Megafires in a Warming World: What Wildfire Risk Factors Led to California’s Largest Recorded Wildfire

Round 1

Reviewer 1 Report

Thank you for this contribution. The authors describe how they parameterized the Prometheus fire spread model to approximate four days of growth on the 2020 Doe Fire in Northern California, which subsequently became the August Complex Fire. They used principal components analysis to characterize the association between hourly weather and modeled rate of spread, which provides non-fire modelers with a high-level view of what drives fire behavior in the model, under the specified conditions for the Doe Fire. They also developed several alternative weather scenarios to illustrate how sensitive burned area and fire radiative power (FRP) are to variation in starting fuel moisture, humidity/temperature/vapor pressure deficit, and what they call “ignition timing”. I find that much of the content is interesting and worth presenting, but that the framing and presentation of the results need improvement so that the content can be fully appreciated by the average reader. I’ll start with my big picture points and follow with specific comments with line numbers.

The biggest limitation I see with your study is that you are really looking at the association between modeled fire behavior and fire behavior inputs. I am content that you calibrated the model to daily area burned, but Prometheus may be approximating reality for the wrong reasons, which warrants presenting your results with less gusto. I suggest that you sprinkle in the word “modeled” in every other or third sentence of the results to emphasize to the reader that you are really testing the sensitivity of a model to its inputs, and under a narrow range of conditions (fuels, topography, and weather) that may not be representative of the broader landscape or how the model functions in very different settings.

Why the emphasis on FRP? I am supportive of increasing its use, and helping our community understand what it means, but I would currently describe it as poorly understood by the average fire scientist. This could be beyond the scope of reasonable revisions for the special issue, but I can’t help but feel that you missed a major opportunity to use the MODIS and VIIRS fire detections for calibration and validation. First, this would help to show that Prometheus can accurately estimate FRP. Second, it would help people contextualize FRP quantities – e.g., help them understand what 100 or 1,000 MW per ha looks like.

As your results are highly dependent on the Prometheus Fire Modeling System, you need to provide more description of the model. At a minimum, you should describe what spread processes it represents (and excludes), and what the assumptions are of Prometheus and the underlying Canadian Forest Fire Behavior Prediction System.

Prometheus documentation:

http://www.firegrowthmodel.ca/prometheus/downloads/Prometheus_Information_Report_NOR-X-417_2010.pdf

Canadian Forest Fire Behavior Prediction (FBP) System documentation: https://www.for.gov.bc.ca/hfd/pubs/Docs/Frh/Frh012.pdf

https://cfs.nrcan.gc.ca/pubwarehouse/pdfs/31414.pdf

An important assumption of FBP, like the Rothermel (1972) model, is that “the fire is wind or wind/slope driven, and spread is not affected by a convection column”. Given that you are using Prometheus to model what you describe as “extreme behavior” with “explosive growth…not expected by firefighters”, I want to know if these conditions hold for the Doe Fire. In other words, should we even expect Prometheus parameterized with the observed environmental conditions to replicate the observed behavior? If not, what does that mean for your results? This is a question I think you should address in the discussion.

Some data points you may want to consider:

The ICS-209 report for the Doe Fire on 8/16 says “significant fire behavior, making own wind” in the “SIGNIF_EVENTS_SUMMARY” field.

Early photos of the Doe Fire are documented in the August Complex Inciweb page - https://inciweb.nwcg.gov/incident/photographs/6983/204/

Several media photos appear to document a well-developed plume in a quick Google image search of “Doe Fire” during some portion of the active burning periods.

I’m not a huge fan of the principal components analysis (PCA), but I can’t think of a better analysis to fit your retrospective view. Again, this should be framed as more of a test of the Prometheus model sensitivity under a narrow set of conditions than an analysis of fire behavior sensitivity to environmental variation.

I found many questionable mixing of fire behavior and effects concepts throughout the paper. Most are not appropriate and could be roughly [in]validated using the burn severity map. I suggest dropping this language to avoid confusion and to avoid reaching beyond what you can reasonably model with confidence.

Structurally, you should think about whether you want to have a “results” section or a “results and discussion” section. You currently trend toward the latter, but your “conclusions” are very discussion-y. I would personally pull all the discussion content from the results and combine it with your current conclusions section as a “discussion”.

