A Machine-Learning Approach to Predicting Daily Wildfire Expansion Rate

Round 1

Reviewer 1 Report

This paper is well presented and the research methods are sound. Overall I accept that the authors applied reasonable approaches and the findings are informative. I would like to see a bit more discussion on the application and the potential benefits of the ML models for wildfire planning and response.

The historical fire data are derived from 250m pixels and some of the variables are at 0.25 degrees. How does the relatively coarse nature of these data affect the ability to predict at the sq km scale? For instance if the MAE = 1 sq km, that is effectively 4, 250 m pixels. Whereas 0.25 degree cells are ~25km at the equator. I think some discussion is warranted around this.

Line-specific comments:

• Line 90 - repeated sentence in caption
• Line 113 - "…include value in previous month, mean value in previous year." Which value for previous month? Is yearly mean informative? How about an accumulating threshold (like GDD) or accumulation above/below median.
• Line 143 - NDVI data from what period?
• Line 160 - "Substantial" How is that defined. DO you mean widespread?
• Line 166 - Figure 2. Move to appendix/supplemental?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Title：Machine-Learning Approach to Predict Daily Wildfire Expansion Rate

By: Assaf Shmuel and Eyal Heifetz

Review comments

The authors utilized three machine learning methods (XGBoost, RF, MLP) to predict daily wildfire expansion rate on global scale, they also compared with LR and found that the machine learning methods performed well. In general, the manuscripts were well written. However, the results were not discussed profoundly. Therefore, I suggest a major revision.

Specific comments

1.     Line 9-10: I can’t understand the dynamics in the sentence, “The dynamics which determine wildfire growth rates are complex and depend on numerous meteorological factors, topography, and fuel loads”. Did you mean factors? I suggest change dynamics to factors.

2.     Line 45: what is ibid?

3.     In the data section, how to deal with the individual fire that burned for consecutive dayst

4.     Line 276-277: you said that the bottom left side of the table includes the entire dataset including observations from the first three days of the fire, however, on line 288-289, you said the bottom left model is trained on the entire dataset and does not include data on previous days of the fire. I am confused with its inconsistence.

5.     The results were not well discussed. You should compare your findings with previous works and more references need to be cited.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

General Comments:

This paper investigates the prediction of daily growth rate of wildfires using a global wildfire dataset. The study applies various machine learning models, and the results indicate that the XGBoost model performs the best in terms of prediction. Generally, the citation format of the article is incorrect. Please modify it according to the reference format of Fire Journal. In addition, some images in the article have low resolution, and it is recommended to set a higher DPI when generating the images. There are some specific areas that require further revisions by the authors:

Specific Comments:

Abstract: Please provide more detailed information to help readers better understand which types of machine learning models you used. Please provide more detailed information to help readers better understand which types of machine learning models you used.

Introduction: When introducing the concept of machine learning, it may be necessary to explain in more detail what machine learning is, its characteristics, and why it is useful for wildfire prediction. In addition, more literature reviews need to be added to strengthen the background of the article.

Methodology: The model evaluation in the paper only used the MAE (Mean Absolute Error) as the metric, which is very sensitive to outliers. A more scientific approach would be to use a variety of evaluation metrics to assess from multiple dimensions. The paper utilizes a simple train-test split and does not include cross-validation or hold-out validation approaches that could make the results more reliable. Furthermore, it is suggested to provide more detailed information on the tuning of hyperparameters.

Results: This section lacks necessary visualizations, such as scatter plots showing partial predictions and true values, and box plots comparing prediction errors of different models.

Discussion: Although some limitations were briefly mentioned, the focus was mainly on future research directions, without a deep discussion of the limitations of this study. It is suggested to add a discussion on the limitations of the research. Besides, The discussion section did not adequately summarize the contribution of this study to the knowledge in this research field, which needs to be strengthened. The innovation and contribution of this study can be summarized from three aspects: data, methods, and results.

The writing of the paper is generally smooth, but in some places, some grammatical corrections and expression clarifications may be needed.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The authors have expanded the abstract and introduction as requested in the comments. New metrics have been added for model performance evaluation and are now displayed in the results figures and tables. Details on model parameters, training and testing have also been supplemented. It is evident that the authors have made adequate modifications and additions in response to the feedback on content, and thus the manuscript appears satisfactory for publication based on substance. However, it is advisable that the authors review the formatting of the manuscript as well as the use of certain punctuation marks, as attention to these details is warranted.

The English quality is good overall, but some details still need revision.

Author Response

We would like to express our gratitude to the reviewer for her or his comments. We agree that the manuscript in its current form has substantially improved thanks to the reviewer's suggestions.

We have corrected the references formatting and several punctuation marks.