather than punctuation quotes. We have also replaced the equations in the dNBR with those commonly used in other papers in the MDPI.

 

Point 5: 5, Please check the language of the paper to make sure it is correct.

• Response 5: We've double-checked the overall word consistency, but we're not native speakers, so there may be some unnatural flow in the grammar. We will make changes as you suggested.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1. Introduction section:

The introduction section (lines 30-35) provides background information on forest fires, but it is recommended that the authors more clearly point out the innovation of this study and the need for the study in lines 120-125 of the introduction. masks that provide filtration of forest fire hotspots (including forest fires, and planned burn-outs) are already very mature technological tools for engineering, so please combine with your own research methodology for a detailed elaboration of the method's innovativeness and the study's Necessity.

 

2. research design:

The study design is described in lines 127-155, but it is recommended that the authors provide more information about the rationale for the study design, including the reasons for choosing the particular algorithm and parameters, and preferably add a detailed elaboration of the formulas for the algorithm.

 

3. methodology description:

The methods section is described in detail in lines 181-300, but it is recommended that the authors add a presentation of the specifics of the study methodology to the process elaboration, ensuring that all method descriptions are closely related to the study objectives and results.

 

4. results presentation:

The results section should be clearly presented in lines 372-427.Figure 10, Figure 11 suggests using text to elaborate on the specific meaning of the symbols within the figure.It is felt that fires have occurred throughout the figure and it is recommended that text and more intuitive visualisations be used to enhance the presentation of the results.The results section should be clearly presented in lines 372-427.The results section should be clearly presented in lines 372-427.

 

5. conclusion support:

The conclusion section is missing, which is a major flaw in the paper. It is recommended that the authors write a conclusion to summarise the main findings of the study and the practical implications for forest fire management policy.

 

6. originality/novelty:

The ST-MASK algorithm proposed in the study is presented in lines 238-247, showing novelty. It is recommended that the authors further discuss the comparison of the method with existing techniques and potential applications.

 

7. importance of content:

The content of the study is considered important for environmental protection and disaster management in lines 1-19. It is recommended that the authors further emphasise the contribution of their research to theory and practice.

 

8. presentation quality:

The authors' presentation flow and images are clear and relevant to the text content.

 

9. scientific soundness:

Scientific soundness needs to be assessed based on the logic of the methodology, the accuracy of the data and the validity of the analysis. It is recommended that the authors provide an in-depth discussion of their contribution to the existing body of knowledge, including the discovery of regularities, in the discussion section in lines 460-495.

 

10 The conclusion section is missing:

The paper lacks a conclusion section, which is an important part of a research paper. It is recommended that the authors add this section to improve the structure of the paper.

 

11. Insufficient refinement of remote sensing data:

The paper mainly belongs to the category of engineering applications, and it is recommended that the authors further refine the remote sensing data in order to demonstrate the contribution to the theoretical development.

 

12. Depth of theoretical research:

It is recommended that the authors strengthen the depth of theoretical research and enhance the theoretical value of the research through more in-depth data analysis, model optimisation or algorithm improvement.

 

13. references:

Ensure that all references are listed correctly according to the journal's citation format and that all references are properly cited in the text.

 

14. data availability statement:

Lines 182-204, should detail how the data used in the study was obtained or accessed. Also detail the currency of the data and its role in this thesis.

 

15. The title clearly conveys the core of the study, but consider whether keywords such as ‘VIIRS-based and machine learning’ could be included to enhance the information. The abstract section provides a comprehensive overview of the research, but it is recommended that it be checked for grammatical or spelling errors and to ensure conciseness.

Overall Evaluation:

This study has potential applications in the field of forest fire monitoring and management. However, in order to improve the quality and impact of the paper, the authors need to revise the above issues in detail.

Comments on the Quality of English Language

English language quality:

Overall, the English language quality of the document is good, but authors are advised to proofread it carefully to exclude any possible grammatical or spelling errors.

Author Response

Point 1: 1. Introduction section:

The introduction section (lines 30-35) provides background information on forest fires, but it is recommended that the authors more clearly point out the innovation of this study and the need for the study in lines 120-125 of the introduction. masks that provide filtration of forest fire hotspots (including forest fires, and planned burn-outs) are already very mature technological tools for engineering, so please combine with your own research methodology for a detailed elaboration of the method's innovativeness and the study's Necessity.

