Using artificial neural networks to forecast economic impact of multi-hazard hurricane-based events

Abstract

In multi-hazard events, it remains difficult to communicate the collective effect these hazards have on the envisioned outcome or impact to the public. Currently, there are multiple models in use by emergency management and other government personnel to predict effects of hazards that put emphasis on wind damage (just as the Saffir–Simpson scale does), which tend to leave out the non-wind driven precipitation hazard. Experts who work with hazard events consistently build a knowledge base over time from experience that accounts for the collective effects of these multiple hazards in relation to locational vulnerabilities. In this study, an original artificial neural network is developed and used in an effort to mimic the previously mentioned learned and experienced-based knowledge. The output from the neural network model is an Impact Level Ranking System that ranks hurricanes based on total economic damage. The use of population affected, landfall location(s), wind speed, pressure, storm surge, and precipitation for inputs with a final Bayesian Regulation training approach allows for an ability to forecast multi-hazard hurricane events in terms the public could comprehend while remaining thorough in all hazards.