Bayesian prediction of mean square errors with covariates

Abstract

Estimation of mean square prediction error of wind components is required in the optimal interpolation (OI) process in numerical prediction of atmospheric variables. Previous work has suggested that statistical models with log-linear scale parameters which include covariates can be used to predict mean square prediction errors. However, the parameters of the statistical relationships appear to change over time. A procedure is described to recursively update the estimated parameters. Data from July of 1991 are used to fit the model parameters and to study the predictive ability of the recursive procedure. This preliminary investigation indicates that observational and first guess wind components can be helpful in predicting mean square prediction error for wind components.... Hierarchical model, Gaussian model with log-linear scale parameters