Optimal Spatial Representations for Numerical Weather Prediction Models Based on Normal Mode Analyses.

摘要

A method is devised to determine the computational errors associated with space truncation in numerical weather prediction models. This method is based on evaluating the normal modes of the linearized form of the prediction model. The modes are separated into their physical and computational parts. A special procedure has been devised to make this separation definitive. It is the computational modes which give rise to computational errors. To reduce the impact of these modes, their contributions are filtered out of the initial data used to begin a forecast. Although computational mode amplitude will develop during integration because of nonlinearity, the impact of the computational errors can be controlled by the filtering technique described here. If initial filtering is not adequate, additional filtering may be applied periodically throughout the integration period. (Author)