Computational Optimization for Climate Prediction

Developing climate prediction model integration schemes which can provide realistic scenarios on long time scales with limited computing resources is the challenge of this research. One method to succeed in this task is to increase the integration timestep. We have tested several techniques which may prove useful. The most successful was applied to the shallow water equations over a spherical surface in which the prediction model was represented in its normal modes, the high frequency modes were balanced while the low frequency modes were predicted. Experiments which we will describe extend this procedure to a state-of-the-art model (the NCAR/CCM3). We have taken the predicted data from each timestep of the model integration, projected it onto Hough modes, separated the modes into fast and slow components, integrated the slow components with a timestep three times longer than that used in the standard model run and balanced the fast modes. The modal data was then reconverted to model format and returned for the next iteration. Seasonal model output using this procedure was compared to the standard model run output and the results of ten realizations showed the both calculations gave identical results within model variability.