Hybrid-Gain Versus Hybrid-Covariance Data Assimilation

The term 'Hybrid data assimilation' refers to blending static and flow-dependent, ensemble-based information when calculating an analysis increment. 'Hybrid-covariance' or 'Hybrid ensemble-variational' techniques use a blend of static and ensemble-based background error covariance estimates to calculate the analysis increment. 'Hybrid-gain' algorithms blend the Kalman gains calculated from static and ensemble-based covariance estimates, which, in certain circumstances, is equivalent to blending the analysis increments themselves. In this talk I will compare hybrid-gain and hybrid-covariance data approaches for atmospheric data assimilation using the operational NOAA global forecast system. Using the same static background-error covariance (B) estimate in each yields nearly identical skill, using a variety of metrics, although the hybrid-gain approach is computationally less expensive. We conclude that, in both cases, the static B acts to stabilize or regularize the cycling EnKF when the ensemble size is too small to adequately sample the space of non-decaying Lyapunov modes, and that, for our experimental configuration, the details of how the static B is applied are of secondary importance.