• Response 1: (Page 3 line 121~132): The procedure that this study follows to filter forest fire hotspots can be made more accurate with additional data, but in essentially, the process of forming the ST-MASK does not require the intervention of other factors. Therefore, the contribution of the ST-MASK to this study is that it is a technique that allows filtering based on a single data source. To highlight this point, we emphasise that the methodology generates filters based on unsupervised learning.

 

Point 2: 2. research design:

The study design is described in lines 127-155, but it is recommended that the authors provide more information about the rationale for the study design, including the reasons for choosing the particular algorithm and parameters, and preferably add a detailed elaboration of the formulas for the algorithm.

• Response 2: (Page 6 line 235~286, Page 10 line 340~360): The algorithm at the core of our study is DBSCAN, but we did not provide the reason for using it and the value of the Eps parameter. Therefore, as you recommended, we have added the reason for using DBSCAN among the various clustering algorithms in our study, the parameters of DBSCAN used in our study.

 

Point 3: 3. methodology description:

The methods section is described in detail in lines 181-300, but it is recommended that the authors add a presentation of the specifics of the study methodology to the process elaboration, ensuring that all method descriptions are closely related to the study objectives and results.

• Response 3: (Page 9 line 316~338): The detailed algorithmic sequence for ST-MASK presented in this study is described in the mathematical form.

 

Point 4: 4. results presentation:

The results section should be clearly presented in lines 372-427.Figure 10, Figure 11 suggests using text to elaborate on the specific meaning of the symbols within the figure.It is felt that fires have occurred throughout the figure and it is recommended that text and more intuitive visualisations be used to enhance the presentation of the results.The results section should be clearly presented in lines 372-427.The results section should be clearly presented in lines 372-427.

• Response 4: (page 16 line 489~495, Page 17 line 509~522): Added textual figure descriptions in the manuscript, split unclear Figure 11 into Figures 14 and 15, and added legends.

 

Point 5: 5. conclusion support:

The conclusion section is missing, which is a major flaw in the paper. It is recommended that the authors write a conclusion to summarise the main findings of the study and the practical implications for forest fire management policy.

• Response 5: (Page 20 line 586~615): We summarise the major outcomes of the study and discuss the implications for forest fire monitoring in the conclusion.

 

Point 6: 6. originality/novelty:

The ST-MASK algorithm proposed in the study is presented in lines 238-247, showing novelty. It is recommended that the authors further discuss the comparison of the method with existing techniques and potential applications.

• Response 6: (page 19 line 550~557, Page 20 line 586~615): We have added a section in the conclusion about the comparison with existing methods. We consider this research to have a high potential application, so we have mentioned the potential use of this method in the conclusion.

 

Point 7: 7. importance of content:

The content of the study is considered important for environmental protection and disaster management in lines 1-19. It is recommended that the authors further emphasise the contribution of their research to theory and practice.

• Response 7: (Page 1 line 16~19): The abstract has been amended to mention that satellites have significant advantages in terms of data acquisition, making them the most effective method for environmental and disaster monitoring.

 

Point 8: 8. presentation quality:

The authors' presentation flow and images are clear and relevant to the text content.

• Response 8: Thank you for your comments.

 

Point 9: 9. scientific soundness:

Scientific soundness needs to be assessed based on the logic of the methodology, the accuracy of the data and the validity of the analysis. It is recommended that the authors provide an in-depth discussion of their contribution to the existing body of knowledge, including the discovery of regularities, in the discussion section in lines 460-495.

• Response 9: (Page 19 line 552~557, Page 20 line 603~615): Based on the ability of DBSCAN to detect outliers, we attempted to cluster VIIRS patterns from commercial activities or light reflections, highlighting the potential for this technique to be used in forest fire monitoring. We also added to the discussion session the potential for this methodology to be applied to a wide range of situations beyond VIIRS-based forest fire extraction, as it can be used to identify areas of outliers or normal values with just a few buffer zones instead of complex filtering using land cover data.

 

Point 10: 10 The conclusion section is missing:

The paper lacks a conclusion section, which is an important part of a research paper. It is recommended that the authors add this section to improve the structure of the paper.

• Response 10: (Page 20 line 586~615): We have added a conclusion, which comments on the significance of the study, improvements over existing methods, directions for improvement and potential applications of the study.

