Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Risk Management and Analytics in Wildfire Response

Risk Management and Analytics in Wildfire Response Purpose of Review The objectives of this paper are to briefly review basic risk management and analytics concepts, describe their nexus in relation to wildfire response, demonstrate real-world application of analytics to support response decisions and orga- nizational learning, and outline an analytics strategy for the future. Recent Findings Analytics can improve decision-making and organizational performance across a variety of areas from sports to business to real-time emergency response. A lack of robust descriptive analytics on wildfire incident response effectiveness is a bottleneck for developing operationally relevant and empirically credible predictive and prescriptive analytics to inform and guide strategic response decisions. Capitalizing on technology such as automated resource tracking and machine learning algorithms can help bridge gaps between monitoring, learning, and data-driven decision-making. Summary By investing in better collection, documentation, archiving, and analysis of operational data on response effectiveness, fire management organizations can promote systematic learning and provide a better evidence base to support response decisions. We describe an analytics management framework that can provide structure to help deploy analytics within organizations, and provide real-world examples of advanced fire analytics applied in the USA. To fully capitalize on the potential of analytics, organizations may need to catalyze cultural shifts that cultivate stronger appreciation for data-driven decision processes, and develop informed skeptics that effectively balance both judgment and analysis in decision-making. . . . Keywords Decision-making Data science Operations research Suppression effectiveness Introduction objectives—is thus an inescapable aspect of wildfire manage- ment [1–3]. This framing of risk makes clear the need to explic- Wildfire presents risks to communities, landscapes, and water- itly address uncertainty, which can manifest in various types and sheds and to those who respond to and attempt to manage it. stem from various sources (e.g., fire weather, suppression re- Risk—broadly defined as the effect of uncertainty on source productivity) [4]. Further, this framing makes clear that This article is part of the Topical Collection on Fire Science Management * Matthew P. Thompson John S. Hogland matthew.p.thompson@usda.gov john.s.hogland@usda.gov Yu Wei Human Dimensions Program, Rocky Mountain Research Station, yu.wei@colostate.edu USDA Forest Service, Fort Collins, CO, USA David E. Calkin david.e.calkin@usda.gov Department of Forest and Rangeland Stewardship, Warner College of Natural Resources, Colorado State University, Fort Collins, CO, Christopher D. O’Connor USA christopher.d.oconnor@usda.gov Human Dimensions Program, Rocky Mountain Research Station, Christopher J. Dunn USDA Forest Service, Missoula, MT, USA chris.dunn@oregonstate.edu Nathaniel M. Anderson Department of Forest Engineering, Resources & Management, nathaniel.m.anderson@usda.gov College of Forestry, Oregon State University, Corvallis, OR, USA Curr Forestry Rep (2019) 5:226–239 227 risk is a value-laden concept predicated on defined objectives long history of application of science to support wildfire and that objectives can be positively or negatively impacted response decisions, with much of the effort focusing on (e.g., fire can enhance or degrade forest health). predicting expected wildfire behavior given existing Risk management (RM)—a set of coordinated activi- fuels, topography, and forecasted weather [e.g., ties to direct andcontrolanorganizationwithregard to [19–21]. Continued advances in spatial risk assessment risk—has become somewhat of an organizing framework methodologies provide additional information on the po- for wildfire management, with applications ranging from tential consequences to highly valued resources and as- programmatic budgeting to fire prevention, fuel reduction, sets, which can help guide fire management prioritiza- community planning, and broader topics such as perfor- tion and decision processes [22, 23]. Further, re- mance, communication, and governance [1, 5–12, 13� ]. searchers are developing tools to improve situational Wildfire management is rich with opportunities to apply awareness and operational responder safety [24–26]. and refine RM acumen—organizations around the globe However, reviews from around the globe on decision implement RM practices as a matter of routine. As one support and fire modeling collectively point to a lack example, the Australasian Fire Authorities Council of systems that provide empirically credible and opera- adopted International Standard 31000 Risk tionally relevant information on the effectiveness of al- management—principles and guidelines [1]as a guidepost ternative suppression strategies and tactics [27–29]. for all firefighting operations [14]. As another, the USDA For the purposes of this review, three interrelated principles Forest Service describes RM as a required core competen- of RM are particularly relevant: that RM is part of decision- cy for fire managers, and promulgates a RM protocol to making; that RM is based on the best available information; guide assessment, analysis, communication, decision- and that RM facilitates learning and continual improvement. making, review, and learning [15]. RM can help these The emphasis on decisions recognizes that performance is organizations increase the likelihood of achieving objec- inexorably tied to decisions. As Blenko et al. [30] note, and tives, establish a reliable basis for decision-making and with clear analogies to wildfire management, “An army’ssuc- planning, efficiently allocate and use resources, improve cess depends at least as much on the quality of the decisions its operational effectiveness and safety, and improve organi- officers and soldiers make and execute on the ground as it zational learning [1–3]. does on actual firefighting power.” The emphasis on decisions Here, we limit our review to evaluating RM concepts is also based on the fact that in an uncertain world, bad out- and principles in the context of wildfire response, i.e., comes can result from well-made decisions, and vice versa the development of a response strategy and its opera- [31], thus the need to emphasize the quality of decisions, tional execution over the duration of an active fire in- and further to understand how characteristics of decisions cident from detection to containment. Strategies can and the environment in which they are made may influence range from full suppression to managing for ecosystem decision quality [32]. benefit, depending on a variety of factors like relevant The emphasis on best available information supports policies, land ownership patterns, potential socioeco- risk-informed decision-making and reinforces decision nomic and ecological impacts, fire growth potential, quality. Three general challenges here are generating and availability of resources. Important components of relevant and actionable information, making that infor- response strategies include mobilizing/demobilizing fire mation available to decision makers, and convincing management resources, allocating and assigning re- them to seek it out and incorporate it into their decision sources to various tasks (e.g., line construction, structure processes [33, 34]. Key information necessary to make protection, mop-up), and monitoring and updating strat- a high-quality strategic response decision is the differ- egies in response to changing conditions. As the com- entiation of response alternatives on the basis of relative plexity, duration, or size of fires increases, response safety, effectiveness, and efficiency [31], and these ele- strategies may increasingly entail blend direct and indi- ments are a primary focus of this paper. Factors such as rect tactics, mobilize a greater amount and diversity of fire weather, fuel type, and total amount of resources ground and aerial resources, and require coordination of have been shown to play a role, but relationships be- a wide variety of activities such as locating drop points tween actual suppression activities and the effectiveness and conducting burnout operations [16]. of suppression resources remain unclear [35–41]. These incident response decisions can be complex, Plucinski [42�� , 43�� ] comprehensively reviewed the uncertain, time-pressured, and require balancing state of knowledge regarding suppression effectiveness tradeoffs across many dimensions (e.g., fire impacts, and identified multiple prominent gaps including the suppression expenditures, and public and responder effectiveness of different suppression resources used in safety), which highlights the need for structured and different suppression techniques, operational fire behav- timely decision support [3, 16–18]. Indeed, there is a ior limits on suppression, use and productivity of all 228 Curr Forestry Rep (2019) 5:226–239 types of suppression resources, how and whether sup- with predictable decision biases and possible operational pression resources can work synergistically, the impact inefficiencies, and gaps in fully putting the three key of fuel management on suppression effectiveness, and RM principles into practice. the impact of protective actions on structure loss. A central thesis of this paper is that a stronger emphasis on In practice, the limited scientific evidence and deci- data-driven decisions and analytics would help bridge those sion support functionality for generation and evaluation gaps. To that end, here, we aim to expose readers to principles of response alternatives mean that these decisions are and insights by drawing from the analytics literature with rele- largely left to expert judgment and intuition. Decisions vance to wildfire management and emergency response. We left to expert judgment can often be excellent, but the also synthesize relevant fire literature from around the globe, dynamic and unpredictable nature of the fire environ- primarily drawing from studies published by authors in North ment coupled with the lack of structured feedback on America, the Mediterranean, and Australia. operational effectiveness likely leads to compromised Ultimately, we argue for a paradigm based on stronger performance of even expert decision makers [44, 45]. adoption of data-driven decisions in fire management that For example, studies from Australia and the USA have we colloquially refer to as “Moneyball for fire” [63], in- demonstrated that fire managers as well as emergency spired by the innovative use of advanced data analytics in managers are susceptible to a number of cognitive errors professional baseball and other sports (Box 1). The im- and biases, and may exhibit wide variation in risk pref- provements in sports analytics started from the ability to erences and judgment [14, 46–52]. Further, empirical conduct more complex analyses of recorded performance observations suggest possible inefficiencies in operations data. Real-time tracking in sports evolved from these ear- due to ineffective or excessive use of resources [53–60]. lier successes, which has opened the door for more anal- Lastly, the emphasis on learning and continual im- ysis, insight, and innovation, and has fundamentally trans- provement relates to developing systems to create and formed the games in unexpected ways (see https://www. transfer knowledge, as well as systems of accountability ted.com/talks/rajiv_maheswaran_the_math_behind_ to monitor performance and correct or reinforce behav- basketball_s_wildest_moves). Although fire management ior to achieve better outcomes over time. Here, the be- organizations collect a considerable amount of data havior of the organization rather than the manager is of related to wildfires, robust data on fire response and even greater importance, requiring concrete steps to de- suppression resource performance is lacking [28, 42�� , velop learning processes that generate, collect, interpret, 43�� , 59]. Thus, the core analogy is not necessarily and disseminate information [61]. One challenge is about making real-time strategic adjustments in response around the organization’s capabilities for improving data to a changing fire environment or a changing adversary’s collection, developing performance indicators, and game strategy, but rather the gains in performance from adopting new technologies and innovations. Meeting preparatory work through investment in real-time moni- this challenge has proven difficult. In the USA, for ex- toring and analysis that lead to a better informational basis ample, federal agencies have been criticized for lacking to support strategic planning through to real-time response sufficient data fidelity and reliability along with limited decisions. analytic capability to understand effectiveness or to in- The remainder of the paper is structured as follows. form decision-making [62]. First, we briefly introduce some core analytics concepts Plucinski [43�� ] succinctly summarized the state of and an analytics management framework, along with affairs, stating that there remain many gaps in the cur- some observations on implementing an analytics agenda rent understanding of suppression effectiveness, that the within organizations. Next, we offer our perspective on data to fill these gaps are not routinely recorded, and the nexus between RM and analytics in wildfire re- that adoption of technologies such as resource tracking sponse, present a stylized figure linking a RM decision will be essential to capture more and better data cycle to the three main types of analytics, and describe streams. In other words, limited collection, documenta- potential application of descriptive, predictive, and pre- tion, archiving, and analysis of operational data on re- scriptive analytics in wildfire. We attempt to ground sponse effectiveness over time has inhibited systematic some of these ideas by relating real-world application learning. This, in turn, has limited meaningful and mea- of analytics to support strategic wildfire decisions, using surable expansion of the knowledge base on the effec- examples with which the authors have direct experience tiveness of alternative response strategies and tactics to in the western USA. We present these real-world exam- inform wildfire response decision-making. The net re- ples in light of the aforementioned analytics manage- sults are bounded operational utility of decision aids ment framework. Lastly, we discuss future opportunities which may lead to disproportionate reliance on expert and challenges for the fire management and science intuition over scientific and organizational evidence, communities. Curr Forestry Rep (2019) 5:226–239 229 Box 1 Timely provision of analytics can be a key driver of success in many areas, and real-time analytics have been developed “Moneyball” for Fire: Lessons from the Sports Analytics Revolution for a range of time-sensitive applications including financial Sports organizations around the world are leveraging analytics and market trading, military operations, smart electrical grids, in- evidence-based management to improve performance. What lessons from the sports analytics revolution can be applied to wildfire telligent transportation systems, and, most relevant, emergen- management? cy response [70, 71]. The increasing use of big data in analyt- Popularized by the book Moneyball: The Art of Winning an Unfair Game ics is often characterized in terms of the volume, variety, and by Michael Lewis [63] (and the movie of the same title), the idea of velocity of data, and technologies are emerging that can rap- “moneyball” is simply the adoption of data-driven decision-making to improve sports performance. Often the brilliance of sports analytics is idly ingest and analyze large quantities of data from sources as in its simplicity. In baseball, for example, a key insight was to evaluate variable as mobile phones and call records, social media ac- players on the basis of their on-base percentage (OBP) rather than their tivity, wearable devices, satellites, remote sensors, and batting average, as it was demonstrated that OBP was a better predictor crowdsourcing platforms [72]. Location-specific data is criti- of ability to score runs. In basketball, it was the realization that the expected number of points per shot was higher for three-point shots cally important for emergency response, for instance, to track than almost all other two-point shots, apart from those taken very close first responders with respect to changing conditions and to the basket. emerging threats [73]. Machine learning in particular seems In a contentious scene in the movie, a scout expresses his dissatisfaction to be growing in emergency response application potential with the analytics approach, stating that it is not possible to put a team together with a computer, and that science could never replace his due its ability to assimilate multiple sources and data types experience and his intuition. This is, of course, a false narrative—the and to generate insights from complex and dynamic environ- whole point of sports analytics is that it is not an either/or situation—it ments. Hong and Akerkar [74] comprehensively review ma- is intended to complement, not replace, expert judgment. Furthermore, chine learning approaches for emergency response, for a range coaches still need to make game-time adjustments, and players still need to execute. Analytics can help teams better prepare for strategic of emergencies including earthquakes, hurricanes, floods, decisions and their execution. landslides, and wildfires, and for a range of tasks including Three Main Types of Analytics Applied to Personnel Decisions in event prediction, early warning, event detection and tracking, Sports, and Analogies to Fire and situational awareness. Real-time analytics are a core com- Player acquisition ➔ Suppression resource acquisition Game strategy ➔ Wildfire response strategy ponent of a broader effort to capitalize on state-of-the-art big Training (fitness, game situations) ➔ Training (fitness, fire simulations) data analytics and advanced technology to provide improved Importantly, these three personnel decisions build upon and are insights for accurate and timely emergency response decision- dependent upon each other. For example, teams select personnel and making [75]. design their training regimen in order to successfully implement the most effective strategies. If not already apparent, key elements of embracing Key Lessons analytics are investing in better data and better science. Engaging experts and analysts to work together to solve organizational Of course, the notion of