Behavior-driven simulations of disasters to foster resilient communities

Although emergency planning is critical to protecting communities, little is known about human systems in disasters. A more thorough understanding of the population dynamics of disaster victims will help policy planners to optimize response efforts. The chaotic nature of emergencies makes simulations a valuable planning tool.

This work is part of an extensive case study of a major insult to the infrastructure and population of a large U.S. city. Constructing the simulation involves modeling the power, transportation, health, and communication infrastructure in addition to the population and behavior. Our agent-based model creates a population statistically indistinguishable from Census data, including demographics like income, occupation, and several other variables. The behavior models describe behavior following a disaster with extensive casualties, widespread destruction, disruption of power and critical services, and ongoing threats to survivors.

From data on sociological studies of human responses following catastrophes, we used a Markov Decision Process framework to model behavior. We created a decision tree of survival goals, where each branch represents behavior options contingent on individual circumstances, such as location and health status, which the simulation updates dynamically. The time horizon for the study is two weeks following the event. Behaviors include evacuating, reconstituting family units, seeking healthcare, and taking shelter. The probabilities of certain behaviors changes over time, so that actions like calling 911 become less likely hours after the event.

This effort allows the study of cellular communication loads, health systems, and traffic burdens post-disaster, and serves as a platform to test interventions. Current emergency plans do not consider behavior, which creates unrealistic assessments of how events unfold. Modeling socially coupled systems not only changes the answers, but the questions we ask. For example, we show that restoring communication systems alters traffic loads. Simulation will enable communities to optimize mitigation and relief efforts.