Wildfires are increasingly impacting ecosystems, particularly in a changing environment. The FPA FOD version 6 data published in 2022 provide tons of fire observations and records. This study developed an extra attribute dataset for the FPA FOD data, which augments the FPA FOD v6 with nearly 270 attributes that coincide with the date and location of each wildfire ignition in the United States. For each wildfire record, the physical, biological, social, and administrative attributes were collected and added. Major classes of these attributes encompass climate, weather and fire danger, topography, land cover and vegetation, jurisdiction and management, infrastructure, and social context. This publicly available dataset can help to answer numerous questions about human- and lightning-caused wildfires and analyze fire impacts on ecosystems.

This work is important. I suggest accepting this manuscript.

Best of luck.

The manuscript describes the database variables and methodology used and then carries out some cursory analyses that describe patterns, trends and differences between naturally and human-ignited fires. I think FPA FOD-Attributes is a formidable effort and useful resource and offer just a few comments for the authors to consider.

L44-45. I see the point, but written as is it reads as if these changes are permanent or constant. Please rephrase.

L48-49. Shouldn’t this be a bit more elaborated? I’m thinking about those with less prior knowledge.

L65. It’s forest floor moisture, or duff moisture, rather than soil moisture.

L140. Why fuel model G, it’s not particularly representative across USA (and in the current NFDRS is included in fuel model Y). And why not based on the specific existing fuel model? (an attribute that I suppose LANDFIRE includes but is absent from the database, why?).

L148. And why not 10-hour and, especially, 1-hour fuel moisture contents? These size classes drive fire spread. VPD has an influence on fine fuel moisture but it’s not quite the same, as it is mostly responsive to air temperature, while 1-hour moisture is much more responsive to relative humidity.

L150. Why this particular window? It would make more sense to have them for the duration of the fire, or at least for the duration of its active spread.

L441. Florida is not mountainous.

L552. And burn severity?

Overall comments

Uniqueness: this is a substantial update to prior FPA-FOD releases, but nonetheless is still v6 of an updated dataset, so the uniqueness is at best moderate.

Usefulness: given the amount of effort a typical user would have to go to in order to access and query the 270 attributes, this is a very useful dataset that assists in more harmonized analysis on the very important topic of fire occurrence.

Completeness: the attributes themselves seem to be very complete.  I would note however about the completeness (or uniformity) or fire data across some state lines, i.e. Figure 4 shows a strong contrast in natural ignition density at the borders of New York state compared to Pennsylvania and New Hampshire.  Is there some bias in tracking methods or thresholds that the user should be warned of?  I did not recall reading any such warning in the text, or to a reference that gives more detail to these sorts of administrative contrasts in adjacent jurisdictions.

Specific comments

Line 649: please revise the reference to the Strategy to match the (likely) permalink on Frames.gov: https://www.frames.gov/catalog/14351 “Wildland Fire Executive Council. 2013. The National Cohesive Wildland Fire Management Strategy: Phase III Western Regional Action Plan. 99 p.”

Table 1: worth differentiating MODIS NDVI (Didan, 2021) vs the Vermote, 2019 AVHRR-sourced NDVI in the “variable category” column or elsewhere on the table. Otherwise, it looks like a duplicate unless one looks at the references.

Line 161: percentiles should be reported at k^th , not k%.

Overall, I feel that the above points constitute only a minor revision, and I'd be happy with the editor confirming their completion, no need to send it back for reconsideration of the peer reviewers.

General Comments to the Authors

The paper is well written stylistically. I think the intent was to describe this impressive dataset, but I am unsure of the intent of the results as I don’t know that there is much that is truly novel in the results. I would have liked more information regarding why you selected the variables that are in the dataset. Also, a discussion of autocorrelation would be good as I think there are likely a number of variables included that are highly correlated. This can lead to overfitting in machine learning models. In addition, why did you leave out some important variables to fire occurrence prediction? The ignition of fires by power lines is mentioned regarding the Camp fire, but that attribute is not in the database. Nor is railways lines, which is also a significant contributor to ignitions. I only mention this because there are so many variables included that I have not seen in the fire occurrence prediction literature and not some of the more common ones.

Specific comments to the Authors

-Has there been any attempt to quantify the error in the ignition data?

-How useful and available are all these variables in real time operationally i. e. What is the operational feasibility of using this dataset?

-How does evacuation time influence the probability of ignition of human or lightning caused fires? I can see the linkage with important decisions that are made after the fire has started, but I don’t see a direct link to prediction of human or lightning caused fires.

-Section 3.1. Manual comparison; why did you only use 100 points? There are 2.3 million fire records, so that is 0.004%.

-Section 3. 2 Seven fires out of 2.3 million also seems a little low, in my opinion. Especially if they are based on the same weather variables. I would expect weather derived indices using the same weather source to be the same. I’m not clear on how this is a validation, maybe I am missing something?

-Line 383 why did you use those attributes at the ignition point from Oct to Dec? The active fire front and areas of spread for a large fire will be far beyond the point of ignition months later. How large was the final fire perimeter?

Figure 2-what are the acronyms, please spell them out.

-Why did you use the media, how reliable is this as a validation source? Are there not fire perimeters and fire records for all suppressed fires from the fire management agencies?

-Were all validation fires recent fires or did you also include older fires? The dataset goes back to 1992, but most validation fires appear to be from more recent years.

Results; why are there references to your findings? Should that not be discussion?

Lines 428-437. Since you are also discussing results. The trends in human and lightning caused fires; is this the same as other studies? In Canada we see the opposite.

Line 428-Where are decadal trends shown? Fig 3 shows annual trends.

Line 433 if human caused fires are increasing how is this related to fire prevention strategies? We can't prevent lightning fires, but we can use fire restrictions etc. to decrease human caused fires.

Line 439- CONUS is used many times before this; define sooner

Line 443-What other subcategories of natural fires are there?

Fig 4 - Nice graphic, I like the addition of histograms

Paragraph 468-475 and Figure 6. I'm still not sure of the take-away here. More people, more small fires except in California? We know already that lightning caused fires occur everywhere and often where people are not.

Line 517 - How easily accessible are these datasets for fire management, again going back to operational availability or is the intent for research only?

Line 549 - 550 Here it is stated that the dataset can’t be used to assess large fire growth, but Section 3.3 is about temporal evolution of fire attributes. I found this really confusing.