needing better data and science problems to support wildfire response decisions is not new [27, Defining approachable and understandable analytics Investing in more and better data 28, 31, 29, 42�� , 43�� , 59]. What is new, arguably, is Keeping the human element front and centerSources: [64, 65] embedding these issues within a coherent, principles- based framework that recognizes better data as but one of many steps towards improved decision-making and performance, i.e., an analytics management framework. Why Analytics? Key drivers of analytics success include clear goals, focused problems to solve, quality data from multiple We organize our discussion around basing decisions on rigor- sources, multidisciplinary analytics teams, accessible an- ous analysis and analytic insight over intuition, and on the key alytics systems, data translators, collaborative decision- linkages between analytics, decision-making, and organiza- making processes, end users as advocates, and iteration tional performance [66� ]. Analytics shares a common under- and continuous improvement [64]. Modern analytics is lying purpose with operations research and management sci- also being driven by new technologies, especially ad- ence, improving business operations and decision-making vances in computer science related to analyzing large, through the utilization of information, quantitative analysis, dynamic datasets in real time, but new technologies can and technology [67]. In general, data-driven decisions tend fail to have impact if they are not deployed in an ef- to be better ones, and organizations with comparatively stron- fective organizational framework. ger analytics capabilities tend to outperform their peers and Table 1 presents the nine components of an analytics man- competitors [68, 69� ]. Analytics can be a tool to facilitate agement framework, which can also be thought of as compris- improved decision-making, to measure performance, and ing an iterative cycle. The framework makes clear required even to measure improvements in performance due to adop- core competencies of a successful analytics program: strategic— tion of analytics-based management. planning for how data will be used to help solve 230 Curr Forestry Rep (2019) 5:226–239 Table 1 Analytics Management Strategic Organizational goal Prioritize goal(s) the organization seeks to achieve Framework 2019 Ben Shields, MIT Sloan School of Problem Define specific problem(s) that align with organizational goal(s) Management Data Identify the data needed to solve key problems Technical People Employ people to direct and manage analytics work Process Capture, manage, model, analyze, and visualize data Technology Adopt technologies to enable analytics work Managerial Communication Translate analytical insights into actionable recommendations for key stakeholders Decision-making Use analytics insights in the decision-making process Iteration Track and improve upon the decision problems and achieve organizational goals; technical— 2 and 3 summarize some core themes of each topic, organizing the people, processes, and technologies required respectively. As we see it, the nexus of both topics to manage and analyze data; and managerial — includes developing fluency with uncertainty and prob- communicating data, applying it in decision-making, and ability, emphasizing structured decision-making, making using it for continuous improvement [64]. This framework a commitment to generate and use the best available emphasizes the broader connections to people and process information, monitoring, and iteratively improving these and even culture, as well as the path dependency of data to core competencies over time. Furthermore, this nexus is insight to value [66� ]. definedbyanemphasisonpeopleand culture. For our Translating analytics insight into action requires more purposes here, we can contextualize and distill the nex- than simply setting up data collection systems connected us as follows: providing more and better operationally to a team of data analysts; embracing analytics may require relevant information on the safety and effectiveness of abroader “data-driven cultural change” predicated on exis- suppression strategies and tactics, the formal and trans- tence of an analytics strategy, strong senior management parent use of that information by fire managers in de- support, and careful change management initiatives [76�� ]. cision processes, and the comprehensive tracking of de- That is, the value of data analytics comes not just from the cisions and actions in relation to strategic response ob- technologies that enable it but also from the organizational jectives and fire outcomes. shifts in behavior and enhanced capabilities for strategic insight and performance measurement [77]. The underpin- ning of this shift is an acknowledgement that analytics is Box 2 needed in addition to expert judgment and experience. What Is Risk Management? Similarly, Davenport [78] stresses that the right technology Risk management (RM) is a set of coordinated processes and activities is but one aspect of a successful analytics initiative; the right that identify, monitor, assess, prioritize, and control risks that an organization faces. focus, the right people, and the right culture are also essen- What Are the Different Levels of Risk Management? tial. A final caveat, perhaps the most important one, is that Enterprise ➔ Strategic ➔ Operational ➔ Real-time better data will not necessarily lead to better decisions. Shah What Are the Main Principles of Risk Management? et al. [79� ]caution that “unquestioning empiricists” who Integrating RM into all organizational processes, including decision-making trust analysis over judgment may be no better than “visceral Explicitly accounting for uncertainty decision makers” who go exclusively with their gut. The Addressing problems in a systematic, structured, and timely manner challenge is to develop “informed skeptics” who possess Basing decisions on the best available information strong analytic skills and who can effectively leverage both Tailoring processes to context, and accounting for human and cultural factors judgment and analysis in decision-making. Human judg- Promoting transparency and inclusiveness ment therefore remains front and center in the context of Being dynamic, iterative, and responsive to change embracing analytics to improve decision-making. Facilitating continual improvement How Is Risk Management Different from “Business as Usual”? Informal ➔ Formal The Risk Management and Analytics Nexus in Wildfire Implicit ➔ Explicit Response Intuitive ➔ Analytical Reactive ➔ Proactive In this section, we outline concepts and applications Short-term perspective ➔ Long-term perspective Sources: [1–3] from RM and analytics to wildfire management. Boxes Curr Forestry Rep (2019) 5:226–239 231 Box 3 learning to assess what might happen in the future. This would include predictions of fire weather, fire behavior, and potential What is Analytics? control locations [80–82]. Prescriptive analytics use operation Analytics is the extensive use of data, statistical and quantitative analysis, research methods such as optimization and simulation to rec- explanatory and predictive models, and fact-based management to ommend efficient solutions. This would include assigning drive decisions and actions How Does Using Analytics Improve Performance? suppression resources to tasks such as asset protection or fire Data ➔ Insight ➔ Value line construction [83–85]. Note that some applications can What Are the Main Principles of Analytics? entail combining multiple types of analytics, for example, Treating fact-based decision-making as not only a best practice but also a comparison of observed fire size and impacts (descriptive) part of culture Recognizing the value of analytics, and making their development and with simulated size and impacts in the absence of suppression maintenance a primary focus (predictive) to estimate the productivity and effectiveness of Applying sophisticated information systems and rigorous analysis to a suppression operations [86]. Table 2 provides an illustrative range of functions set of examples of descriptive, predictive, and prescriptive Considering analytics to be so important it is managed at the enterprise level analytics in the context of wildfire response strategy and op- Avidly consuming data and seizing every opportunity to generate erational execution. information The usefulness of predictive and prescriptive analytics is Emphasizing the importance of analytics internally predicated on reliable descriptive analytics on fire and fire Making quantitative capabilities part of the organization’sstory Creating a workforce with strong analytical skills and considering it a key operations. Descriptive analytics can be used to validate pre- to organizational success dictive models, which in turn can be used to parameterize How Do Data-Driven Organizations Act Differently? prescriptive models. Developing a road map to enhance de- The first question a data-driven organization asks itself is not “what do we scriptive analytics is therefore critically important. Thompson think?” but “what do we know?” Decision makers move away from acting solely on hunches and instinct, et al. [31] and Plucinski [42�� ] outline operational data collec- and move away from citing data to support decisions already made tion needs, much of which is reliant on obtaining information Sources: [69� , 78] from fire crews. However, there is a broader horizon for data capture that could capitalize on advanced technologies like the internet of things and machine-to-machine communication. Figure 1 presents a styled RM decision cycle and its rela- These ideas are related to the Industry 4.0 initiative tionship to descriptive, predictive, and prescriptive analytics. (referencing a 4th industrial revolution); the goals of which Descriptive analytics use statistical methods such as statistical are to achieve higher levels of operational efficiency, produc- modeling and data mining to provide insight into what hap- tivity, and automation [88]. These ideas are already being pened in the past. This would include real-time and post hoc integrated into forest operations and supply chain manage- monitoring of suppression operations to gauge effectiveness ment, for instance, the installation of sensors and on-board and development performance measures related to resource computers onto harvest equipment to provide decision support use, productivity, and effectiveness [54, 56, 59]. Predictive and streamline operations [89]. Gingras and Charette [90]in- analytics use techniques such as forecasting and machine troduce a Forestry 4.0 initiative seeking to, among other Fig. 1 Stylized three-stage RM decision cycle and relationship to the three main types of analytics; adapted from [31, 67] 232 Curr Forestry Rep (2019) 5:226–239 Table 2 Examples of analytics in context of wildfire response strategy analytics that are either known knowledge gaps (descriptive) or that and operational execution, adopted from [16, 42�� , 87]. Italic text have had limited or no application in real-world contexts (predictive indicates predictive analytics highlighted in this paper that have recently and prescriptive) been developed and applied in the western USA. Bold text indicates Descriptive Predictive Prescriptive Fire weather Weather forecasts Control line construction, holding, and mop-up Fire size and shape Fire danger ratings (location, type, length, timing) Burn severity Burn probabilities Point protection (location, timing, type) Daily perimeter growth Fire intensity probabilities Logistical feature creation (e.g., safety zones, drop zones) Fire duration Fire arrival times Burnout operations Suppression expenditures Estimated containment time Resource ordering (type and amount) Resource use Estimated suppression costs Resource mobilization and demobilization Resource productivity Potential control locations Resource allocation to assignments Resource effectiveness Suppression difficulty Resource movement Safety zones and escape routes Resource productivity Resource availability things, improve communication networks in remote locations machines) could provide rich opportunities for improving predic- and enable real-time data exchange between operators and tive models. Machine learning techniques have some advantages decision centers. Through combination of technologies like over traditional approaches like generalized linear models due to radio frequency identification, remote sensing through satel- their ability to handle complex problems with multiple lites and drones, and adding sensors and wireless communi- interacting elements, and, increasingly, big data streams [99, cations to suppression machinery, it may be possible to simi- 100]. They can also require large amounts of data, some of which larly create a Fire 4.0 initiative. may come from modern data capture technology, but accounting In the context of supporting strategic and tactical response for extremely rare events in models may also benefit from inter- decisions, predictive analytics might provide the most value. national collaboration and data sharing. Clearly fire behavior predictions are essential, and the literature Table 4 presents an illustrative set of machine learning appli- is rich with descriptions of fire behavior models, their applica- cations in wildfire. One observation is the lack of models on fine- tions, and their limitations [19, 20, 91–93]. Here, we focus in- scale suppression operations, which presents a possible opening stead on analytics more specifically related to operations, which for future machine learning applications when such data are in broad strokes can help assess factors related to safety and available. Future applications could, for example, train models effectiveness, for instance, determination of locations to avoid to predict where fire managers will build line, or where operators for safety reasons or of locations where control opportunities will locate water and retardant drops. At larger scales, models might be most successful. Researchers have sought to incorpo- could predict emerging resource needs for prepositioning of re- rate factors such as fire intensity, heat exposure, safety zones, sources and dispatch prioritization. snag hazard, egress, accessibility, and mobility into these analyt- Because of nested dependencies, prescriptive analytics are ics. Table 3 briefly summarizes some of these recent studies. currently perhaps the most constrained and have the least di- As described earlier, the growing use machine learning tech- rect connection to on-the-ground fire management operations niques (including classification and regression trees, artificial [27, 29]. Prescriptive modeling often serves the role of inte- neural networks, evolutionary computation, and support vector grating descriptive data and predictive results through Table 3 Illustrative set of studies Topic Source(s) that generate predictive analytics to guide and inform safe and Suppression difficulty index based on flame length, heat per unit area, fire line productivity, road [17, 26] effective response operations and trail density, accessibility, and mobility Fire control probability surface used to predict potential control locations [82� ] Spatiotemporal patterns of post-fire snag hazard [94] Travel impedance maps to map efficient egress routes from crew locations to safety zones [95] Determination of safety zone suitability as a function of size, geometry, and height of surrounding [96] vegetation Travel time prediction in variable terrain based on GPS tracked data [97] Expected aviation accident rates by aircraft type and workload [98] Curr Forestry Rep (2019) 5:226–239 233 Table 4 Illustrative set of studies Topic Source(s) that apply machine learning techniques to wildfire Modeling the effect of suppression on large fire spread [39] Predicting exposure of populations to fine particular matter during wildfires [101] Predicting post-fire debris flow events [102] Predicting fire occurrence and burnt area [103–106] Modeling initial attack success [107] Modeling wildfire spread dynamics [108] Estimating vegetation biomass, cover, and other characteristics [109, 110] Modeling impact of wildfires on river flows [111] mathematical equations to reflect operational constraints and line, based on manager preferences regarding damage, cost, to optimally achieve management objectives. By design, com- and firefighter safety [112� ]. At present, the descriptive model plicated decision rules and logic could be built in the model to is informing agency discussions regarding development of guide the search process for optimal solutions, and uncertainty key performance indicators, the predictive model is being ac- can be accounted for with probabilistic rather than determin- tively used to support real-time decision making and strategic istic frameworks. Such a system could be difficult for humans planning throughout the western USA, and the prescriptive to track due to its complexity and potential tradeoffs between model is a prototype. alternative solutions. However, parameterizing a prescriptive Figure 2 illustrates these interrelated concepts from a post model with unreliable data can lead to infeasible, unreason- hoc analysis of the 2018 Ferguson Fire (39,200 ha) that able, or suboptimal decisions. Once a decision is suggested by burned in the Sierra National Forest, Stanislaus National the system, it is often difficult for managers to intuitively Forest, and Yosemite National Park in California, USA. understand the quality of the decision and all the reasons for Panel a displays the fire perimeter in relation to locations of it to be chosen due to unknown causal relationships among constructed fire line, and summarizes fire line effectiveness statistically weighted variables, and especially due to data per the framework of Thompson et al. [113� ]. Panel b displays quality. A role for prescriptive models, in the near future at the fire perimeter in relation to an underlying fire control least, might be for exploring tradeoffs, generating efficient probability surface [82� ], with a histogram showing the per- frontiers, identifying hypotheses to potentially improve deci- centage of fire perimeter length in one of five control proba- sion quality, and performing scenario and sensitivity analyses bility categories. Panel c displays results of a spatial optimi- [e.g., 85, 112� ]. Using prescriptive models to directly support zation model developed by Wei et al. [112� ] that outlines one fire management decisions not only requires improved data possible response strategy based on landscape risk, potential quality and model design but may also require that the models control locations, and manager-specified preferences. are easy to