Generation and Mapping of Fuel Types for Fire Risk Assessment
Round 1
Reviewer 1 Report
Main comments
The paper aims at developing a methodology to produce a homogeneous fuel map over Europe for fire risk assessment. In the present paper, the methodology is first applied to Spain and Portugal, so that no European map is provided yet. The fuel types are determined from Sentinel 3 (multi seasonal “Synergy” products in 5 observation bands + NDVI), horizontal vegetation continuity (existing product from MODIS), biogeographic region (existing from EEA), as well as fuel biomasses extrapolated from the US classification system (FBFT). The final product is then compared to two existing fuel maps (the global fuel map and the EFFIS fuel map).
The paper is very well written, easy to follow and cites abundant and relevant literature. The methods sound and the discussion is interesting. Most of the innovative part of this work is done on the vegetation map, which is based on the “Synergy” product of sentinel 3 and a classification algorithm (SVM) to create a new vegetation map (from 5 observation bands and corresponding NDVI). The paper could be slightly improved by briefly reviewing the existing applications of this relatively recent product, especially in the context of fuel modelling, and by emphasizing more on the specificity of this product to highlight the novelty of their work.
Regarding the technical work, I have two important concerns with this paper.
- Regarding the classification algorithm, the training is based on 403 training pixels assessed with Google Earth in 14 categories. 85% success is very encouraging and apparently most errors were caused by mixed pixels. However, it would be necessary to check the performance on a “true” validation dataset (not used for model adjustment), which was apparently not the case. Indeed, machine-learning performance is often much lower on independent validation dataset and this would be good to verify by adding something like 50 to 100 new random google points.
- The second issue is more problematic as it is related to a very strong assumption of the present work: there would be an equivalence between sub-humid/humid categories of the FBFT US fuel types and the Atlantic and alpine of the European fuel types, as well as a match between FBFT US arid/semi-arid to Euro-Mediterranean.
This assumption already raises a problem in Spain and Portugal -as noted by the authors in the discussion-, since they cannot distinguish Atlantic and Alpine fuels. I think this methodological issue would be all the more important when the authors will try to extend their approach to the rest of Europe. Ideally, the authors should improve this in the present case, but if they consider it requires too much work to improve this here, they should at least point how they want to handle this in the future for their European map.
I think that it would be interesting to determine or check the correspondence between the 45 categories of the new European fuel maps and the 40 of the Behave system in a more objective manner, for example by comparing not only theoretical ecotype labels, but also their environmental specifications (based on -ideally- evapotranspiration, available water, drought index, etc. and at least rainfalls and temperatures). I realize that this would require additional work (extracting these metrics in US continent and for Spain/Portugal and comparing them). Another approach could be to apply the SVM classification to US territories to look for some links between FBFT and FBFT map.
Beyond these improvements on the rationale of the method – i.e. linking more rigorously fuels types among continents-, I think it would also be nice to propose in perspective the development of a European database with homogeneous field sampling in Europe to help validating and selecting remote sensing products in the future, and to avoid any tricky reference to FBFT.
Detailed comments
- Ln 37: “as well as oxygen and chain reactions to maintain the fire ”. Akward. Mention heat transfer to ignition thresholds rather that chain reactions.
- Ln 39: Fuel by itself is not a category. Fuel types?
- Ln 43: easily and regularly updated
- Ln 90: for sentinel you can cite this recent paper too https://doi.org/10.3390/rs12142251
- Figure 2: The use of the Sentinel-2 mosaic, here and also in the text is a little bit unclear
- Line 280: “For each fuel type, biomass, spread rate, and flame length values (Table 2) were extracted from the original FBFT fuel descriptions.” This ignores variations in fire danger (wind, fuel moisture)! This statement should modulated… By the way, table 2 could typically go in the supplementary.
- Table 4: and corresponding text: same as my previous comment
- Line 395 and next: I would not say “overestimated” and “underestimated” as there is no reason why the global map or the European would be more accurate. Just say higher and lower…
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Author Comments
Thank you for this wonderful submission, it was a pleasure to read. I have provided line-based comments below, and whenever possible I also included direct quotes to make it easier to find where I’m referring to in the text – at no point do the quotation marks indicate sarcasm.
29 – maybe consider rephrasing so that wildfires are not portrayed in a negative light regarding society. Fire under the proper context is not a threat to society and is in fact beneficial.
36 – include cultural burning or prescribed fire in your natural category. Human caused fire can be natural too.
49 – “work scale’s detail” sounds awkward, maybe consider rewording. Also, try to avoid use of possessives at all, there are many instances throughout the paper that could be reworded to avoid using them. Example in Line 58 – reword to say “The vegetation characteristics that are…” instead of how it’s written with the apostrophe.
60 – “crown height” is repeated in the list, and you will need to describe the difference between “density” and “canopy bulk density”, and how “number of trees” is a different characteristic of the fuel than “density”. Some of the items are variants of others and the list could be condensed.
73-85 – this paragraph has grammar errors and typos that need fixed to provide clarity to arguments. Additionally, the details in Lines 73-105 could be condensed into a single paragraph given the audience of the journal.
113-118 seem unnecessary given the next immediate section is the methods. Maybe consider combining with lines 106-112 to make a single, concise paragraph on your objectives. Definitely stating a research question might be a good idea too, although your objectives description is well done so it may not be necessary.
