A new statistical model for predicting seasonal north Atlantic Hurricane activity

Statistical, dynamical, and statistical-dynamical hybrid models have been developed in past decades for the seasonal prediction of North Atlantic hurricane numbers. These models' prediction skills show considerable decadal variability, with particularly poor performance in the past few years. Here, environmental factors that affect hurricane activities are reevaluated to develop a new statistical model for seasonal prediction by 1 June of each year. The predictors include the April-May multivariate ENSO index (MEI) conditioned upon the Atlantic multidecadal oscillation (AMO) index, the power of the average zonal pseudo-wind stress across the North Atlantic in May, and the average March-May tropical Atlantic sea surface temperature. When compared to the actual number of hurricanes each year from 1950 to 2013, this model has a root-mean-square error (RMSE) of 1.91 with a correlation coefficient of 0.71. It shows a 39% improvement in RMSE over a no-skill metric (based on the 5-yr running mean of seasonal hurricane counts) for the period 2001-13. It also outperforms three statistical-dynamical hybrid models [CPC, Colorado State University (CSU), and Tropical Storm Risk (TSR)] by more than 25% for the same period. Furthermore, two approaches are developed to provide the uncertainty ranges around the predicted (deterministic) hurricane number per season that better encompass the range of uncertainty than does the standard method of adding/subtracting a standard deviation of the errors.