Background Error Specification for the 127-Layer GFS

In June 2019, NCEP implemented Version 15 of the GFS into operations as part of the Next Generation Global Prediction System (NGGPS). This implementation replaced the Global Spectral Model (GSM) dynamic core with the Finite Volume on a Cubed Sphere core (FV3) of GFDL and was the first step towards the Unified Forecast System (UFS) to be used across all spatial and temporal scales. The goal for this implementation was to adapt as much of the existing system’s infrastructure to the new dynamic core as possible, using a comparable resolution and much of the same physics. The following Version 16 implementation of the GFS (GFSv16) looks to build upon the previous one, introducing advanced physics and increasing the vertical resolution and model top.

One of the biggest data assimilation challenges for the GFSv16 implementation concerns raising the model top from approximately 55 km to 80 km, with the vertical resolution increasing from 64 layers to 127 layers. New static background error statistics need to be derived as there is currently no climatological background error information above the operational model top. The traditional method of using lagged forecast pairs, known as the NMC method, proves challenging due to the extensive model spin up in the highest levels. An Ensemble Kalman Filter (EnKF) only system could also be used to generate climatological statistics, though care needs to be taken with the presence of large ensemble spread towards the model top.

In this presentation, we will compare static background error statistics derived from the NMC method with those derived from an EnKF only system and discuss their impacts on the assimilation within the GFSv16 framework. Results from single observation impact tests and low resolution cycling experiments will be shown.