Computational Evacuation Modeling in Wildfires
- 325 Downloads
Evacuation refers to the dispersal or removal of people from dangerous areas and their arrival at a place of relative safety. In the context of wildfires, this involves the movement of people away from a wildfire, the forest, or the wildland-urban interface (WUI) to a safer place of shelter. Evacuations may be mandatory, advised, or spontaneous. An evacuation model is a qualitative or quantitative conceptual framework used to depict evacuee response, assist decision-making, and finally improve people safety. Computational evacuation modeling refers to the implementation of this conceptual framework into a computer, typically to quantify evacuee performance.
Wildfires are a global issue which affects numerous communities around the world, with an associated economic burden estimated to be between 71 and 348 billion dollars (the estimated impact is challenging because of the numerous direct/indirect, tangible/intangible, short-term/long-term impacts that a wildfire might have) in the USA alone (Thomas et al. 2017). Wildfires pose significant evacuation challenges to residential populations, as they require a fast relocation of large numbers of people subject to time constraints while affecting the availability and capacity of the network available for evacuation. This is particularly an issue for wildland-urban interfaces (WUI) where large populations may live in the proximity of the fire threat. The evacuation might involve subpopulations with different resources, capabilities, objectives, and information. Evacuation might also take place through different modes of transportation, including the use of vehicles or on foot.
Authorities may adopt different approaches concerning the instructions given to a community exposed to a wildfire threat, i.e., a mandatory or recommended evacuation order. These policies will have different implications on the safety of the community, i.e., if/when people would leave and their level of exposure to the fire threat (the reader may see the contribution on “Evacuation,” “Stay and Defend,” and “Community Response to Wildfire”).
Design: community design and evacuation planning (the assessment of the evacuation in relation to different hypothetical what-if scenarios involving different fire hazards, populations, and resources)
Incident response: real-time evacuation decision-making (identifying trigger and buffer areas influencing the timing and type of order, such as defend-in-place, mandatory, or recommended evacuation)
The effectiveness and appropriateness of evacuation modeling tools depend on their modeling capabilities and the temporal and spatial scales of the scenarios being investigated. Evacuation modeling tools have benefits linked to their ability to provide decision support during vulnerability analysis as well as during emergency response, given sufficiently representative model functionality. It is clear that such functionality would need to address fire, people, and traffic aspects to capture wildfire evacuation.
Models currently exist that address the functionality highlighted above. This includes fire models and their associated sub-models (e.g., models for the representation of fire and smoke spread, spotting, etc., which are discussed in the contribution on “Modeling Approaches” and “Physics-Based Modeling”) and evacuation models of various types. To date, most of the evacuation modeling tools address different evacuation aspects in isolation. People response models represent the time needed to initiate protective actions (typically called the pre-evacuation time in building evacuations). People movement models represent the subsequent evacuee movement and response to a fire threat. Traffic models represent traffic movement on the road network possibly considering different transportation means for evacuation (Pel et al. 2012) (e.g., public or private vehicles).
Once the evacuation modelers have identified which aspects of the evacuation process they want to model implicitly or explicitly and the data available for calibration, the next step is to identify suitable models to achieve this. This entry presents an overview of different modeling approaches and tools available for the representation of three modeling layers, people response modeling, people movement modeling, traffic modeling, and what data and data exchanges between them are needed to produce the desired results. Although an important aspect of evacuation modeling, fire modeling is only briefly presented. For more information, the reader may see the contribution on “Fire Modeling.” The last section outlines some of the limitations and key challenges of evacuation modeling.
Fire models are generally made of different sub-models which may use an empirical, semiempirical, or physics-based approach (Sullivan 2009a, b, c) (the reader may see the contribution on “Physics-Based Modeling”). Given computational constraints, the fire models used by fire and emergency agencies (commonly titled operational fire models) include empirical and semiempirical approaches. Fire models take into consideration different factors such as fuel type, rate of spread, spotting, plume, and smoke transport adopting one or more sub-models. Weather and topology also play a significant role in the fire spread. Fire models allow the estimation of the evolution of the fire perimeter over time, thus being able to inform trigger points for evacuation orders during the incident timeline. A comprehensive fire model that is generally applicable to all types of vegetation does not currently exist. This is linked to the enormous variation in vegetation, requiring regionally specific fire models (Ronchi et al. 2017; Sullivan 2009a).
People Response Modeling
Evacuation modeling in case of wildfires should contain both movement calculations (typically linked to people/traffic modeling) and a behavioral perspective (Cova et al. 2009). The latter addresses the factors that influence human decisions to evacuate as well as compliance to the instructions received by emergency responders. Existing evacuation simulation tools do not generally incorporate wildfire-specific dedicated models for people response. They generally instead make use of simplified implicit representations of human behavior within people or traffic models (see Fig. 2). Nevertheless, information and models developed for different types of incidents at small or large scales (e.g., building evacuations or other large disasters, e.g., hurricanes) can be used as a starting point to represent wildfire evacuation response.
