A state-of-the-art ensemble data assimilation system has been implemented for several versions of NCAR's Community Atmosphere Model (CAM) using the Data Assimilation Research Testbed (DART). This allows CAM to be run in a numerical weather prediction (NWP) framework where it assimilates the observations used by operational NWP centers and produces short range forecasts. The ability to confront a climate model with observations in this way has not been available for models developed exclusively for climate prediction and simulation. The costs involved in developing more traditional variational-type data assimilation methods have precluded their implementation for models like CAM or the GFDL AM models. However, only about a person month is required to implement a DART ensemble assimilation system for an existing climate model. This makes it practical to use data assimilation as part of any climate modeling system. Evaluating the weather prediction capabilities of a climate model can provide information that complements and enhances the results from long climate simulation runs.

The presentation begins by comparing assimilations from DART/CAM to operational assimilations from NCEP. The CAM assimilations are comparable in quality and are better in the lower troposphere. A discussion of how a data assimilation capability can assist climate model evaluation and development makes up the majority of the presentation. Identifying the systematic errors in the model's short-term forecasts can elucidate the relative quality of different parameterization schemes without requiring long model runs. It is also possible to tune parameters in model physics schemes directly with data assimilation. Results using the DART/CAM facility to tune a parameter of a CAM gravity wave drag scheme are presented.