Seasonal prediction of landfalling hurricanes along Eastern Seaboard of the United States

Abstract

Statistical models for the seasonal prediction of hurricane strikes on the Eastern Seaboard of the United States (ESUS) are developed with the application of the statistical software (SAS) generalized linear model (GENMOD) to perform a Poisson regression linked by a logarithmic function. Preseason climatic/oceanic signals of the El Nino South Oscillation (ENSO), the Atlantic Meridional Mode (AMM), the Atlantic Multi-decadal Oscillation (AMO), Sahel rainfall (20°–10°N, 20°W–10°E) and the North Atlantic sea surface temperature (SST) across a selected domain, all of which are closely associated with the dominant principal components of the North Atlantic Hurricane Track Density Function (HTDF), are applied as predictors. While using the Poisson regression without dividing the season into differing levels of activity types greatly degrades the model, classification of season types [based on accumulated cyclone energy (ACE)] before using the Poisson regression, can significantly improve the model performance. With this proposed new methodology, landfall counts in hyperactive and above-normal season type years can be predicted with 56.01 and 71.36% skill improvements respectively, compared with merely using climatology.