Convective Meteorology (Mesoscale Dynamics)

Development of a Hybrid En3DVar Data Assimilation System and Comparisons with 3DVar, and EnKF for Radar Data Assimilation with Observing System Simulation Experiments

Rong Kong

School of Meteorology

11 March 2016, 3:00 PM

National Weather Center, Room 5600
120 David L. Boren Blvd.
University of Oklahoma
Norman, OK

A hybrid ensemble-variational (EnVar) data assimilation system is developed and tested based on the ARPS 3DVar and EnKF systems. The extended control variable method is used to combine the static and flow-dependent ensemble background error covariances. Particular emphasis is placed on direct assimilation of radar reflectivity and radial velocity data and the data are processed to the radar elevation levels. Observing System Simulation Experiments (OSSEs) are used to test and demonstrate the performance of the system, where radar data are assimilated for a supercell storm. To ensure fair comparisons among the EnKF, 3DVar, and En3DVar, each system is tuned to its “optimal configuration” by adjusting background error decorrelation scales (in 3DVar) or covariance localization radius (in EnKF). Overall, the hybrid En3DVar scheme outperforms 3DVar and EnKF when ensemble size is small (<40). For larger ensemble sizes (>=40), hybrid En3DVar is comparable to deterministic EnKF (DfEnKF, in which a deterministic forecast is used in place of the ensemble mean forecast in the background), and both are better than pure En3DVar (full weight given to the ensemble covariance in the cost function) and 3DVar. DfEnkF and pure En3DVar results are not the same, though theoretically they should be. A series of tests was conducted to understand the difference, serial versus simultaneous assimilation of observations, localization difference, as well as the high nonlinearity of the reflectivity observation operator are found to be the main contributing factors that cause the difference.

Convective Meteorology (Mesoscale Dynamics) Seminar Series website