Specific comments:

L18-19: The lack of “modeled” anywhere in this sentence gave me the impression that you are referring to the observed behavior, which is not what you present in the paper.

L19-20: Is “are” more appropriate here than “were” because you are referencing a model that is presumably unchanged since you used it?

L19-22: I would drop ignition timing here since it is really just an indicator for a different set of environmental conditions during a different time period. Instead of saying “for example”, you could lead off with something like “despite the impressive early growth of the fire, shifting the ignition date to…was predicted to…”.

L22-25: I would drop the managed wildfire innuendo here since it is not central to your paper. Ideally, an empirical analysis to define suitable managed wildfire conditions would include many fires and study both their behavior and effects.

L30: Don’t use consume in this sense given its more specific meaning in wildland fire.

Figure 1: You should define meters above sea level in the caption.

L34-35: “Variety of conditions” is vague. I think you mean to say something like “multiple fire behavior drivers aligned to make this fire siege unprecedented”.

L42: What about “causing” instead of “to provide enough energy for”?

L50-53: I think most readers understand your use of “resources” here, but I suggest that personnel may be easier to communicate the scale of effort given the mix of resources on the fire. I’m guessing that your intent of including this is to suggest that suppression effects were minimal. I think you could more clearly communicate this by mentioning earlier that you will look at the early growth period of one of the component fires, that had minimal suppression because of resource limitations.

L54-63: I don’t think that much of this paragraph contributes to your core narrative, especially since you note the lack of homes damaged by the wildfire. The Carbon emissions feedback sounds like you are grasping to justify the importance of the fire. It is enough to mention that the size was enormous. Most readers will understand that this could lead to undesirable impacts to natural resources and human assets on most landscapes.

L55: I know this number sounds more significant than hectares, but you should keep the units consistent. You can put 1,000,000 acres in parentheses.

L60: Remove extra period.

L62: I would pick something different than “constructive”. Benign? Controllable? Manageable?

L65-66: Maybe this is just poorly worded, but it does not align with my understanding of “fire regime” or the definition used by the source you cite as “the spatial and temporal patterns and ecosystem impacts of fire on the landscape”. If you Google “fire regime triangle” you will see a figure that suggests it occurs at coarser spatial and temporal scale than the fire behavior triangle, with the dominant controls being vegetation, climate, and ignitions.

L66-68: Like the last comment, I think you are making this too complicated by suggesting that there is a simple link between surface/crown fire behavior and severity.

L68: I think you are supposed to use an author name with this reference when starting a sentence. Check MDPI guidelines.

L68-73: Your wording makes this section unnecessarily complicated. Something like “Research at scales ranging from individual fires to regional fire activity over multiple seasons has demonstrated that…”

Also, I suggest including this relevant reference, which documents that weather variables are very important controls on high severity fire in Northern California:

Parks SA, Holsinger LM, Panunto MH, Jolly WM, Dobrowski SZ, Dillon GK (2018) High-severity fire: evaluating its key drivers and mapping its probability across western US forests. Environmental Research Letters 13, 044037.

L78: Drop “scenarios”.

L101: This wording implies that it (alone) transformed in the August Complex. This is where it may make more sense to mention that the rapid growth of this fire early in the incident meant that it was minimally influenced by suppression actions.

L102: “the” August Complex.

L106-107: Like my general comments, I’m curious why the growth was unexpected? Your uncalibrated run of Prometheus suggested that it would get even bigger than it did. I wish that you had used FARSITE for this, so a clearer comparison could be made to what they modeled on the incident. I’m guessing that a raw run of FARSITE or WFDSS Near Term Fire Behavior would also overpredict growth. Unexpected to who and why?

Table 1 and others: Reformat to MDPI standards.

L115-127: Like my general comments, I think you need to beef up this section. You place too much emphasis on the “GIS part” of Prometheus without any discussion of what spread process it models (e.g., surface fire, crown fire, spotting, etc.) and its limitations (meant for wind/slope driven fires).

L146-155: Again, I wish you had used FARSITE to avoid this step. Your crosswalk technique seems reasonable, but I think it would have been easier for you to approximate the weather stream you need for FARSITE then to do this extra work.

L165-166: What about “are not met” instead of “do not exceed determined thresholds”?

