Our results have several implications for how weather information and forecast models can be communicated more effectively to support tactical decision-making. We consider two implications in greater detail. First, our results highlight the importance of considering how information is used in light of the potentially heuristic decision strategies of fire managers. Tools will be more effective when designed with the decision strategies of fire managers in mind, either by supporting heuristic-based decision-making or by debiasing and encouraging more deliberative decision-making. Second, our results point to possible areas of improvement for weather forecast models that might improve confidence. Wind and precipitation forecasts merit particular attention, either by improving model accuracy directly, improving confidence in existing models, or both.
Supporting heuristic versus deliberative decision-making
Weather information can be an important determinant of tactical decision-making and success in wildfire management (Rapp et al. 2020; Countryman 1972). However, our results highlight that weather information may not be used or interpreted consistently across decision-makers. Rather, what information fire managers use and what they learn from it depends on the context; weather information does not exist in a vacuum. This is consistent with the concept of preference construction, i.e., the phenomenon where decision-makers do not have pre-defined, immutable preferences going into the decision-making process. Instead, decision-makers form their preferences “on the spot” in response to cues that are available throughout the decision process. As a result, preferences are not revealed but rather constructed (Slovic 1995; Gregory et al. 2012). Specifically, our results show that the relative importance of a given piece of weather information may depend on prior decisions. For example, wind was the most important piece of weather information when switching to direct attack, but wind was less important when switching to indirect attack. The tactical decision made previously influenced how weather informed future tactical preferences. Similarly, we saw some situations where the tendency was to stick with the status quo, regardless of what the status quo decision was, which may suggest when the best decision is ambiguous, fire managers lean on previous decisions (Wilson et al. 2011). This is not necessarily a maladaptive or inefficient decision-making strategy if there are non-negligible costs to switching tactics, but the benefits of switching are unclear or uncertain.
While it is not clear from these results alone why the initial attack tactics shaped fire manager preferences, there are several theoretical explanations to consider. In the context of this experiment and in decision-making in the field, fire managers may be interpreting information holistically or comparing it to previous experience rather than integrating and weighing information through a deliberative process (Drews et al. 2015; Klein 2008). In that case, the initial team’s decision is a piece of information in and of itself, as respondents compare the current scenario to previous experience where the initial attack team either directly or indirectly engaged the fire.
Furthermore, individual pieces of information may not be considered separately but rather in light of each other. Indeed, in the context of weather, this is likely an adaptive and appropriate strategy where weather factors can be more than the sum of their parts and reach critical thresholds for extreme fire behavior (Young et al. 2019). Although examining interaction effects or non-linear thresholds was outside the scope of our study, it is worth exploring in the future to understand not only how weather components physically interact to create fire behavior, but how fire managers combine pieces of weather information to infer expected fire behavior and how this may influence their tactical decisions. For example, while more extreme projected fire behavior is related to fire managers ordering more resources, certain weather situations such as extremely high winds may pose unique risks or challenges that factor into tactical decision-making (Bayham et al. 2020).
The decision strategy that a fire manager chooses to use can make decision support and the provision of critical information more or less difficult. For example, decision-makers can use compensatory or non-compensatory decision strategies. Non-compensatory strategies do not deal directly with tradeoffs across attributes of a decision, while compensatory strategies do. A non-compensatory strategy would consider each attribute separately (e.g., if rain is forecasted, directly attack the fire; otherwise, consider the wind forecast), whereas a compensatory strategy would consider each attribute in combination (e.g., consider the precipitation and wind forecast information in light of each other when deciding on a preferred strategy). Non-compensatory strategies may be more common and are challenging to address through utility-maximizing decision support tools (i.e., decision support tools that assume decision-makers are utility-maximizers and, therefore, seek to calculate the maximum utility of each possible alternative with the assumption that the highest utility alternative is the best or most preferred). For example, a fire manager may use the Trade-off Analysis Exercise risk management tool to clarify and consider tradeoffs between risks to firefighters, the public, and identified values for several potential courses of action (Schultz et al. 2021). A compensatory decision-maker is willing to make tradeoffs between acceptable levels of risk across different values while a non-compensatory decision-maker seeks to minimize risk to one value, regardless of how much that may put other values at risk. Because the non-compensatory decision-maker is not seeking to maximize utility, but rather maximize the value of one attribute (e.g., minimize risk to a particular value, only attack directly if it is raining, etc.), utility-maximizing decision support tools may be less useful (Retief et al. 2013; Payne et al. 1993). Indeed, utility-maximizing decision support tools may be the least trusted where they are the most needed, for decisions that include painful or undesirable tradeoffs in which decision-makers have an incentive to ignore or deny the tradeoffs and make non-compensatory decisions (Beattie and Barlas 2001).
