Gaussian linear state-space model for wind fields in the North-East Atlantic

A multisite stochastic generator for wind speed is proposed. It aims at simulating realistic wind conditions with a focus on reproducing the space-time motions of the meteorological systems. A Gaussian linear state-space model is used where the latent state may be interpreted as regional wind conditions and the observation equation links regional and local scales. Parameter estimation is performed by combining a method of moments and the expectation-maximization algorithm. The model is fitted to 6-hourly reanalysis data in the North-East Atlantic. It is shown that the fitted model is interpretable and provides a good description of important properties of the space-time covariance function of the data, such as the non-full symmetry induced by prevailing flows in this area. Copyright