Proposed Improvements to the NCEP GFS Hybrid 4DEnVar Configuration

The operational data assimilation system at the National Centers for Environmental Prediction (NCEP) for the Global Forecast System (GFS) was recently successfully updated to a hybrid 4DEnVar. Following this initial implementation, work is ongoing to improve several components of the data assimilation configuration. The current hybrid system uses a dual-resolution framework (T1534/T574) with an 80 member ensemble at a reduced horizontal resolution of T574 to prescribe a flow-dependent background error covariance to update the analysis at a much higher deterministic control forecast at T1534. Work done at lower resolution T670/T254 by Lei and Whitaker (2016), comparing the relative benefits of increasing ensemble size to ensemble resolution suggests greater benefits in increasing the ensemble resolution compared to ensemble size. Experiments at full operational resolution suggest neutral to little improvement in increasing the ensemble resolution from T574 to T878. Further increase in ensemble resolution is currently under testing at NCEP.

The hybrid scheme gives a weight of 87.5% to the ensemble derived background error covariance and 12.5% to the static component at all model levels. Work is underway to estimate these weights in the vertical following a method described in Ménétrier and Augliné (2015) along with horizontal and vertical localization scales.

The GFS forecast model uses a digital filter initialization (DFI) technique when the model is initialized from the analysis. Several studies have shown benefit in using an incremental analysis update (IAU) to gently introduce a 4D analysis increment as the model integrates. Results from replacing the DFI with the 4DIAU will also be presented.