Stephen G. Penny

Two methods for incorporating a time‐invariant, high‐rank covariance estimate in an ensemble‐based data assimilation system for global weather prediction are compared. The hybrid‐covariance approach linearly combines the static and ensemble‐based covariance estimate in a four‐dimensional variational solver, whereas the hybrid‐gain approach blends analysis increments computed separately using a three‐dimensional variational solution and an ensemble Kalman filter solution. Results show that the simpler and less expensive hybrid‐gain approach performs similarly if the incremental normal‐mode balance constraint applied to the ensemble‐part of the hybrid‐covariance update is turned off.