Effect of ENSO on Queensland seasonal landfalling tropical cyclone activity

A statistical model for predicting seasonal tropical cyclone landfalls in Queensland, Australia using an index of the El Nino-Southern Oscillation (ENSO) is presented. The approach uses a generalised linear model (GLM) to relate seasonal counts to the Southern Oscillation Index (SOI). A Bayesian methodology is employed to estimate parameters of the model. The available tropical cyclone record is first separated into historical (1910/1911-1959/1960) and instrumental (1960/1961-2004/2005) eras. Historical counts, which are considered to be less reliable observational sources, are used to specify informative prior distributions when fitting the GLM to instrumental counts with the Bayesian approach. The inclusion of historical information is found to lead to increased certainty in parameter estimates when compared to a model where historical counts are excluded and a non-informative prior model used. Predictive distributions are given, which allow inferences on seasonal landfall activity, given pre-seasonal values of the SOL A cross-validation procedure shows that the model incorporating historical information outperforms, in terms of mean-squared prediction error, both the non-informative prior model and a model without the SOI-predictor (climatology). A trend analysis highlights possible decadal variability in the relationship between ENSO and seasonal tropical cyclone activity. Copyright (C) 2007 Royal Meteorological Society.