Quantification and Climatic Attribution of Covariance in Tropical Cyclone Activity across Teleconnected Basins

Rigorous statistical quantification of natural hazard covariance across regions has important implications for risk management, and is also of fundamental scientific interest. We present a multivariate Bayesian Poisson regression model for inferring the covariance in tropical cyclone (TC) counts across multiple ocean basins and across Saffir-Simpson intensity categories. Our working hypothesis is that such covariability results from the influence of large-scale modes of climate variability on local environments that can alternately suppress or enhance TC genesis and intensification, so our model also quantifies the covariance of TC counts with various climatic modes in order to deduce the source of inter-basin TC covariability. The model explicitly treats the time-dependent uncertainty in observed maximum sustained wind data, and hence the nominal intensity category of each TC. Differences in annual TC counts as measured by different agencies are also formally addressed. The Monte Carlo output of the model can be probed for probabilistic answers to such questions as: - Does the relationship between different categories of TCs differ statistically by basin?

- Which climatic predictors have significant relationships with TC activity in each basin?

- Are the relationships between counts in different basins conditionally independent given the climatic predictors, or are there other factors at play affecting inter-basin covariability?

- How can a portfolio of insured property be optimized across space to minimize risk?

Although we present results of our model applied to TCs, the framework is generalizable to covariance estimation between multivariate counts of natural hazards across regions and/or across peril types.