Digital twin simulations of theory-driven crisis messaging during hurricane evacuations in synthetic populations: a Miami-Dade County case study

Background Digital twin and agentic artificial intelligence technology provide innovative systems for testing behavioral science theory, which can improve emergency communication in crisis situations. More advanced and effective evidence-based messaging is needed for better safety preparation for extreme weather and more trusted evacuation communication. Methods This study developed a digital twin of Miami-Dade County populated with a synthetic population embedded with behavioral theory (Extended Parallel Process Model, Theory of Planned Behavior) and the development of a Message Assessment Framework (MAF) to systematically test theory-based crisis messages. Agents were exposed to fear-only, efficacy-only, norm-only, combined fear+efficacy, combined fear+efficacy+norm, and a neutral control message. Results Messages grounded in behavioral theory were more effective than the control message at encouraging evacuation. Messages that combined fear and efficacy provided the best results in the synthetic population’s decision to evacuate (OR = 15.45, p < 0.001), while adding social cues did not produce a statistically distinguishable added benefit. Discussion This research demonstrates a proof-of-concept approach for using agentic AI and digital twins to pre-test communication strategies, offering a scalable method for optimizing emergency messaging prior to real-world implementation.