understand to decision makers and are not viewed Table 5 displays how the fire assessment and planning pro- as a “black box” with unknown and mysterious inner cesses using predictive modeling of potential control locations workings. fits into the analytics management framework. The predictive model developed by O’Connor et al. [82� ] has been widely Analytics Demonstration and Real-World Application: deployed in the USA for pre-fire planning applications as well Potential Control Locations and Fire Line as for real-time decision support on more than two dozen Effectiveness wildfires during the 2017 and 2018 fire seasons. As of this writing, fire ecologists and analysts have delivered potential control location map products to National Forest System staff, To further ground these analytics concepts in fire operations, we consider the construction of fire containment line and pres- partners, and stakeholders on over thirty landscapes across the ent recent research and application on that topic. A descriptive western USA. The pairing of such information with quantita- model of fire line effectiveness analyzes fire perimeter and tive risk assessments and the delineation of strategic response operations data to describe where line was built, whether it Tragically, there was a tree strike fatality on the Ferguson Fire. The engaged the fire, and if so, whether it was effective at stopping Corrective Action Plan (https://wildfiretoday.com/documents/ fire [113� ]. A predictive model analyzes historical fire perim- CorrectiveActionPlanHughes.pdf) recommended evaluation of how eters in relation to environmental and landscape characteris- changing environmental conditions, such as extensive tree mortality, are tics, and trains a machine learning algorithm to estimate the being factored into response strategies and tactics. We believe operationally relevant analytics that speak to firefighter safety could help meet that need, and probability of a given location being a suitable control line to ultimately reduce the occurrence of such tragedies in the future. Nothing in this stop fire [82� ]. A prescriptive model recommends strategies to brief case study should be construed as commentary on the decisions or actions managers that combine the most suitable locations to construct taken on the Ferguson Fire. 234 Curr Forestry Rep (2019) 5:226–239 Fig. 2 Case study analysis of the 2018 Ferguson Fire, showing a model run with a user-specified set of preferences regarding cost and descriptive analytics of fire line effectiveness, b predictive analytics of safety. For panel a: Tr = ratio of total length of line to perimeter; Er = potential control locations in relation to the final perimeter, and c ratio of engaged line to total line; HEr = ratio of held line to engaged line; prescriptive analytics that recommend combinations of control locations HTr = ratio of held line to total line [59, 113� ] to create overarching response strategies, here showing the results of one zones are not only enhancing cross-boundary planning pro- Discussion cesses but also helping managers achieve desirable fire out- comes that protect assets while enhancing ecosystem health A core idea from this paper is the need to develop a broader [114]. Following an intensive strategic wildfire risk planning analytics strategy for wildfire management, and to infuse an- process in the spring of 2017, the Tonto National Forest in alytics into planning, decision, and learning processes. The central Arizona has managed six large wildfires in accordance aims of such an effort would be to align actions with manage- with pre-identified strategic wildfire response zones that help ment objectives, enhance transparency and accountability, im- to align incident response objectives with land and resource prove decision support, create learning opportunities, and ul- management planning direction. The 2017 Pinal, Highline, timately improve response safety and effectiveness [115]. and Brooklyn fires were managed for ecological restoration, Doing so would help fire management organizations redeem asset protection, and wildfire process maintenance objectives, some of their core RM responsibilities, such as generating respectively. During the 2018 fire season, when milder fire better information to support risk-informed decisions, and weather conditions were prevalent, the Bears, Daisy, and continually improving. Cholla fires were each managed for ecological restoration We recognize there are a broader range of important and future risk reduction objectives consistent with pre- incident-related decisions than that considered here, such as identified strategic wildfire response zones and corresponding routing detection aircraft, stationing and prepositioning sup- fire response objectives. pression resources, and transferring or reassigning resources Table 5 Analytics Management Strategic Organizational goal Safe and effective response Framework as applied for predictive model of potential Problem Identifying locations on landscape where responder hazard is lower and control locations, adapted from likelihood of control success is higher [82� , 114] Data Historical fire perimeters, landscape and environmental attributes Technical People Fire ecologists, modelers, analysts, local fire managers Process Predictive modeling combined with local expert judgment Technology Machine learning algorithm, GIS software Managerial Communication Workshops with local managers and stakeholders Decision-making Planning: guided designation of fire containers and strategic response zones; incident response: guided choices of line construction location Iteration Refinement of model, feedback from managers Curr Forestry Rep (2019) 5:226–239 235 to different regions or different fires [29, 116–118]. Some of metrics of performance and risk. Embracing analytics could these decisions are inexorably linked to response strategies, have a lot to offer regarding issues of overreliance on expert for example, allocating scarce resources between fires is pre- judgment and decision quality. This includes decomposing the mised on some understanding of how resources would be used problem into manageable sub-problems and developing oper- and with what degree of efficacy if assigned to different fires. ationally relevant decision aids. Through more proactive ac- We opted to focus on incident-level response strategies be- quisition and use of more reliable and more trustworthy infor- cause these can be high-impact decisions, and because we mation, fire management organizations could help dampen believe there is great potential for stronger adoption of RM some of the biases described earlier and improve fire manage- and analytics principles within this decision context. ment decision-making [33, 34]. Looking to the future, fire management organizations Although challenges exist regarding capture, interpre- will need to modernize data collection and analysis sys- tation of fire operations data, and advances in scientific tems associated with fire activity and wildfire suppression understanding, the bigger challenges may well be orga- operations. The emphasis will be on measuring and mon- nizational and cultural. Effectuating a data-driven cultur- itoring quantitative performance metrics to evaluate oper- al change is a known barrier to wider spread adoption ational capabilities, safety, efficiency, and effectiveness. of analytics [76�� ]. How to convince fire managers that Digital technology integration, including automated re- enhanced monitoring and collection (i.e., descriptive an- source tracking, is likely to be a key component of new alytics) will not be used for “Monday morning data collection systems. Operations research and industrial quarterbacking,” and furthermore, how to demonstrate engineering, combined with data science, are perfectly the value of predictive and prescriptive analytics to suited to this task. These disciplines offer well-defined managers will be some of the fire science community’s scientific approaches for performance measurement, pro- challenges moving forward. Stronger collaboration be- cess engineering, logistics, and operations management, tween the fire science and management communities including health and safety. As stated earlier, similar mod- could help here. As summarized by Martell [87]: pre- ernization is already occurring in the forest sector under a dictive analytics specialists cannot develop useful pre- variety of precision forestry initiatives. dictive models unless they collaborate with fire manage- That many of these technologies have been around for ment organizations that are willing to share their prob- years does raise questions of why fire management organiza- lems and data, and prescriptive analytics specialists can- tions have not already turned to them. For example, re- not develop useful prescriptive models unless they col- sponders might find tracking technology that allows close laborate with predictive analysts and with fire manage- monitoring of all of their actions to be invasive, or key deci- ment organizations that are willing to share their prob- sion makers may fear increased liability risks. Organizations lems and test their models. To the degree that analytics are likely to face issues of data governance, privacy and secu- results in demonstrably better outcomes in terms of op- rity, leading to data policy questions such as which data to erational safety, efficiency, and effectiveness, it is likely make available, to whom, through what channels, and for to help incentivize its own adoption over time. what purposes [72, 73, 75]. Limited abilities to effectively The emphasis on people and culture in analytics frame- manage and standardize the complex data streams collected works also highlights the need for stronger collaboration with from various sources decrease the utility and increase the cost social scientists. There are opportunities to convert experien- of data capture [119]. Another barrier is the need to build tial evidence into organizational and scientific evidence (i.e., workforce capacity to effectively use data science [120]. In informing the development of meaningful performance met- addition to addressing data concerns, organizations may not rics), to improve knowledge exchange, to identify potential have supportive learning environments in place, where barriers to meaningful organizational change, and to design learners have psychological safety to express disagreement communication strategies in light of how information is cur- or own up to mistakes, where opposing ideas and competing rently shared within fire response networks [121, 122]. outlooks are welcomes, and where taking risks is appreciated Beyond rolling out the analytics management framework [61]. for an expanded set of fire operations, we see a number of We see a future with increased automation and prescriptive productive pathways forward. As a possible interim strategy, analytics recommendations; however, fire management will the fire community could seek to more comprehensively cat- always be a question of human judgment. Every fire event alog lessons learned [86] and develop more decision support has unique circumstances and potentially unresolvable uncer- systems based on elicitation of expert judgment [123]. On the tainties, rendering completely prescriptive approaches infeasi- technical side, analysts can capitalize on the growing use of ble, hence the focus on decision makers. In analytics, many machine learning techniques in forest and fire modeling. The but not all aspects of expert judgment and organizational widespread use of open-source technology also presents op- wisdom may track closely with measurable and predictable portunities for sharing data and code [124]. On the 236 Curr Forestry Rep (2019) 5:226–239 Summit & 4th Human Dimensions of Wildland Fire Conference. organizational side, performance management systems could pp. 92–113, Boise, ID. be redesigned with risk in mind to better account for non- 3. Thompson, M.P., MacGregor, D.G. and Calkin, D., 2016. Risk deterministic relationships between decisions and outcomes management: core principles and practices, and their relevance [125]. Performance measurement could also expand to a to wildland fire. Gen. Tech. Rep. RMRS-GTR-350. Fort Collins, CO: US Department of Agriculture, Forest Service, Rocky systems-based approach that conceptualizes effective re- Mountain Research Station. 29 p. sponse not just in operations terms but also factors such as 4. Riley K, Thompson M. An uncertainty analysis of wildfire model- evacuations, road closures, and public information [13� ]. ing . Natural hazard uncertainty assessment: modeling and deci- sion support. Monograph. 2017;223:193–213. 5. Ager AA, Evers CR, Day MA, Preisler HK, Barros AM, Nielsen- Pincus M. Network analysis of wildfire transmission and implica- Conclusion tions for risk governance. PLoS One. 2017;12(3):e0172867 Beaver. 6. Calkin DE, Cohen JD, Finney MA, Thompson MP. How risk The translation of scientific ideas and insights into prac- management can prevent future wildfire disasters in the tice can be essential for improving decision quality and wildland-urban interface. Proc Natl Acad Sci. 2014;111(2):746– cultivating stronger RM acumen. A number of hurdles remain: increasing investment in data capture and analysis 7. Chung W. Optimizing fuel treatments to reduce wildland fire risk. Curr For Rep. 2015;1(1):44–51. technology, dealing with data privacy and security issues, 8. Fischer AP, Spies TA, Steelman TA, Moseley C, Johnson BR, developing better models and decision support tools, and Bailey JD, et al. Wildfire risk as a socioecological pathology. catalyzing organizational and cultural change. Despite Front Ecol Environ. 2016;14(5):276–84. these challenges, we believe there are ample opportunities 9. Hesseln H. Wildland fire prevention: a review. Curr For Rep. 2018;4(4):178–90. for use of descriptive, predictive, and prescriptive analyt- 10. Mavsar R, Cabán AG, Varela E. The state of development of fire ics in fire. We provided one example of a fire line effec- management decision support systems in America and Europe. tiveness framework and its role in serving as a connective Forest Policy Econ. 2013;29:45–55. thread between the three types of analytics. Similar cycles 11. McCaffrey S. Community wildfire preparedness: a global state-of- of the analytics strategy could be pursued for mop-up, the-knowledge summary of social science research. Curr For Rep. 2015;1(2):81–90. aviation, burn out operations, and structure protection. 12. Steelman TA, McCaffrey S. Best practices in risk and crisis com- Developing core competencies with analytics and cultivat- munication: implications for natural hazards management. Nat ing a “Moneyball for fire” paradigm may help fire man- Hazards. 2013;65(1):683–705. agement organizations on their RM journey. 13.� Steelman T, Nowell B. Evidence of effectiveness in the Cohesive Strategy: measuring and improving wildfire response. Int J Wildland Fire. 2019;28(4):267–74 Extends conceptualization Funding Information Dr. Wei and Dr. Dunn received funding from the of wildfire response performance measurement to a systems- USDA Forest Service through Joint Venture Agreements. based perspective considering factors beyond operational effectiveness. Compliance with Ethical Standards 14. Penney G. Exploring ISO31000 risk management during dynamic fire and emergency operations in Western Australia. Fire. Conflict of Interest All authors declare they have no conflict of interest. 2019;2(2):21. 15. National Interagency Fire Center. 2019. Interagency Standards for Open Access This article is distributed under the terms of the Creative Fire and Fire Aviation Operations 2019. Chapter 5: USDA Forest Commons Attribution 4.0 International License (http:// Service Wildland Fire and Aviation Program Organization and creativecommons.org/licenses/by/4.0/), which permits unrestricted use, Responsibilities. Available at: https://www.nifc.gov/ distribution, and reproduction in any medium, provided you give appro- PUBLICATIONS/redbook/2019/Chapter05.pdf. Accessed 14 priate credit to the original author(s) and the source, provide a link to the Nov 2019 Creative Commons license, and indicate if changes were made. 16. Dunn CJ, Thompson MP, Calkin DE. A framework for developing safe and effective large-fire response in a new fire management References paradigm. For Ecol Manag. 2017;404:184–96. 17. O’Connor C, Thompson M, Rodríguez y Silva F. Getting ahead of the wildfire problem: quantifying and mapping management chal- lenges and opportunities. Geosciences. 2016;6(3):35. Papers of particular interest, published recently, have been 18. Thompson M, Bowden P, Brough A, Scott J, Gilbertson-Day J, highlighted as: Taylor A, et al. Application of wildfire risk assessment results to � Of importance wildfire response planning in the southern Sierra Nevada, �� Of major importance California, USA. Forests. 2016;7(3):64. 19. Plucinski MP, Sullivan AL, Rucinski CJ, Prakash M. Improving 1. International Organization for Standardization. ISO 31000:2018 the reliability and utility of operational bushfire behaviour predic- Risk Management. Geneva: ISO; 2018. tions in Australian vegetation. Environ Model Softw. 2017;91:1– 2. Beaver A 2015. Applying the ISO 31000:2009 risk management – principles and guidelines for improved risk management decision 20. Cruz MG, Alexander ME, Sullivan AL, Gould JS, Kilinc M. making. Proceedings of the 13th Annual Wildland Fire Safety Assessing improvements in models used to operationally predict Curr Forestry Rep (2019) 5:226–239 237 wildland fire rate of spread. Environ Model Softw. 2018;105:54– 19 Outlines a framework for evaluating suppression effective- 63. ness and synthesizes findings at flame and fire line scales. 21. Kalabokidis K, Ager A, Finney M, Athanasis N, Palaiologou P, 43.�� Plucinski MP. Contain and control: wildfire suppression effective- Vasilakos C. AEGIS: a wildfire prevention and management in- ness at incidents and across landscapes. Curr For Rep. 2019;5(1): formation system. Nat Hazards Earth Syst Sci. 2016;16:643–61. 20–40 Comprehensive review of studies of suppression effec- 22. Pacheco AP, Claro J, Fernandes PM, de Neufville R, Oliveira TM, tiveness and roadmap for enhanced operational data collec- Borges JG, et al. Cohesive fire management within an uncertain tion, and synthesizes research on effectiveness at incident and environment: a review of risk handling and decision support sys- landscape scales. tems. For Ecol Manag. 2015;347:1–17. 44. Shanteau J. Competence in experts: the role of task characteristics. 23. Molina JR, Moreno R, Castillo M, y Silva FR. Economic suscep- Organ Behav Hum Decis Process. 