A traditional method of improving cloud parameterization employs detailed studies using data collected during intensive field campaigns over a limited area and a short duration. The nonlinearity of cloud processes requires that observations be made on all the relevant modeling scales. Such field data are clearly not adequate. A traditional method of evaluating the performance of cloud parameterizations in GCMs is to verify the simulation of GCM using monthly mean global and regional satellite and surface data. This is not adequate to isolate the errors associated with cloud parameterization from those associated with other components in a GCM.
In order to close the gap between these two traditional methods, we propose a new method of analyzing satellite data for evaluating the performance of cloud parameterizations in GCMs, numerical weather prediction models, and cloud resolving models (hereafter, cloud models). Instead of gridding satellite data to some standard GCM grids, we classify them into distinct cloud systems or "cloud objects" defined by cloud types (e.g., trade cumulus, stratus, and deep convection). We identify a large ensemble of cloud systems from NASA Earth Observing System (EOS) satellite cloud and radiation data and match them with nearly simultaneous atmospheric state data from ECMWF. The atmospheric state data (temperature, water vapor, wind and advective tendencies) are used to provide the initial and forcing information for cloud model simulations. The merged satellite/ECMWF cloud system data will then be saved as a function of cloud type and will be used to evaluate and improve current performance of cloud models. The advantage of this new data analysis method is to take cloud model evaluation beyond the more traditional methods into tests of large statistically robust ensembles of matched atmosphere=> cloud model=> satellite cloud data comparisons. For example, frequency distributions of various cloud and radiation parameters for a given atmospheric state and cloud type can be directly compared between ensemble satellite data and cloud model simulations. The results of these comparisons can be used to further improve cloud model performance.