Figure 2.1 – did you have vegetation classes/categories at this point, and if so, how many? You list the number of categories of fuel types below, so it might be good to stay consistent. – Ah, I see in lines 169-171 you have the categories described, still would be a good idea to have on the figure (assuming it’s not too much work to edit in, in which case, it’s not that important)
170-174 could be removed. The audience of the journal will know what the terms conifer, broadleaved, and evergreen mean. The descriptions of the categories are good, but the description of terms can probably be removed to save space if needed.
181-219 – fantastic descriptions here. Clear, concise, and easily replicable. Only one minor comment, maybe make the Table 1 column for the Initials fit tight to the text in that column. Right now there’s a little extra white space that’s pushing the second column far to the right on the page.
242 – might need another sentence to provide further description of your fire spread occurrence possibility classes. You cite the sources, but a quick bit on why the percent classes were split like they were might be a good idea.
245-252 – this part is unclear. It might improve clarity to make the connection between biogeographic region (primarily the climate aspect) and fuel type stronger here.
261-268 – Seems a little out of place here, maybe include in the introduction instead? Or maybe cut it out entirely given the audience of the journal. You also explain spread rate and flame length, but not the biomass load.
280-285 – unclear on how the parameterization occurred or why it was necessary. Is the parameterization not a part of the adaptation/translation described in the paragraph previous? It seems like once the Iberian fuel types were adapted to the FBFT, the description and characteristics of biomass, spread, flame length per type would already be known because those parameters are what define each fuel type. If this is not so or there is more nuance, it might be beneficial to provide clarity here.
286-299 – This seems out of place and it’s unclear what is meant by extracted on line 287. Why compare the biomass values to another dataset but not the other three parameters? If there are reasons for not including descriptions of the other two fuels parameters, then I might consider removing this section or condensing into another.
302-315 – This needs more details and descriptions. – the other methods sections are heavily detailed, but this section feels rushed. Specifically, how were the zonal statistics calculated?
**General Methods comment – some sections have too much detail, others have too little. I might consider only including details on methods that you created or are novel. Anything based on previous methodology can be condensed and then cited. **
318 – use Support Vector Machine first here and then use the acronym. Not everyone will know what it stands for up front.
327-328. – unclear what is meant by “basic vegetation cartography to fuel mapping.”
Table 3 may be more beneficial as a part of the supplemental materials.
Figure 6 – awesome figure. I love how you blended the map with the table and used it to help form the legend. Really cool geoviz!
Figure 7 – another solid map, and the inset is a nice touch. My only comment would be to add in the “adapted” text in the figure title. Maybe something like “FBFT-adapted fuel map for the Iberian…” I say this only because that was the wordage used in the text. Some folks may look at Figure 7 and mistakenly think it’s the standard FBFT map and not the one adapted specifically for this region.
363-373. – it’s unclear at times in this paragraph when you’re referring to the standard FBFT and the FBFT-adapted.
Table 4 – is this table for FBFT-adapted fuels, based on what your maps found? Title may need clarified. Overall, it’s a little hard to determine with the asterisks what values below to the CCI Biomass and from the standard FBFT. It’s a kind of busy table, maybe consider condensing or splitting into multiple tables. Also, maybe part of the supplement materials?
Figure 8 – what do you mean by FBFT-defined? It might help with overall clarity of your datasets, the ones you derived/adapted, and the ones that come standard or defined already, and the ones you compared your results with (e.g. CCI) if you used standard terms for each throughout.
394-405 – the over and underestimations appear quite large in some cases (e.g. tree mean is 542%). Additionally the correlation values are low for grass and shrubs, almost non-existent. It might be beneficial to have a statement or two in the results to lay the groundwork for describing the results in the next section. The results may benefit from support up front so the reader doesn’t have to wait to the discussion to see your interpretations.
Table 5 and Figure 9 seem tacked on at the end of the results. If they are necessary for the paper, I might consider them part of the supplemental materials.
426 – unclear what you mean here – you didn’t use object based image analysis (use the full term here too, not all will know the acronym, similar to the SVM above) because of computational costs, but other authors found that OBIA was better than SVM? Or other authors found OBIA wasn’t worth the time/costs either?
429-437 – but mixed pixels were not chosen in the training and validation though, right? It seems like mixed pixels would be an issue with every image classification.
463-471 – it does not make sense that intercomparison was not possible if the purpose of the new method was to be scalable from local to regional to global scales. Unclear why a comparison could not be made here, especially given the issues with estimations of classes.
498 – not sure you can make the claim this methodology is applicable at multiple scales, especially at the global scale.
** General discussion comments: a few of the sections here need more detail and hashing out. You have a great start to the discussion, but weaknesses in the methodology are not fully explained and reasoning behind some of the results are not fully explained. **
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Dear authors,
I think you have done a thorough and relevant work to improve your already solid manuscript based on my comments. Sorry for missing the training vs validation points in the first version. Everything is now very clear.
I also appreciated your honest and careful responses to several points.
Best regards
F Pimont
Reviewer 2 Report
Ok, I've reviewed all the comments and I'd like to say they did a great job explaining and providing details on things they chose not to update or edit and the ones they did. I have just one minor disagreement regarding the cultural burning comment, but based on their comments in return, I think it's more opinion and interpretation rather than convention so I don't feel like it needs pressed. They also turned the edits around incredibly fast given all the suggestion I made. I don't see any reason to delay acceptance here.
