A Comparative Study of Genetic Algorithm-Based Ensemble Models and Knowledge-Based Models for Wildfire Susceptibility Mapping

Round 1

Reviewer 1 Report

This manuscript has well investigated and compared multi-criteria decision-making methods with machine learning methods (to prepare a forest fire susceptibility map). This manuscript has enough innovation and different parts of the article are well written. After considering the following corrections, this manuscript can go to the next steps.

1. Research innovation should be stated at the end of the introduction.

2. The fire occurrence points should be shown in Figure 1.

3. The font of all the tables should be checked.

4. The percentage of susceptibility classes using each method should be provided.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

57 - In my opinion, there is no need to explain this general background to fires and their impacts - unless it is specific to this study area and the modeling you have undertaken

88 - fuel load cannot be ignored - it varies across space and time

174 - I wonder how useful is location data for this purpose. It doesn't appear to capture the extent or severity of each fire. I expect that fires in the forested areas would be more extensive, and should have greater influence on the models, but if they are recorded as a single location, they are treated equally as a small fire in other areas

236 - it would be better to have a fractional cover variable that discriminated between living and dead vegetation, and bare ground - the dead fraction providing an indication of fuel load. Together with climate updates, this would provide a dynamic FFS, which is useful. However, this product may not be available for your region. See https://cmi.ga.gov.au/data-products/dea/630/dea-fractional-cover-percentiles-landsat

 

270 - 'was developed'

FIG 3- 'Valedating' should be spelt 'validating'; and 'hyprid' is spelt 'hybrid'. Also 'Final maps' should be 'final maps'

558 - yes, but only because it indicates where there is something (fuel) to burn, but its not a very good surrogate for fuel 

561 - this is counter-intuitive! How would the presence of fuel act as a barrier against fire spread? Also, moisture is not always present - major fire events occur in times when the vegetation is dry - surely this is the case in this region. I suspect that the fire location data is under-reporting fire in forested areas (not picking up the extent and severity of fires), or there are more fires reported elsewhere due to higher ignition rates near human activities 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

The study titled “A Comparative Study of Genetic Algorithm-based Ensemble Models and Knowledge –based Models for wildfire Susceptibility Mapping” provides information on the application of AI/ML models to predict wildfires in Jordan.  The study applied data of fire occurrences obtained from official sources between 1991 and 2021, simultaneously, several predictor variables, such elevation, slope, aspect, land use, distance to roads, wind speed, rainfall, temperature, population density, NDVI, and topographic wetness index were applied to predict fire occurrences at various locations. 

 

Specific comments below

Line 172, can the authors provide information on how fire was mapped, was this obtained from topographic maps, satellite data sources, or some combination.  A table to show the fires that occurred during that time period (1991 to 2021) would be helpful.  Their extent, spatial pattern, would be insightful, further, main causes of these fires and use of fire in the region would be interesting to the reader.

 

Line 177, can the authors provide some indication of how the 70% locations were identified for training the models, similar question for the validation.  Were these chosen randomly or otherwise. 

 

Currently the authors use random partitioning of the dataset for testing and training.  Can the authors also perform a similar training and testing by conducting a temporal division, so for example, exclude one year from the model training and then test the models on the withheld year.  This may establish robust models, that are more reliable and consistent. 

 

Line 192, can the authors follow standard units and abbreviations such as m for meters and msl not amsl.   

 

Line 209, How was the land use map obtained, more details, whether from remote sensing data or some other source.  Similar comments for the population density map, wind speed, rainfall, temperature, NDVI, TWI.  My suggestion, please place this in tabular form, more easier for reader to understand the same.  You can have several columns and include the minimum, maximum, for these various independent variables.

 

Line 269, the authors mention weights for each factor, can they provide some sense for variables with higher and lower weights.

Lines 290, computation of the fire ratios, this needs more clarity, so is this the method used for obtaining the weights, you need to make this clear.  Anyways, the earlier comments regarding the weights remain, did the authors test for multicollinearity, to select the most important variables.

 

Line 299, the flowchart, there are a few typos, please check, such as “valedating”, “hyprid”.

 

Line 317, I don’t see a Table 1, are they referring to table 2 instead.

Figure 10, you have four classes, very low, low, moderate, and high, unfortunately the class intervals are different in the four figures (a, b, c, d), it makes it harder for the reader to follow.  Can you have consistent interval classes for all the four maps of fire susceptibility, that way, we can assess the model results better.

 

I find the authors have included lot more information, they need to present the results better.  They can have the detailed results in the supporting document, that way, the manuscript would be more engaging for the reader.  Right now, they do not present, the tables, figures in an optimal manner, it thus suffers from indiscriminate presentation of the results.  For example can they combine figures 4  to 7 into a single figure or table??

 

Another consistent problem in the manuscript, is the including of the methods and results interchangeably, I don’t know whether this is ok with the journal.  Ideally, one would like to see the methods and results in separate sections.  For example see lines 480 to 495, 516 and 517, 530 to 533 suffer from the same problem. 

