Abstract
Time and frequency domain models that relate a vector time series at any location to a measured series at another location are presented and evaluated by application to the prediction of coastal currents. The time domain model belongs to the general class of cross-autoregressive integrated (CARI) models based on classical time series analysis methods. The frequency domain model is a spectral component (SC) model, which relates the Fourier spectrum at different stations. The errors resulting from the use of both model types to predict measured currents off San Francisco, California, were computed. It was found that of the CARI models, a six parameter model was best, although a simpler three parameter model was almost as good. The main drawback to the CARI model was the nonstationarity of the formulation, requiring several years of data to develop a stationary model for a particular season. The SC model was more stationary and had smaller errors than the best CARI model. It is concluded that frequency domain models are the best method of predicting two-dimensional vector time series from other series.
