A BEHAVIORALLY-INTEGRATED INDIVIDUAL-LEVEL STATE-TRANSITION MODEL THAT CAN PREDICT RAPID CHANGES IN EVACUATION DEMAND...

Area of Science:

• Disaster Management
• Computational Social Science
• Epidemiological Modeling

Background:

• Current evacuation demand forecasting models (behavior-based and flow-based) have limitations in real-time adaptation and prediction windows.
• Static behavioral models cannot incorporate unfolding disaster data, while flow-based models offer short-term predictions.
• Existing models struggle to anticipate sudden surges or drops in evacuation demand.

Purpose of the Study:

• To develop a behaviorally-integrated, individual-level state-transition model for online evacuation demand prediction.
• To enable accurate, long-term forecasting of evacuation demand, including sudden changes.
• To improve disaster response planning through advanced evacuation modeling.

Main Methods:

• Developed an individual-level state-transition model with survival analysis formulation for history-dependent probabilities and heterogeneity.
• Employed a Bayesian updating approach for real-time assimilation of observed evacuation data.
• Integrated behavioral curves for long-term trend insights and used a likelihood-based approach for forecast updates.

Main Results:

• The model accurately predicted rapid surges or drops in evacuation demand at least two days in advance across six Hurricane Harvey scenarios.
• Demonstrated robustness in forecasting evacuation demand using mobile app data.
• Successfully integrated real-time data with behavioral insights for improved prediction accuracy.

Conclusions:

• The proposed behaviorally-integrated state-transition model enhances evacuation demand forecasting accuracy and lead time.
• This approach bridges the gap between behavior-based and flow-based modeling using real-world data.
• The model offers a significant advancement for emergency response planning and disaster management.