A Reduced Spectral Transform for the NCEP Seasonal Forecast Global Spectral Atmospheric Model

Abstract

A reduced spectral transformation is applied to the NCEP atmospheric global spectral model for operational seasonal forecasts. The magnitude of the associated Legendre coefficient provides a basis for this new transformation, which is a simple modification of a traditional reduced grid spectral transform. This transformation can be called a ?reduced spectral? method because its Fourier and Legendre transformations need less computation than the traditional uniform full grid or reduced grid methods. In addition, the reduced spectral method saves an extra 50% on Legendre transformations and is easy to load balance for massively parallel computing under certain decompositions. A comparison, without model physics, among reduced spectral, reduced grid, and full grid transforms indicates that they have negligible differences up to more than a half-month integration and small differences up to a 1- month integration. Extended integrations without physics for up to 4 months show that there is proximity of zonal symmetry between reduced spectral and full grid transforms. When the comparison includes model physics, the results show negligible differences up to 7 days; but the chaotic nature, known as an internal variability, is amplified by physical parameterizations and produces significant differences among these methods after a 1- month integration, which is expected. The seasonally averaged results from 10 years of AMIP-type runs are similar between the reduced spectral method and the full grid method, indicating that they have similar model climatology. These experiments indicate that this reduced spectral transform can be used for short-range as well as seasonal or climate predictions.