Agent-based Modeling of Social Influence, Traffic Patterns, and Warning Strategies during Hurricane Evacuation
As coastal populations experience the growing threat of hurricanes as a consequence of global climate change, research on hurricane evacuation has drawn increasing attention from scientists, emergency management planners, and the public. This study models the effects of social influence on human evacuation behaviors, warning strategies, and evacuation traffic patterns, issues that have been insufficiently addressed in previous research. To contribute to knowledge about social influence on evacuation, an agent-based model is developed to simulate household decisions about evacuation in response to Hurricane Georges, using the Florida Keys as the geographical context. Data sources informing model initialization include census block group data, business databases, and statistics from several evacuation surveys. The model consists of three main components: social contact network, diffusion of evacuation orders, and diffusion of household evacuation behavior. The social network provides a basis for the two diffusion processes. Households who are informed of the evacuation order begin a decision-making process about evacuation. This evacuation decision depends in part on the evacuation behavior of households in their social network. Information about the hurricane evacuation order spreads through both mass media and inter-household communication. Social influence among households is modeled using a threshold approach, in which an individual household experiences peer pressure to evacuate from surrounding households and is convinced to evacuate if such pressure exceeds individualized thresholds. The simulation results are aligned with empirical observations from a previous evacuation survey and the observed evacuation traffic. Therefore, the baseline simulation of the model provides a representation of the spatio-temporal patterns of evacuation behavior in the Florida Keys study area. Based on the model, two warning strategies are simulated to examine their effectiveness. These strategies are mass media campaigns and door-to-door notice. A sensitivity analysis indicates that the major impact of media campaigns is to shorten the time required to inform the whole community, thus improving the evacuation rate. The door-to-door notice strategy that targets households with the highest evacuation thresholds (i.e., least likely to evacuate) is the most effective choice given limited resources, followed by targeting households that are socially active. In addition to social influence and warning strategies, this study models hurricane evacuation traffic patterns based on the travel demands generated by simulated behaviors of households. Through simulation, this study examines whether households that are willing to evacuate can do so without delay due to traffic congestion. The predicted travel demands are input into the software package TRANSIMS to simulate on-road traffic. A mixed travel mode situation is simulated in which public transportation via bus is added as an alternative to private automotive transport. Public transit is an option of particular utility for low-mobility populations. The simulation results demonstrate that adding public transportation capacity reduces the traffic load significantly and provides a practical option to the public. With the inclusion of social influence and public transportation, this dissertation research theoretically bridges previous evacuation models from the engineering, natural, and social sciences. The network-based evacuation model developed in this study adds authenticity to evacuation modeling. This study explicitly addresses the potential role of public transportation in hurricane evacuation and suggests feasible plans for bus route scheduling to evacuate low-mobility populations. Results indicate ways in which policy makers can design more efficient and effective evacuation plans.