The representation of people response modeling is necessary to produce an accurate estimation of the time needed to evacuate an area threatened by a wildfire. This is because people may start their movement toward safety at different times and with different movement modalities, in relation to the information available and instructions received. Models should also be able to represent different evacuation modalities. A mandatory evacuation is likely to be associated with greater compliance and shorter response times, while the effectiveness of self-delayed evacuation heavily relies on the warnings and the type of information received. For this reason, it is important to understand if warnings in use are sufficiently informative or if it is needed to model the additional delay associated with information seeking (McLennan et al. 2018). Regardless of the type of instructions provided by the authorities and the available information (e.g., information from peers, social media, visual cues, neighbor behavior, etc.), the first step should include the assessment of the population involved in the threat as well the scenario conditions (what are generally called behavioral scenarios in building evacuation research (Nilsson and Fahy 2016)) to model evacuation decisions.
An example of a commonly used framework to represent people response during disasters is the Protective Action Decision Model (Lindell and Perry 2012). This is a comprehensive behavioral model which describes evacuee performance in response to a threat. This model has been implemented in an algorithmic formulation for evacuation simulation tool (Cova et al. 2009). In such type of frameworks, several factors need to be implemented for an accurate representation of people response for evacuation modeling purposes. Examples of such factors are presented in Murray-Tuite and Wolshon (2013), Trainor et al. (2012), and Wachinger et al. (2013) for general evacuation modeling applications.
Categorization of factors affecting evacuation response (Ronchi et al. 2017)
Environmental conditions (e.g., existence of fire front, severity/size, wind speed, distance)
Physical conditions (e.g., availability of access route, vegetation, topography)
Procedural conditions (e.g., existence of warning system, information provided, fire protection measures, transport provisions)
Social conditions (e.g., actions of others, status, role and attributes of others, location of others, size of community, peer information)
Structural conditions (e.g., building design/construction, damage/status, etc.)
Role in social/organizational structure
Time at/relationship with location
Situation awareness (warning received, information gathered, perceived credibility/consistency)
Available modes of transport
Proximity to incident
Access to resources
Confidence in own abilities/others/plan/science/policies
Perceived Risk Response Identification
Wait (new information, rescue, arrival of other people, conditions to improve, etc.)
Prepare (to leave, to defend in place, etc.)
Protect (address fire, etc.)
Maintain (continue as before)
Search (for information, for people, for animals for equipment, etc.)
Communicate (call others, receive information, etc.)
Leave current location (leave on foot to perform tasks beyond current location, leave on foot to place of safety, move to nearby vehicle, leave vehicle to return to property/second vehicle, etc.)
A comprehensive people response model should be able to represent or account for the factors listed above either implicitly or explicitly, reflecting the key stages of the evacuation decision-making process. To date, only a few attempts have been conducted to develop people response models dedicated to wildfire evacuations (Ronchi et al. 2017). Nevertheless, modeling approaches employed for other types of large-scale disasters (e.g., hurricane evacuations (Murray-Tuite et al. 2012; Dixon et al. 2017)) can serve as a basis for wildfire people response model development. Such approaches employ regression models which consider how different factors affect negatively or positively the likelihood of people evacuating (Dixon et al. 2017; Lovreglio et al. 2015). An alternative approach is the use of agent-based modeling, which allow for the representation of human behavior. Such approaches allow the identification of a set of rules which drive individual or aggregated agent response that is then implemented into a simulation tool to represent time-based responses. The resulting function obtained by these rules can be used to predict (generally in a probabilistic fashion) when an agent will start its evacuation journey (Yin et al. 2014; Scerri et al. 2010; Chen et al. 2006) and the results of them executing this response.
Evacuee decisions can be modeled at an aggregate level (e.g., considering a household as a group, community response, etc.) or individual level. These decisions are updated in the modeling framework adopting an event-based (i.e., the decisions are re-evaluated in the model when an event happens) or time-step-based approach (the decisions are re-evaluated at fixed time-steps). Such approaches depend on the level of granularity of the model and the basic assumptions it employs to represent human behavior.
People response can be represented in evacuation modeling either (1) at a macroscopic level by identifying a suitable function that accounts for the key factors affecting the evacuation response of a given population in a given scenario or (2) at a microscopic level by having a user representing the sequence of actions that people would do before starting the evacuation. Such representation can be performed using a dedicated sub-model to be coupled with other modeling layers or being embedded into one of the other modeling layers (people or traffic modeling).
