Round 1

Reviewer 1 Report

I am not sure that values-at-risk is adequately measured by calls for evacuations. Evacuations are not necessarily determined by incident command teams (other public officials or public safety agencies can make this call), and it is not clear how to distinguish values-at-risk from fire growth potential when they overlap. In a large fire, fire growth may reach several different values at risk, but are their qualitative differences in these values that influence the amount, kind, and placement of different suppression resources? It was a creative approach to use evacuations as a surrogate for values-at-risk, but I'm not convinced that it is a valid or reliable measure for determining the influence on suppression resource orders.

The use of predicted weather data was an a great data source, and the results well supported the hypothesis. This study was not a case of "proving the obvious," because finding that incident command teams are more forward-looking and proactive with the use of weather predictions than backwards-looking with the use of observed fire behavior is a genuine research contribution that may help usher in improvements in suppression operational planning and decision-making.

Author Response

Reviewer 1

I am not sure that values-at-risk is adequately measured by calls for evacuations. Evacuations are not necessarily determined by incident command teams (other public officials or public safety agencies can make this call), and it is not clear how to distinguish values-at-risk from fire growth potential when they overlap. In a large fire, fire growth may reach several different values at risk, but are their qualitative differences in these values that influence the amount, kind, and placement of different suppression resources? It was a creative approach to use evacuations as a surrogate for values-at-risk, but I'm not convinced that it is a valid or reliable measure for determining the influence on suppression resource orders.

The reviewer raises several important issues regarding the measure of values-at-risk (VAR) that warrant additional explanation.  First, the reviewer questions the validity or reliability of evacuations as a surrogate for VAR.  Second, the reviewer suggests that growth potential may conflate the effect of evacuation when they overlap.

We agree that evacuations are an incomplete measure of VAR and are influenced by many parties involved in the wildfire response operation.  We use evacuation status to capture a state of the response operation, which we hypothesized could influence resource ordering patterns.  Regardless of who declares the evacuation, an evacuation indicates that people and property are likely at risk, and should impact response strategies.  Even if evacuations are an incomplete measure of VAR, the statistical consequence is insignificance. Perhaps, this incomplete measure of VAR is why the coefficient estimates for evacuation and interactions with fire growth are not statistically distinguishable from zero.  We added the following sentence to the discussion of these results in section 4, “In addition, our risk metric may be an incomplete measure of the full set of values at risk that may influence resource ordering patterns.”

We now address the second concern.  The regression model estimates correlation coefficients conditional on the other covariates in the model, which should distinguish between the effects of VAR and growth potential unless the two variables are highly correlated.  We plot the distribution of growth potential and evacuation potential in the following figure.  While the figure suggests that managers perceive an increasing potential for evacuation as the growth potential increases, there are still a relatively large number of instances where growth potential is high and there is no perceived evacuation potential.  One could imagine situations where growth potential is high because of weather (as demonstrated in Figure 3 of the manuscript), but the fire is in an uninhabited area (i.e., National Forest). 

Our regression results in Table 4 also suggest that growth potential and potential evacuation are sufficiently uncorrelated so as not to render the model unreliable.  One consequence of highly correlated regressors (multicollinearity) in finite samples is that the regression coefficient is sensitive to the inclusion of the correlated regressor in the model.  However, we find that both coefficient estimates on growth potential (high and low) are stable across columns (1) and (2) in Table 4.  Potential evacuation is only included in column (2).  These results suggest that the correlation between growth potential and evacuation potential is not so large that the model is incapable of distinguishing between their effects on resource ordering patterns. We have added the following sentence to the end of the 3 paragraph in section 3.2, “Moreover, the stability of the growth potential coefficient estimates in columns (1) and (2) suggest that while potential evacuation and growth potential may be correlated, it is not severe enough to substantially affect the growth potential coefficient estimates.”

The use of predicted weather data was an a great data source, and the results well supported the hypothesis. This study was not a case of "proving the obvious," because finding that incident command teams are more forward-looking and proactive with the use of weather predictions than backwards-looking with the use of observed fire behavior is a genuine research contribution that may help usher in improvements in suppression operational planning and decision-making.

Thank you.

Reviewer 2 Report

The present article deals with an interesting and current subject. The work falls within the objectives of the journal (Forests) since it studies how weather is a primary driver of resource orders over the course of extended attack efforts on large fires. How weather-driven fire behavior and risks Influence IMT is innovative. The subject also has a great importance at an international level, responding to a phenomenon that affects many countries and is on the agenda.

Although pertinent, there are some aspects to be considerer:

The word risk appears in the title and in the keywords, but the article is not about risk, it is about how weather is a primary driver of resource orders over the course of extended attack efforts on large fires. There is not a background about risk neither a discussion about it.

Fire growth is a very important issue in the paper, but it is missing in the title and in the keywords.

Some gaps are identified in the definition of the concept IMTs. It corresponds to Incident Management Teams. The first time cited, it should be mentioned in brackets and in the title or as a Keyword. And it should be defined.