L168-173: Probably more of the suppression effects facilitated by lighter fuels and flatter topography. Much of this flank was contained with Dozer Line.

https://data-nifc.opendata.arcgis.com/datasets/operational-data-archive-2020/about

L184-187: I agree that the final matches, but you don’t really describe any daily metrics. I don’t think this is necessary if you rephrase as “reasonable approximation of the fire footprint at the end of the 8/22 burning period”. Otherwise, I would present the information in daily area burned – observed vs. simulated. [Edit: you do present this in figure 5, maybe you can foreshadow this?]

L189-195: So, all the behavior metrics are averaged by time period over the entire fire? It is often clarifying to mention the sample size – 96?

L199-204: Personally, I don’t get much value from the climatology scenario. Is there a common application this speaks to? FSPro uses climatology, but in a Monte Carlo simulation framework to express uncertainty in potential fire spread. Basically, let me (the reader) know why the climatology scenario is relevant.

L206-216: You should provide more reasoning about why you are adding the observed historical temperature trend to the “control” weather stream and what you intend it to represent. Is this meant to project to future warming in XXXX year and reverse to what may have been normal in XXXX?

L217-221: Why even include this? Perhaps I don’t understand. My guess is that you see little variation between the >= 90^th percentiles, especially if you didn’t constrain your calculations to the fire season.

L234: You include the text reference instead of a number. Also, it is missing from the references.

L240: “Night and early morning”?

L243: What is “non-wind driven”? Slope driven? Plume driven?

L249-252: I’m not sure I want to see the Canadian fuel models, but I find it somewhat odd that you refer to the Scott and Burgan (2005) fuel models in the text and figure. Awkward, but OK?

L257-261: I wouldn’t use this information to suggest beneficial (or less harmful) effects. What you modeled was lower total fuel consumption in forests compared to shrubland. Is this due to their being a lower surface fuel load in these fuel models compared to the Canadian equivalent of SH7? The BAER map seems to contradict this idea, with most of the high severity in the forests.

https://inciweb.nwcg.gov/incident/map/7228/5/109797

Again, I think you should avoid the generalizations you are making between behavior and effects. Effects are complicated.

L262-263: Remove filler sentence.

L279-280: This sentence does not compute.

L282: FMC?

L284-288: Not results material.

Table 2: U and V will not be obvious to some readers. You should define new abbreviations in the caption.

Table 3 and associated text: I don’t understand why you are presenting hourly growth and other metrics normalized by this. These stats don’t seem to make sense, especially ROS. Your map (Figure 3) shows maximum ROS of 42 m/min (which sounds reasonable), but reversing your normalization, produces ROS in the range of 88 to 1,050 m/min. Is my logic wrong? This seems like an attempt to control for the effect of geometry (more perimeter = more growth despite no change in ROS), but I think it adds confusion. I would give the raw stats without normalizing.

L291-292: This is what you should have said in the methods. Remove here.

L309-311: Discussion content. Would 5% less growth on this incident make it meaningfully easier to control? I doubt it. Your point is reasonable when thinking about long-term, landscape-scale risk where smaller fires translate to lower exposure and loss.

L315-332: I now understand what this scenario is. See my earlier note about clarifying the point of this scenario in the methods. I still don’t like it: 1) I think you are probably picking up seasonal extremes that may not align with the fire season, and 2) manipulating only the starting conditions does not provide a very controlled test. I suggest dropping this from your paper to simplify the narrative.

Figure 5: You may want to mention that you do not have hourly observations of fire size. Some people may think that the control simulation does not capture the daily dynamics, but this is just a lack of temporal resolution. You could emphasize this visually by making the observed line solid black with large black points for each observation.

Figure 6: Make the caption independent of the Figure 5 caption.

L354-355: Management implications for discussion. Relevant to mention what CalFire is already doing in California?

https://www.techwire.net/news/cal-fire-uses-new-modeling-tech-to-outflank-wildfires.html

L356-369: Why is this not described in the methods? All you need to do is explain why you are looking at this range of increases earlier and the presentation here will be simpler.

L363-369: Discussion content.

L383: Why not just put the final for the four days? Is this the average of the later three days? Introducing new stats raises new questions unless carefully presented.