 

Point 11: 11. Insufficient refinement of remote sensing data:

The paper mainly belongs to the category of engineering applications, and it is recommended that the authors further refine the remote sensing data in order to demonstrate the contribution to the theoretical development.

• Response 11: (Page 20 line 594~596): This study utilized the best data available in the private sector. Further refinement of the study would be possible with more advanced data. We have added this as a possible improvement to the study in the conclusion.

 

Point 12: 12. Depth of theoretical research:

It is recommended that the authors strengthen the depth of theoretical research and enhance the theoretical value of the research through more in-depth data analysis, model optimisation or algorithm improvement.

• Response 12: (Page 20 line 596~602): We have described the algorithmic process in more detail to optimise it for future research, and we have described the model selection process, which we have added to mention as a possible improvement to this study.

 

Point 13: 13. references:

Ensure that all references are listed correctly according to the journal's citation format and that all references are properly cited in the text.

• Response 13: (Page 20 line 628~751): References have been reviewed and errors in some references have been corrected. The citations on the webpage have been revised to conform to the MDPI-ACS citation format.

 

Point 14: 14. data availability statement:

Lines 182-204, should detail how the data used in the study was obtained or accessed. Also detail the currency of the data and its role in this thesis.

• Response 14: (Page 5 line 199~234): We have reconfirmed the availability of the data and added accessible links to the data used as citations. We have also revised the text to explain the role of the data in more detail.

 

Point 15: 15. The title clearly conveys the core of the study, but consider whether keywords such as ‘VIIRS-based and machine learning’ could be included to enhance the information. The abstract section provides a comprehensive overview of the research, but it is recommended that it be checked for grammatical or spelling errors and to ensure conciseness.

• Response 15: (Page 1 line 2~30): As the name of journal is 'remote sensing', we have modified the title to refer directly to VIIRS rather than to an inclusionary word 'satellite'. The abstract has been reviewed for simplicity.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The study proposes an algorithm to reduce false positive fires, using hot spots, land cover layers and Landsat scenes. The innovation of the study is to propose a spatio-temporal algorithm that allows grouping hot spots into polygons to prevent misidentification of hot spots as wildfires (ST-MASK). However, I believe that it is necessary to better explain the assumption that nearby hot spots should be grouped as similar entities, when there may be false positives combined with wildfires. Polygons are used as a mask to eliminate hot spots due to industrial or other human activity. In this regard, hot spot elimination is based on the land cover layer and ST-DBSCAN polygons. However, with an accurate satellite classification this removal would not require the ST-MASK.

Table 2. The upper range of the unburned severity level is overlapped with the low and moderate levels (high and high).

Section 2.5.3 should be justified in terms of the algorithm to obtain ST-MASK.

Line 388. Analysis not explained in methods.

Figure 9 corresponds to methods.

395. Corresponds to methods, since it specifies the classification to be used.

405 The increase (16%) of non-forest fire events due to ST-MASK is not clear, as the same land use layer is used. Grouping hot spots by their geographic position may increase the risk of overestimating non-forest fires, since forest fires can be grouped with non-forest fires because they are close to each other.

Line 413. Corresponds to methods.

Line 422. The contribution of the study to eliminate false forest fires, therefore the analysis would have to validate these results, not the forest fires.

Figure 11 include the legend of the points and polygons presented.

The discussion is insufficient, as the interpretation of the results is incomplete. The authors highlight the advantages of the algorithm but omit to present the factors that may compromise their results, the studies that should be carried out to support their results, as well as the weaknesses of the study and how they could be reduced.

Author Response

Point 1: Table 2. The upper range of the unburned severity level is overlapped with the low and moderate levels (high and high).

• Response 1: (Page 11 line 385): We have checked Table2 for errors, but there is no overlap. This table uses the same values as Table LA_2 on page 256 in the 39th citation (https://doi.org/10.2737/rmrs-gtr-164).

 

Point 2: Section 2.5.3 should be justified in terms of the algorithm to obtain ST-MASK.

• Response 2: (Page 17 line 498~507): We added that the results were improved by using ST-MASK to justify.

 

Point 3: Line 388. Analysis not explained in methods.

• Response 3: (Page 5 line 182~188): While there was mention of performing numerical analysis of VIIRS, there was no mention of kernel density estimation (KDE), so we added a section on KDE to the methodology.

