Improving balance in the NCEP Hybrid Ensemble-Var data assimilation system

Data assimilation strives to optimally combine observations with a model estimate to provide a more accurate representation of the current state. However, the corrections made to individual state variables may not be in dynamic balance with one another. These imbalances in the initial conditions can create unphysical inertial-gravity waves that propagate in the model and degrade the forecast. Imbalances in the moisture and cloud variables can also prohibit clouds that are created in the analysis to be sustained throughout the following forecast.

NCEP's Gridpoint Statistical Interpolation data assimilation system (GSI) employs the tangent-linear normal-mode constraint (TLNMC) to improve the balance of the analysis increments from within the cost function, rather than acting as a post-processor. It calculates the incremental time tendencies using a simplified tangent-linear version of the non-linear forecast model that are projected onto fast gravity wave modes from which correction terms to the analysis increment are computed. Modifications to the time tendency model, including the introduction of moist physics, have been applied as a method to improve balance within the initial state and thereby improve forecast skill through the production of more representative tendencies. Expanding from the hybrid 3DEnVar to 4D, we also explore the impact of applying the TLNMC over multiple time levels. Results from other techniques, such as incremental analysis update and the full field digital filter in the 3D and 4D hybrid settings within the GFS, will also be discussed. Particular interest is taken to explore the effect of these initialization techniques on the spin-up and spin-down of balanced clouds within the analysis.