Defining probabilistic evacuation triggers for wildfires using an ensemble of flame spread models: The Fort McMurray case study

A central challenge in wildfire evacuation planning is defining evacuation triggers, which determine when evacuation remains safe and when it becomes dire. These triggers are inherently probabilistic, as they depend on estimates of wildfire behaviour, human response and evacuation dynamics. In this paper, we present a framework for defining probabilistic evacuation triggers for wildfires using an ensemble of flame spread models, with different computational approaches, applied at the landscape scale. By using an ensemble the resulting evacuation trigger boundaries are more robust and comprehensive than those derived from any single model. Results show that individual models produce different evacuation trigger boundaries, reflecting differences in fuel representation, fire dynamics, and spread rate predictions. The models in this study are Farsite, Prometheus/WISE, ELMFIRE, Google EPD and EPD-ConvLSTM The Fort McMurray wildfire in Canada is used as the case study, motivated by the large-scale and dire evacuation that occurred in 2016. The results indicate that Fort McMurray’s triggers extend beyond practical detection distances under rapid fire spread conditions. This study demonstrates the value of ensemble modelling for evacuation planning and highlights the importance of also considering shelter-in-place strategies alongside evacuation plans for highly exposed communities.