To improve the predictive capability of current general circulation models, a new approach of systematic evaluation and improvement of cloud parameterizations has been proposed (Xu et al. 2002). This new technique classifies satellite data into distinct cloud systems defined by their types (e.g., trade cumulus, stratus and deep convective systems). These cloud systems are then matched with nearly simultaneous atmospheric state from ECMWF data. The atmospheric data are used to provide inputs for cloud model (e.g., cloud parameterizations of single column models, cloud-resolving models and large-eddy simulation models) simulations. This approach takes cloud model evaluation beyond the traditional methods into tests of large statistically robust ensembles of matched atmospheric states==> cloud model==> satellite cloud system data comparisons and emphasizes the comparison of the higher-order distributions of some subgrid-scale characteristics of cloud systems between satellite observations and cloud models, instead of the grid-mean characteristics.

This study presents a comprehensive comparison of EOS satellite cloud and radiative observations with those produced by the ECMWF forecasting model for tropical convective systems during the March 1998 period. To facilitate a direct comparison, the ECMWF predicted cloud fields over a grid size of 0.5625 degree x 0.5625 degree are distributed in one hundred subcolumns using the maximum-random overlap assumption (Klein and Jacob 1999). The cloud optical properties and the radiative fluxes are then calculated for each subcolumn with the delta-four-stream Fu-Liou radiation parameterization. The same selection criteria (i.e., optical depth greater than 10 and cloud top height above 10 km) as used in the satellite analysis of cloud systems (Xu et al. 2002) are used to select the subcolumns within the ECMWF grids nearly collocated with satellite observed cloud systems for producing the cloud and radiative statistics. The probability density functions (PDFs) of cloud and radiative properties are compared against the satellite observations.

The PDFs of all variables for the combined 29 cloud systems observed during March 1998 are quantitatively similar between satellite observations and ECMWF cloud fields, except for TOA albedo which exhibits longer tail on small albedos for ECMWF cloud fields. Most of the PDFs from both satellite observations and ECMWF cloud fields are rather close to the Gaussian distribution (cloud ice diameter, longwave outgoing radiation, albedo, cloud effective height, cloud effective pressure and cloud effective temperature) or to the exponential distribution (cloud optical depths, liquid water paths, ice water path). Small differences in the PDFs between satellite observations and ECMWF cloud fields are present for most variables. For example, the ECMWF optical depth PDF exhibits a much longer tail than the satellite observations, due to the limitation in the measurements. The opposite is true for the ice water path. These differences result in larger ice diameters and smaller albedos in the ECMWF cloud fields. On the other hand, the PDFs of cloud effective height, pressure and temperature are sensitive to the chosen threshold for determining the cloud tops. A threshold of 0.7 on the visible optical depth, equivalent to infrared emissivity of 0.5, which is typical of thin cirrus clouds, results in PDFs of cloud effective height, pressure and temperature rather similar to the satellite observations.

The PDFs for a few individual cloud systems will also be contrasted to pinpoint the causes of the differences between the satellite observations and the ECMWF cloud fields discussed above.