Hurricane Forecasting Using a Multivariate Spatial Functional Linear Model

• Activity Number: 81
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Christopher Krut*+ and Montserrat Fuentes and Brian J. Reich
• Companies: North Carolina State University and North Carolina State University and North Carolina State University
• Keywords: Spatial Statistics ; Functional Linear Model ; Hurricanes ; Functional Data ; Bayesian Statistics

Abstract

Hurricanes are massive weather systems with potentially large costs in terms of human life and property damage. Timely and accurate forecasts are a critical component in mitigating the impact of these storms. The official forecast is currently generated by experienced forecasters who use output from deterministic numerical models and other available information. The following paper proposes a multivariate spatial functional linear model for refining existing forecasts. In the spirit of the functional data approach we view the tracks and intensity of storms as a 3 dimensional function of time. Using this functional data representation we construct a multivariate spatial functional linear model. Spatially varying coefficient functions are obtained, in the usual way, by placing Gaussian process priors on the coefficients from a basis expansion of the coefficient functions. Working within the Bayesian framework we perform inference using standard MCMC methods. Model performance is evaluated via a simulation study. It is demonstrated, using historical data, that the multivariate spatial functional linear model can be an effective tool for refining existing hurricane forecasts.