Agent-based modelling of evacuation scenarios for a landslide-generated tsunami in Milford Sound, New Zealand

Agent-based modelling is a useful tool for evacuation planning as it can increase understanding of the factors affecting potential evacuation outcomes. In New Zealand, Milford Sound has been shown to have a high risk from landslide-generated tsunami, with an estimated 1-in-1000-year wave runup of ∼17 m arriving on shore within 2–7 min. With an annual average of >1500 people visiting a day, there is potential for widespread loss of life. However, the number of people present varies substantially with time of day and season, yet how this affects the ability to evacuate remains unknown. This research developed an agent-based model to understand how many people can be safely evacuated in Milford Sound and explored how the number of people initially exposed affected the evacuation outcome alongside the effect of potential changes to evacuation messaging. Assuming a 17 m wave, the results suggest that currently no one can safely evacuate before the shortest wave arrival time regardless of the number of people present. Altering evacuation messaging results in minimal gains, with only ∼5 % of the exposed population reaching safety in time. This work demonstrates the importance of evacuation modelling for understanding risk in isolated tourism destinations where the population exposure can fluctuate dramatically across multiple timescales. Accounting for changing population exposure is essential to understand whether evacuation is a suitable risk treatment and can provide valuable information for determining safe levels of population exposure in locations with high hazard but limited evacuation options.