Asymptotic properties of multivariate tapering for estimation and prediction
Parameter estimation for and prediction of spatially or spatio--temporally correlated random processes are used in many areas and often require the solution of a large linear system based on the covariance matrix of the observations. In recent years, the dataset sizes to which these methods are applied have steadily increased such that straightforward statistical tools are computationally too expensive to be used. In the univariate context, tapering, i.e., creating sparse approximate linear systems, has been shown to be an efficient tool in both the estimation and prediction settings. The asymptotic properties are derived under an infill asymptotic setting. In this paper we use a domain increasing framework for estimation and prediction using multivariate tapering. Under this asymptotic regime we prove that tapering (one-tapered form) preserves the consistency of the untapered maximum likelihood estimator and show that tapering has asymptotically the same mean squared prediction error as using the corresponding untapered predictor. The theoretical results are illustrated with simulations.