L383-384: I don’t understand the emphasis on fine fuels given their low reliance on the long-term trend. Although logical, the next sentence is also describing a level of detail that you didn’t observe.

L385-387: Is this referring to the simulation period, or later? I’m not catching why this is a major point to emphasize from the simulation output you presented.

L396-398: This seems like a discussion point. What is the FBP not capturing about VPD? Is this because FBP doesn’t represent plume dominated mega fires?

L409-429: I don’t feel that these paragraphs fit with what you did. Again, the emphasis on smoke impacts seems tangential. Similarly, you didn’t do anything to address fuel treatment, except by mentioning the desire that this flavor of work could help identify appropriate conditions for managed wildfire. Keep to your core narrative. Think about how policy makers and managers will read this. Would you cite the core content of this paper in another publication to support these points? I think your strongest emphasis should be on the implications for incident management: 1) suppression was only effective on one flank of this fire, 2) [besides the lightning] these weather conditions are not that rare in Northern California, and 3) the climate trends suggest that growth events will continue to outpace suppression capabilities. Stepping back, I think you are warranted to discuss the implications for how we might: 1) pre-position resources based on weather forecasts, 2) allocate resources across active incidents based on weather forecasts/growth projections, and 3) reduce ignitions during high-risk weather. I encourage you to frame these mitigations based on what is already happening, because my impression is that all three of these are currently happening in California.

L409-429: There are also many value laden words in these paragraphs that do little to communicate your narrative. To many readers, use of these words are red flags for bias, especially when discussing content that reaches beyond the work presented. A simple, clean story will provide more value to the reader and you.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The study simulates a few days of spread of the 2020 August Complex fire event in California using the spatially explicit version of the Canadian fire behaviour prediction system and then run some experiments to examine (mostly) the effects of changes in weather-related variables. While it has good ideas it also suffers from conceptual problems, oversimplification and establishes connections that are not warranted; it also fails to identify and discuss drivers of extreme fire behaviour that are not accounted for by Prometheus (see specific comments).

The assumptions regarding fuel types/models and fuel consumption are particularly worrying. I understand the ease of use of ERA5 reanalysis data, but this option is not without issues. Why use Prometheus when you could have used FARSITE without the nuisance/problem of assuming fuel types that greatly depart (physically/physiognomically) from US fuel models? The Canadian FBPS has limited options for fuels in a Mediterranean-type climate setting where shrub fuels are relevant. I’m certain that the required weather data would not be hard to acquire. Fire spread calibration may account for this but still, the authors are assuming fuel consumptions from completely different fuel complexes and this will impact the calculations of energy release, or is fuel consumption based on Scott & Burgan fuel loads?. If it’s the latter case it brings about another issue: the authors are assuming that all fuels were available, i.e., all fuels were consumed; calculations of heat release with Rothermel’s model would have produced different heat release rates.

Specific comments

L65. Suggested modification. “Over time, the combination of these factors influences constitutes …”.

L152. The CFFBPS does not use “fuel model” concept. Replace by fuel type. Unlike a fuel type in the Canadian system a fuel model is defined by a number of parameters.

L153-155. This needs an expansion for detail and more information as is difficult to understand what exactly was done. How good were the fitted equations? Perhaps add to the appended table some descriptors of fit.

L159. The DMC is not indicative of live fuel moisture content, as it does not have enough “memory” for that, but the Drought Code can be.

L161. Byram, not Byrum.

L1651. FMC indexes, not FMC values.

L183. I searched in Tymstra but did not find the default required burning conditions. I’m surprised with most values, e.g. RH needs to be lower than 25% for a fire to spread? Or are these the required threshold conditions in the presence of active fire suppression? Please check/explain.

L234. Rivera et al. missing from ref. list. Also, it should be a number.

L235. This is not FRP, as FRP does not capture all the heat released by combustion. In what units is TFC? Typically, it would be in kg/m2 and so the expression would represent reaction intensity (kJ/m2/s or kW/m2) but here it seems it’s in kg. Because duration is constant (1 hour) I suggest “FRP” recalculation to be just heat of combustion * TFC and so it can be called heat release, or heat release per unit area if TFC is in units of kg/m2.