1992;53(2):252–66. tibility of fire-prone landscapes in natural protected areas of the 45. Kahneman D, Klein G. Conditions for intuitive expertise: a failure southern Andean range. Sci Total Environ. 2018;619:1557–65. to disagree. Am Psychol. 2009;64(6):515. 24. Jolly WM, Freeborn PH. Towards improving wildland firefighter 46. Wilson RS, Winter PL, Maguire LA, Ascher T. Managing wildfire situational awareness through daily fire behaviour risk assess- events: risk-based decision making among a group of federal fire ments in the US northern Rockies and northern Great Basin. Int managers. Risk Analysis: An International Journal. 2011;31(5): J Wildland Fire. 2017;26(7):574–86. 805–18. 25. Campbell MJ, Page WG, Dennison PE, Butler BW. Escape route 47. Hand MS, Wibbenmeyer MJ, Calkin DE, Thompson MP. Risk index: a spatially-explicit measure of wildland firefighter egress preferences, probability weighting, and strategy tradeoffs in wild- capacity. Fire. 2019;2(3):–40. fire management. Risk Anal. 2015;35(10):1876–91. 26. Rodríguez y Silva F, Martínez JRM, González-Cabán A. A meth- 48. Wibbenmeyer MJ, Hand MS, Calkin DE, Venn TJ, Thompson odology for determining operational priorities for prevention and MP. Risk preferences in strategic wildfire decision making: a suppression of wildland fires. Int J Wildland Fire. 2014;23(4): choice experiment with U.S. wildfire managers. Risk Anal. 544–54. 2013;33(6):1021–37. 27. Duff TJ, Tolhurst KG. Operational wildfire suppression model- 49. McLennan J, Holgate AM, Omodei MM, Wearing AJ. Decision ling: a review evaluating development, state of the art and future making effectiveness in wildfire incident management teams. J directions. Int J Wildland Fire. 2015;24(6):735–48. Conting Crisis Manag. 2006;14(1):27–37. 28. Filkov A, Duff T, Penman T. Improving fire behaviour data ob- 50. Calkin DE, Venn T, Wibbenmeyer M, Thompson MP. Estimating tained from wildfires. Forests. 2018;9(2):81. US federal wildland fire managers’ preferences toward competing 29. Martell DL. A review of recent forest and wildland fire manage- strategic suppression objectives. Int J Wildland Fire. 2013;22(2): ment decision support systems research. Curr For Rep. 2015;1(2): 212–22. 128–37. 51. Roberts PS, Wernstedt K. Decision biases and heuristics among 30. Blenko MW, Mankins MC, Rogers P. The decision-driven orga- emergency managers: just like the public they manage for? Am nization. Harv Bus Rev. 2010;88(6):54–62. Rev Public Adm. 2019;49(3):292–308. 31. Marcot BG, Thompson MP, Runge MC, Thompson FR, McNulty 52. Wernstedt K, Roberts PS, Arvai J, Redmond K. How emergency S, Cleaves D, et al. Recent advances in applying decision science to managing national forests. For Ecol Manag. 2012;285:123–32. managers (mis?) interpret forecasts. Disasters. 2019;43(1):88– 32. Spetzler C, Winter H, Meyer J. Decision quality: value creation 109. from better business decisions: John Wiley & Sons; 2016. 53. Hand M, Katuwal H, Calkin DE, Thompson MP. The influence of 33. Pfeffer J, Sutton RI. Evidence-based management. Harv Bus Rev. incident management teams on the deployment of wildfire sup- 2006;84(1):62. pression resources. Int J Wildland Fire. 2017;26(7):615–29. 34. Barends E, Rousseau DM, Briner RB. Evidence-based manage- 54. Holmes TP, Calkin DE. Econometric analysis of fire suppression ment: the basic principles. Amsterdam: Center for Evidence- production functions for large wildland fires. Int J Wildland Fire. Based Management; 2014. 2013;22(2):246–55. 35. Plucinski MP. Modelling the probability of Australian grassfires 55. Stonesifer CS, Calkin DE, Thompson MP, Stockmann KD. escaping initial attack to aid deployment decisions. Int J Wildland Fighting fire in the heat of the day: an analysis of operational Fire. 2013;22(4):459–68. and environmental conditions of use for large airtankers in 36. Beverly JL. Time since prior wildfire affects subsequent fire con- United States fire suppression. Int J Wildland Fire. 2016;25(5): tainment in black spruce. Int J Wildland Fire. 2017;26(11):919– 520–33. 56. Katuwal H, Dunn CJ, Calkin DE. Characterising resource use and 37. Collins KM, Price OF, Penman TD. Suppression resource deci- potential inefficiencies during large-fire suppression in the western sions are the dominant influence on containment of Australian US. Int J Wildland Fire. 2017;26(7):604–14. forest and grass fires. J Environ Manag. 2018;228:373–82. 57. Calkin DE, Stonesifer CS, Thompson MP, McHugh CW. Large 38. Finney M, Grenfell IC, McHugh CW. Modeling containment of airtanker use and outcomes in suppressing wildland fires in the large wildfires using generalized linear mixed-model analysis. For United States. Int J Wildland Fire. 2014;23(2):259–71. Sci. 2009;55(3):249–55. 58. Donovan GH, Prestemon JP, Gebert K. The effect of newspaper 39. Fernandes PM, Pacheco AP, Almeida R, Claro J. The role of fire- coverage and political pressure on wildfire suppression costs. Soc suppression force in limiting the spread of extremely large forest Nat Resour. 2011;24(8):785–98. fires in Portugal. Eur J For Res. 2016;135(2):253–62. 59. Thompson M, Lauer C, Calkin D, Rieck J, Stonesifer C, Hand M. 40. Tremblay PO, Duchesne T, Cumming SG. Survival analysis and Wildfire response performance measurement: current and future classification methods for forest fire size. PLoS One. 2018;13(1): directions. Fire. 2018;1(2):21. e0189860. 60. Plucinski MP, Pastor E. Criteria and methodology for evaluating 41. Cardil A, Lorente M, Boucher D, Boucher J, Gauthier S. Factors aerial wildfire suppression. Int J Wildland Fire. 2013;22(8):1144– influencing fire suppression success in the province of Quebec (Canada). Can J For Res. 2018;49(5):531–42. 61. Garvin DA, Edmondson AC, Gino F. Is yours a learning organi- 42.�� Plucinski MP. Fighting flames and forging firelines: wildfire sup- zation? Harv Bus Rev. 2008;86(3):109. pression effectiveness at the fire edge. Curr For Rep. 2019;5(1):1– 238 Curr Forestry Rep (2019) 5:226–239 62. Booz Allen Hamilton. 2014 Quadrennial fire review: final report; judgment” and argues for developing “informed skeptics” as 2015. Available online: https://www.forestsandrangelands.gov/ desired decision makers. QFR/documents/2014QFRFinalReport.pdf 80. Katuwal H, Calkin DE, Hand MS. Production and efficiency of 63. Lewis MM. Moneyball: the art of winning an unfair game. New large wildland fire suppression effort: a stochastic frontier analy- York: W. W; 2003. sis. J Environ Manag. 2016;166:227–36. 64. Shields B. Analytics management: business lessons from the 81. Monedero S, Ramirez J, Cardil A. Predicting fire spread and be- sports data revolution at MIT Sloan School of management. haviour on the fireline. Wildfire analyst pocket: a mobile app for March. 2019;19-20:2019. wildland fire prediction. Ecol Model. 2019;392:103–7. 65. Rein I, Shields B, Grossman A. The sports strategist: developing 82.� O’Connor CD, Calkin DE, Thompson MP. An empirical machine leaders for a high-performance industry. USA: Oxford University learning method for predicting potential fire control locations for Press; 2014. pre-fire planning and operational fire management. Int J Wildland 66.� Sharma R, Mithas S, Kankanhalli A. Transforming decision- Fire. 2017;26(7):587–97 Predictive analytics tool for mapping making processes: a research agenda for understanding the impact potential fire control locations, now widely implemented on of business analytics on organisations. Eur J Inf Syst. 2014;23(4): National Forest System lands in the western USA. 433–41 Reviews literature on role of business analytics in or- 83. van der Merwe M, Minas JP, Ozlen M, Hearne JW. A mixed ganizations, and emphasizes relationships between business integer programming approach for asset protection during escaped analytics, decision-making processes, and organizational wildfires. Can J For Res. 2014;45(4):444–51. performance. 84. Belval EJ, Wei Y, Bevers M. A stochastic mixed integer program 67. Mortenson MJ, Doherty NF, Robinson S. Operational research to model spatial wildfire behavior and suppression placement de- from Taylorism to Terabytes: a research agenda for the analytics cisions with uncertain weather. Can J For Res. 2015;46(2):234– age. Eur J Oper Res. 2015;241(3):583–95. 68. McAfee A, Brynjolfsson E, Davenport TH, Patil DJ, Barton D. 85. Wei Y, Thompson MP, Haas JR, Dillon GK, O’Connor CD. Big data: the management revolution. Harv Bus Rev. 2012;90(10): Spatial optimization of operationally relevant large fire confine 60–8. and point protection strategies: model development and test cases. 69.� LaValle S, Lesser E, Shockley R, Hopkins MS, Kruschwitz N. Big Can J For Res. 2018;48(5):480–93. data, analytics and the path from insights to value. MIT Sloan 86. Rodríguez y Silva FR, González-Cabán A. Contribution of sup- Manag Rev. 2011;52(2):21 Review of how organizations are pression difficulty and lessons learned in forecasting fire suppres- using analytics to gain insight and guide actions and argues sion operations productivity: a methodological approach. J For that the biggest challenges are managerial and cultural and Econ. 2016;25:149–59. offers recommendations for successfully building an analytics 87. Martell D. A forest and wildland fire management analytics frame- program. work. For Wildland Fire Manag: a risk management perspective 70. Mohamed, N. and Al-Jaroodi, J., 2014. Real-time big data analyt- workshop, Banff International Research Station, Banff, Canada. ics: applications and challenges. In 2014 international conference 2017;5-10:2017. on high performance computing & simulation (HPCS) (pp. 305- 88. Lu Y. Industry 4.0: A survey on technologies, applications and 310). IEEE. open research issues. J Ind Inf Integr. 2017;6:1–10. 71. Galindo G, Batta R. Review of recent developments in OR/MS 89. Müller F, Jaeger D, Hanewinkel M. Digitization in wood supply–a research in disaster operations management. Eur J Oper Res. review on how Industry 4.0 will change the forest value chain. 2013;230(2):201–11. Comput Electron Agric. 2019;162:206–18. 72. Qadir J, Ali A, ur Rasool R, Zwitter A, Sathiaseelan A, Crowcroft 90. Gingras JF, Charette F. FP innovations forestry 4.0 initiative. J. Crisis analytics: big data-driven crisis response. J Int Human Bangor: 2017 Council on Forest Engineering Annual Meeting; Action. 2016;1(1):12. 2017. July 30-August 2, 2017 73. Akter, S. and Wamba, S.F., 2017. Big data and disaster manage- 91. Sá AC, Benali A, Fernandes PM, Pinto RM, Trigo RM, Salis M, ment: a systematic review and agenda for future research. Ann et al. Evaluating fire growth simulations using satellite active fire Oper Res, pp1–21. data. Remote Sens Environ. 2017;190:302–17. 74. Hong M, Akerkar R. Analytics and evolving landscape of ma- 92. Arca B, Ghisu T, Casula M, Salis M, Duce P. Aweb-based wildfire chine learning for emergency response. In: In Machine learning simulator for operational applications. Int J Wildland Fire. paradigms. Cham: Springer; 2019. p. 351–97. 2019;28(2):99–112. 75. Shah SA, Seker DZ, Hameed S, Draheim D. The rising role of big 93. Duff T, Cawson J, Cirulis B, Nyman P, Sheridan G, Tolhurst K. data analytics and IoT in disaster management: recent advances, Conditional performance evaluation: using wildfire observations taxonomy and prospects. IEEE Access. 2019;7:54595–614. for systematic fire simulator development. Forests. 2018;9(4):189. 76.�� Vidgen R, Shaw S, Grant DB. Management challenges in creating 94. Dunn CJ, O’Connor CD, Reilly MJ, Calkin DE, Thompson MP. value from business analytics. Eur J Oper Res. 2017;261(2):626– Spatial and temporal assessment of responder exposure to snag 39 Similar to [62], reviews literature on business analytics, hazards in post-fire environments. For Ecol Manag. 2019;441: focusing in part on how to create value through actions based 202–14. on better decisions based on better data, and the roles of data 95. Campbell MJ, Dennison PE, Butler BW. A LiDAR-based analysis and value, organization and process, and people and of the effects of slope, vegetation density, and ground surface technology. roughness on travel rates for wildland firefighter escape route 77. Mikalef P, Pappas IO, Krogstie J, Giannakos M. Big data analytics mapping. Int J Wildland Fire. 2017;26(10):884–95. capabilities: a systematic literature review and research agenda. 96. Campbell MJ, Dennison PE, Butler BW. Safe separation distance IseB. 2018;16(3):547–78. score: a new metric for evaluating wildland firefighter safety zones 78. Davenport, T.H., 2006. Competing on analytics. Harvard Business using Lidar. Int J Geogr Inf Sci. 2017b;31(7):1448–66. Review, 84(1), p.98. 97. Campbell MJ, Dennison PE, Butler BW, Page WG. Using 79.� Shah S, Horne A, Capellá J. Good data won’t guarantee good crowdsourced fitness tracker data to model the relationship be- decisions. Harv Bus Rev. 2012;90(4):23–5 Cautions against tween slope and travel rates. Appl Geogr. 2019;106:93–107. overvaluing “big data” at the expense of cultivating “big Curr Forestry Rep (2019) 5:226–239 239 98. Stonesifer CS, Calkin DE, Thompson MP, Kaiden JD. Developing 113.� Thompson MP, Freeborn P, Rieck JD, Calkin DE, Gilbertson-Day an aviation exposure index to inform risk-based fire management JW, Cochrane MA, et al. Quantifying the influence of previously decisions. J For. 2014;112(6):581–90. burned areas on suppression effectiveness and avoided exposure: a case study of the Las Conchas Fire. Int J Wildland Fire. 99. Liu Z, Peng C, Work T, Candau JN, DesRochers A, Kneeshaw D. 2016;25(2):167–81 Descriptive analytics model to evaluate fire Application of machine-learning methods in forest ecology: recent line effectiveness, subsequently expanded on in Thompson et progress and future challenges. Environ Rev. 2018;26(4):339–50. al [59]. 100. Olden JD, Lawler JJ, Poff NL. Machine learning methods without 114. O’Connor CD, Calkin DE. Engaging the fire before it starts: a case tears: a primer for ecologists. Q Rev Biol. 2008;83(2):171–93. study from the 2017 Pinal Fire (Arizona). Wildfire. 2019;28, 101. Reid CE, Jerrett M, Petersen ML, Pfister GG, Morefield PE, Tager 14(1):–18. IB, et al. Spatiotemporal prediction of fine particulate matter dur- 115. Thompson MP, MacGregor DG, Dunn CJ, Calkin DE, Phipps J. ing the 2008 northern California wildfires using machine learning. Rethinking the wildland fire management system. J For. Environ Sci Technol. 2015;49(6):3887–96. 2018;116(4):382–90. 102. Kern AN, Addison P, Oommen T, Salazar SE, Coffman RA. 116. Belval EJ, Calkin DE, Wei Y, Stonesifer CS, Thompson MP, Machine learning based predictive modeling of debris flow prob- Masarie A. Examining dispatching practices for interagency hot- ability following wildfire in the intermountain Western United shot crews to reduce seasonal travel distance and manage fatigue. States. Math Geosci. 2017;49(6):717–35. Int J Wildland Fire. 2018;27(9):569–80. 103. de Bem PP, de Carvalho Júnior OA, Matricardi EAT, Guimarães 117. Chow JY, Regan AC. Resource location and relocation models RF, Gomes RAT. Predicting wildfire vulnerability using logistic with rolling horizon forecasting for wildland fire planning. regression and artificial neural networks: a case study in Brazil’s INFOR: Information Systems and Operational Research. Federal District. Int J Wildland Fire. 2019;28(1):35–45. 2011;49(1):31–43. 104. Rodrigues M, de la Riva J. An insight into machine-learning al- 118. Belval EJ, Wei Y, Calkin DE, Stonesifer CS, Thompson MP, gorithms to model human-caused wildfire occurrence. Environ Tipton JR. Studying interregional wildland fire engine assign- Model Softw. 2014;57:192–201. ments for large fire suppression. Int J Wildland Fire. 2017;26(7): 105. Sayad YO, Mousannif H, Al Moatassime H. Predictive modeling 642–53. of wildfires: a new dataset and machine learning approach. Fire 119. Wyngaard J, Barbieri L, Thomer A, Adams J, Sullivan D, Crosby Saf J. 2019;104:130–46. C, et al. Emergent challenges for science sUAS data management: 106. Leuenberger M, Parente J, Tonini M, Pereira MG, Kanevski M. fairness through community engagement and best practices devel- Wildfire susceptibility mapping: deterministic vs. stochastic ap- opment. Remote Sens. 2019;11:1797. proaches. Environ Model Softw. 2018;101:194–203. 120. Gibert K, Horsburgh JS, Athanasiadis IN, Holmes G. 107. Rodrigues M, Alcasena F, Vega-García C. Modeling initial attack Environmental data science. Environ Model Softw. 2018;106:4– success of wildfire suppression in Catalonia, Spain. Sci Total Environ. 2019;666:915–27. 121. Nowell B, Steelman T, Velez ALK, Yang Z. The structure of 108. Ganapathi Subramanian S, Crowley M. Using spatial reinforce- effective governance of disaster response networks: insights from ment learning to build forest wildfire dynamics models from sat- the field. Am Rev Public Adm. 2018;48(7):699–715. ellite images. Frontiers in ICT. 2018;5:6. 122. Butler BW, Brown S, Wright V, Black A. Bridging the divide 109. Riley KL, Grenfell IC, Finney MA. Mapping forest vegetation for between fire safety research and fighting fire safely: how do we the western United States using modified random forests imputa- convey research innovation to contribute more effectively to wild- tion of FIA forest plots. Ecosphere. 2016;7(10):e01472. land firefighter safety? Int J Wildland Fire. 2017;26(2):107–12. 110. Anderson KE, Glenn NF, Spaete LP, Shinneman DJ, Pilliod DS, 123. Plucinski MP, McCarthy GJ, Hollis JJ, Gould JS. The effect of Arkle RS, et al. Estimating vegetation biomass and cover across aerial suppression on the containment time of Australian wildfires large plots in shrub and grass dominated drylands using terrestrial estimated by fire management personnel. Int J Wildland Fire. Lidar and machine learning. Ecol Indic. 2018;84:793–802. 2012;21(3):219–29. 111. Hallema, D.W., Sun, G., Caldwell, P.V., Norman, S.P., Cohen, 124. Wang X, Wotton BM, Cantin AS, Parisien MA, Anderson K, E.C., Liu, Y., Bladon, K.D. and McNulty, S.G., 2018. Burned Moore B, et al. cffdrs: an R package for the Canadian forest fire forests impact water supplies. Nature communications, 9(1), danger rating system. Ecol Process. 2017;6, 5(1). p.1307. 125. Thekdi S, Aven T. An enhanced data-analytic framework for inte- 112.� WeiY,ThompsonMP, ScottJH, O’Connor CD, Dunn CJ. grating risk management and performance management. Reliab Designing operationally relevant daily large fire containment Eng Syst Saf. 2016;156:277–87. strategies using risk assessment results. Forests. 2019;10(4):311 Prescriptive analytics model to develop optimal response strategy based on quantitative risk assessment and potential Publisher’sNote Springer Nature remains neutral with regard to control locations; expands upon Wei et al. [85]. jurisdictional claims in published maps and institutional affiliations. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Current Forestry Reports Springer Journals

Risk Management and Analytics in Wildfire Response

Download PDF
14 pages

Loading next page...
