A Brief Review of Machine Learning Algorithms in Forest Fires Science

Round 1

Reviewer 1 Report

Thank you so much for considering me as a potential reviewer of this manuscript. I really enjoyed  reviewing it. This Review is so interesting and match the topics of Applied Science Journal.

This Review is well written an can be publish with minor revisions.

Authors should organize all cited references in the entire Review.

Author Response

We would like to thank the reviewer for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript.

1. Authors should organize all cited references in the entire Review.

Thank you for your observation; (Done)

Reviewer 2 Report

The article is interesting and concerns on important topic, which is predition of forest fires. Nevertheless, I have some suggestion to impove the paper :

1. There are still some unmodified part of "template": at the beginning of the paper (before the title) its type is not indicated, you should also remove the first paragraph of The 3rd chapter "The section may be divided by subheadings...."

2. Fig.1 and 2 are not cited in the text, therefore it is not clear with what it is connected with.

3. In 2.2. chapter there is following sentence " The picture is the feedback loops between Data Preparation and Modelling to depict these iterations" It is not clear which picture authors' mentioned. Additionally, should it be "picture" or "figure"?

4. In my opinion chapter number 3 is the most important, so it should be prepared better. First of all, there is a lot of method's abbrevation no everyone has to be familiar with. For many methods accuraties are given in the text, which make it difficult to make some comparition. I suggest to create the table with appropiate values). I have impression that in this chapter there is a list of publication connected with AI and ML methods used for predicting fire, but in my opinion the very valuable would be some authors' comments about prons and cons of mentioned methods (even part of them, authors' find the most important).

5. References are not prepared according the template (it should be numbered in order of appearance), it should also be cited by numbers.

Author Response

We would like to thank the reviewer for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript.

1. There are still some unmodified part of "template": at the beginning of the paper (before the title) its type is not indicated, you should also remove the first paragraph of The 3rd chapter "The section may be divided by subheadings...."

Thank you for your observation; (Done)

2. 1 and 2 are not cited in the text, therefore it is not clear with what it is connected with.

Thank you for your note. It's added

3. In 2.2. chapter there is following sentence " The picture is the feedback loops between Data Preparation and Modelling to depict these iterations" It is not clear which picture authors' mentioned. Additionally, should it be "picture" or "figure"?

Thank you for your observation. It has been explained.

4. In my opinion chapter number 3 is the most important, so it should be prepared better. First of all, there is a lot of method's abbrevation no everyone has to be familiar with. For many methods accuraties are given in the text, which make it difficult to make some comparition. I suggest to create the table with appropiate values). I have impression that in this chapter there is a list of publication connected with AI and ML methods used for predicting fire, but in my opinion the very valuable would be some authors' comments about prons and cons of mentioned methods (even part of them, authors' find the most important).

Thank you for your observation. Some method's abbreviations have been defined but unfortunately there are many others are actually named as abbreviations.

Thank you for your valuable observation about methods accuracies, which highlights one of the main challenges researchers face, namely the absence of a benchmark dataset, that can give us correct comparisons. In our research, we will strive to construct and adopt a benchmark dataset and then perform comparisons between pros and cons of mentioned methods as you suggested, as it represents a crucial aspect of future work.

5. References are not prepared according to the template (it should be numbered in order of appearance), it should also be cited by numbers.

Thank you for your note; (Done)

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear authors,

I have thoroughly examined your review paper and I must regretfully inform you that I found the content to be rather complex and poorly organized, despite its intention to review ML-based studies in Forest Fire research. Allow me to outline my concerns:

1) The flow of section 3, where you present the body of your review, lacks methodological or thematic organization. The current organization seems to involve adding numerous paragraphs related to each study, without any thematic coherence. Although you have listed some studies on occurrence/likelihood/susceptibility/risk, conditioning factors, burned areas mapping, fire detection, and spatial evolution of fire, these studies are presented in a disorganized and mixed manner. Instead, it would be advisable to select relevant sub-headings with brief explanations of each theme, followed by an overview of key trends in machine learning within those themes. Then, please provide a mention of recent research papers related to each theme.

2) Your review covers a broad time period, despite the existence of many review papers that have already covered those periods. It would be prudent to limit the time period of your review, for example, to the last five years. Some of the papers you mention may be valuable but are quite dated.

3) Section 2 requires editing. Within your review, some studies employ tabular data within a GIS framework, while others delve into computer vision and pattern recognition using remotely sensed products. It is essential to provide a clear distinction between the machine learning methods employed for tabular data and the deep neural networks utilized for imagery data.

4) The studies using ML methods primarily focus on mapping the forest fire susceptibility of specific areas. However, despite the multitude of studies conducted by scholars in recent years, your review falls short in examining these studies comprehensively. I would highly recommend incorporating highly-cited susceptibility studies based on machine learning techniques published since 2021. Here are a few examples:

10.1016/j.ecolind.2021.107869

10.1016/j.ecoinf.2022.101647

10.1016/j.ecoinf.2021.101397

10.1016/j.ecoinf.2021.101292

5) Although you do mention active fire products in the literature review, there is not a significant emphasis placed on the utilization of these products. Additionally, fire detection studies predominantly employ spaceborne or UAV-borne hyperspectral imagery. It is crucial for you to delve further into this preference within your literature review.