L257. “Interestingly, some areas that experienced high ROS did not experience high TFC”. If TFC is constant, as it seems to be, this just reflects the nature of the fuel complex, i.e., the fuels consumed are all the existing fuels, which renders the sentence irrelevant. See the general comments section: how was TFC calculated? Constant values per fuel model, US fuel loads adjusted as per the Canadian BUI or Canadian fuel loads adjusted as per the BUI?

L272. The fire spread predictions are dependent of fuel moisture content surrogates (the Canadian codes), so you cannot state that the outcomes are caused by the VPD.

L272. “Fine fuel moisture content (…) fed the fire enough to combust larger fuels”. While this might be true (at least as observed in a fireplace) it is not something resulting from either the US or the Canadian fire spread models.

L274. “dry FMC” or low FMC?

L281 (and lines 385-387 are also wrong). “The high VPD on day 1 led to drier fine fuel moisture codes through days 1-3”. No (and shown by Table 3), because the FFMC (main driver, as the DMC and DC were saturated and do not change significantly at daily time scales) is a short-response variable with a very short memory, as it was conceived to express the dynamic response of fine fuel moisture content to atmospheric conditions. Consequently the “connection” sentence is wrong. As wind speeds were low for the type of fire growth observed, I think the authors should discuss other factors involved, namely pyroconvection and fire-atmosphere feedbacks. I did a little search and found that pyroCb were formed, indicating atmospheric influence (instability, vertical structure) beyond surface weather: https://lethalheating.blogspot.com/2020/12/2020-review-earth-looked-like-hell-from.html?spref=tw
The results reported in lines 301-314 reinforce my impression: very small response to variation in temperature.

L286. Did not check, but I think these references pertain to fire seasons or, anyway, do longer time scales, not to the behaviour of individual fires. Scale matters a lot when one examines fire activity drivers.

Table 3: Why ROS per unit area? It’s difficult to grasp … Suggest replacing by HG per hour.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The  aithors have used percentile approaches for weather data. The article relies on 43 years of data over which, it is assumed, that climaye change has been occurring, maning that percentiles of the 90-99% will under-estimate the impacts of a warming climate. The study is useful in considering the use of, and validation of Prometheus, however, a more robust approach should have included an extreme value analysis, over 20 year windows, for the 43 years of data to show where the current Doe Fire sits within the annual exceedanace probabilities. I do not suggest a rewrite, just that it should be noted as a limitation of the current methodology. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thank you for addressing my comments. I’m still not a fan of the principal components analysis, but I can live with it.

Per your comment about Farsite not being maintained, that refers only to the standalone Farsite software. Farsite and FlamMap have been combined into a single software package. Farsite is also commonly used through the Wildland Fire Decision Support System – it is just called “near-term behavior” in this platform. Farsite is definitely not going away. The choice of model doesn’t really matter in this case. It would have been nice to stick with a US modeling system to make the case for how unexpected the Doe behavior was, i.e., did the near-term fire behavior run significantly underpredict the growth? I buy that your contact found it unexpected, but my experience is that many of the US fire modeling systems overpredict growth with default settings and fuels from LANDFIRE…so this big blow ups should not be unexpected to someone using fire growth models.

I am glad that another reviewer corrected your use of fire radiative power. I should have recognized that issue.

I have a few minor suggestions to improve the text below. Most of them focus on my earlier point about clarifying that fire behavior was modeled instead of observed.

Specific comments:

L60: I suggest starting with “weather driven” instead of “fire weather driven”. All fires are driven by fire weather.

L74: “RH” is undefined at this point, and you haven’t mentioned the acronym table yet.

L109: Missing “fire” in the model name. Same elsewhere?

L116: Same as last comment. Why has it not been abbreviated yet?

L234: This is an example where adding “model” helps. Try “[t]hese additional scenarios were used to illustrate the effect of warming on modeled fire growth”.

L255-256: This sounds circular. And can’t you get fireline intensity from the model? It seems better to say something like “HR influences suppression difficulty and fire effects”.

L269: Drop “for”.

L295: Lowest?

L307-308: Here is another example where adding “model” helps. Try “[t]hese differences indicate how sensitive fire growth predictions are to initial model conditions.”

L339: Add “could”!

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors improved the manuscript sufficiently for publication

Author Response

Thank you!