 
/lp/springer-journals/risk-management-and-analytics-in-wildfire-response-tAT84EIr6u

References (132)

Publisher
Springer Journals
Copyright
Copyright © 2019 by The Author(s)
Subject
Environment; Sustainable Development; Environmental Management; Nature Conservation; Forestry; Forestry Management; Ecology
eISSN
2198-6436
DOI
10.1007/s40725-019-00101-7
Publisher site
See Article on Publisher Site

Abstract

Purpose of Review The objectives of this paper are to briefly review basic risk management and analytics concepts, describe their nexus in relation to wildfire response, demonstrate real-world application of analytics to support response decisions and orga- nizational learning, and outline an analytics strategy for the future. Recent Findings Analytics can improve decision-making and organizational performance across a variety of areas from sports to business to real-time emergency response. A lack of robust descriptive analytics on wildfire incident response effectiveness is a bottleneck for developing operationally relevant and empirically credible predictive and prescriptive analytics to inform and guide strategic response decisions. Capitalizing on technology such as automated resource tracking and machine learning algorithms can help bridge gaps between monitoring, learning, and data-driven decision-making. Summary By investing in better collection, documentation, archiving, and analysis of operational data on response effectiveness, fire management organizations can promote systematic learning and provide a better evidence base to support response decisions. We describe an analytics management framework that can provide structure to help deploy analytics within organizations, and provide real-world examples of advanced fire analytics applied in the USA. To fully capitalize on the potential of analytics, organizations may need to catalyze cultural shifts that cultivate stronger appreciation for data-driven decision processes, and develop informed skeptics that effectively balance both judgment and analysis in decision-making. . . . Keywords Decision-making Data science Operations research Suppression effectiveness Introduction objectives—is thus an inescapable aspect of wildfire manage- ment [1–3]. This framing of risk makes clear the need to explic- Wildfire presents risks to communities, landscapes, and water- itly address uncertainty, which can manifest in various types and sheds and to those who respond to and attempt to manage it. stem from various sources (e.g., fire weather, suppression re- Risk—broadly defined as the effect of uncertainty on source productivity) [4]. Further, this framing makes clear that This article is part of the Topical Collection on Fire Science Management * Matthew P. Thompson John S. Hogland matthew.p.thompson@usda.gov john.s.hogland@usda.gov Yu Wei Human Dimensions Program, Rocky Mountain Research Station, yu.wei@colostate.edu USDA Forest Service, Fort Collins, CO, USA David E. Calkin david.e.calkin@usda.gov Department of Forest and Rangeland Stewardship, Warner College of Natural Resources, Colorado State University, Fort Collins, CO, Christopher D. O’Connor USA christopher.d.oconnor@usda.gov Human Dimensions Program, Rocky Mountain Research Station, Christopher J. Dunn USDA Forest Service, Missoula, MT, USA chris.dunn@oregonstate.edu Nathaniel M. Anderson Department of Forest Engineering, Resources & Management, nathaniel.m.anderson@usda.gov College of Forestry, Oregon State University, Corvallis, OR, USA Curr Forestry Rep (2019) 5:226–239 227 risk is a value-laden concept predicated on defined objectives long history of application of science to support wildfire and that objectives can be positively or negatively impacted response decisions, with much of the effort focusing on (e.g., fire can enhance or degrade forest health). predicting expected wildfire behavior given existing Risk management (RM)—a set of coordinated activi- fuels, topography, and forecasted weather [e.g., ties to direct andcontrolanorganizationwithregard to [19–21]. Continued advances in spatial risk assessment risk—has become somewhat of an organizing framework methodologies provide additional information on the po- for wildfire management, with applications ranging from tential consequences to highly valued resources and as- programmatic budgeting to fire prevention, fuel reduction, sets, which can help guide fire management prioritiza- community planning, and broader topics such as perfor- tion and decision processes [22, 23]. Further, re- mance, communication, and governance [1, 5–12, 13� ]. searchers are developing tools to improve situational Wildfire management is rich with opportunities to apply awareness and operational responder safety [24–26]. and refine RM acumen—organizations around the globe However, reviews from around the globe on decision implement RM practices as a matter of routine. As one support and fire modeling collectively point to a lack example, the Australasian Fire Authorities Council of systems that provide empirically credible and opera- adopted International Standard 31000 Risk tionally relevant information on the effectiveness of al- management—principles and guidelines [1]as a guidepost ternative suppression strategies and tactics [27–29]. for all firefighting operations [14]. As another, the USDA For the purposes of this review, three interrelated principles Forest Service describes RM as a required core competen- of RM are particularly relevant: that RM is part of decision- cy for fire managers, and promulgates a RM protocol to making; that RM is based on the best available information; guide assessment, analysis, communication, decision- and that RM facilitates learning and continual improvement. making, review, and learning [15]. RM can help these The emphasis on decisions recognizes that performance is organizations increase the likelihood of achieving objec- inexorably tied to decisions. As Blenko et al. [30] note, and tives, establish a reliable basis for decision-making and with clear analogies to wildfire management, “An army’ssuc- planning, efficiently allocate and use resources, improve cess depends at least as much on the quality of the decisions its operational effectiveness and safety, and improve organi- officers and soldiers make and execute on the ground as it zational learning [1–3]. does on actual firefighting power.” The emphasis on decisions Here, we limit our review to evaluating RM concepts is also based on the fact that in an uncertain world, bad out- and principles in the context of wildfire response, i.e., comes can result from well-made decisions, and vice versa the development of a response strategy and its opera- [31], thus the need to emphasize the quality of decisions, tional execution over the duration of an active fire in- and further to understand how characteristics of decisions cident from detection to containment. Strategies can and the environment in which they are made may influence range from full suppression to managing for ecosystem decision quality [32]. benefit, depending on a variety of factors like relevant The emphasis on best available information supports policies, land ownership patterns, potential socioeco- risk-informed decision-making and reinforces decision nomic and ecological impacts, fire growth potential, quality. Three general challenges here are generating and availability of resources. Important components of relevant and actionable information, making that infor- response strategies include mobilizing/demobilizing fire mation available to decision makers, and convincing management resources, allocating and assigning re- them to seek it out and incorporate it into their decision sources to various tasks (e.g., line construction, structure processes [33, 34]. Key information necessary to make protection, mop-up), and monitoring and updating strat- a high-quality strategic response decision is the differ- egies in response to changing conditions. As the com- entiation of response alternatives on the basis of relative plexity, duration, or size of fires increases, response safety, effectiveness, and efficiency [31], and these ele- strategies may increasingly entail blend direct and indi- ments are a primary focus of this paper. Factors such as rect tactics, mobilize a greater amount and diversity of fire weather, fuel type, and total amount of resources ground and aerial resources, and require coordination of have been shown to play a role, but relationships be- a wide variety of activities such as locating drop points tween actual suppression activities and the effectiveness and conducting burnout operations [16]. of suppression resources remain unclear [35–41]. These incident response decisions can be complex, Plucinski [42�� , 43�� ] comprehensively reviewed the uncertain, time-pressured, and require balancing state of knowledge regarding suppression effectiveness tradeoffs across many dimensions (e.g., fire impacts, and identified multiple prominent gaps including the suppression expenditures, and public and responder effectiveness of different suppression resources used in safety), which highlights the need for structured and different suppression techniques, operational fire behav- timely decision support [3, 16–18]. Indeed, there is a ior limits on suppression, use and productivity of all 228 Curr Forestry Rep (2019) 5:226–239 types of suppression resources, how and whether sup- with predictable decision biases and possible operational pression resources can work synergistically, the impact inefficiencies, and gaps in fully putting the three key of fuel management on suppression effectiveness, and RM principles into practice. the impact of protective actions on structure loss. A central thesis of this paper is that a stronger emphasis on In practice, the limited scientific evidence and deci- data-driven decisions and analytics would help bridge those sion support functionality for generation and evaluation gaps. To that end, here, we aim to expose readers to principles of response alternatives mean that these decisions are and insights by drawing from the analytics literature with rele- largely left to expert judgment and intuition. Decisions vance to wildfire management and emergency response. We left to expert judgment can often be excellent, but the also synthesize relevant fire literature from around the globe, dynamic and unpredictable nature of the fire environ- primarily drawing from studies published by authors in North ment coupled with the lack of structured feedback on America, the Mediterranean, and Australia. operational effectiveness likely leads to compromised Ultimately, we argue for a paradigm based on stronger performance of even expert decision makers [44, 45]. adoption of data-driven decisions in fire management that For example, studies from Australia and the USA have we colloquially refer to as “Moneyball for fire” [63], in- demonstrated that fire managers as well as emergency spired by the innovative use of advanced data analytics in managers are susceptible to a number of cognitive errors professional baseball and other sports (Box 1). The im- and biases, and may exhibit wide variation in risk pref- provements in sports analytics started from the ability to erences and judgment [14, 46–52]. Further, empirical conduct more complex analyses of recorded performance observations suggest possible inefficiencies in operations data. Real-time tracking in sports evolved from these ear- due to ineffective or excessive use of resources [53–60]. lier successes, which has opened the door for more anal- Lastly, the emphasis on learning and continual im- ysis, insight, and innovation, and has fundamentally trans- provement relates to developing systems to create and formed the games in unexpected ways (see https://www. transfer knowledge, as well as systems of accountability ted.com/talks/rajiv_maheswaran_the_math_behind_ to monitor performance and correct or reinforce behav- basketball_s_wildest_moves). Although fire management ior to achieve better outcomes over time. Here, the be- organizations collect a considerable amount of data havior of the organization rather than the manager is of related to wildfires, robust data on fire response and even greater importance, requiring concrete steps to de- suppression resource performance is lacking [28, 42�� , velop learning processes that generate, collect, interpret, 43�� , 59]. Thus, the core analogy is not necessarily and disseminate information [61]. One challenge is about making real-time strategic adjustments in response around the organization’s capabilities for improving data to a changing fire environment or a changing adversary’s collection, developing performance indicators, and game strategy, but rather the gains in performance from adopting new technologies and innovations. Meeting preparatory work through investment in real-time moni- this challenge has proven difficult. In the USA, for ex- toring and analysis that lead to a better informational basis ample, federal agencies have been criticized for lacking to support strategic planning through to real-time response sufficient data fidelity and reliability along with limited decisions. analytic capability to understand effectiveness or to in- The remainder of the paper is structured as follows. form decision-making [62]. First, we briefly introduce some core analytics concepts Plucinski [43�� ] succinctly summarized the state of and an analytics management framework, along with affairs, stating that there remain many gaps in the cur- some observations on implementing an analytics agenda rent understanding of suppression effectiveness, that the within organizations. Next, we offer our perspective on data to fill these gaps are not routinely recorded, and the nexus between RM and analytics in wildfire re- that adoption of technologies such as resource tracking sponse, present a stylized figure linking a RM decision will be essential to capture more and better data cycle to the three main types of analytics, and describe streams. In other words, limited collection, documenta- potential application of descriptive, predictive, and pre- tion, archiving, and analysis of operational data on re- scriptive analytics in wildfire. We attempt to ground sponse effectiveness over time has inhibited systematic some of these ideas by relating real-world application learning. This, in turn, has limited meaningful and mea- of analytics to support strategic wildfire decisions, using surable expansion of the knowledge base on the effec- examples with which the authors have direct experience tiveness of alternative response strategies and tactics to in the western USA. We present these real-world exam- inform wildfire response decision-making. The net re- ples in light of the aforementioned analytics manage- sults are bounded operational utility of decision aids ment framework. Lastly, we discuss future opportunities which may lead to disproportionate reliance on expert and challenges for the fire management and science intuition over scientific and organizational evidence, communities. Curr Forestry Rep (2019) 5:226–239 229 Box 1 Timely provision of analytics can be a key driver of success in many areas, and real-time analytics have been developed “Moneyball” for Fire: Lessons from the Sports Analytics Revolution for a range of time-sensitive applications including financial Sports organizations around the world are leveraging analytics and market trading, military operations, smart electrical grids, in- evidence-based management to improve performance. What lessons from the sports analytics revolution can be applied to wildfire telligent transportation systems, and, most relevant, emergen- management? cy response [70, 71]. The increasing use of big data in analyt- Popularized by the book Moneyball: The Art of Winning an Unfair Game ics is often characterized in terms of the volume, variety, and by Michael Lewis [63] (and the movie of the same title), the idea of velocity of data, and technologies are emerging that can rap- “moneyball” is simply the adoption of data-driven decision-making to improve sports performance. Often the brilliance of sports analytics is idly ingest and analyze large quantities of data from sources as in its simplicity. In baseball, for example, a key insight was to evaluate variable as mobile phones and call records, social media ac- players on the basis of their on-base percentage (OBP) rather than their tivity, wearable devices, satellites, remote sensors, and batting average, as it was demonstrated that OBP was a better predictor crowdsourcing platforms [72]. Location-specific data is criti- of ability to score runs. In basketball, it was the realization that the expected number of points per shot was higher for three-point shots cally important for emergency response, for instance, to track than almost all other two-point shots, apart from those taken very close first responders with respect to changing conditions and to the basket. emerging threats [73]. Machine learning in particular seems In a contentious scene in the movie, a scout expresses his dissatisfaction to be growing in emergency response application potential with the analytics approach, stating that it is not possible to put a team together with a computer, and that science could never replace his due its ability to assimilate multiple sources and data types experience and his intuition. This is, of course, a false narrative—the and to generate insights from complex and dynamic environ- whole point of sports analytics is that it is not an either/or situation—it ments. Hong and Akerkar [74] comprehensively review ma- is intended to complement, not replace, expert judgment. Furthermore, chine learning approaches for emergency response, for a range coaches still need to make game-time adjustments, and players still need to execute. Analytics can help teams better prepare for strategic of emergencies including earthquakes, hurricanes, floods, decisions and their execution. landslides, and wildfires, and for a range of tasks including Three Main Types of Analytics Applied to Personnel Decisions in event prediction, early warning, event detection and tracking, Sports, and Analogies to Fire and situational awareness. Real-time analytics are a core com- Player acquisition ➔ Suppression resource acquisition Game strategy ➔ Wildfire response strategy ponent of a broader effort to capitalize on state-of-the-art big Training (fitness, game situations) ➔ Training (fitness, fire simulations) data analytics and advanced technology to provide improved Importantly, these three personnel decisions build upon and are insights for accurate and timely emergency response decision- dependent upon each other. For example, teams select personnel and making [75]. design their training regimen in order to successfully implement the most effective strategies. If not already