A comparison of 3D-Var and Ensemble Data Assimilation Methods Using the NCEP GFS

A comparison is presented between 3D-Var and an ensemble-based data assimilation method using real observations in a global forecast model. The observation data consists of NCEP's operational data stream with the exception of the satellite radiance data. The forecast model is a 2004 version of NCEP's GFS model, run at T62 resolution. The ensemble data assimilation method is the ensemble square-root filter (EnSRF), run with a 100-member ensemble and tested with three different model error parameterizations. We demonstrate that the EnSRF produces slightly improved analyses relative to 3D-Var, measured in the fit of 6-h to radiosondes or 2-day forecasts to analyses. This demonstrates the potential operational feasibility of ensemble filters for operational atmospheric data assimilation.