Because fire managers must make time-constrained decisions with considerable risk and uncertainty (Thompson et al. 2017b), it may be important to consider fire managers as adaptive decision-makers in the context of tactical decision-making. Adaptive decision-makers must make tradeoffs between accuracy and effort when choosing decision strategies (Payne et al. 1993). In other words, they might choose a more effortful strategy (i.e., a compensatory and tradeoff-focused strategy) to ensure a more accurate decision when the stakes are high but may do the opposite when the stakes are low. As their goals shift over the course of a fire, the same piece of information may be used in more deliberative or heuristic ways as the need for accuracy versus effort shifts. For example, when a fire first ignites and the probability of containment is high, precipitation forecasts may be used heuristically to quickly determine what resources should be sent to respond to an ignition. Later during an extended attack when it is clear the fire will not be easily contained, precipitation may be just one piece of information weighed against a host of other factors (e.g., current wind conditions, resources available, etc.). Furthermore, even in the context of one decision, fire managers may shift between decision-making strategies over time. This may occur when the decision context is uncertain and a different strategy seems more appropriate as information about possible alternatives is uncovered (Mintz et al. 1997; Mintz 2004).
These results have important implications for the design and evaluation of decision support tools for operational personnel. Understanding the impact of decision support tools on fire outcomes is difficult because the information these tools provide is only one consideration among many for fire managers (Canton-Thompson et al. 2008; Rapp et al. 2020). During pre-fire planning, decision support tools can help decision-makers make more informed and defensible decisions as they consider information in a collaborative and deliberative setting (Thompson et al. 2020). Successful decision support prior to a fire igniting may improve tactical decisions and outcomes in two ways. First, it can clarify objectives and goals for an area including what role fire may play on that landscape should one ignite. Second, it can provide insight into the relative ease of containment of a fire based on the climate, topography, and fuels (O’Connor, Thompson, and Rodríguez y Silva 2016; Wei et al. 2018). However, during a wildfire event, tactical decisions made in response to changing conditions may be more time-constrained and decision-makers may have fewer resources to dedicate to systematic decision-making or the type of compensatory decision-making intended to be supported by most existing tools.
While previous researchers have highlighted the types of information necessary for an operations-focused decision support tool (Dunn et al. 2017), results here emphasize that decision support tools should be designed and evaluated with the decision strategies used by fire managers in mind. For example, fire managers may consider some weather information more deliberatively or heuristically based on how it influences fire behavior. Weather conditions have both a direct and indirect impact on wildfires. For example, wind speed directly influences fire behavior by providing additional oxygen to the combustion zone and also by improving convective heat transfer to un-burned fuel ahead of the flaming front; therefore, increases in wind speed directly cause fires to spread faster and with higher intensity (Werth et al. 2011). Thus, all else equal, information on wind forecasts may be easier to analyze deliberatively given its incremental and direct effect on fire behavior. In comparison, other weather variations, such as temperature, humidity, and rainfall, indirectly influence fire behavior by their effects on fuel moisture content. Fire spread is determined by a simple energy balance: heat is used to either raise the temperature of adjacent fuels or it is used to evaporate water within that fuel. Variations in weather can either wet or dry fuels depending on the gradient between the fuel and air in the boundary layer around the fuel and these fuel moisture fluctuations can slow or accelerate fire spread. Thus, humidity and temperature have an indirect and incremental effect on fire behavior and may be neglected as information when making decisions rapidly. In comparison, rainfall has the strongest and most direct impact on these fuel moisture variations because it can quickly saturate fuels as well as leave additional water on the surface of the fuels. The strong influence of precipitation on fire behavior leads to a discrete and relatively concrete reduction in fire behavior, making it a useful indicator for heuristic-based decision-making, while wind, temperature, and humidity variations are more incremental and gradual.
With that in mind, decision support tools can be designed to support compensatory or non-compensatory decision-making depending on how they frame and provide weather information. Importantly, fire managers cannot be neatly demarcated as either compensatory or non-compensatory, but rather, fire managers likely change and adapt their decision-making strategy depending on the importance of the decision and the time constraints they face. Thus, it is may be helpful to provide decision-makers with a variety of tools or sources of information they can choose from based on their capacity to make deliberative versus heuristic decisions. For example, tools for compensatory decision-makers should seek to simplify and summarize information while tools for non-compensatory decision-makers should seek to reduce the arbitrariness of cutoff levels or decision thresholds (e.g., at what change of precipitation do fire managers act as if it will rain, at what ERC do fire managers switch to direct attack) (Cook 1993). In the case of previous decisions having undue influence on future planning, or the effects of anchoring to previous strategies and insufficiently adapting to new weather information, decision support tools should incorporate things like “consider the opposite” (Hirt and Markman 1995). To consider the opposite, decision support tools ask the decision-maker to consider if their decision would change with a different status quo in place, and if so, why, as a means of balancing out any effect of a particular preexisting strategy.
Improving confidence in and use of fire weather forecasts
Our results also provide insight into what conditions fire managers find most appropriate for direct and indirect attacks. Broadly speaking, fire managers preferred to directly attack fires occurring early in the season with mild fire behavior but preferred indirect attack on fires occurring late in the season with extreme fire behavior. For some fires occurring in the early or middle of the fire season where it is not raining, fire managers prefer to continue with the status quo, regardless of what it is. Fire managers were more sensitive to wind when switching to direct attack and more sensitive to precipitation when switching to indirect attack. Although the importance of different pieces of weather information varied in their influence on decision-making depending on the prior decision, our results still point to several practical needs when it comes to improving the weather information available to support decisions.
