FURTHER TESTS OF A GRID SYSTEM FOR GLOBAL NUMERICAL PREDICTION

Abstract

Two smoothing techniques are tested as a practical means of allowing a larger time step in the numerical integration of a primitive equation free-surface model. The numerical integration uses a finite-difference grid and operators based on the method of Kurihara and Holloway. A time step six times larger can be used with a corresponding six-fold decrease in computer time, by implementing the weighted averaging procedure given by Langlois and Kwok in their description of the Mintz-Arakawa general circulation model. A Fourier filtering scheme permits the use of a time step 10 times larger, and results in a five-fold improvement in computer time. After 10 days, the geopotential and wind fields obtained with these techniques still closely resemble the unsmoothed fields, the closest correspondence being found with the Fourier filtering technique. In another set of experiments, steady-state solutions to special cases of the governing analytic equations are used as initial conditions in a test of the accuracy of the grid and operators. These steady-state solutions are preserved satisfactorily for the 10-day integration period.