A Comparison of Three Global Grids Used in Numerical Prediction Models

Abstract

Three finite-difference global grids [the original Kurihara (OK), a modified Kurihara (MK) and latitude- longitude (LL)] are tested by comparing numerical solutions with a barotrpic free-surface model to a high-resolution control run and by comparing forecasts with a general circulation model to observations. With the free surface model, 30-day integrations are made for three different resolutions of each grid and with three initial conditions two mathematical patterns and one 500 mb observed field. The LL grid performed well on the mathematical patterns, especially the case with zonal wavenumber 4. The numerical solutions with the high resolution MK grid also performed satisfactorily for both mathematical patterns. The OK grid did not perform as well, particularly on the case with zonal wavenumber 1. For the observed case, the LL grid in general had lower rms errors although the solutions did not depend as strongly on the three different grid types as the solutions for the mathematical patterns. For the three-dimensional cases, the GFDL nine-level model was used for 14-day forecasts for observed conditions in March and 3-day forecasts in November. Forecast sensitivity to the different grids is low for short range. The MK grid had the lowest 500 mb rms errors for the duration of both forecasts. Both the MK and LL grids were free of the problem of anomalously high geopotential heights over the North Pole that occurred with the OK grid.