 

Point 4: Figure 9 corresponds to methods.

• Response 4: (Page 4 line 160~160): Moved this figure to Methodology as suggested.

 

Point 5: 395. Corresponds to methods, since it specifies the classification to be used.

• Response 5: (Page 4 line 161~168): Moved this section to Methodology as suggested.

 

Point 6: 405 The increase (16%) of non-forest fire events due to ST-MASK is not clear, as the same land use layer is used. Grouping hot spots by their geographic position may increase the risk of overestimating non-forest fires, since forest fires can be grouped with non-forest fires because they are close to each other.

• Response 6: Even when using the same land use layer, the hotspots have a resolution of up to 375 metres, so it is possible that scattered fire hotspots may appear to be located in forest areas of the land use layer. This study also aims to remove additional hotspots that cannot be distinguished by land cover to reflect the geographical characteristics of Korea, where industrial facilities and forests are bound to be in close proximity. Therefore, the 16% represents hotspots detected in forests in close proximity to industrial facilities.

 

Point 7: Line 413. Corresponds to methods.

• Response 7: Forest fire found within the time period mentioned in the method is considered as "Result" so it is validated in the results section.

 

Point 8: Line 422. The contribution of the study to eliminate false forest fires, therefore the analysis would have to validate these results, not the forest fires.

• Response 8: (Page 16 line 476~531): As this is a study of false forest fires, we have modified the order of presentation and analysis of the results. We have added a section on the process of verifying the accuracy of forest fire detection and removing false forest fires with dNBR results, and a section on the results of applying and not applying ST-MASK.

 

Point 9: Figure 11 include the legend of the points and polygons presented.

• Response 9: (Page 17 line 508~531): We split the unclear Figure 11 into Figures 14 and 15 and added a legend.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

A lot of improvements have already been made based on the suggestions, giving us a further understanding of the article, which is somewhat innovative, but the conclusions are written more like the discussion part of the article, and it is recommended to make changes to get some general conclusions and patterns.

Comments on the Quality of English Language

The quality of English writing has improved dramatically

Author Response

Point 1: A lot of improvements have already been made based on the suggestions, giving us a further understanding of the article, which is somewhat innovative, but the conclusions are written more like the discussion part of the article, and it is recommended to make changes to get some general conclusions and patterns.

• Response 1: We appreciate your suggestions and have rewritten the discussion and conclusions. The discussion now fits the flow of the paper and we have added a discussion of what was mentioned in the results but was missing. The conclusions are more general and include the discovery of patterns. We also checked the quality of the English, including a first review with an expert.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors clarified the observations that I made and expanded their discussion, which improved the clarity of the document. The authors omitted to answer the first observation that I made to their study: “Polygons are used as a mask to eliminate hot spots due to industrial or other human activities. In this sense, hot spot removal is based on the ground cover layer and the ST-DBSCAN polygons. However, with accurate satellite classification this removal would not require the ST-MASK”. I suggest that authors could discuss the limitation of using a remote sensor that has a spatial resolution of 375m, which can lead to false positives. In addition to explaining in the discussion why the method they use, besides being automatic, is better. In this sense, the authors refer in line 591 and 592 that "NBR-based techniques can take a long time for satellites to scan the desired area and are highly affected by cloud cover". However, this applies to types of coverage that change in a short time, but industrial or urban installations present changes that can be identified over months or years. In addition, there are open access satellite scenes, such as Sentinel, which have a resolution of 10 m and can be used to create a layer of current industrial or urban areas. Such layers can be updated annually. 

The discussion highlights the weaknesses of noise in forest fire detection, but its self-criticism is weak. As well as the factors that may compromise their results, the studies that should be carried out to support their results, as well as the weaknesses of the study and how they could be reduced.

Author Response

Point 1: The authors clarified the observations that I made and expanded their discussion, which improved the clarity of the document. The authors omitted to answer the first observation that I made to their study: “Polygons are used as a mask to eliminate hot spots due to industrial or other human activities. In this sense, hot spot removal is based on the ground cover layer and the ST-DBSCAN polygons. However, with accurate satellite classification this removal would not require the ST-MASK”. I suggest that authors could discuss the limitation of using a remote sensor that has a spatial resolution of 375m, which can lead to false positives. In addition to explaining in the discussion why the method they use, besides being automatic, is better. In this sense, the authors refer in line 591 and 592 that "NBR-based techniques can take a long time for satellites to scan the desired area and are highly affected by cloud cover". However, this applies to types of coverage that change in a short time, but industrial or urban installations present changes that can be identified over months or years. In addition, there are open access satellite scenes, such as Sentinel, which have a resolution of 10 m and can be used to create a layer of current industrial or urban areas. Such layers can be updated annually.

