Modeling the Dependence of Tropical Storm Counts in the North Atlantic Basin on Climate Indices

The authors analyze and model time series of annual counts of tropical storms lasting more than 2 days in the North Atlantic basin and U.S. landfalling tropical storms over the period 1878–2008 in relation to different climate indices. The climate indices considered are the tropical Atlantic sea surface temperature (SST), tropical mean SST, the North Atlantic Oscillation (NAO), and the Southern Oscillation index (SOI). Given the uncertainties associated with a possible tropical storm undercount in the presatellite era, two different time series of counts for the North Atlantic basin are employed: one is the original (uncorrected) tropical storm record maintained by the National Hurricane Center and the other one is with a correction for the estimated undercount associated with a changing observation network. Two different SST time series are considered: the Met Office’s HadISSTv1 and NOAA’s Extended Reconstructed SST. Given the nature of the data (counts), a Poisson regression model is adopted. The selection of statistically significant covariates is performed by penalizing models for adding extra parameters and two penalty functions are used. Depending on the penalty function, slightly different models, both in terms of covariates and dependence of the model’s parameter, are obtained, showing that there is not a “single best” model. Moreover, results are sensitive to the undercount correction and the SST time series. Suggestions concerning the model to use are provided, driven by both the outcomes of the statistical analyses and the current understanding of the underlying physical processes responsible for the genesis, development, and tracks of tropical storms in the North Atlantic basin. Although no single model is unequivocally superior to the others, the authors suggest a very parsimonious family of models using as covariates tropical Atlantic and tropical mean SSTs.