Concluding the 2017 Hurricane Season: Evaluation of Impact Level Forecasts with Varied Meteorological Hazards

Abstract

The 2017 hurricane season was a costly multi-billion-dollar season for the United States. Communicating the real-time risk of these events to the public is usually performed through explanations from meteorologists via media coverage. However, despite meteorologists’ expert knowledge, the public will sometimes not heed warnings because they have developed mistrust in the accuracy of the science used and sometimes believe that media coverage exaggerates. Additionally, the public relies heavily on the Saffir-Simpson Wind Scale even when the most dangerous factors typically are not wind speed. Recently, a new hurricane impact level ranking system was developed using an artificial neural-network model in order to communicate the impending risk of an event from its multiple hazards. In the last three hurricane seasons, the real-time use of this neural network–based model has been tracked, evaluated, and proven relatively accurate. This study assesses what changes in the meteorological forecasts are causing changes in the predicted impact level and how the change in the impact level compares with the language used to communicate risk within the advisories themselves. The results show that the changing language in the National Hurricane Center advisories matched the 2017 forecasted impact levels and that the impact level forecast is most accurate approximately 30 h out from landfall. The most influential meteorological parameter that changes the predicted impact level for most hurricanes over the last three seasons was found to be the wind speed and population affected. For a track shift to result in an impact level change, the affected population is required to approximately double. For Hurricane Harvey, however, rapid intensification of all variables led to its eventual higher impact level forecast.