Assimilation of Reflectivity Data within the ARPS Hybrid Three-Dimensional Ensemble-Variational (3DEnVar) Data Assimilation Framework

Chengsi Liu* and Ming Xue, Rong Kong
University of Oklahoma

A hybrid En3DVar algorithm is implemented and tested within the ARPS variational DA framework, coupled with the EnSRF-based ARPS EnKF DA system. Recursive filter is used for the horizontal and vertical localizations of the ensemble covariances.

To enable physically reasonable analysis of liquid and ice phase hydrometeors from the reflectivity data within the 3DVar framework, temperature-dependent background error profiles are designed for the hydrometeors and used in the ARPS 3DVar. The procedure is found to produce better hydrometeor analyses than the so-called hydrometeor-classification method, which uses a temperature-dependent reflectivity observation operator. To help suppress spurious storms, clear-air reflectivity is also assimilated. When using the default background error correlation scale that is constant everywhere, the 3DVar analysis in the precipitation region is degraded by the inclusion of clear air reflectivity. To deal with the problem, a double-pass procedure is developed, in which the precipitation region and clear air reflectivity data are assimilated separately in the first and second pass, respectively. A reduced spatial correlation scale is used in the second pass. The analysis of a convective storm in both clear-air and precipitation regions is improved with this procedure.

The hybrid En3DVar system based on the improved ARPS 3DVar is tested through Observing System Simulation Experiments (OSSEs) for a supercell storm with and without model error. Optimal spatial localization radii for EnKF and En3DVar as well as the optimal background error correlation scale for 3DVar are obtained through sensitivity experiments and used in the experiments. Hybrid experiments with different combinations of the static and ensemble covariances are conducted and compared to the performance of 3DVar and EnKF. Results show that the performance of pure En3DVar is comparable to that of EnKF in the absence of model error. Under imperfect model condition, hybrid En3DVar outperforms EnKF. In both cases, 3DVar performs the worst.

*Preference: Oral