Abstract
A case study applies state space models and the Kalman filter to forecasting the path of hurricanes. The track of a hurricane is forecast by using its previous motion together with the historical record of storms which follow a similar path. Criteria are described for selecting the predictor storms, which are then incorporated into the measurement equation of the state space model. Linear and quadratic trend structural time series models are used in the transition equation of the state space model and forecasts are produced using the Kalman filter. Forecasts are obtained for hurricanes Andrew (1992), Bob (1991), Hugo (1989) and Gilbert (1988). The Kalman filter approach compares very favourably with the HURRAN and CLIPER benchmark models used by the National Hurricane Center.