Ricardo Todling, NASA

A Second Look at a Filter-Free Approach for Hybrid Ensemble Assimilation

Amal El Akkraoui


For over a year and a half the Global Modeling and Assimilation Office has been running a hybrid 3dVar as its quasi-operational system. This system relies on an ensemble of analyses generated by an ensemble Kalman filter (EnKF) solver. The benefits of this system over our previous 3dVar system have been clearly demonstrated, and corroborate what other enters have found when making similar transition from 3dVar to hybrid 3dVar.

Examination of ensemble spread, and considerations of the role played by ensemble re-centering and additive inflation have motivated us to implement a filter-free approach to generate the analysis members. In it, the EnKF solver is removed, and an ensemble of analysis is created by simply using the additive inflation mechanism applied to the central, hybrid, analysis. A couple of years ago, we have shown results of this simple approach to work rather well; being acceptably comparable to those obtained when using the EnKF.

The present work represents a second examination of the issue. This time around, we plan to examine the significance of the kinds of perturbations used in the additive inflation mechanism. Namely, we study the impact of replacing the typical 48-minus-24 hours forecast differences used for perturbations with simply randomly generated perturbations. Interpretation of the results and what they imply to hybrid ensemble data assimilation approaches will be discussed at the time of the presentation.

File available here