Assessment of Ensemble-Based Observation Impact Estimation for the FV3GFS

The ensemble forecast sensitivity to observations (EFSO) formulation has been implemented for application within the context of ensemble square root filter (EnSRF) data assimilation at the National Centers for Environmental Prediction (NCEP). As with the adjoint-based forecast sensitivity observation impact (AFSOI) approach, the ensemble-based observation impact approach effectively enables a simultaneous computation of estimated forecast impacts and sensitivities for any and all observations assimilated in a numerical weather prediction (NWP) system. The NCEP FV3GFS, which is scheduled to become operational in early 2019, applies 4D ensemble-variational (4DEnVar) data assimilation, and as such, requires an ensemble of short range forecasts to provide flow-dependent background uncertainty information. As currently configured, the EnSRF data assimilation algorithm is applied to assist in the assignment of initial conditions for the aforementioned ensemble of short range forecasts. Following the EFSO approach described in Kalnay et al. 2012, the ensemble of analyses resulting from the EnSRF update are used in representation of analysis-error covariance, and accordingly in approximation of the EnSRF Kalman gain. EFSO calculations are then based on the projection of this approximate Kalman gain to an evaluation forecast time using the FV3GFS forecast model.

Estimates of observation analysis impact and 24 hour forecast error reduction impact based on EFSO are calculated for several weeks of 4DEnVar and experimental pure EnSRF FV3GFS cycling. The extent to which true analysis impact and true 24 hour forecast error reduction agree with the aforementioned estimates is assessed for the moist total and dry total energy norm metrics. Furthermore, we examine differences between estimates of observation analysis impact and 24 hour forecast error reduction impact based on EFSO, with the intent of highlighting complementary information.