I would suggest a more in-depth background about IMT concept and fire growth concept and model. There is no bibliographic reference on fire growth models used.

It would be fine to refer to the latest episodes of fires (Chile, Portugal, California, Australia) that justify such studies. Also a contextualization of current fires which exceed the capabilities of extinction.

Maps are completely missing in this paper, which would be very interesting to have an idea of the magnitude of the fires we are dealing with.

In the methodology section, the presence of a scheme or graphic would be appreciated to show how the process was done.

“We limit the data to 2013 because changes to the ICS-209 implemented in 2014 eliminate one of our key variables”. This sentence is very relevant since the temporal choice of fires is a key factor. At the beginning of the Introduction say “Extreme, erratic, unpredictable are all words used to describe the behavior of some of the most devastating fires over the last decade”. Your period of study covers from 2007 to 2013. Therefore, the first sentence should be better clarified because it involves removing a variable. Which of the four variables, fire size, growth potential, terrain accessibility, and evacuation status was eliminated?  How does the model vary by removing this variable?

How did you managed geographical data in order to introduce to the models. Did you used GIS?

Page 5 “(see supplement for details)”. Where are these supplements?

“We are not concerned about the difficulty of interpreting these coefficients as the variables included in the principle components have been previously investigated (e.g., [7,39]) and are not the focus of this study”. Should be better clarify.

Author Response

Reviewer 2

The present article deals with an interesting and current subject. The work falls within the objectives of the journal (Forests) since it studies how weather is a primary driver of resource orders over the course of extended attack efforts on large fires. How weather-driven fire behavior and risks Influence IMT is innovative. The subject also has a great importance at an international level, responding to a phenomenon that affects many countries and is on the agenda.

Although pertinent, there are some aspects to be considerer: 

The word risk appears in the title and in the keywords, but the article is not about risk, it is about how weather is a primary driver of resource orders over the course of extended attack efforts on large fires. There is not a background about risk neither a discussion about it. 

Respectfully, we would argue that the article is very much about risk. As stated clearly in the introduction, “The objective of this study is to understand how incident management teams (IMTs) respond to weather- and value-driven fire risk.” We built our model around perceptions of risk based on fire growth and values-at-risk, and throughout are consistent in discussing these as risk factors of concern.

We agree we could have provided more background and discussion on risk, and have done so.

See abstract: “Our secondary objective is to test how an additional risk factor, evacuation, as well as a constructed risk metric combining fire growth and evacuation, influences resource ordering.”

See introduction: “Some of this work has a direct and logical connection to resource ordering, although none has examined the how risk perceptions and preferences influence dynamics of resource ordering directly. For instance, in studies of strategic decision making, [13] find that managers are more sensitive to risk to homes and watersheds than to cost and personnel exposure, [14] find that managerial risk preferences are inconsistent with minimizing expected economic loss, and [15] find that managers exhibit risk aversion and nonlinear probability weighting. In all cases, however, choice experiments compared strategies only coarsely using variables such as expenditures and personnel hours, and did not address resource ordering.

Fire growth is a very important issue in the paper, but it is missing in the title and in the keywords.

The reviewer makes another good point.  We have added “Fire Growth” and “Incident Management Teams” to the keywords.  We clarify the connection between fire growth and risk in addressing the previous comment.  We have chosen to keep the title as is because we are trying to show how weather feeds through expectations about and realizations of fire growth, and how it then effects resource ordering.  

Some gaps are identified in the definition of the concept IMTs. It corresponds to Incident Management Teams. The first time cited, it should be mentioned in brackets and in the title or as a Keyword. And it should be defined.

I would suggest a more in-depth background about IMT concept and fire growth concept and model. There is no bibliographic reference on fire growth models used.

The reviewer correctly points out that we assumed a level of familiarity with US incident management structure than is appropriate for the audience.  We have added the following description to the first paragraph of section 2.1: “IMTs are the groups of fire managers who manage the suppression response for each fire.  During smaller or less complex fires, the IMT may consist of a single person while for larger fires IMTs may consist of define. Resources are ordered by the leader of the IMT (the incident commander), but on large fires may also be driven by the operations section of the team, thus, we refer to the decision makers as “IMTs” in this paper.”

We do not use a fire growth model in the paper; rather, we calculate fire growth as the difference in fire size between each ICS-209 report. We have modified paragraph 3 of section 2.2 to clarify this point by adding, “Fire size is the area burned at the time and date of the ICS-209 report.  We calculate fire growth by subtracting fire size from the previous report (t-1) from the current report (t).  Consequently, fire growth is an empirical measure rather than an outcome derived from a fire spread model.”

It would be fine to refer to the latest episodes of fires (Chile, Portugal, California, Australia) that justify such studies. Also a contextualization of current fires which exceed the capabilities of extinction.

The reviewer makes a nice suggestion that expands the relevance of the paper.  We have added the following sentence to the first paragraph of the introduction: “Extreme wildfire events in Chile, Portugal, Australia, Canada, and the USA, among other locations, have exceeded control capacity and resulted in high costs and losses, and provide context and justification for research into weather, risk, and suppression decisions.