The discussion highlights the weaknesses of noise in forest fire detection, but its self-criticism is weak. As well as the factors that may compromise their results, the studies that should be carried out to support their results, as well as the weaknesses of the study and how they could be reduced.

 

• Response 1-1: (Page 20 line 551~556): In the first paragraph of the missing answer, Round1, the reviewer pointed out that " with an accurate satellite classification this removal would not require the ST-MASK.". In response, the reviewer's suggestion is in the same direction as Response6 in Round1, and we can use it. The additional discussion is as follows:

As of 2024, the highest resolution of VIIRS to detect hotspots is 375 m (SUOMI NPP). If the resolution of the hotspot is lower than land cover, false positive hotspots can remain even when masked by land cover. On the other hand, ST-MASK filtering can remove 94% of the false positive hotspots, which represents a 16.33% noise reduction compared with LULC masking (based on August 2022 Gangwon-do data).

• Response 1-2: (Page 19 line 536~594, Page 20 line 624~637): The general advantages of ST-MASK are mentioned in the newly written discussion as a comparison to other techniques. An advantage of ST-MASK other than its performance and benefits from automation is highlighted in Conclusion[624-637] as the potential for the key pattern recognition algorithms to be used in a variety of studies.

 The application of multiple algorithms could enhance research depth and segment roles, leading to techniques applicable across diverse fields. These advantages are expected to address various analytical challenges. Contemporary data analysis extends be-yond remote sensing to a wide array of collected data. While some data can be easily distinguished through inherent attributes, spatial data often lack structured attributes. For three-dimensional data sets comprising location and time elements, ST-MASK can efficiently create representative regions, detect noise, and filter or extract information. This approach is applicable even to non-spatial data. A typical application of outlier detection algorithms is machine fault or failure detection. ST-MASK can generate masks of normal machine operating ranges using x and y coordinates and create polygons representing outlier patterns for failure analysis. This innovative approach offers the potential for enhancing data analysis and outlier detection across various domains, from remote sensing to industrial applications, by providing a more efficient and adaptable method for pattern recognition and anomaly detection.

• Response 1-3: (Page 6 line 224): Regarding the point about the shortcomings of NBR, as shown on page 224 of the text, NBR is not relevant to the land cover used in the study. In this study, NBR was used to track the area of forest fires and compare with VIIRS-based area tracking by calculating the changing land vegetation burned as dNBR. Therefore, it should be applied to coverage types that change in a short period of time, as ‘forest fire monitoring’ is intended to be a quick picture of the situation rather than ‘forest fire investigation’, which requires precision.
• Response 1-4: (Page 19 line 579~594): Thanks for pointing out the lack of criticism in this study. We have added the limitations of ST-MASK's algorithm as well as the areas we would like to address in future research.

The findings in (1)–(4) suggest that the ST-MASK is useful for determining forest fires from VIIRS hotspots and eliminating hotspots that cannot be removed with LULC data. However, this study has a potential limitation in that recently in-stalled heat emitting facilities can be recognized as fire due to absence of ST-mask, because of not enough false positive hotspot data to form it. This limitation is due to the shortcoming of the technique to detect false positives using long-term hotspots, which cannot collect enough clusters from the false positive hotspot data for recently installed facilities. Therefore, to overcome this limitation, this study proposes a complementary method, forming an ST-MASK with fewer than three hotspots, i.e., the minimum number of hotspots that form a convex hull polygon. To validate false positive areas with fewer hotspots, this study simultaneously uses hyperspectral vegetation indices NDVI and NBR. Generally, high-resolution satellites such as Sentinel or Landsat are used to calculate these indices as in this study, but VIIRS can also obtain coarse-resolution NDVI and NBR, while includes Red (I1;0.64 µm), SWIR (I2;0.865 µm) and NIR (I3;1.61 µm) bands, which can be used to justify false-positive hotspots.

Author Response File: Author Response.pdf