The authors present a novel approach to model burned area, combining two frameworks that have previously been used to model species distribution (Maximum Entropy) and fire (Bayesian Inference). Applying FLAME 1.0 to Brazilian biomes, they evaluate the model performance, differentiating fire categories, and analyze the contribution of various predictor variables to burned area. Additionally, they provide a detailed analysis of the uncertainties in the relationship between fire drivers and fire occurrences.  This paper presents a very relevant and comprehensive study, and makes a strong case for future research to give greater consideration to regional nuances in fire behavior.

General comments

This work is nicely done, and my comments are mostly editorial. Throughout the manuscript, I would suggest to more carefully differentiate between statistical/ data-driven fire modelling vs process-based modelling (I highlighted some of those occurrences in the specific comments below). Similarly, I would suggest to consistently refer to ‘independent variables’/ ‘explanatory variables’ or ‘predictors’ as opposed to just calling them ‘variables’. These are really minor suggestions, and I provide some more specific comments of similar nature below. 

Specific comments

L39: I would suggest to write ‘particularly for the Pampas and Pantanal regions’

L58: There is probably no right or wrong with this, but I would suggest to use Terrestrial Biosphere Model (TBM) as an umbrella term for both LSMs and DGVMs

L76-81: Referring to Bayesian Inference in relation to fire the should also have a reference - e.g. Kelley et al., 2019?

L98: Maybe you could also link to Bayesian Inference here to highlight that Max Entropy alone is not sufficient to model fire, but the combination of both can. The description of Bayesian Inference is a bit sandwiched between descriptions of Max Entropy as it is anyway - maybe it would work to move that part here? But I don’t insist on this suggestion.

L130:  I think there needs to be a space between 500 and m. How did you regrid the data?

L132: You’re not consistent when citing Map Biomas throughout the manuscript

L132: You haven’t defined the abbreviation LULC yet

L143-148: I would first describe how biomes were defined (i.e. move this part to L139 so that it reads ‘[...] across different vegetation types. We based our vegetation categorization on Hardesty et al., 2005 [...]’ or similar) and then at the end describe the specific biomes you came up with.

L158: Are you also including lagged variables? I wonder whether, if you only use August, September, and October values, you lose memory effects in the fuel properties

L168: Can you specify whether you use a sub-daily/ daily or monthly timestep for the climate variables?

L169: ‘We obtained soil and vegetation carbon and soil moisture’ (?) Later in the manuscript you mention you used dead vegetation carbon. If that is the case, you should mention it here.

L195: How did you interpolate the data? Presumably just a linear interpolation?

L197: Maybe rephrase a bit? ‘were calculated for each grid cell’ or similar?

L197-212: Are the forest metrics constant or do they change in time?

Table1: How did you calculate the dry days? I’m mostly wondering in terms of the temporal dimension (e.g. do you restart consecutive dry days every August, or [...]?) 

L222: I think you could be a bit more precise here: You’re assessing the relationships among the predictor variables, not between the predictor variables and fire in this part of your analysis. Of course there are likely non-linear relationships here as well but I’m not sure arguing with the non-linear relationship between predictors AND fire is the best justification here.

L228-232: Then why did you include lightning in the first place:)

L234: I would refer to Table 1 after ‘explanatory variables’

L238-240: Did you change the number of predictor variables later on? ‘Initially’ implies that you did. If you did not, do you expect that changing the number of predictor variables would have a strong impact on your results? If it’s not a lot of work it could be interesting to see in the supplement. However, your manuscript is already very comprehensive and if this would be a lot of work that wouldn’t add a lot, feel free to ignore my comment.

L243: I like that you explain how the predictors relate to fire. Maybe you could include that Tmax and precip relate to fire weather?

L245: See my comment earlier: Did you exclusively use carbon in dead vegetation? If so, how did you derive it (or is it a direct JULES output?)

L248: Can you specify how you aggregated the explanatory variables over time? Does group 2 vary with time or is it constant?

L463-465: I think it would be useful to give more detail on what the different columns (ALL, NAT, NON) depict in the figure caption so it is easier to understand the figure on its own.

L468-494: You mention a lot of values for the mean bias in these two paragraphs, and I’m not sure where they are coming from? Are they listed anywhere? 

L521-524: I would suggest to describe the interpretation of all colors in your figures in the figure caption, including why pixels are white - even if it sounds painfully obvious.

L536-538: I find it a bit confusing here, and also later on, that you don’t specify the direction of the deviation. Is that because your framework doesn’t allow that? Intuitively I would expect that values below the median would have a different impact than values higher than the median, but I might have just misunderstood the metric.

L561-562: ‘The remaining 4.53% of the area remains uncertain’ this is a bit vague. Can rephrase it?

L683-684: I found this also a bit unclear

Figures 8 - 10: Could you mention either in the figure itself or in the figure title (at least in Figure 8) what the different groups are again?

L777: Here for example, I would refer to ‘process-based’ fire models here rather than 'conventional fire models' given you’re citing two papers looking at process-based fire models coupled (also in L283 - 'global fire models' can be process-based or statistical or data-driven [...])

L821-833: This paragraph is a bit unorganized and I’m not clear which aspect you’re focusing on here: That your model is able to better capture temporal variability in fire patterns, or that it has a better regional representation? I assume the last one, but L822 sounds like you tested how well FLAME did in capturing temporal variability. Maybe I missed but did you really test the performance over time? You then also make a bit of a jump to coupled Earth System Modelling in my opinion but the point you make is of course very valid. I would suggest to rewrite it to something like ‘Additionally, while fire-enabled Earth System Models can integrate feedback mechanisms between land and atmosphere, therefore enabling the evaluation of inter-variable effects, offline global fire models do not. Similarly, FLAME is not designed to [...]’ or something along those lines

L867-868: ‘Despite being a combination with land use’ - this is sounds a bit confusing

L880: Would suggest to rewrite to ‘[...] reaching a level of fragmentation that impedes forest fires from spreading’ 

L897: I think it would be nice to repeat here what Group 3 is, the way you did in L847 and L871 with Group 1 and 2

L918: ‘alternative metrics’ - can you give an example?

L924-926: Why was land use change not included? 

L943: Would suggest to rewrite to ‘Understanding the factors that drive fires [...]’ or similar

In this manuscript, the authors describe a clever adjustment of a species distribution model to make it work for burned area. They apply it to the different biomes in Brazil, generating parameter set distributions that can be used to quantify uncertainty in burned area drivers, for both natural and non-natural fire as well as their combination. The authors then go through the biomes discussing the various ways they are similar or different in terms of burned area drivers.

I think this is quite a cool method and application, but the paper itself could use some work in terms of organization, presentation, and clarification. I also have some substantive critiques of the work itself—nothing disqualifying, but still things that should at least be discussed. For that reason I’m recommending major revisions and would be happy to review the next version.

See supplementary PDF for full comments.