apparent, key elements of embracing Key Lessons analytics are investing in better data and better science. Engaging experts and analysts to work together to solve organizational Of course, the notion of needing better data and science problems to support wildfire response decisions is not new [27, Defining approachable and understandable analytics Investing in more and better data 28, 31, 29, 42�� , 43�� , 59]. What is new, arguably, is Keeping the human element front and centerSources: [64, 65] embedding these issues within a coherent, principles- based framework that recognizes better data as but one of many steps towards improved decision-making and performance, i.e., an analytics management framework. Why Analytics? Key drivers of analytics success include clear goals, focused problems to solve, quality data from multiple We organize our discussion around basing decisions on rigor- sources, multidisciplinary analytics teams, accessible an- ous analysis and analytic insight over intuition, and on the key alytics systems, data translators, collaborative decision- linkages between analytics, decision-making, and organiza- making processes, end users as advocates, and iteration tional performance [66� ]. Analytics shares a common under- and continuous improvement [64]. Modern analytics is lying purpose with operations research and management sci- also being driven by new technologies, especially ad- ence, improving business operations and decision-making vances in computer science related to analyzing large, through the utilization of information, quantitative analysis, dynamic datasets in real time, but new technologies can and technology [67]. In general, data-driven decisions tend fail to have impact if they are not deployed in an ef- to be better ones, and organizations with comparatively stron- fective organizational framework. ger analytics capabilities tend to outperform their peers and Table 1 presents the nine components of an analytics man- competitors [68, 69� ]. Analytics can be a tool to facilitate agement framework, which can also be thought of as compris- improved decision-making, to measure performance, and ing an iterative cycle. The framework makes clear required even to measure improvements in performance due to adop- core competencies of a successful analytics program: strategic— tion of analytics-based management. planning for how data will be used to help solve 230 Curr Forestry Rep (2019) 5:226–239 Table 1 Analytics Management Strategic Organizational goal Prioritize goal(s) the organization seeks to achieve Framework 2019 Ben Shields, MIT Sloan School of Problem Define specific problem(s) that align with organizational goal(s) Management Data Identify the data needed to solve key problems Technical People Employ people to direct and manage analytics work Process Capture, manage, model, analyze, and visualize data Technology Adopt technologies to enable analytics work Managerial Communication Translate analytical insights into actionable recommendations for key stakeholders Decision-making Use analytics insights in the decision-making process Iteration Track and improve upon the decision problems and achieve organizational goals; technical— 2 and 3 summarize some core themes of each topic, organizing the people, processes, and technologies required respectively. As we see it, the nexus of both topics to manage and analyze data; and managerial — includes developing fluency with uncertainty and prob- communicating data, applying it in decision-making, and ability, emphasizing structured decision-making, making using it for continuous improvement [64]. This framework a commitment to generate and use the best available emphasizes the broader connections to people and process information, monitoring, and iteratively improving these and even culture, as well as the path dependency of data to core competencies over time. Furthermore, this nexus is insight to value [66� ]. definedbyanemphasisonpeopleand culture. For our Translating analytics insight into action requires more purposes here, we can contextualize and distill the nex- than simply setting up data collection systems connected us as follows: providing more and better operationally to a team of data analysts; embracing analytics may require relevant information on the safety and effectiveness of abroader “data-driven cultural change” predicated on exis- suppression strategies and tactics, the formal and trans- tence of an analytics strategy, strong senior management parent use of that information by fire managers in de- support, and careful change management initiatives [76�� ]. cision processes, and the comprehensive tracking of de- That is, the value of data analytics comes not just from the cisions and actions in relation to strategic response ob- technologies that enable it but also from the organizational jectives and fire outcomes. shifts in behavior and enhanced capabilities for strategic insight and performance measurement [77]. The underpin- ning of this shift is an acknowledgement that analytics is Box 2 needed in addition to expert judgment and experience. What Is Risk Management? Similarly, Davenport [78] stresses that the right technology Risk management (RM) is a set of coordinated processes and activities is but one aspect of a successful analytics initiative; the right that identify, monitor, assess, prioritize, and control risks that an organization faces. focus, the right people, and the right culture are also essen- What Are the Different Levels of Risk Management? tial. A final caveat, perhaps the most important one, is that Enterprise ➔ Strategic ➔ Operational ➔ Real-time better data will not necessarily lead to better decisions. Shah What Are the Main Principles of Risk Management? et al. [79� ]caution that “unquestioning empiricists” who Integrating RM into all organizational processes, including decision-making trust analysis over judgment may be no better than “visceral Explicitly accounting for uncertainty decision makers” who go exclusively with their gut. The Addressing problems in a systematic, structured, and timely manner challenge is to develop “informed skeptics” who possess Basing decisions on the best available information strong analytic skills and who can effectively leverage both Tailoring processes to context, and accounting for human and cultural factors judgment and analysis in decision-making. Human judg- Promoting transparency and inclusiveness ment therefore remains front and center in the context of Being dynamic, iterative, and responsive to change embracing analytics to improve decision-making. Facilitating continual improvement How Is Risk Management Different from “Business as Usual”? Informal ➔ Formal The Risk Management and Analytics Nexus in Wildfire Implicit ➔ Explicit Response Intuitive ➔ Analytical Reactive ➔ Proactive In this section, we outline concepts and applications Short-term perspective ➔ Long-term perspective Sources: [1–3] from RM and analytics to wildfire management. Boxes Curr Forestry Rep (2019) 5:226–239 231 Box 3 learning to assess what might happen in the future. This would include predictions of fire weather, fire behavior, and potential What is Analytics? control locations [80–82]. Prescriptive analytics use operation Analytics is the extensive use of data, statistical and quantitative analysis, research methods such as optimization and simulation to rec- explanatory and predictive models, and fact-based management to ommend efficient solutions. This would include assigning drive decisions and actions How Does Using Analytics Improve Performance? suppression resources to tasks such as asset protection or fire Data ➔ Insight ➔ Value line construction [83–85]. Note that some applications can What Are the Main Principles of Analytics? entail combining multiple types of analytics, for example, Treating fact-based decision-making as not only a best practice but also a comparison of observed fire size and impacts (descriptive) part of culture Recognizing the value of analytics, and making their development and with simulated size and impacts in the absence of suppression maintenance a primary focus (predictive) to estimate the productivity and effectiveness of Applying sophisticated information systems and rigorous analysis to a suppression operations [86]. Table 2 provides an illustrative range of functions set of examples of descriptive, predictive, and prescriptive Considering analytics to be so important it is managed at the enterprise level analytics in the context of wildfire response strategy and op- Avidly consuming data and seizing every opportunity to generate erational execution. information The usefulness of predictive and prescriptive analytics is Emphasizing the importance of analytics internally predicated on reliable descriptive analytics on fire and fire Making quantitative capabilities part of the organization’sstory Creating a workforce with strong analytical skills and considering it a key operations. Descriptive analytics can be used to validate pre- to organizational success dictive models, which in turn can be used to parameterize How Do Data-Driven Organizations Act Differently? prescriptive models. Developing a road map to enhance de- The first question a data-driven organization asks itself is not “what do we scriptive analytics is therefore critically important. Thompson think?” but “what do we know?” Decision makers move away from acting solely on hunches and instinct, et al. [31] and Plucinski [42�� ] outline operational data collec- and move away from citing data to support decisions already made tion needs, much of which is reliant on obtaining information Sources: [69� , 78] from fire crews. However, there is a broader horizon for data capture that could capitalize on advanced technologies like the internet of things and machine-to-machine communication. Figure 1 presents a styled RM decision cycle and its rela- These ideas are related to the Industry 4.0 initiative tionship to descriptive, predictive, and prescriptive analytics. (referencing a 4th industrial revolution); the goals of which Descriptive analytics use statistical methods such as statistical are to achieve higher levels of operational efficiency, produc- modeling and data mining to provide insight into what hap- tivity, and automation [88]. These ideas are already being pened in the past. This would include real-time and post hoc integrated into forest operations and supply chain manage- monitoring of suppression operations to gauge effectiveness ment, for instance, the installation of sensors and on-board and development performance measures related to resource computers onto harvest equipment to provide decision support use, productivity, and effectiveness [54, 56, 59]. Predictive and streamline operations [89]. Gingras and Charette [90]in- analytics use techniques such as forecasting and machine troduce a Forestry 4.0 initiative seeking to, among other Fig. 1 Stylized three-stage RM decision cycle and relationship to the three main types of analytics; adapted from [31, 67] 232 Curr Forestry Rep (2019) 5:226–239 Table 2 Examples of analytics in context of wildfire response strategy analytics that are either known knowledge gaps (descriptive) or that and operational execution, adopted from [16, 42�� , 87]. Italic text have had limited or no application in real-world contexts (predictive indicates predictive analytics highlighted in this paper that have recently and prescriptive) been developed and applied in the western USA. Bold text indicates Descriptive Predictive Prescriptive Fire weather Weather forecasts Control line construction, holding, and mop-up Fire size and shape Fire danger ratings (location, type, length, timing) Burn severity Burn probabilities Point protection (location, timing, type) Daily perimeter growth Fire intensity probabilities Logistical feature creation (e.g., safety zones, drop zones) Fire duration Fire arrival times Burnout operations Suppression expenditures Estimated containment time Resource ordering (type and amount) Resource use Estimated suppression costs Resource mobilization and demobilization Resource productivity Potential control locations Resource allocation to assignments Resource effectiveness Suppression difficulty Resource movement Safety zones and escape routes Resource productivity Resource availability things, improve communication networks in remote locations machines) could provide rich opportunities for improving predic- and enable real-time data exchange between operators and tive models. Machine learning techniques have some advantages decision centers. Through combination of technologies like over traditional approaches like generalized linear models due to radio frequency identification, remote sensing through satel- their ability to handle complex problems with multiple lites and drones, and adding sensors and wireless communi- interacting elements, and, increasingly, big data streams [99, cations to suppression machinery, it may be possible to simi- 100]. They can also require large amounts of data, some of which larly create a Fire 4.0 initiative. may come from modern data capture technology, but accounting In the context of supporting strategic and tactical response for extremely rare events in models may also benefit from inter- decisions, predictive analytics might provide the most value. national collaboration and data sharing. Clearly fire behavior predictions are essential, and the literature Table 4 presents an illustrative set of machine learning appli- is rich with descriptions of fire behavior models, their applica- cations in wildfire. One observation is the lack of models on fine- tions, and their limitations [19, 20, 91–93]. Here, we focus in- scale suppression operations, which presents a possible opening stead on analytics more specifically related to operations, which for future machine learning applications when such data are in broad strokes can help assess factors related to safety and available. Future applications could, for example, train models effectiveness, for instance, determination of locations to avoid to predict where fire managers will build line, or where operators for safety reasons or of locations where control opportunities will locate water and retardant drops. At larger scales, models might be most successful. Researchers have sought to incorpo- could predict emerging resource needs for prepositioning of re- rate factors such as fire intensity, heat exposure, safety zones, sources and dispatch prioritization. snag hazard, egress, accessibility, and mobility into these analyt- Because of nested dependencies, prescriptive analytics are ics. Table 3 briefly summarizes some of these recent studies. currently perhaps the most constrained and have the least di- As described earlier, the growing use machine learning tech- rect connection to on-the-ground fire management operations niques (including classification and regression trees, artificial [27, 29]. Prescriptive modeling often serves the role of inte- neural networks, evolutionary computation, and support vector grating descriptive data and predictive results through Table 3 Illustrative set of studies Topic Source(s) that generate predictive analytics to guide and inform safe and Suppression difficulty index based on flame length, heat per unit area, fire line productivity, road [17, 26] effective response operations and trail density, accessibility, and mobility Fire control probability surface used to predict potential control locations [82� ] Spatiotemporal patterns of post-fire snag hazard [94] Travel impedance maps to map efficient egress routes from crew locations to safety zones [95] Determination of safety zone suitability as a function of size, geometry, and height of surrounding [96] vegetation Travel time prediction in variable terrain based on GPS tracked data [97] Expected aviation accident rates by aircraft type and workload [98] Curr Forestry Rep (2019) 5:226–239 233 Table 4 Illustrative set of studies Topic Source(s) that apply machine learning techniques to wildfire Modeling the effect of suppression on large fire spread [39] Predicting exposure of populations to fine particular matter during wildfires [101] Predicting post-fire debris flow events [102] Predicting fire occurrence and burnt area [103–106] Modeling initial attack success [107] Modeling wildfire spread dynamics [108] Estimating vegetation biomass, cover, and other characteristics [109, 110] Modeling impact of wildfires on river flows [111] mathematical equations to reflect operational constraints and line, based on manager preferences regarding damage, cost, to optimally achieve management objectives. By design, com- and firefighter safety [112� ]. At present, the descriptive model plicated decision rules and logic could be built in the model to is informing agency discussions regarding development of guide the search process for optimal solutions, and uncertainty key performance indicators, the predictive model is being ac- can be accounted for with probabilistic rather than determin- tively used to support real-time decision making and strategic istic frameworks. Such a system could be difficult for humans planning throughout the western USA, and the prescriptive to track due to its complexity and potential tradeoffs between model is a prototype. alternative solutions. However, parameterizing a prescriptive Figure 2 illustrates these interrelated concepts from a post model with unreliable data can lead to infeasible, unreason- hoc analysis of the 2018 Ferguson Fire (39,200 ha) that able, or suboptimal decisions. Once a decision is suggested by burned in the Sierra National Forest, Stanislaus National the system, it is often difficult for managers to intuitively Forest, and Yosemite National Park in California, USA. understand the quality of the decision and all the reasons for Panel a displays the fire perimeter in relation to locations of it to be chosen due to unknown causal relationships among constructed fire line, and summarizes fire line effectiveness statistically weighted variables, and especially due to data per the framework of Thompson et al. [113� ]. Panel b displays quality. A role for prescriptive models, in the near future at the fire perimeter in relation to an underlying fire control least, might be for exploring tradeoffs, generating efficient probability surface [82� ], with a histogram showing the per- frontiers, identifying hypotheses to potentially improve deci- centage of fire perimeter length in one of five control proba- sion quality, and performing scenario and sensitivity analyses bility categories. Panel c displays results of a spatial optimi- [e.g., 85, 112� ]. Using prescriptive models to directly support zation model developed by Wei et al. [112� ] that outlines one fire management decisions not only requires improved data possible response strategy