Time Series Prediction of Hurricane Landfall.
Abstract
Greater accuracy is required in predicting hurricane landfall in order to insure timely evacuation. A significant result of this research is the classification of past storms by time series stationarity category which relates to direction of movement. Also, a psi-weight representation of the forecast is used to develop a bivariate Normal confidence ellipse for the threshold autoregressive model. It is shown that the landfall of North Atlantic hurricane and tropical storms can be accurately predicted by modeling the storm track as a bivariate (latitude and longitude) fifth-order autoregressive process. A threshold approach is used to allow model parameters to change as the storm moves to a new region of the ocean. For test cases, operational average 72 hours prediction error is at least three standard deviations below the average error of the official forecasts issued by the National Hurricane Center.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1986
- Accession Number
- ADA170742
Entities
People
- Thomas F. Curry
Organizations
- Air Force Institute of Technology