Bayesian Multiple Changepoint Analysis of Hurricane Activity in the Eastern North Pacific: A Markov Chain Monte Carlo Approach

Abstract

Abstract A Bayesian framework is developed to detect multiple abrupt shifts in a time series of the annual major hurricanes counts. The hurricane counts are modeled by a Poisson process where the Poisson intensity (i.e., hurricane rate) is codified by a gamma distribution. Here, a triple hypothesis space concerning the annual hurricane rate is considered: “a no change in the rate,” “a single change in the rate,” and “a double change in the rate.” A hierarchical Bayesian approach involving three layers—data, parameter, and hypothesis—is formulated to demonstrate the posterior probability of each possible hypothesis and its relevant model parameters through a Markov chain Monte Carlo (MCMC) method. Based on sampling from an estimated informative prior for the Poisson rate parameters and the posterior distribution of hypotheses, two simulated examples are illustrated to show the effectiveness of the proposed method. Subsequently, the methodology is applied to the time series of major hurricane counts over the eastern North Pacific (ENP). Results indicate that the hurricane activity over ENP has very likely undergone a decadal variation with two changepoints occurring around 1982 and 1999 with three epochs characterized by the inactive 1972–81 epoch, the active 1982–98 epoch, and the inactive 1999–2003 epoch. The Bayesian method also provides a means for predicting decadal major hurricane variations. A lower number of major hurricanes are predicted for the next decade given the recent inactive period of hurricane activity.