Predicting Extreme Hurricane Winds in the United States

We use a POT (Peaks over Threshold) model used to evaluate distribution
of extreme winds within near coastal regions of the United States. Using
maximum likelihood methods we estimate the return levels for various long
range return periods for each coastal region. We show that the return levels
depend on climate variables such as ENSO, and NAO. Finally we demonstrate
the usefulness of the Bayesian approach with the POT model. We use this
model to analyze the relationship between several sets of predictor and
extreme hurricane winds. This model allows us to extend the historical
record by incorporating measurement error and the use of a reference
distribution for missing predictor values.

ALSO, I will be discussing:

1. Alternative prior formulations for GPD parameters. (Non-informative
priors are informative).
2. WinBUGS program and sampling methods
3. GPD/GEV sampling module in WinBUGS (If I get it working...)

Dr. Thomas Jagger
Florida State University
Department of Geography