Citation

Pilkington SF, Mahmoud HN. Front. Built Environ. 2017; 3: e67.

Copyright

(Copyright © 2017, The Author(s), Publisher Frontiers Media)

DOI

10.3389/fbuil.2017.00067

PMID

unavailable

Abstract

Tropical cyclones are an example of a multi-hazard event with impacts that can highly vary depending on landfall location, wind speed, storm surge, and inland flooding from precipitation. These storms are typically categorized by their wind speed and pressure, while evacuation orders are typically given based on storm surge. The general public relies on these single hazard assessment parameters when attempting to understand the risk of an oncoming event. However, after the fact, these events are ranked by economic damage and death toll. Therefore, it is imperative that when these events are communicated to the public, during the forecast period, the multiple hazards are incorporated in terms the public can easily associate with, such as economic damage. This paper provides an evaluation on the potential for real-time use of artificial neural networks, through the utilization of an already developed Hurricane Impact Level Model, to forecast a range of economic damage from tropical cyclone events, during the 2015 and 2016 United States Hurricane Season. The Hurricane Impact Level Model is built prior to the start of each season and simulated every three hours, in conjunction with National Hurricane Center issued advisories, for oncoming tropical cyclones forecasted to make landfall. Weaker and more common tropical cyclones have a less varied forecast and produce more accurate Impact Level predictions. More complicated and uncertain events, such as 2016 Hurricane Matthew, require the user's discretion in communicating varying landfall locations for a complex track forecast to the model. As National Hurricane Center (NHC) forecasts change with respect to both track and meteorological hazards affecting land, the estimated Impact Level and the Hurricane Impact Level (HIL) model confidence will also change. In other words, if a track shifts to a more vulnerable location, or to more locations, or the meteorological hazards increase, the Impact Level will subsequently increase. All tropical cyclones from the 2015 and 2016 seasons demonstrate the validity of the Hurricane Impact Level Model with a forecast confidence of at least 60% for up to 30 hours out from an impending landfall as well as reliability for real-time use, if data is available.

Language: en

Keywords

Forecasting; Hurricane; Hurricane Matthew; Neural Network; tropical cyclone