Assessment of Ensemble Forecast Sensitivity to Observation (EFSO) Quantities for Satellite Radiances Assimilated in the 4DEnVar GFS

The Ensemble Forecast Sensitivity to Observation (EFSO) formulation for estimating the forecast impact of assimilating individual observations has been implemented at NCEP. For the GFS, this approach requires Ensemble Kalman Filter (EnKF) products as input, and has been implemented within the current source code that provides EnKF functionality at NCEP. The EFSO formulation incorporates the relationship between kalman gain and analysis-error covariance to construct observational increments that can be projected forward in time with a forecast model, enabling an estimate of the forecast impact due to assimilating observations. The ensemble of analyses resulting from the EnKF update can be used in the representation of analysis-error covariance, and accordingly the kalman gain. The expectation for the outer product of the difference between the analysis and background in observation space (d_ab) and the innovation (d_ob) is a source of information for specifying analysis-error covariance. As such, observational diagnostics are used in this study as a basis for determining the extent to which inflated or non-inflated EnKF analyses from GFS cycled experiments are appropriate in representing analysis-error covariance.

Three months of EFSO calculations are performed based on EnKF products output from a low resolution 3 month (DJF 2014/2015) cycling experiment that is intended to be representative of the 4dEnVar GFS. Based on the aforementioned experiment, IASI and CrIS channels with EFSO suggested net detriment are identified. The low resolution 3 month (DJF 2014/2015) cycling experiment is repeated, with the exception that the EFSO suggested detrimental IASI and CrIS channels are not assimilated. Standard forecast verification statistics for the two aforementioned cycling experiments are compared to quantify the impact of removing the EFSO suggested detrimental IASI and CrIS channels from the control observing system.

In the second part of this presentation, relationships between innovation and 4dEnVar GFS EFSO quantities for AIRS, IASI, CrIS, AMSUA and ATMS observations are identified. There is no basis for attributing these relationships to data quality; as such the innovation dependent contribution to EFSO variability may serve as guidance for improved assimilation of satellite radiances.