Rahul Mahajan, NOAA/NCEP/EMC

Hybrid 4DEnVar implementation for the NCEP GFS and future directions

Daryl Kleist, Catherine Thomas, Yanqiu Zhu and John Derber


NCEP operations for the Global Forecast System (GFS) and Global Data Assimilation System (GDAS) recently implemented a 4D extension of the hybrid 3DEnVar that includes temporally evolving error estimates of the background. The hybrid 4D EnVar algorithm has a few attractive qualities for an operational center relative to traditional 4DVAR, most notably the lack of need for developing and maintaining a tangent linear and adjoint of the forward model as well as reduced computational cost. Results at reduced resolution have been extremely encouraging. The results from various sensitivity experiments will be presented. Also, results from a full resolution parallel experiment with additional upgrades will also be presented.

Beyond this initial implementation of hybrid 4D EnVar, various enhancements are still in their early phase such as improved data selection, the role of the static error covariance, incremental analysis update (IAU) and outer loop configuration. 4DIAU has been explored as a means of gently forcing in the 4D increment into the model and will serve as a replacement for the full field digital filter for initialization. Using the IAU will hopefully allow the gradual spinup of moisture related quantities as well as retaining the increment. The IAU will also be used to initialize the ensemble. Analogous to the outer loop configuration in 4DVAR and so-called Quasi-Outer Loop of the EnKF, an outer loop can be run as part of 4D EnVar whereby the non-linear model is re-run using the increment at the beginning of the assimilation window. This may prove to be particularly important for dual-resolution configurations that use low-resolution ensemble members within high-resolution deterministic assimilation.

File available here