Prediction of the Diurnal Cycle Using a Multimodel Superensemble. Part II: Clouds

Abstract

This study addresses the issue of cloud parameterization in general circulation models utilizing a twofold approach. Four versions of the Florida State University (FSU) global spectral model (GSM) were used, including four different cloud parameterization schemes in order to construct ensemble forecasts of cloud covers. Next, a superensemble approach was used to combine these model forecasts based on their past performance. It was shown that it is possible to substantially reduce the 1?5-day forecast errors of phase and amplitude of the diurnal cycle of clouds from the use of a multimodel superensemble. Further, the statistical information generated in the construction of a superensemble was used to develop a unified cloud parameterization scheme for a single model. This new cloud scheme, when implemented in the FSU GSM, carried a higher forecast accuracy compared to those of the individual cloud schemes and their ensemble mean for the diurnal cycle of cloud cover up to day 5 of the forecasts. This results in a 5?10 W m?2 improvement in the root-mean-square error to the upward longwave and shortwave flux at the top of the atmosphere, especially over deep convective regions. It is shown that while the multimodel superensemble is still the best product in forecasting the diurnal cycle of clouds, a unified cloud parameterization scheme, implemented in a single model, also provides higher forecast accuracy compared to the individual cloud models. Moreover, since this unified scheme is an integral part of the model, the forecast accuracy of the single model improves in terms of radiative fluxes and thus has greater impacts on weather and climate time scales. This new cloud scheme will be tested in real-time simulations.