based on landscape risk, potential quality and model design but may also require that the models control locations, and manager-specified preferences. are easy to understand to decision makers and are not viewed Table 5 displays how the fire assessment and planning pro- as a “black box” with unknown and mysterious inner cesses using predictive modeling of potential control locations workings. fits into the analytics management framework. The predictive model developed by O’Connor et al. [82� ] has been widely Analytics Demonstration and Real-World Application: deployed in the USA for pre-fire planning applications as well Potential Control Locations and Fire Line as for real-time decision support on more than two dozen Effectiveness wildfires during the 2017 and 2018 fire seasons. As of this writing, fire ecologists and analysts have delivered potential control location map products to National Forest System staff, To further ground these analytics concepts in fire operations, we consider the construction of fire containment line and pres- partners, and stakeholders on over thirty landscapes across the ent recent research and application on that topic. A descriptive western USA. The pairing of such information with quantita- model of fire line effectiveness analyzes fire perimeter and tive risk assessments and the delineation of strategic response operations data to describe where line was built, whether it Tragically, there was a tree strike fatality on the Ferguson Fire. The engaged the fire, and if so, whether it was effective at stopping Corrective Action Plan (https://wildfiretoday.com/documents/ fire [113� ]. A predictive model analyzes historical fire perim- CorrectiveActionPlanHughes.pdf) recommended evaluation of how eters in relation to environmental and landscape characteris- changing environmental conditions, such as extensive tree mortality, are tics, and trains a machine learning algorithm to estimate the being factored into response strategies and tactics. We believe operationally relevant analytics that speak to firefighter safety could help meet that need, and probability of a given location being a suitable control line to ultimately reduce the occurrence of such tragedies in the future. Nothing in this stop fire [82� ]. A prescriptive model recommends strategies to brief case study should be construed as commentary on the decisions or actions managers that combine the most suitable locations to construct taken on the Ferguson Fire. 234 Curr Forestry Rep (2019) 5:226–239 Fig. 2 Case study analysis of the 2018 Ferguson Fire, showing a model run with a user-specified set of preferences regarding cost and descriptive analytics of fire line effectiveness, b predictive analytics of safety. For panel a: Tr = ratio of total length of line to perimeter; Er = potential control locations in relation to the final perimeter, and c ratio of engaged line to total line; HEr = ratio of held line to engaged line; prescriptive analytics that recommend combinations of control locations HTr = ratio of held line to total line [59, 113� ] to create overarching response strategies, here showing the results of one zones are not only enhancing cross-boundary planning pro- Discussion cesses but also helping managers achieve desirable fire out- comes that protect assets while enhancing ecosystem health A core idea from this paper is the need to develop a broader [114]. Following an intensive strategic wildfire risk planning analytics strategy for wildfire management, and to infuse an- process in the spring of 2017, the Tonto National Forest in alytics into planning, decision, and learning processes. The central Arizona has managed six large wildfires in accordance aims of such an effort would be to align actions with manage- with pre-identified strategic wildfire response zones that help ment objectives, enhance transparency and accountability, im- to align incident response objectives with land and resource prove decision support, create learning opportunities, and ul- management planning direction. The 2017 Pinal, Highline, timately improve response safety and effectiveness [115]. and Brooklyn fires were managed for ecological restoration, Doing so would help fire management organizations redeem asset protection, and wildfire process maintenance objectives, some of their core RM responsibilities, such as generating respectively. During the 2018 fire season, when milder fire better information to support risk-informed decisions, and weather conditions were prevalent, the Bears, Daisy, and continually improving. Cholla fires were each managed for ecological restoration We recognize there are a broader range of important and future risk reduction objectives consistent with pre- incident-related decisions than that considered here, such as identified strategic wildfire response zones and corresponding routing detection aircraft, stationing and prepositioning sup- fire response objectives. pression resources, and transferring or reassigning resources Table 5 Analytics Management Strategic Organizational goal Safe and effective response Framework as applied for predictive model of potential Problem Identifying locations on landscape where responder hazard is lower and control locations, adapted from likelihood of control success is higher [82� , 114] Data Historical fire perimeters, landscape and environmental attributes Technical People Fire ecologists, modelers, analysts, local fire managers Process Predictive modeling combined with local expert judgment Technology Machine learning algorithm, GIS software Managerial Communication Workshops with local managers and stakeholders Decision-making Planning: guided designation of fire containers and strategic response zones; incident response: guided choices of line construction location Iteration Refinement of model, feedback from managers Curr Forestry Rep (2019) 5:226–239 235 to different regions or different fires [29, 116–118]. Some of metrics of performance and risk. Embracing analytics could these decisions are inexorably linked to response strategies, have a lot to offer regarding issues of overreliance on expert for example, allocating scarce resources between fires is pre- judgment and decision quality. This includes decomposing the mised on some understanding of how resources would be used problem into manageable sub-problems and developing oper- and with what degree of efficacy if assigned to different fires. ationally relevant decision aids. Through more proactive ac- We opted to focus on incident-level response strategies be- quisition and use of more reliable and more trustworthy infor- cause these can be high-impact decisions, and because we mation, fire management organizations could help dampen believe there is great potential for stronger adoption of RM some of the biases described earlier and improve fire manage- and analytics principles within this decision context. ment decision-making [33, 34]. Looking to the future, fire management organizations Although challenges exist regarding capture, interpre- will need to modernize data collection and analysis sys- tation of fire operations data, and advances in scientific tems associated with fire activity and wildfire suppression understanding, the bigger challenges may well be orga- operations. The emphasis will be on measuring and mon- nizational and cultural. Effectuating a data-driven cultur- itoring quantitative performance metrics to evaluate oper- al change is a known barrier to wider spread adoption ational capabilities, safety, efficiency, and effectiveness. of analytics [76�� ]. How to convince fire managers that Digital technology integration, including automated re- enhanced monitoring and collection (i.e., descriptive an- source tracking, is likely to be a key component of new alytics) will not be used for “Monday morning data collection systems. Operations research and industrial quarterbacking,” and furthermore, how to demonstrate engineering, combined with data science, are perfectly the value of predictive and prescriptive analytics to suited to this task. These disciplines offer well-defined managers will be some of the fire science community’s scientific approaches for performance measurement, pro- challenges moving forward. Stronger collaboration be- cess engineering, logistics, and operations management, tween the fire science and management communities including health and safety. As stated earlier, similar mod- could help here. As summarized by Martell [87]: pre- ernization is already occurring in the forest sector under a dictive analytics specialists cannot develop useful pre- variety of precision forestry initiatives. dictive models unless they collaborate with fire manage- That many of these technologies have been around for ment organizations that are willing to share their prob- years does raise questions of why fire management organiza- lems and data, and prescriptive analytics specialists can- tions have not already turned to them. For example, re- not develop useful prescriptive models unless they col- sponders might find tracking technology that allows close laborate with predictive analysts and with fire manage- monitoring of all of their actions to be invasive, or key deci- ment organizations that are willing to share their prob- sion makers may fear increased liability risks. Organizations lems and test their models. To the degree that analytics are likely to face issues of data governance, privacy and secu- results in demonstrably better outcomes in terms of op- rity, leading to data policy questions such as which data to erational safety, efficiency, and effectiveness, it is likely make available, to whom, through what channels, and for to help incentivize its own adoption over time. what purposes [72, 73, 75]. Limited abilities to effectively The emphasis on people and culture in analytics frame- manage and standardize the complex data streams collected works also highlights the need for stronger collaboration with from various sources decrease the utility and increase the cost social scientists. There are opportunities to convert experien- of data capture [119]. Another barrier is the need to build tial evidence into organizational and scientific evidence (i.e., workforce capacity to effectively use data science [120]. In informing the development of meaningful performance met- addition to addressing data concerns, organizations may not rics), to improve knowledge exchange, to identify potential have supportive learning environments in place, where barriers to meaningful organizational change, and to design learners have psychological safety to express disagreement communication strategies in light of how information is cur- or own up to mistakes, where opposing ideas and competing rently shared within fire response networks [121, 122]. outlooks are welcomes, and where taking risks is appreciated Beyond rolling out the analytics management framework [61]. for an expanded set of fire operations, we see a number of We see a future with increased automation and prescriptive productive pathways forward. As a possible interim strategy, analytics recommendations; however, fire management will the fire community could seek to more comprehensively cat- always be a question of human judgment. Every fire event alog lessons learned [86] and develop more decision support has unique circumstances and potentially unresolvable uncer- systems based on elicitation of expert judgment [123]. On the tainties, rendering completely prescriptive approaches infeasi- technical side, analysts can capitalize on the growing use of ble, hence the focus on decision makers. In analytics, many machine learning techniques in forest and fire modeling. The but not all aspects of expert judgment and organizational widespread use of open-source technology also presents op- wisdom may track closely with measurable and predictable portunities for sharing data and code [124]. On the 236 Curr Forestry Rep (2019) 5:226–239 Summit & 4th Human Dimensions of Wildland Fire Conference. organizational side, performance management systems could pp. 92–113, Boise, ID. be redesigned with risk in mind to better account for non- 3. Thompson, M.P., MacGregor, D.G. and Calkin, D., 2016. Risk deterministic relationships between decisions and outcomes management: core principles and practices, and their relevance [125]. Performance measurement could also expand to a to wildland fire. Gen. Tech. Rep. RMRS-GTR-350. Fort Collins, CO: US Department of Agriculture, Forest Service, Rocky systems-based approach that conceptualizes effective re- Mountain Research Station. 29 p. sponse not just in operations terms but also factors such as 4. Riley K, Thompson M. An uncertainty analysis of wildfire model- evacuations, road closures, and public information [13� ]. ing . Natural hazard uncertainty assessment: modeling and deci- sion support. Monograph. 2017;223:193–213. 5. Ager AA, Evers CR, Day MA, Preisler HK, Barros AM, Nielsen- Pincus M. Network analysis of wildfire transmission and implica- Conclusion tions for risk governance. PLoS One. 2017;12(3):e0172867 Beaver. 6. Calkin DE, Cohen JD, Finney MA, Thompson MP. How risk The translation of scientific ideas and insights into prac- management can prevent future wildfire disasters in the tice can be essential for improving decision quality and wildland-urban interface. Proc Natl Acad Sci. 2014;111(2):746– cultivating stronger RM acumen. A number of hurdles remain: increasing investment in data capture and analysis 7. Chung W. Optimizing fuel treatments to reduce wildland fire risk. Curr For Rep. 2015;1(1):44–51. technology, dealing with data privacy and security issues, 8. Fischer AP, Spies TA, Steelman TA, Moseley C, Johnson BR, developing better models and decision support tools, and Bailey JD, et al. Wildfire risk as a socioecological pathology. catalyzing organizational and cultural change. Despite Front Ecol Environ. 2016;14(5):276–84. these challenges, we believe there are ample opportunities 9. Hesseln H. Wildland fire prevention: a review. Curr For Rep. 2018;4(4):178–90. for use of descriptive, predictive, and prescriptive analyt- 10. Mavsar R, Cabán AG, Varela E. The state of development of fire ics in fire. We provided one example of a fire line effec- management decision support systems in America and Europe. tiveness framework and its role in serving as a connective Forest Policy Econ. 2013;29:45–55. thread between the three types of analytics. Similar cycles 11. McCaffrey S. Community wildfire preparedness: a global state-of- of the analytics strategy could be pursued for mop-up, the-knowledge summary of social science research. Curr For Rep. 2015;1(2):81–90. aviation, burn out operations, and structure protection. 12. Steelman TA, McCaffrey S. Best practices in risk and crisis com- Developing core competencies with analytics and cultivat- munication: implications for natural hazards management. Nat ing a “Moneyball for fire” paradigm may help fire man- Hazards. 2013;65(1):683–705. agement organizations on their RM journey. 13.� Steelman T, Nowell B. Evidence of effectiveness in the Cohesive Strategy: measuring and improving wildfire response. Int J Wildland Fire. 2019;28(4):267–74 Extends conceptualization Funding Information Dr. Wei and Dr. Dunn received funding from the of wildfire response performance measurement to a systems- USDA Forest Service through Joint Venture Agreements. based perspective considering factors beyond operational effectiveness. Compliance with Ethical Standards 14. Penney G. Exploring ISO31000 risk management during dynamic fire and emergency operations in Western Australia. Fire. Conflict of Interest All authors declare they have no conflict of interest. 2019;2(2):21. 15. National Interagency Fire Center. 2019. Interagency Standards for Open Access This article is distributed under the terms of the Creative Fire and Fire Aviation Operations 2019. Chapter 5: USDA Forest Commons Attribution 4.0 International License (http:// Service Wildland Fire and Aviation Program Organization and creativecommons.org/licenses/by/4.0/), which permits unrestricted use, Responsibilities. Available at: https://www.nifc.gov/ distribution, and reproduction in any medium, provided you give appro- PUBLICATIONS/redbook/2019/Chapter05.pdf. Accessed 14 priate credit to the original author(s) and the source, provide a link to the Nov 2019 Creative Commons license, and indicate if changes were made. 16. Dunn CJ, Thompson MP, Calkin DE. A framework for developing safe and effective large-fire response in a new fire management References paradigm. For Ecol Manag. 2017;404:184–96. 17. O’Connor C, Thompson M, Rodríguez y Silva F. Getting ahead of the wildfire problem: quantifying and mapping management chal- lenges and opportunities. Geosciences. 2016;6(3):35. Papers of particular interest, published recently, have been 18. Thompson M, Bowden P, Brough A, Scott J, Gilbertson-Day J, highlighted as: Taylor A, et al. Application of wildfire risk assessment results to � Of importance wildfire response planning in the southern Sierra Nevada, �� Of major importance California, USA. Forests. 2016;7(3):64. 19. Plucinski MP, Sullivan AL, Rucinski CJ, Prakash M. Improving 1. International Organization for Standardization. ISO 31000:2018 the reliability and utility of operational bushfire behaviour predic- Risk Management. Geneva: ISO; 2018. tions in Australian vegetation. Environ Model Softw. 2017;91:1– 2. Beaver A 2015. Applying the ISO 31000:2009 risk management – principles and guidelines for improved risk management decision 20. Cruz MG, Alexander ME, Sullivan AL, Gould JS, Kilinc M. making. Proceedings of the 13th Annual Wildland Fire Safety Assessing improvements in models used to operationally predict Curr Forestry Rep (2019) 5:226–239 237 wildland fire rate of spread. Environ Model Softw. 2018;105:54– 19 Outlines a framework for evaluating suppression effective- 63. ness and synthesizes findings at flame and fire line scales. 21. Kalabokidis K, Ager A, Finney M, Athanasis N, Palaiologou P, 43.�� Plucinski MP. Contain and control: wildfire suppression effective- Vasilakos C. AEGIS: a wildfire prevention and management in- ness at incidents and across landscapes. Curr For Rep. 2019;5(1): formation system. Nat Hazards Earth Syst Sci. 2016;16:643–61. 20–40 Comprehensive review of studies of suppression effec- 22. Pacheco AP, Claro J, Fernandes PM, de Neufville R, Oliveira TM, tiveness and roadmap for enhanced operational data collec- Borges JG, et al. Cohesive fire management within an uncertain tion, and synthesizes research on effectiveness at incident and environment: a review of risk handling and decision support sys- landscape scales. tems. For Ecol Manag. 2015;347:1–17. 44. Shanteau J. Competence in experts: the role of task characteristics. 23. Molina JR, Moreno R, Castillo M, y Silva FR. Economic suscep- Organ Behav Hum Decis Process. 