Minutes matter for life safety and risk exposure in Milford Sound, New Zealand.

The number of people killed by rapid-onset, highly-lethal natural hazards is usually proportional to the number of people directly exposed yet population mobility makes such exposure temporally variable. In hazardous tourist destinations that are dominated by highly mobile transient visitors, this makes it especially challenging to assess and manage risk. While the temporal dynamics of population exposure have been identified as a necessary input to risk analysis, lack of data and methods means such assessments are limited to using simplified assumptions at low temporal resolution (e.g. day-night variations). Here we show that population can fluctuate by an order of magnitude within minutes, meaning the timing of a hazard event could impact the number of resulting fatalities to a similar degree. At Piopiotahi Milford Sound, we show that the population exposed to landslide-triggered tsunami hazards had a maximum recorded change of ~ 1000 people over a 5-min period. This increased the modelled number of fatalities in a tsunami from 71 to 1134 within the corresponding 5-min period. Monitoring these minute-by-minute changes longitudinally over a period of 790 days that included New Zealand's COVID-19 lockdown measures and their removal demonstrates that the level of societal risk here was only acceptable during the strictest lockdown measures and has since become increasingly unacceptable. Our approach shows that incorporating longitudinal high-resolution data on dynamic exposure substantially improves assessment accuracy and reduces inherent uncertainty of dynamic disaster risk, especially in popular tourist destinations and where population shifts are frequent and significant.