Overall, I suggest a resubmission to deliver a more refined review, as it would require additional time and effort. However, the decision ultimately rests with the authors and the editor.

Author Response

We would like to thank the reviewer for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript.

• The flow of section 3, where you present the body of your review, lacks methodological or thematic organization. The current organization seems to involve adding numerous paragraphs related to each study, without any thematic coherence. Although you have listed some studies on occurrence/likelihood/susceptibility/risk, conditioning factors, burned areas mapping, fire detection, and spatial evolution of fire, these studies are presented in a disorganized and mixed manner. Instead, it would be advisable to select relevant sub-headings with brief explanations of each theme, followed by an overview of key trends in machine learning within those themes. Then, please provide a mention of recent research papers related to each theme.

Thank you for your valuable observation, therefore we have reformulated the review of the use of ML in wildfire science as categorized into three main problem domains, including fire detection, fire mapping and fire prediction. Despite of some papers appear in multiple problem domains or subdomains.

• Your review covers a broad time period, despite the existence of many review papers that have already covered those periods. It would be prudent to limit the time period of your review, for example, to the last five years. Some of the papers you mention may be valuable but are quite dated.

Our goal of the review of old papers was that some researchers may invest the data used in the old methods with new algorithms, or vice versa.

• Section 2 requires editing. Within your review, some studies employ tabular data within a GIS framework, while others delve into computer vision and pattern recognition using remotely sensed products. It is essential to provide a clear distinction between the machine learning methods employed for tabular data and the deep neural networks utilized for imagery data.
• The studies using ML methods primarily focus on mapping the forest fire susceptibility of specific areas. However, despite the multitude of studies conducted by scholars in recent years, your review falls short in examining these studies comprehensively. I would highly recommend incorporating highly-cited susceptibility studies based on machine learning techniques published since 2021. Here are a few examples:

10.1016/j.ecolind.2021.107869

10.1016/j.ecoinf.2022.101647

10.1016/j.ecoinf.2021.101397

10.1016/j.ecoinf.2021.101292

• Although you do mention active fire products in the literature review, there is not a significant emphasis placed on the utilization of these products. Additionally, fire detection studies predominantly employ spaceborne or UAV-borne hyperspectral imagery. It is crucial for you to delve further into this preference within your literature review.

Thank you for your observation. A very useful point of view, our current review is a brief review, we will strive to present a scoping comprehensive review of ML applications in wildfir

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

I express my gratitude for extending to me the invitation to assess the revised revision of the manuscript. The authors have made certain alterations in Section 3, primarily in the form of segregating the various themes through the use of sub-sections. Nonetheless, it has come to my attention that certain issues I raised during the previous review have not been adequately addressed. The authors posit that this manuscript constitutes a "brief" review, yet it is crucial to note that review articles ought not to be excessively concise; rather, they should strive for comprehensiveness.

-          The sub-sections within Section 3 still require succinct explanations at the outset for each respective theme, succeeded by a discussion of the key trends encompassing said theme.

-          The review continues to overlook papers of considerable influence (I recommended) within the domain of wildfire susceptibility. A thorough examination of such literature is imperative.

-          The present rendition of Section 2 persists in presenting general information regarding machine learning, despite its requisite focus on expounding upon how ML/DL techniques have been embraced by forestry scholars. It is paramount to elucidate the manner in which these methods have augmented research capabilities. As previously emphasized, Section 2 still lacks a clear differentiation between the machine learning methods employed for tabular data and the deep neural networks utilized for imagery data.

-          I implore the authors to ensure the accuracy of their in-text citation technique in adherence to the guidelines. The usage of an in-text citation such as "[25] did that" may be erroneous. Instead, a more fitting approach might be "ABC et al. [25] did that."

Author Response

We would like to thank the reviewer for careful and thorough reading of this manuscript and for the thoughtful comments and constructive suggestions, which help to improve the quality of this manuscript.

The Paper has been modified according to your directions and notes.

• The sub-sections within Section 3 still require succinct explanations at the outset for each respective theme, succeeded by a discussion of the key trends encompassing said theme.
• Done
•  The review continues to overlook papers of considerable influence (I recommended) within the domain of wildfire susceptibility. A thorough examination of such literature is imperative.
• Done refernces 109-110-111

•  

  The present rendition of Section 2 persists in presenting general information regarding machine learning, despite its requisite focus on expounding upon how ML/DL techniques have been embraced by forestry scholars. It is paramount to elucidate the manner in which these methods have augmented research capabilities. As previously emphasized, Section 2 still lacks a clear differentiation between the machine learning methods employed for tabular data and the deep neural networks utilized for imagery data.

• Done (Section 2.3)

• I implore the authors to ensure the accuracy of their in-text citation technique in adherence to the guidelines. The usage of an in-text citation such as "[25] did that" may be erroneous. Instead, a more fitting approach might be "ABC et al. [25] did that."

• Done