Maps are completely missing in this paper, which would be very interesting to have an idea of the magnitude of the fires we are dealing with.

We have created a map that displays the ignition point and complexity of the incident.  The interactive version of the map located in Supplement.html (also here: https://jbayham.github.io/maps/fire_weather_map.html) displays additional information when the user clicks on a point.

In the methodology section, the presence of a scheme or graphic would be appreciated to show how the process was done.

We have added Figure 1 to section 2.2 to clarify the dataset construction process. 

“We limit the data to 2013 because changes to the ICS-209 implemented in 2014 eliminate one of our key variables”. This sentence is very relevant since the temporal choice of fires is a key factor. At the beginning of the Introduction say “Extreme, erratic, unpredictable are all words used to describe the behavior of some of the most devastating fires over the last decade”. Your period of study covers from 2007 to 2013. Therefore, the first sentence should be better clarified because it involves removing a variable. Which of the four variables, fire size, growth potential, terrain accessibility, and evacuation status was eliminated?  How does the model vary by removing this variable?

Perhaps the sentence regarding the change to the ICS-209 form is unclear.  In 2014, the question about growth potential was removed from the ICS-209.  We cannot use data from 2014 to present because one of our key variables is omitted.  We have revised the sentence to clarify this point, “We limit the data to 2013 because changes to the ICS-209 implemented in 2014 removed the question about growth potential.

We updated the first sentence to say “in recent decades” to clarify time scope of the analysis.

How did you managed geographical data in order to introduce to the models. Did you used GIS?

Yes, we used GIS to merge spatially explicit weather (gridMET and Jolly), vegetation, and topographic (Landfire) data with fire ignition point.  We have clarified this throughout section 2.2 in the manuscript. 

Page 5 “(see supplement for details)”. Where are these supplements?

The supplement was uploaded with the original submission and should be available to reviewers.  The file is called Supplement.html.  We will confirm with the editor that the supplement is available for review.

“We are not concerned about the difficulty of interpreting these coefficients as the variables included in the principle components have been previously investigated (e.g., [7,39]) and are not the focus of this study”. Should be better clarify.

We have clarified this statement by integrating it with a sentence that appears earlier in the same paragraph. The sentence now reads, “Our goal is to control for these factors in our regression rather than study their effects on fire growth, which have been documented in previous research (e.g., [8,41]).”  We have removed the previous sentence that appeared at the end of the paragraph.

Reviewer 3 Report

Bayham et al. present an interesting study integrating diverse wildfire-relevant datasets and regression models to estimate how weather conditions influence resource ordering patterns during wildfires. Based on their investigation of 1,125 wildfires in the Western US they conclude that fire managers are more proactive in making resources requests for fire suppression. They found that incident commanders rely on forecast weather conditions to infer expected fire behavior and order fire suppression resources in accordance with the expected fire behavior.

This is an interesting study of direct relevance to fire suppression efforts and should pass for publication without difficulty.

Minor Comments

This sentence seems to be missing something: “Our secondary objective is to test”.

Consider reframing it to better communicate your message to readers.

In abstract and introduction, indicate full meaning of abbreviations such as IC, MIT, etc. the first time they are mentioned. Again, in discussion consider full meaning of abbreviations like ERC etc. the first time they are mentioned. Many readers may have forgotten them by the time they arrive at the discussion. In Tables 3, 4 and 5 authors indicate that their final fire dataset contains 1,125 fires but in abstract they mention over 1,200 fires. Why different these different values? Figure 2 looks pale and a little hard to see. Can it be improved? Also indicate, by the figures, which of them is figure a or b

Author Response

Reviewer 3

This sentence seems to be missing something: “Our secondary objective is to test”.

Consider reframing it to better communicate your message to readers.

This was a mistake.  The sentence now reads, “Our secondary objective is to test how an additional risk factor, evacuation, as well as a constructed risk metric combining fire growth and evacuation, influences resource ordering.”

In abstract and introduction, indicate full meaning of abbreviations such as IC, MIT, etc. the first time they are mentioned. Again, in discussion consider full meaning of abbreviations like ERC etc. the first time they are mentioned. Many readers may have forgotten them by the time they arrive at the discussion.

We have defined the abbreviation the first time they are used in each section of the paper.

In Tables 3, 4 and 5 authors indicate that their final fire dataset contains 1,125 fires but in abstract they mention over 1,200 fires. Why different these different values?

This was a typo.  The figure in the tables are correct and we have corrected the figure in the abstract.

Figure 2 looks pale and a little hard to see. Can it be improved? Also indicate, by the figures, which of them is figure a or b

We have recreated the figures and fixed the labels indicating which is a) and b).  The lines in the figures are thicker and should be clearer to readers now.

Round 2

Reviewer 2 Report

The authors presented a new improved version and introduced the suggestions proposed.

Figure 2, if it is a map, should have the graphical scale.

Author Response

We have added the graphical scale to the map in figure 2.