The language needs rewriting to bring it up to the journal standard.  There are numerous typos, that I have already highlighted in the comments to the authors.  

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I can't see any evidence that my comments on version 1 have been addressed in version 2

Author Response

Dear esteemed reviewer,
We've addressed all of your valuable comments in our manuscript, but we're not sure why you haven't received our revised response to the reviewer file accurately. Kindly find our responses below and also see the revised manuscript:

Response to reviewer #2

Reviewer's suggestions:

Comment 1: 57 - In my opinion, there is no need to explain this general background to fires and their impacts - unless it is specific to this study area and the modeling you have undertaken

Authors' response to comments 1: Thank you for your fruitful comment. We removed general background to fires and their impact from the second paragraph. The removed text is as below: “In line with Abedi Gheshlaghi's findings in 2019, the natural factors contributing to forest fires include spontaneous combustion, sparks resulting from falling rocks, lightning strikes on trees, and volcanic eruptions (Abedi Gheshlaghi, 2019). Additionally, anthro-pogenic causes involve the proximity of thoroughfares to forests, leading to the heat and fumes emitted by vehicle exhausts, as well as human settlements near forested areas, which can lead to accidental fires (Sari, 2021).”

 

Comment 2: 88 - fuel load cannot be ignored - it varies across space and time

Authors' response to comments 2: Thank you for your comment. We added explanation about fuel load to the introduction section. Please see lines 72 to 74. The added text is as below: “Fuel load plays a critical role in the occurrence and severity of wildfires, as it determines the amount and type of flammable material available to sustain and spread the fire. Understanding and managing fuel load levels is essential for mitigating fire incidents and reducing their impact on ecosystems and human communities.”

Comment 3: 174 - I wonder how useful is location data for this purpose. It doesn't appear to capture the extent or severity of each fire. I expect that fires in the forested areas would be more extensive, and should have greater influence on the models, but if they are recorded as a single location, they are treated equally as a small fire in other areas.

Authors' response to comments 3: Thank you for your comment. The Civil Defense Directorate in the Northern Mazar District has generated these samples by analyzing the polygons that have been affected by fire using the "Create Random Points" technique in the GIS environment. We added this to the section 2.1. Please see lines 184 to 187. The added text is as follow: “The Civil Defense Directorate in the Northern Mazar District has generated these samples by analyzing the polygons that have been affected by fire using the "Create Random Points" technique in the GIS environment.”

 

Comment 4: 236 - it would be better to have a fractional cover variable that discriminated between living and dead vegetation, and bare ground - the dead fraction providing an indication of fuel load. Together with climate updates, this would provide a dynamic FFS, which is useful. However, this product may not be available for your region. See https://cmi.ga.gov.au/data-products/dea/630/dea-fractional-cover-percentiles-landsat

Authors' response to comments 4: Thank you for your wise comment. Having a fractional cover variable that distinguishes between living and dead vegetation, as well as bare ground, would be beneficial. The dead fraction can provide valuable insight into fuel load, especially when combined with climate updates. It would create a dynamic Fire Danger Index (FFS), which is highly useful for assessing fire risks. However, unfortunately this product is not available for our study area.

 

Comment 5: 270 - 'was developed'

Authors' response to comments 5: Done.

 

Comment 6: FIG 3- 'Valedating' should be spelt 'validating'; and 'hyprid' is spelt 'hybrid'. Also 'Final maps' should be 'final maps'

Authors' response to comments 6: Thank you for your comment. The mentioned corrections was done. Please see figure 3.

 

Comment 7: 558 - yes, but only because it indicates where there is something (fuel) to burn, but its not a very good surrogate for fuel

Authors' response to comments 7: Thank you for your wise comment. We added this point to discussion section. Please see lines 593 to 594. The added text is as below: “However, NDVI can identify areas with potential fuel sources for burning, but it is not an ideal substitute for directly measuring fuel quantity or quality.”

 

Comment 8: 561 - this is counter-intuitive! How would the presence of fuel act as a barrier against fire spread? Also, moisture is not always present - major fire events occur in times when the vegetation is dry - surely this is the case in this region. I suspect that the fire location data is under-reporting fire in forested areas (not picking up the extent and severity of fires), or there are more fires reported elsewhere due to higher ignition rates near human activities

Authors' response to comments 8: Sorry for confusion. We removed the confused sentences.

 

Author Response File: Author Response.docx

Reviewer 3 Report

The authors have made an attempt to address several of the suggestions provided in the first review.

Author Response

Comments and Suggestions for Authors:

The authors have made an attempt to address several of the suggestions provided in the first review.

Reply: 

Dear respected reviewer,
I wanted to take a moment to express my sincere gratitude for your time and effort in reviewing our paper. Your insightful comments and constructive feedback have been invaluable in improving the quality of our work. We truly appreciate your dedication to the field and your commitment to helping us enhance our research.

Thank you once again for your valuable contribution.

Best regards,