Online Program

Return to main conference page
Thursday, May 17
Machine Learning Applications
Thu, May 17, 6:15 PM - 7:15 PM
Regency Ballroom B
 

Diagnosing and Predicting the Eyewall Replacement Cycle: Learning from Hurricane Irma (304690)

*Martha Lisbeth Christino, T.C. Williams High School 

Keywords: Machine Learning, Hurricanes, Eyewall Replacement Cycle,

The 2017 Atlantic hurricane season was historic, in both the number and intensity of hurricanes. The intensity of a hurricane is closely linked to the strength of its eyewall. When the eyewall strength fluctuates during the naturally occuring Eyewall Replacement Cycle (ERC), the potential damage can change by a factor of five. Because of limitations in available data, NOAA’s National Hurricane Center has not been able to predict the timing of ERCs. The purpose of this project is to determine how machine learning techniques can be used to diagnose and predict the ERC. Two machine learning programs were created using a single decision tree algorithm, one for diagnosis and one for prediction. The two programs were trained based on data from Hurricane Irma and tested based on Hurricanes Harvey, Jose and Maria. The diagnostic program achieved 77% accuracy in determining if an ERC was occuring at that moment based on 24 hours of previous data. The prediction program achieved 69% accuracy in determining if an ERC would occur in the next six hours based on 18 hours of previous data. These results show that machine learning is a viable method of ERC prediction and can be used in order to better forecast hurricane intensity, which will allow governments and private citizens to better protect lives and property.