Hybrid 4D EnVar for the NCEP GFS: Sensitivity experiments and plans for operational implementation

The ability to incorporate flow-dependent, ensemble-based representations of background error covariances into variational data assimilation has recently been developed and tested for use in the NCEP data assimilation system (Gridpoint Statistical Interpolation, GSI) by utilizing the augmented control variable method. Experiments with the hybrid 3D-Ensemble-var (EnVar) system for the NCEP Global Forecast System (GFS) model have shown that the hybrid paradigm can yield substantial forecast error reduction relative to a 3DVAR-based control system in both single and dual-resolution paradigms. The hybrid 3D EnVar algorithm was implemented into NCEP operations for the GFS in May 2012, relying on an EnKF to update the ensemble.

By taking the existing 3D EnVar algorithm in GSI and allowing for four-dimensional ensemble perturbations, coupled with the 4DVAR infrastructure already in place, a 4D EnVar capability has been developed. The 4D EnVar algorithm has a few attractive qualities relative to 4DVAR, including the lack of need for tangent-linear and adjoint model as well as reduced computational cost. Results for the GFS from an observing system simulation experiment (OSSE) show that analysis error was reduced when going from 3D- to 4DEnVar.

The results from various sensitivity experiments using real data will be presented. In particular, emphasis will be placed on the use of an outer loop as is done within traditional 4DVAR. Additionally, balance and noise related issues will be explored and addressed through the use of a variety of dynamic constraints. Lastly, plans will be discussed regarding the operational implementation 4D EnVar as well as potential long term development path for the NCEP GDAS/GFS.