Global Wildfire Susceptibility Mapping Based on Machine Learning Models

Round 1

Reviewer 1 Report

The manuscript entitled “Global Wildfire Prediction based on Machine Learning Models” proposes the application of various Machine learning models to predict wildfires at the global level. Some influencing factors are analysed. The topic would be moderately interesting for journal readers.

However, the paper presents important weaknesses in structure and contents.

1) The crucial problem presented in section 1 is wildfire occurrence prediction. Although, from L35 to L43 fire behaviour simulation models are discussed. These models, either physical or empirical or semi-empirical (see  Sullivan 2009a, 2009b, 2009c), were never developed to predict fire occurrence (i.e. the number of fires) but to simulate fire behavior (rate of spread, flame length, etc). Thus their aim is completely different from the goal of this paper. So all this introduction, in my opinion, is not providing the right context and base to further elaborate the paper. The Authors should focus on papers concerning the use of ML models in fire occurrence prediction, such as Stojanova et al. (2012),  Vecín-Arias et al. (2016), or Van Beusekom et al. (2018). Indeed, some of them are mentioned in Lines 59-72; but not to build up the first part of the paper’s introduction.

2) At L83, Zhang et al (2021) is mentioned: “While [45] is similar 83 to the current study in some respects, our study differs from it in several aspects”. before illustrating the differences between the approach followed in this work and that of Zhang et al (2021), perhaps it would be better to briefly state the purpose of the work. Instead, this aspect is presented only in line 92.

3) The specific objective of your paper is presented on line 92, but the main purposes are not stated. Why are you using various ML models at the global level?

4) The authors use global data, but it is not clear the time extent. Was only 1 year considered for prediction, or was the dataset longer? (at L167 the authors stated that the mean and median monthly burned area from 2003 up to the month prior to that observation). Please clarify this point

5) Following the specific objective of the articles, the authors should better specify the results of section 3 Results. So far, the results are almost the captions of the figures and tables that are presented, but no data or number is presented. The results might include a summary of numerical and graphical results.

6) On the other hand, the discussion presents some results but does not compare the obtained results with results obtained in other similar studies (which is ultimately the goal of the discussion).

7) A conclusion section might be included in the manuscript, to briefly summarize the main results of the article. Also, the authors should add considerations about the usefulness of the study (in the current version some considerations are provided at L380 within the discussion session). Highlight the significant contribution in the current study in relation to the previous studies.

 

Some other considerations are provided to the authors in order to improve the manuscript for further submission.

8) L165, please indicate the site where Regional fire history was obtained.

9) L133, including the mean monthly value of various fire weather indices to predict the fire occurrence and the burned area for the same month of the observation can be useful but clearly but it makes the model lose value because it could never be applied in the future, precisely to predict fire occurrence.

10) L240, please provide more information on SHAP and how to read the related graphs

 

11) L380 “Accurate wildfire predictions by ML models present a promising opportunity to improve wildfire alerts and provide forest managers with tools of assessing regional wildfire risk.” I would suggest better discussing the limits of the scale and the time.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Global Wildfire Prediction based on Machine Learning Models

Dear Authors

The basic science of this paper has been conducted to a good and appropriate standard. The author and his team wrote this paper according to the journal scope. I am glad to review this paper because the author and his/her team write very well. The Authors used machine learning methods to predict wildfires at the global level. There is also fine/minor English language problems. I found some minor mistakes. But the conclusion part is missing. So, I recommended major revision.  I hope, the authors will solve these issues and resubmit as soon as possible. All these are given below;

General Major Comments

1. Line 11- various machine learning methods?????????

2. Modify the abstract section and explain the significance or novelty of the manuscript at the end of the abstract section.

3. Rewrite the abstract section.

4. In the introduction, the Authors should particularly pay attention to the literature review which should be more critical. 

5. Author should clearly explain the main objectives of this study with the central hypothesis which is missing in the Introduction.

6. https://doi.org/10.3390/f13050715

7. https://doi.org/10.3390/f12091211  

8. Rewrite lines 104 and 105

9. Check the quality of figure 6 and figure 8

10. Line 335, why do you bold km2?

11. Conclusion part is missing

In the end, I would like to say about your study. I believe you did a great job but we still need some improvement in your paper. There are still some English grammar and typo errors. I hope you will modify it very soon and resubmit it again in this journal. I will just focus on my comments.

Best Regards 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I thank the authors for their efforts in meeting the reviewers suggestions.

I’ve some other suggestions to further improve the paper, still.

1)   The introduction could still mislead the reader. Indeed, even if the Authors deleted the main parts mentioning fire behavior, this concept is still present at line 54. I suggest to substitute “fire behaviour” with “fire occurrence” wich is the ultimate goal of your paper.

2)   What do you mean with magnitude? If you mean burned area, then magnitude is not the correct term. Please note that The magnitude of a natural event like wildland fire—its potential to cause effects—is a function of energy release rate (Scott 2006). Please modify accordingly.

3)   At L499 the Author stated “The results of this study demonstrate the advantage of ML models over traditional fire weather indices in wildfire prediction.” Actually, fire weather indices are not used for fire prediction but for fire danger estimation (which is completely different). As the Authors know, fire weather indices such as FWI consist of different components that account for the effects of fuel moisture and wind on fire behaviour and spread. The higher the FWI is, the more favourable the meteorological conditions to trigger a wildfire are. So favourable conditions not predictions. With this in mind, I would suggest the Authors to modify the conclusions and all the other parts of the paper were this confusion is provided.

4) I really appreciate how the Authors schematized the paper objectives (L104-110), although nor the results or the discussion are provided following this scheme. I would suggest the Authors to slightly modify the result section in order to be in agreement with the listed objectives and to provide the reader with an immediate feedback

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear Authors

I am glad to see all changes in the revised version of this manuscript.

Best Wishes

Author Response

Please see the attachment.

Author Response File: Author Response.docx