Stochastic Representations for Model Uncertainty in the Ensemble Data Assimilation System

The ensemble data assimilation system is beneficial to express flow-dependent model errors. Furthermore, the effectiveness of this system depends on the accuracy of the flow-dependent background error covariance. However, the background error covariance is often underestimated due to limited ensemble size, sampling errors and model errors, which causes a filter divergence problem—the analysis state diverges from the nature stage ignoring the observation influence. As one of the remedies to solve this problem, the stochastic representations address the model-related uncertainties by perturbing the model tendency or parameters using a random forcing to replenish the insufficient model errors. In this study, we implemented a stochastic perturbation hybrid tendencies (SPHT) scheme, which perturbs both physical tendency and dynamical tendency using the random forcing, and assessed its impact on the spread of ensemble forecast and ensemble mean error.