While we have recently experienced an explosive growth in the number of satellite observations being assimilated in global atmospheric models, it still remains that a large number of cloud affected pixels are currently not being utilized. Considering that clouds are an essential element of weather and climate and very challenging to forecast, it is unfortunate that a large portion of the observing system containing cloud information goes without being assimilated. In response to this demand, several efforts for assimilating cloudy radiances are currently under way in the major operational NWP centers.

The main challenge for cloud data assimilation stems from the fact that historically large scale models have not resolved the individual cloud systems being depicted by the satellite measurements. Until routine global cloud resolving models become feasible, dealing with this scale mismatch becomes imperative for assimilating cloud observations. In recent years, moist parameterizations in GCMs have evolved towards advanced statistical cloud parameterizations in which the statistical properties of cloud within a model grid-column are represented using distributions of sub-grid scale total water. This statistical formulation provides a very natural path for utilizing high resolution cloud data, one in which the cloud measurements are used to constrain the probability distribution of water vapor and cloud condensate.

Building upon the earlier work of Norris and da Silva (2007) with the GEOS-4 Data Assimilation System, we are developing practical methods to constrain the GEOS-5 statistical parameterization with satellite-retrieved cloud data. In this talk we present a novel approach based on Markov Chain Monte Carlo methods to estimate the total water probability distribution using MODIS retrievals of cloud parameters from both Terra and Aqua satellites. Unlike classical 3D and 4D-Var methods, this non-parametric approach requires no tangent linear models of moist processes, providing posterior probabilities of statistical parameters consistent with the cloud measurements.

We will present results from the assimilation of MODIS cloud retrievals to constrain the GEOS-5 global cloud distributions, using CloudSat, CALIPSO and other independent measurements for validation. As a case study, we will demonstrate how the representation of Hurricane Bill is much improved after MODIS cloud data is assimilated. The impact of cloud data assimilation on the global cloud radiative forcing will be validated using independent CERES observations as in Norris and da Silva (2007). We will also discuss the impact of MODIS observations on the forecast of cloud cover and on the land surface response.