People Movement Modeling
People movement models are used to represent the movement of pedestrians on foot. The movement can occur directly to a place of safety or to an intermediate location (including the case of the movement on foot to a private or shared vehicle). Currently, there is no people movement model which has been specifically designed for the explicit representation of the different movement patterns that people can have in response to a wildfire. Nevertheless, different actions can be represented implicitly with existing people movement models, at different levels of granularity and sophistication. This is typically enabled by people movement being driven by user intervention, i.e., the model user configuring the movement of people to represent behavior and actions.
It should be noted that people movement may ordinarily be considered of secondary importance during large-scale wildfires (as its contribution to the total evacuation time may be minor). Nevertheless, the output from a people movement model represents a key input to the traffic model as it allows representing explicitly the departure time of the vehicles. There are also rare cases of WUI fires where walking is a mode of movement which contributes to the evacuation process (Ronchi et al. 2017; Toledo et al. 2018).
People movement models for evacuation scenarios have been the subject of several reviews in both indoor (Gwynne et al. 1999; Kuligowski et al. 2010) and outdoor scenarios (Bellomo et al. 2016; Duives et al. 2013). Such reviews generally classify people movement models in relationship to their main modeling assumptions (i.e., model mutability, content, scale, credibility, and accessibility) (Ronchi et al. 2017). Key aspects related to people movement modeling characteristics for wildfires relate to their spatial representation and level of granularity (Ronchi and Nilsson 2016). Such assumptions allow identifying the level at which people movement is represented (individual or aggregate) and how people move in the simulated space (with a discretization of the space in cells/networks or in a free continuous space). Macroscopic people movement models generally make an analogy between people and fluid particles, which move in space as a fluid with a certain density. Empirical relationships between speed, densities, and flows are then used to represent the aggregated movement of people, with the main limitations of focusing mostly on the physical (rather than behavioral) aspects of movement. Microscopic models represent people movement adopting two different simulation levels: (1) pathfinding, i.e., the simulation of route choice in space and time at a global or local level, and (2) local movement, i.e., the simulation of space navigation including the interactions between different people and between people and the environment. Microscopic approaches are today the most common methods used to represent evacuees in computer models at a building scale (Ronchi and Kinsey 2011), mostly due to their ability to represent autonomous agents with different properties. Nevertheless, the larger spatial scale associated with wildfires makes mesoscopic and macroscopic approaches attractive, given their faster computational times and simpler configuration required.
The key outputs obtained by people movement models typically include the times needed to reach different areas, the routes adopted, the congestion experienced, as well as the movement itineraries and actions performed during their journey on foot.
Given the fact that evacuation takes place in the majority of cases through the road network, this section focuses on existing modeling tools for such type of scenarios. However, it should be acknowledged that evacuation can take place also by other means (i.e., via sea or airlifting Westhaver 2017) and that scarce attention has been given to the simulation of these types of evacuation scenarios.
Several reviews exist today on the sub-models adopted for the representation of the four steps adopted for traffic modeling (Barceló 2010), including specific applications to evacuation modeling (Pel et al. 2012; Intini et al. 2018). Dedicated models are used to represent the generation, distribution, and modal split steps. These can be modeled sequentially (through independent models) or within a single comprehensive modeling framework including those choices in an overall choice structure. The estimation of the travel demand is followed by the assignment of the trips to the network (i.e., the traffic assignment step). Several methods are adopted today for this step, based on different hypotheses on travel demand variability, time scales, interactions, algorithms, and methods employed (Ronchi et al. 2017). The case of wildfire evacuation simulations requires an iterative process of calibration, as adjustments are required by looking back at the travel demand stages (as these can vary in relation to the evolution of the fire threat).
The key outputs of a traffic model are the variables describing the traffic on the network (i.e., flows, travel times, costs, delays, etc.) and the resulting evacuation times. As for the case of people movement modeling, the resolution of the results obtained relies on the modeling methods adopted for the representation of the space as well as the movement (i.e., microscopic, mesoscopic, macroscopic). The results of traffic evacuation models allow predicting the time to reach a safer place/shelter as well as evaluate measures aimed at improving evacuation efficiency (i.e., reducing congestions).
