Full resolution cycled data assimilation with FV3-JEDI

The Joint Effort for Data assimilation Integration (JEDI) -- led by the Joint Center for Satellite Data Assimilation (JCSDA) -- is an inter-organizational endeavor to develop a common framework for performing data assimilation on model native grids. FV3-JEDI provides an interface between the generic components of JEDI and weather prediction models based on the Finite Volume Cubed-Sphere (FV3) dynamical core. These models include NOAA’s Global Forecast System (GFS) and NASA’s Goddard Earth Observing System (GEOS).

FV3-JEDI implements all the necessary components to perform 3DVar, 3DVar-FGAT, 4DVar and 4DEnVar data assimilation in their static, pure ensemble and hybrid forms. In this paper the efforts towards using FV3-JEDI to perform high resolution cycled experiments with GFS and GEOS are outlined. Cycled experiments are performed over a month-long period and using the full resolution model states for the background and to compute departures. A realistic number of observations are ingested using the Unified Forward Operator (UFO) and using a range of quality control filters similar to those used in production systems. Short range forecasts are produced and compared with operational forecasts of the same period. For hybrid flavors of data assimilation, the experiments replay to existing production ensembles rather than using JEDI to generate a new ensemble on the fly.

As well as the basic linear algebra operations required for data assimilation the strategy for implementing localization and covariance using B matrix on Unstructured Mesh Package (BUMP) is outlined. Further, the issue of computing balance for a cubed sphere grid is discussed. NASA’s Global Modeling and Assimilation Office (GMAO) has invested considerable effort to develop the tangent linear and adjoint of the FV3 dynamical core as well as certain GEOS physics routines. This tangent linear and adjoint model is available through FV3-JEDI and allows for 4DVar data assimilation with GFS (dry physics) and GEOS (moist physics). Cycled 4DVar experiments have been undertaken for GFS and GEOS and are shown here against experiments with other flavors of data assimilation. Attempts to use the GEOS physics with GFS background trajectories are also described.

A considerable saving can be achieved when performing four-dimensional data assimilation in a single executable. The prevents expensive reading and writing of high-resolution model states and, in the case of multiple outer loop 4DVar, prevents repeated model initialization steps. Since forecast models are typically developed independently of the data assimilation algorithms some effort is required to harmonize the two systems. The work that has been undertaken to run FV3-JEDI with GFS and GEOS in a single executable is discussed.