A generalized algorithm for inferring wildfire evacuation decisions and departure times using large-scale mobile device location data

Abstract

Understanding wildfire evacuation behavior is essential for effective disaster planning, yet current approaches often rely on event-specific methods that limit broader applicability. This paper proposes a generalized algorithm for inferring wildfire evacuation decisions and departure times by leveraging large-scale mobile device location data. The framework relaxes the common assumption that all evacuees depart from home, allows customization of spatial and temporal parameters, and introduces refined behavioural categories that capture partial compliance and reduce unclassified cases. The algorithm is presented using two U.S. wildfires: the 2020 Silverado Fire in California and the 2021 Marshall Fire in Colorado. Key parameters were adapted to local contexts to assess generalizability. Results indicate that over 77% of evacuees departed from home in both events, and although behavioural patterns differed, overall evacuation rates were similar. This study improves understanding of wildfire evacuation behavior and supports policymakers and emergency managers in enhancing evacuation plans and community resilience.