1992;53(2):252–66. tibility of fire-prone landscapes in natural protected areas of the 45. Kahneman D, Klein G. Conditions for intuitive expertise: a failure southern Andean range. Sci Total Environ. 2018;619:1557–65. to disagree. Am Psychol. 2009;64(6):515. 24. Jolly WM, Freeborn PH. Towards improving wildland firefighter 46. Wilson RS, Winter PL, Maguire LA, Ascher T. Managing wildfire situational awareness through daily fire behaviour risk assess- events: risk-based decision making among a group of federal fire ments in the US northern Rockies and northern Great Basin. Int managers. Risk Analysis: An International Journal. 2011;31(5): J Wildland Fire. 2017;26(7):574–86. 805–18. 25. Campbell MJ, Page WG, Dennison PE, Butler BW. Escape route 47. Hand MS, Wibbenmeyer MJ, Calkin DE, Thompson MP. Risk index: a spatially-explicit measure of wildland firefighter egress preferences, probability weighting, and strategy tradeoffs in wild- capacity. Fire. 2019;2(3):–40. fire management. Risk Anal. 2015;35(10):1876–91. 26. Rodríguez y Silva F, Martínez JRM, González-Cabán A. A meth- 48. Wibbenmeyer MJ, Hand MS, Calkin DE, Venn TJ, Thompson odology for determining operational priorities for prevention and MP. Risk preferences in strategic wildfire decision making: a suppression of wildland fires. Int J Wildland Fire. 2014;23(4): choice experiment with U.S. wildfire managers. Risk Anal. 544–54. 2013;33(6):1021–37. 27. Duff TJ, Tolhurst KG. Operational wildfire suppression model- 49. McLennan J, Holgate AM, Omodei MM, Wearing AJ. Decision ling: a review evaluating development, state of the art and future making effectiveness in wildfire incident management teams. J directions. Int J Wildland Fire. 2015;24(6):735–48. Conting Crisis Manag. 2006;14(1):27–37. 28. Filkov A, Duff T, Penman T. Improving fire behaviour data ob- 50. Calkin DE, Venn T, Wibbenmeyer M, Thompson MP. Estimating tained from wildfires. Forests. 2018;9(2):81. US federal wildland fire managers’ preferences toward competing 29. Martell DL. A review of recent forest and wildland fire manage- strategic suppression objectives. Int J Wildland Fire. 2013;22(2): ment decision support systems research. Curr For Rep. 2015;1(2): 212–22. 128–37. 51. Roberts PS, Wernstedt K. Decision biases and heuristics among 30. Blenko MW, Mankins MC, Rogers P. The decision-driven orga- emergency managers: just like the public they manage for? Am nization. Harv Bus Rev. 2010;88(6):54–62. Rev Public Adm. 2019;49(3):292–308. 31. Marcot BG, Thompson MP, Runge MC, Thompson FR, McNulty 52. Wernstedt K, Roberts PS, Arvai J, Redmond K. How emergency S, Cleaves D, et al. Recent advances in applying decision science to managing national forests. For Ecol Manag. 2012;285:123–32. managers (mis?) interpret forecasts. Disasters. 2019;43(1):88– 32. Spetzler C, Winter H, Meyer J. Decision quality: value creation 109. from better business decisions: John Wiley & Sons; 2016. 53. Hand M, Katuwal H, Calkin DE, Thompson MP. The influence of 33. Pfeffer J, Sutton RI. Evidence-based management. Harv Bus Rev. incident management teams on the deployment of wildfire sup- 2006;84(1):62. pression resources. Int J Wildland Fire. 2017;26(7):615–29. 34. Barends E, Rousseau DM, Briner RB. Evidence-based manage- 54. Holmes TP, Calkin DE. Econometric analysis of fire suppression ment: the basic principles. Amsterdam: Center for Evidence- production functions for large wildland fires. Int J Wildland Fire. Based Management; 2014. 2013;22(2):246–55. 35. Plucinski MP. Modelling the probability of Australian grassfires 55. Stonesifer CS, Calkin DE, Thompson MP, Stockmann KD. escaping initial attack to aid deployment decisions. Int J Wildland Fighting fire in the heat of the day: an analysis of operational Fire. 2013;22(4):459–68. and environmental conditions of use for large airtankers in 36. Beverly JL. Time since prior wildfire affects subsequent fire con- United States fire suppression. Int J Wildland Fire. 2016;25(5): tainment in black spruce. Int J Wildland Fire. 2017;26(11):919– 520–33. 56. Katuwal H, Dunn CJ, Calkin DE. Characterising resource use and 37. Collins KM, Price OF, Penman TD. Suppression resource deci- potential inefficiencies during large-fire suppression in the western sions are the dominant influence on containment of Australian US. Int J Wildland Fire. 2017;26(7):604–14. forest and grass fires. J Environ Manag. 2018;228:373–82. 57. Calkin DE, Stonesifer CS, Thompson MP, McHugh CW. Large 38. Finney M, Grenfell IC, McHugh CW. Modeling containment of airtanker use and outcomes in suppressing wildland fires in the large wildfires using generalized linear mixed-model analysis. For United States. Int J Wildland Fire. 2014;23(2):259–71. Sci. 2009;55(3):249–55. 58. Donovan GH, Prestemon JP, Gebert K. The effect of newspaper 39. Fernandes PM, Pacheco AP, Almeida R, Claro J. The role of fire- coverage and political pressure on wildfire suppression costs. Soc suppression force in limiting the spread of extremely large forest Nat Resour. 2011;24(8):785–98. fires in Portugal. Eur J For Res. 2016;135(2):253–62. 59. Thompson M, Lauer C, Calkin D, Rieck J, Stonesifer C, Hand M. 40. Tremblay PO, Duchesne T, Cumming SG. Survival analysis and Wildfire response performance measurement: current and future classification methods for forest fire size. PLoS One. 2018;13(1): directions. Fire. 2018;1(2):21. e0189860. 60. Plucinski MP, Pastor E. Criteria and methodology for evaluating 41. Cardil A, Lorente M, Boucher D, Boucher J, Gauthier S. Factors aerial wildfire suppression. Int J Wildland Fire. 2013;22(8):1144– influencing fire suppression success in the province of Quebec (Canada). Can J For Res. 2018;49(5):531–42. 61. Garvin DA, Edmondson AC, Gino F. Is yours a learning organi- 42.�� Plucinski MP. Fighting flames and forging firelines: wildfire sup- zation? Harv Bus Rev. 2008;86(3):109. pression effectiveness at the fire edge. Curr For Rep. 2019;5(1):1– 238 Curr Forestry Rep (2019) 5:226–239 62. Booz Allen Hamilton. 2014 Quadrennial fire review: final report; judgment” and argues for developing “informed skeptics” as 2015. Available online: https://www.forestsandrangelands.gov/ desired decision makers. QFR/documents/2014QFRFinalReport.pdf 80. Katuwal H, Calkin DE, Hand MS. Production and efficiency of 63. Lewis MM. Moneyball: the art of winning an unfair game. New large wildland fire suppression effort: a stochastic frontier analy- York: W. W; 2003. sis. J Environ Manag. 2016;166:227–36. 64. Shields B. Analytics management: business lessons from the 81. Monedero S, Ramirez J, Cardil A. Predicting fire spread and be- sports data revolution at MIT Sloan School of management. haviour on the fireline. Wildfire analyst pocket: a mobile app for March. 2019;19-20:2019. wildland fire prediction. Ecol Model. 2019;392:103–7. 65. Rein I, Shields B, Grossman A. The sports strategist: developing 82.� O’Connor CD, Calkin DE, Thompson MP. An empirical machine leaders for a high-performance industry. USA: Oxford University learning method for predicting potential fire control locations for Press; 2014. pre-fire planning and operational fire management. Int J Wildland 66.� Sharma R, Mithas S, Kankanhalli A. Transforming decision- Fire. 2017;26(7):587–97 Predictive analytics tool for mapping making processes: a research agenda for understanding the impact potential fire control locations, now widely implemented on of business analytics on organisations. Eur J Inf Syst. 2014;23(4): National Forest System lands in the western USA. 433–41 Reviews literature on role of business analytics in or- 83. van der Merwe M, Minas JP, Ozlen M, Hearne JW. A mixed ganizations, and emphasizes relationships between business integer programming approach for asset protection during escaped analytics, decision-making processes, and organizational wildfires. Can J For Res. 2014;45(4):444–51. performance. 84. Belval EJ, Wei Y, Bevers M. A stochastic mixed integer program 67. Mortenson MJ, Doherty NF, Robinson S. Operational research to model spatial wildfire behavior and suppression placement de- from Taylorism to Terabytes: a research agenda for the analytics cisions with uncertain weather. Can J For Res. 2015;46(2):234– age. Eur J Oper Res. 2015;241(3):583–95. 68. McAfee A, Brynjolfsson E, Davenport TH, Patil DJ, Barton D. 85. Wei Y, Thompson MP, Haas JR, Dillon GK, O’Connor CD. Big data: the management revolution. Harv Bus Rev. 2012;90(10): Spatial optimization of operationally relevant large fire confine 60–8. and point protection strategies: model development and test cases. 69.� LaValle S, Lesser E, Shockley R, Hopkins MS, Kruschwitz N. Big Can J For Res. 2018;48(5):480–93. data, analytics and the path from insights to value. MIT Sloan 86. Rodríguez y Silva FR, González-Cabán A. Contribution of sup- Manag Rev. 2011;52(2):21 Review of how organizations are pression difficulty and lessons learned in forecasting fire suppres- using analytics to gain insight and guide actions and argues sion operations productivity: a methodological approach. J For that the biggest challenges are managerial and cultural and Econ. 2016;25:149–59. offers recommendations for successfully building an analytics 87. Martell D. A forest and wildland fire management analytics frame- program. work. For Wildland Fire Manag: a risk management perspective 70. Mohamed, N. and Al-Jaroodi, J., 2014. Real-time big data analyt- workshop, Banff International Research Station, Banff, Canada. ics: applications and challenges. In 2014 international conference 2017;5-10:2017. on high performance computing & simulation (HPCS) (pp. 305- 88. Lu Y. Industry 4.0: A survey on technologies, applications and 310). IEEE. open research issues. J Ind Inf Integr. 2017;6:1–10. 71. Galindo G, Batta R. Review of recent developments in OR/MS 89. Müller F, Jaeger D, Hanewinkel M. Digitization in wood supply–a research in disaster operations management. Eur J Oper Res. review on how Industry 4.0 will change the forest value chain. 2013;230(2):201–11. Comput Electron Agric. 2019;162:206–18. 72. Qadir J, Ali A, ur Rasool R, Zwitter A, Sathiaseelan A, Crowcroft 90. Gingras JF, Charette F. FP innovations forestry 4.0 initiative. J. Crisis analytics: big data-driven crisis response. J Int Human Bangor: 2017 Council on Forest Engineering Annual Meeting; Action. 2016;1(1):12. 2017. July 30-August 2, 2017 73. Akter, S. and Wamba, S.F., 2017. Big data and disaster manage- 91. Sá AC, Benali A, Fernandes PM, Pinto RM, Trigo RM, Salis M, ment: a systematic review and agenda for future research. Ann et al. Evaluating fire growth simulations using satellite active fire Oper Res, pp1–21. data. Remote Sens Environ. 2017;190:302–17. 74. Hong M, Akerkar R. Analytics and evolving landscape of ma- 92. Arca B, Ghisu T, Casula M, Salis M, Duce P. Aweb-based wildfire chine learning for emergency response. In: In Machine learning simulator for operational applications. Int J Wildland Fire. paradigms. Cham: Springer; 2019. p. 351–97. 2019;28(2):99–112. 75. Shah SA, Seker DZ, Hameed S, Draheim D. The rising role of big 93. Duff T, Cawson J, Cirulis B, Nyman P, Sheridan G, Tolhurst K. data analytics and IoT in disaster management: recent advances, Conditional performance evaluation: using wildfire observations taxonomy and prospects. IEEE Access. 2019;7:54595–614. for systematic fire simulator development. Forests. 2018;9(4):189. 76.�� Vidgen R, Shaw S, Grant DB. Management challenges in creating 94. Dunn CJ, O’Connor CD, Reilly MJ, Calkin DE, Thompson MP. value from business analytics. Eur J Oper Res. 2017;261(2):626– Spatial and temporal assessment of responder exposure to snag 39 Similar to [62], reviews literature on business analytics, hazards in post-fire environments. For Ecol Manag. 2019;441: focusing in part on how to create value through actions based 202–14. on better decisions based on better data, and the roles of data 95. Campbell MJ, Dennison PE, Butler BW. A LiDAR-based analysis and value, organization and process, and people and of the effects of slope, vegetation density, and ground surface technology. roughness on travel rates for wildland firefighter escape route 77. Mikalef P, Pappas IO, Krogstie J, Giannakos M. Big data analytics mapping. Int J Wildland Fire. 2017;26(10):884–95. capabilities: a systematic literature review and research agenda. 96. Campbell MJ, Dennison PE, Butler BW. Safe separation distance IseB. 2018;16(3):547–78. score: a new metric for evaluating wildland firefighter safety zones 78. Davenport, T.H., 2006. Competing on analytics. Harvard Business using Lidar. Int J Geogr Inf Sci. 2017b;31(7):1448–66. Review, 84(1), p.98. 97. Campbell MJ, Dennison PE, Butler BW, Page WG. Using 79.� Shah S, Horne A, Capellá J. Good data won’t guarantee good crowdsourced fitness tracker data to model the relationship be- decisions. Harv Bus Rev. 2012;90(4):23–5 Cautions against tween slope and travel rates. Appl Geogr. 2019;106:93–107. overvaluing “big data” at the expense of cultivating “big Curr Forestry Rep (2019) 5:226–239 239 98. Stonesifer CS, Calkin DE, Thompson MP, Kaiden JD. Developing 113.� Thompson MP, Freeborn P, Rieck JD, Calkin DE, Gilbertson-Day an aviation exposure index to inform risk-based fire management JW, Cochrane MA, et al. Quantifying the influence of previously decisions. J For. 2014;112(6):581–90. burned areas on suppression effectiveness and avoided exposure: a case study of the Las Conchas Fire. Int J Wildland Fire. 99. Liu Z, Peng C, Work T, Candau JN, DesRochers A, Kneeshaw D. 2016;25(2):167–81 Descriptive analytics model to evaluate fire Application of machine-learning methods in forest ecology: recent line effectiveness, subsequently expanded on in Thompson et progress and future challenges. Environ Rev. 2018;26(4):339–50. al [59]. 100. Olden JD, Lawler JJ, Poff NL. Machine learning methods without 114. O’Connor CD, Calkin DE. Engaging the fire before it starts: a case tears: a primer for ecologists. Q Rev Biol. 2008;83(2):171–93. study from the 2017 Pinal Fire (Arizona). Wildfire. 2019;28, 101. Reid CE, Jerrett M, Petersen ML, Pfister GG, Morefield PE, Tager 14(1):–18. IB, et al. Spatiotemporal prediction of fine particulate matter dur- 115. Thompson MP, MacGregor DG, Dunn CJ, Calkin DE, Phipps J. ing the 2008 northern California wildfires using machine learning. Rethinking the wildland fire management system. J For. Environ Sci Technol. 2015;49(6):3887–96. 2018;116(4):382–90. 102. Kern AN, Addison P, Oommen T, Salazar SE, Coffman RA. 116. Belval EJ, Calkin DE, Wei Y, Stonesifer CS, Thompson MP, Machine learning based predictive modeling of debris flow prob- Masarie A. Examining dispatching practices for interagency hot- ability following wildfire in the intermountain Western United shot crews to reduce seasonal travel distance and manage fatigue. States. Math Geosci. 2017;49(6):717–35. Int J Wildland Fire. 2018;27(9):569–80. 103. de Bem PP, de Carvalho Júnior OA, Matricardi EAT, Guimarães 117. Chow JY, Regan AC. Resource location and relocation models RF, Gomes RAT. Predicting wildfire vulnerability using logistic with rolling horizon forecasting for wildland fire planning. regression and artificial neural networks: a case study in Brazil’s INFOR: Information Systems and Operational Research. Federal District. Int J Wildland Fire. 2019;28(1):35–45. 2011;49(1):31–43. 104. Rodrigues M, de la Riva J. An insight into machine-learning al- 118. Belval EJ, Wei Y, Calkin DE, Stonesifer CS, Thompson MP, gorithms to model human-caused wildfire occurrence. Environ Tipton JR. Studying interregional wildland fire engine assign- Model Softw. 2014;57:192–201. ments for large fire suppression. Int J Wildland Fire. 2017;26(7): 105. Sayad YO, Mousannif H, Al Moatassime H. Predictive modeling 642–53. of wildfires: a new dataset and machine learning approach. Fire 119. Wyngaard J, Barbieri L, Thomer A, Adams J, Sullivan D, Crosby Saf J. 2019;104:130–46. C, et al. Emergent challenges for science sUAS data management: 106. Leuenberger M, Parente J, Tonini M, Pereira MG, Kanevski M. fairness through community engagement and best practices devel- Wildfire susceptibility mapping: deterministic vs. stochastic ap- opment. Remote Sens. 2019;11:1797. proaches. Environ Model Softw. 2018;101:194–203. 120. Gibert K, Horsburgh JS, Athanasiadis IN, Holmes G. 107. Rodrigues M, Alcasena F, Vega-García C. Modeling initial attack Environmental data science. Environ Model Softw. 2018;106:4– success of wildfire suppression in Catalonia, Spain. Sci Total Environ. 2019;666:915–27. 121. Nowell B, Steelman T, Velez ALK, Yang Z. The structure of 108. Ganapathi Subramanian S, Crowley M. Using spatial reinforce- effective governance of disaster response networks: insights from ment learning to build forest wildfire dynamics models from sat- the field. Am Rev Public Adm. 2018;48(7):699–715. ellite images. Frontiers in ICT. 2018;5:6. 122. Butler BW, Brown S, Wright V, Black A. Bridging the divide 109. Riley KL, Grenfell IC, Finney MA. Mapping forest vegetation for between fire safety research and fighting fire safely: how do we the western United States using modified random forests imputa- convey research innovation to contribute more effectively to wild- tion of FIA forest plots. Ecosphere. 2016;7(10):e01472. land firefighter safety? Int J Wildland Fire. 2017;26(2):107–12. 110. Anderson KE, Glenn NF, Spaete LP, Shinneman DJ, Pilliod DS, 123. Plucinski MP, McCarthy GJ, Hollis JJ, Gould JS. The effect of Arkle RS, et al. Estimating vegetation biomass and cover across aerial suppression on the containment time of Australian wildfires large plots in shrub and grass dominated drylands using terrestrial estimated by fire management personnel. Int J Wildland Fire. Lidar and machine learning. Ecol Indic. 2018;84:793–802. 2012;21(3):219–29. 111. Hallema, D.W., Sun, G., Caldwell, P.V., Norman, S.P., Cohen, 124. Wang X, Wotton BM, Cantin AS, Parisien MA, Anderson K, E.C., Liu, Y., Bladon, K.D. and McNulty, S.G., 2018. Burned Moore B, et al. cffdrs: an R package for the Canadian forest fire forests impact water supplies. Nature communications, 9(1), danger rating system. Ecol Process. 2017;6, 5(1). p.1307. 125. Thekdi S, Aven T. An enhanced data-analytic framework for inte- 112.� WeiY,ThompsonMP, ScottJH, O’Connor CD, Dunn CJ. grating risk management and performance management. Reliab Designing operationally relevant daily large fire containment Eng Syst Saf. 2016;156:277–87. strategies using risk assessment results. Forests. 2019;10(4):311 Prescriptive analytics model to develop optimal response strategy based on quantitative risk assessment and potential Publisher’sNote Springer Nature remains neutral with regard to control locations; expands upon Wei et al. [85]. jurisdictional claims in published maps and institutional affiliations.

Journal

Current Forestry Reports – Springer Journals

Published: Nov 20, 2019

There are no references for this article.