Statistical Corrections to Numerical Predictions III

Abstract

A simplified numerical model is used to test new proposals for the statistical correction of numerical predictions. Two procedures are reported here first, time interpolations to allow application of the previously reported STAT method (Faller and Schemm, 1977) when correct data for prediction verification are available only after several time steps of a prediction; and second, the MUST (multi-step statistical) method in which corrections are applied every few time steps, when correct verification data are available. The interpolations were found to introduce systematic errors in the regression equations, thus rendering the statistical correction procedure useless. The MUST-II procedure (in which individual regression equations are determined for each grid point) was found to give excellent results, computationally stable and accurate for long periods and superior in every way to all previously tried methods. Application of MUST-I, in which a single regression equation is used for all grid points, to the NMC Barotropic-Mesh model gave poor results, probably because of the large spatial variability of the atmospheric dynamics associated with geographical factors.