A Bayesian agent-based model and software for wildfire safe evacuation planning and management

Abstract

Efficient evacuation of wildfire-threatened communities is a pressing challenge. A reliable evacuation planning and execution requires a comprehensive understanding of the diverse and interdependent physical, social, and behavioral components, and advanced, yet easy to use decision support system. This paper proposes the Wildfire Safe Egress (WiSE) framework, which integrates the fire dynamics, human behavior, and traffic model to predict the chance of safe egress by any given community during a wildfire evacuation. WISE framework presents a unified dependency diagram and workflow offering consistent granularity between sub-models and creates comparable evacuation scenarios. A human behavior model is proposed to predict the community decision making and action based on their socio-demographic vulnerability profile. An agent-based stochastic approach generates evacuation departure times. The travel times are calculated through a congestion-informed traffic simulation. Finally, a Bayesian Network is used to combine the sub-models and to predict community safety (probability of successful evacuation) via probabilistic inference based on the integrated model. A proof-of-concept software implementation of the WiSE framework is also presented. To demonstrate the model and platform capabilities the evacuation of the entire city of Paradise during the California Camp Fire 2018 is simulated. The simulation results are qualitatively validated by the firefighters who served in this disaster. A sensitivity analysis of the parameters is performed to compare several evacuation scenarios and provide insights for future wildfire evacuation plannings.