The inherent inaccuracy within each modeling layers, e.g., lack of comprehensive models able to be used, as well as limited validation of existing behavioral sub-models
Limited coupling between different modeling layers
People response/movement models and traffic models are rarely explicitly linked to fire models in a way that the evolving fire can affect evacuee behavior. Recent developments in wildfire evacuation modeling have examined the explicit integration of different modeling layers, including coupled people/traffic models (Fellendorf and Vortisch 2010), as well as coupled traffic/fire models (Beloglazov et al. 2016; Li et al. 2018) and even coupled people/traffic/fire models (Ronchi et al. 2017). Nevertheless, existing coupled models rarely present a consistent level of crudeness in the representation of different modeling layers, generally putting greater emphasis on one of the modeling aspects being considered. Currently, only operational fire models can be coupled with evacuation models for real-time decision support, given the reduced computational resource they require. Nevertheless, they have a prediction error estimated to be approximately to 40% regarding the rate of spread (Cruz and Alexander 2013). This error seems to indicate that they can be used to identify evacuation trigger points but that improvements are needed to allow for a more consistent, representative and refined approach to modeling fire, people, and traffic movement during an evacuation. Model granularity affects the applicability of the models to different scales given time and computational resources available. This is strictly linked to their use, i.e., evacuation planning or real-time decision support. The choice for different modeling approaches should not be based on the assessment of the tool for an individual aspect (fire, people, or traffic). It should instead address the influence of one modeling domain on another. This is linked to possible propagation of inaccuracies between models, a well-known issue in the building fire evacuation modeling domain (Ronchi et al. 2013).
Research gaps in the field of evacuation modeling in wildfires still remain. In particular, the accuracy of behavioral predictions would greatly benefit from a higher number of datasets reflecting human behavior in different conditions and considering populations with different characteristics. This includes an increased understanding of the impact of external factors on evacuation response (e.g., traffic management, real-time information, dynamic evolution of the fire conditions, impact of warning systems, social media influence).
Despite these limitations, existing evacuation models provide a quantitative measure of evacuation performance ahead of time, with a relatively small cost. This can be used to enhance the situational awareness of decision-makers and hopefully then the effectiveness of their response (Seppänen and Virrantaus 2015).
Evacuation models can provide useful information to planners and to emergency responders (i.e., contributing evidence to a decision-making process), given that the user understands the limitations of the model employed. Although extensive validation efforts are required to increase the reliability of the predictive capabilities of evacuation models, they can provide an increased understanding of current and future evacuees’ behavior and the consequences of such behavior on overall performance, i.e., evidence of what is happening and where. Such models enable an array of scenarios to be examined and the resultant outcomes produced, providing insights into scenarios (that might otherwise not be examined), and evaluated at a number of different levels of refinement.
Spatial and temporal information on evacuation response has a fundamental role in wildfire emergency management. Decision-making heavily relies on the information available, along with its accuracy and credibility. The emergency response relies on the ability of a community to prepare for the wildfire hazard and the response in relation to its evolution. The effectiveness of these decisions can be significantly improved if supported by quantitative information provided by an evacuation model. Such insights do not solve the problem, but provide evidence that better informs the decision-making process in relation to the problem.
- Barceló J (2010) Fundamentals of traffic simulation, vol 145. Springer, New YorkGoogle Scholar
- Bellomo N, Clarke D, Gibelli L, Townsend P, Vreugdenhil BJ (2016) Human behaviours in evacuation crowd dynamics: from modelling to ‘big data’ toward crisis management. Phys Life Rev 18:1–21. https://doi.org/10.1016/j.p
- de Araujo MP, Casper C, Lupa MR, Brinckerhoff P, Waters B, Hershkowitz P (2011) Adapting a four-step MPO travel model for wildfire evacuation planning: a practical application from Colorado Springs. Transportation Research Board 90th Annual Meeting, Washington DC, United States, 2011-1-23 to 2011-1-27. https://trid.trb.org/view/1091428
- Kuligowski ED, Peacock RD, Hoskins BL (2010) A review of building evacuation models, 2nd edn. NIST technical note 1680. National Institute of Standards and TechnologyGoogle Scholar
- Nilsson D, Fahy R (2016) Selecting scenarios for deterministic fire safety engineering analysis: life safety for occupants. Springer, New York, pp 2047–2069Google Scholar
- Ronchi E, Kinsey M (2011) Evacuation models of the future: insights from an online survey on user’s experiences and needs, presented at the advanced research workshop evacuation and human behaviour in emergency situations EVAC11, Santander, pp 145–155Google Scholar
- Ronchi E, Nilsson D (2016) Basic concepts and modelling methods. In: Cuesta A, Abreu O, Alvear D (eds) Evacuation modeling trends. Springer International Publishing, Cham, pp 1–23Google Scholar
- Ronchi E, Rein G, Gwynne S, Wadhwani R, Intini P, Bergstedt A (2017) e-Sanctuary: open multi-physics framework for modelling wildfire urban evacuation. Fire Protection Research Foundation, QuincyGoogle Scholar
- Scerri D, Gouw F, Hickmott S, Yehuda I, Zambetta F, Padgham L (2010) Bushfire BLOCKS: a modular agent-based simulation. In: Proceedings of the 9th international conference on autonomous agents and multiagent systems, vol 1, pp 1643–1644Google Scholar
- Westhaver A (2017) Why some homes survived: learning from the Fort McMurray wildland/urban interface fire disaster. Institute for Catastrophic Loss Reduction, TorontoGoogle Scholar