An Agent-Based Modeling Approach to Protective Action Decision-Related Travel during Tornado Warnings

Abstract

Tornadoes represent a significant threat to life and property and tend to evoke protective action in most people. Studies have suggested that many people travel to the nearest storm shelter or flee the area, rather than sheltering-in-place as recommended by the National Weather Service (NWS). While shelter-in-place is the recommendation of the NWS, for tornado safety, few studies have quantified the risk reduction when compared to traveling to a storm shelter or fleeing the area. To address this knowledge gap, we developed an agent-based model, the tornado warning-induced shelter, travel, and evacuation response agent-based model (TWISTER ABM), to simulate protective action behaviors in the city of Norman, Oklahoma, under eight protective action scenarios including: (1) everyone who responds to the warning (responders) seeks refuge in the nearest sturdy building (seek refuge), seeks shelter in a FEMA-rated shelter (seek shelter), or flees the area, (2) all responders flee the area, (3), all responders seek refuge (shelter-in-place), (4) all responders seek shelter, (5) all agents flee the area, (6) all agents seek refuge, (7) all agents seek shelter, and (8) all agents do nothing. We found that, for an EF5 tornado hitting Norman at rush hour, the overall fatality rates by protective action type were 0.6% for those who took no action, 0.3% for those who sought refuge, 1.5% for those who sought shelter, and 1.1% for those fleeing the area. We also found that fatality rates were reduced by a factor of 6.6 for Scenario 6 (shelter-in-place) over Scenario 7 (travel to a FEMA-rated shelter). We believe that models such as TWISTER ABM can be used by the NWS and emergency managers in their attempts at communicating the effectiveness of shelter-in-place. Tornadoes are dangerous windstorms that can cause serious injury or death to people who do not take protective action. The NWS states that sheltering-in-place is the safest form of protective action, but no studies to date have shown how much it can reduce casualties. We developed an agent-based model to study how changes in protective action type can influence the fatality rate (fatalities per 1,000 residents) caused by a tornado in the city of Norman, Oklahoma. We found that, for an EF5 tornado hitting Norman at rush hour, the overall fatality rates, for all model runs, were lowest for agents who sheltered-in-place (0.3%) and highest for those who traveled to public shelters (1.53%). We also found that fatality rates were lowest when all agents sheltered-in-place (0.24%) and highest when every agent responding to the warning traveled to public shelters (1.54%), a 6.6 times reduction for shelter-in-place. We believe that models such as TWISTER ABM can be used by the NWS and emergency managers in their attempts at communicating the effectiveness of shelter-in-place.