Estimation of Clouds in Atmospheric Models

Tomislava Vukicevic
Cooperative Institute for Research in Atmosphere, Colorado State University

Accurate estimates of cloud properties are required for improved understanding of feedbacks within the atmospheric system and the system's predictability on wide range of spatial and temporal scales from individual storms to climate. Quantitative observations of clouds and precipitation are typically obtained by indirect, remote sensing methods. Although considerable progress has been made in remotely sensing and retrieving the cloud properties, complex 3D cloud structure and interaction with the atmospheric environment is not well specified from observations alone. We propose that accurate estimates of evolving 3D cloudy atmosphere on cloud resolving scales could be derived by assimilation of remote sensing observations of the cloudy atmosphere into a cloud resolving dynamical model such that they optimally constrain what controls the model solution. The model solution is controlled by initial and boundary conditions and model parameterizations.

The complex problem is initially approached by applying a new 4D-variational (4DVAR) research data assimilation system with a cloud resolving model to the assimilation of the GOES (Geostationary Operational Environmental Satellites) imager observations. The study results show that the observations can significantly improve modeled cloud, resulting in a 3D distribution of the hydrometeor mixing ratio and number concentration parameters with nearly zero mean error and a small standard deviation in the observation space. The strength of constraint by the selected observations depends upon conditions in the model to support the cloud formation and on the information content of the observations.

The sensitivity of data assimilation results to the assimilation technique was also studied. The experiments with the model error indicate that this error in the form of a generic linear forcing in the governing equations, which was adopted from other 4DVAR data assimilation studies, is not suitable for the cloud resolving data assimilation. This is because the actual, unknown error, is likely highly nonlinear in the state space and consequently cannot be well represented with Gaussian statistics. The estimation of so called physical parameters may be more appropriate in this case. An illustrative example is produced with the Lorenz 3-component chaos model.

Last Update: September 29, 2005