Statistical Modeling of Storm Counts

Abstract

A statistical model is presented of a recently compiled record of monthly extratropical storm counts for the mid-Atlantic coast of the United States for the period 1942?83. The counts are modeled as a Poisson process with nonstationary mean function. The mean function is decomposed into a secular component and a seasonal cycle. Because the form of the secular component is unknown, a nonparametric regression approach suitable for Poisson data is used to estimate it. The estimated secular component is generally constant through the 1950s, then declines through the 1970s. The estimate is found to be statistically significant. A Fourier series involving two harmonics is fit to the seasonal cycle. A preliminary check indicates that the seasonal cycle remains stable through time. Some diagnostics based on suitably defined residuals are presented that generally confirm the goodness-of-fit and distributional assumptions underlying the model.