Predictions of Damages from Atlantic Tropical Cyclones in Two Scales: A Hierarchical Bayesian Study

We propose joint distributional models of tropical cyclone characteristics and the economic damages they inflict, using a hierarchical Bayesian framework. This is done in two temporal scales: (i) at the scale of a hurricane season, where we propose to study the joint distribution of tropical cyclone frequencies, the probability that they may cause economic damages and the monetary value of the damages, and (ii) at the individual cyclone level where we model the monetary value of damage caused by hurricanes. Standard climatic features and known climate indices are used as covariates. The hierarchical Bayesian models obtain predictive distributions, and we report excellent predictive performances at both the seasonal and individual hurricane scales. Additional robustness studies are conducted to establish that the predictions are insensitive to modeling or algorithmic choices. Our analysis projects average annual damages of approximately 10.27 billion dollars for the United States, primarily due to high-intensity tropical cyclones.