First, wind and precipitation were the most important pieces of weather information for decision-making yet respondents expressed lower confidence in the reliability of wind and precipitation forecasts. Thus, we suggest prioritizing efforts to improve the forecast accuracy where possible for these variables and increase confidence in the resulting forecast as appropriate. Typical fire weather forecasts are derived from the National Digital Forecast Database (NDFD) which are produced continuously across the USA by the US National Weather Service (Glahn and Ruth 2003). A recent study has shown that the NDFD consistently underpredicts windspeeds when the winds are stronger than about 4 m/s (~9 mi/h) (Page et al. 2018). Winds are particularly difficult to forecast due in part to local terrain influences, and extensive work is ongoing to improve wind forecasts in complex terrain. Models that downscale wind forecasts to correct for terrain influences, such as WindNinja (Wagenbrenner et al. 2016), show promise in improving local-scale wind forecasts.
Quantitative precipitation forecasts provided to wildland fire decision-makers are commonly derived from the NDFD, and they are often modified by forecasters prior to issuance. However, investigators are continually exploring ways to improve precipitation forecast skill and spatial resolution using models such as the High Resolution Rapid Refresh (HRRR) (Benjamin et al. 2016). Continual improvements to the HRRR model physics and data assimilation show promise in improving precipitation forecasts over the next 18 h (Bytheway et al. 2017). This interval generally conforms to an operational period for wildland fire operations. Other improvements to precipitation forecasts, such as ensembling, can provide uncertainty estimates of forecasts that may also be useful for decision-makers. Ultimately, given the importance of precipitation forecasts on decision-making, any efforts to improve skill or characterize uncertainty in precipitation forecasting will likely influence wildfire outcomes.
That said, improving model accuracy may not be sufficient on its own. While a certain threshold of accuracy and quality is necessary for model forecasts to have value to decision-makers, model quality is multi-faceted and not the same as the utility of a model to decision-makers (Murphy 1993). While it would be reasonable to expect some correlation between accuracy and confidence in wind and precipitation forecasts, it is not a given that improvements in forecast accuracy will automatically lead to increased confidence. Thus, distinct efforts may be necessary to improve confidence in the models. These efforts could be informed by better understanding what aspects of the model lead stakeholders to use or ignore the resulting forecast. For example, in some cases, personnel may be resistant to using models due to cultural ideas surrounding technology and models (Rapp et al. 2020; Noble and Paveglio 2020). In those cases, it may be more fruitful or even necessary to focus on changing how users relate using models and being competent at their jobs. In other cases, stakeholders and users may be disconnected from the development process for models, and communicating or demonstrating improvements may be helpful. In other cases still, the problem may not be with the models, but the perceived competence of the modelers (Noble and Paveglio 2020; Rapp et al. 2020). In these instances, investing additional resources and attention towards training modelers and establishing relationships between modelers and end users may contribute to improving confidence in the resulting forecasts.
Seasonality was the most important non-weather-related attribute across conditions, with roughly half of respondents explicitly highlighting it as an important decision criterion in the open-ended response. Across both conditions, respondents preferred direct attack early in the season and indirect attack later. Although direct and indirect attacks can be used on all fires regardless of the over-arching strategy, this preference appears borne out by the data which suggests that a greater proportion of fires are managed for suppression early in the season while the proportion being managed for other reasons increases later in the season (Young et al. 2020). In terms of tactics and strategy, the decision space of fire managers is likely larger later in the season as seasonal changes associated with the onset of autumn are likely to aid containment and reduce the severity of fire behavior. Additionally, fire managers may be able to justify using more resources to manage or indirectly attack a fire later in the season because these resources are less likely to be needed on a later fire during the same fire season. A key follow-up question is therefore how does weather information interpretation change over time? For example, while wind may be an important driver of fire manager decision-making regardless of the time of year, the interpretation of precipitation may depend on the time in season, where precipitation earlier in the season may have less of an impact (or indeed may make fires worse through lightning strikes) but late-season precipitation may signal a season-ending event. Similarly, weather information may vary in importance over the course of a fire event. This work examines a pivotal moment in fire management, when fires transition from initial to extended attack, but other key decision points are worth considering, such as the decision to manage for resource benefit or suppression. Indeed, as more forests utilize pre-identified operation control points, it will be important to understand how weather shapes which control points are selected and what tactics are used. It is worth exploring in greater detail how fire managers personally understand and estimate wind, rainfall, humidity, and other fire behavior drivers and thus how information on these drivers influences the perception of fire behavior over the course of events and seasons (i.e., to what extent do fire managers’ mental models of the effect of fire weather conditions on fire behavior mimic actual fire behavior model predictions?).