This research examines historical burned area (2006-2022) in French Mediterranean, Atlantic Pine and temperature forests.  2022 was an exceptionally large fire year which led to higher than usual burned area to occur in the Atlantic Pine and temperature forests compared to more historically frequent burning in the Mediterranean systems.  Burning in the old-growth Atlantic Pine and temperature forests lead to higher biomass loss than the Mediterranean forests, and by using higher resolution satellite imagery, less burned area was reported compared to EFFIS and MODIS. Additionally, Lidar based biomass estimates are combined with burned area in a novel approach.

Comments:

1.

Line 85. This is a 0.25 degree product I believe. 

2.

Line 134.  I am a little confused how the pre and post-burn periods are defined temporally. Are NDVI, NBR and NBR2 acquired 1 year pre fire and 1 year post-fire or some other method used?

3.

Line 136. What are the parameters in your random forest?  How many trees, depth of the trees etc. How is your random forest validated?  Cross validation of some sort?  What are the evaluation metrics?  Without knowing how well the model is performing it is hard to know if the classifier is any good. I realize you compare burned area to ERFFIS and MODIS, but the actual random forest validation metrics will be useful to include.

4.

Line 143.  In general it would be better if your figures went in order, they jump from 1 to 4 here.

5.

Line 144.  How are you designating the forest/shrubland/pasture/grasslands?  Is this an ancillary product that should be cited?

6.

Line 157.  What type of resampling? 

Line 160.  Which cloud mask?  Citation needed.

8.

Line 285.  Space between 66,393 and ha needed.

9.

Line 398.  Space needed, 2022,before.

The authors have prepared a comprehensive and interesting manuscript about characterizing the biomass loss across a selection of fire seasons in France, through a diverse set of ecozones or forest types. The paper is generally well written, and the methods appear sound. I enjoyed the discussion at the end about how to appropriately characterize fire seasons increasingly viewed as ‘exceptional’ or ‘unprecedented’ by the media and public, with real data and analyses. I have a few minor suggestions for clarity of the work, mostly around the methods. One topic I think requires some clarification in the text is what level of ecological fire severity is detected by the BAMTs polygons? For example, are unburned islands excluded? What about low-severity (no tree mortality but burning underneath)? Since the biomass being affected by fire is all considered dead as a result of the burn, it would be helpful to better understand whether the polygons produced from the BAMT method are including only high severity, or some sort of mix (i.e. surface fires, or ‘underburning’), as this would be a source of error, and lead to overstating biomass loss, if mixed and low severity fire is included within the burned areas, and this would need to be clearly acknowledged. This difference could also account for some of the differences in area burned estimates with other fire mapping products, which are described as overestimations by the other products.

Title: possibly revise “seizing” to another word? Maybe “characterizing” or “contextualizing”, instead?

Line 17: hyphenate fire-prone?

Line 62: replace has with have.

Line 73: AGB-L is defined as the acronym, but for most of the rest of the introduction section AGB is used alone or Loss is spelled out. I also feel that this definition of AGB-L is very important. This is not combustion but rather is combustion combined with mortality of live biomass. This is why I feel that some clarification about whether there is any mixed-mortality wildfire captured in the burned polygons is needed.

Line 134: Please define the range of dates considered for pre-fire/post-fire. Were these ‘initial’ assessments (immediately before and after fire)? Multi-year (extended assessment)? Either is okay but the methods are not replicable without these details.

Line 137: Can the authors add to the text to explain how the BAMTs determines burned/unburned? Is the training data supplied by the user of the tool, and specific to the region, or is automatically supplied by the tool? What area or region is the training data from?

Line 190: How many NFI plots were used for each patch/model parameter p? Was there a minimum number of plots used?  What was the range?

Line 365: Although I recognize that the old-growth forests are likely the highest biomass stands on the landscape, I’m not sure that the differences between the two distributions fully support the statement “affected a higher proportion of old-growth forests than were available to burn”. I don’t think that’s possible, since even if it affected 100% of them that is still the maximum that would be available to burn. A couple suggestions to address this would be to introduce statistical tests that compare distributions to determine whether they are significantly different (e.g., K-S test), which would then support the authors saying something like “burned stands that had a significantly higher biomass than was typical for the region”. Alternatively, or additionally, another option would be to use something like a chi-square test or likelihood ratio tests of the spatial data to compare how much old growth was available on the landscape, and whether they preferentially burned, relative to their availability.

Line 420: Typo “three” instead of tree.

I found the figure caption for Figure 10 a bit hard to follow, specifically what the reference product being compared to was and what comparison was being made.