Forecast sensitivity to the observation error covariance in variational data assimilation

The development of the adjoint of the forecast model and of the adjoint of the data assimilation system (adjoint-DAS) make feasible the evaluation of the derivative-based forecast sensitivity to DAS input parameters in numerical weather prediction (NWP). The adjoint estimation of the forecast sensitivity to the observation error covariance in the DAS is considered as a practical approach to provide all-at-once first order estimates to the forecast impact as a result of variations in the specification of the observation error statistics and guidance for tuning of error covariance parameters. The proposed methodology extends the capabilities of the adjoint modeling tools currently in place at major NWP centers for observation sensitivity and observation impact analysis. Illustrative numerical results are presented with the fifth-generation NASA Goddard Earth Observing System (GEOS-5) atmospheric DAS and its adjoint.