Assimilating WSR-88D Radial Velocity Data into WRF Model with Ensemble Kalman Filter: the 20 May 2013 Moore, Oklahoma Tornadic Supercell

The EF-5 tornado that struck Moore, Oklahoma and adjacent areas on the afternoon of 20 May 2013 is one of the strongest and deadliest tornados in 2013 in United States. During its 39-minutes touch down from 2:56 p.m. to 3:35 p.m. CDT with a path length of 14 miles (23 km), this tornado reached a peak winds estimated at 210 mph (340 km/h) and maximum width of 1.1 miles (1.8 km), killed 25 people including 2 indirectly, and 377 others were injured. The related supercell developed within detection range of two WSR-88D radars of KTLX (Oklahoma City) and KFDR (Fredrick) as well as one Terminal Doppler Weather Radar (TDWR) of KOUN/TOKC (Oklahoma City), and it is only several tens of miles away from KTLX during the lifespan of the tornado, which provide a unique and rare set of WSR-88D radar observations of a tornadic supercell.

Because of the insufficient water condensate and the inability to solve fine-scale wind field of the coarse resolution of the initial conditions, 1-km horizontal resolution simulations using WRF/ARW V3.5 directly from GFS analysis are unable to produce the convection neither at the right location nor at the right time. Further more, its location and development are also sensitive to physical parameterization schemes including microphysics, planetary boundary layer, land-surface and surface layer processes.

In order to provide adequate initial conditions, three one-way nested domains of 27-, 9-, 3-km horizontal resolution respectively were utilized, and conventional surface and upper air observations, satellite-retrieved winds, aircraft reports are assimilated using ensemble Kalman filter (EnKF), and a 1-km storm-scale domain with EnKF assimilating radial velocity from KTLX and KFDR every 5 minutes is employed from 1800 UTC with initial conditions from the analysis of the nested EnKF cycles. With additional reflectivity control that removes spurious water condensates according to reflectivity observations from these two radars, the 1-km EnKF analysis shows development of the tornadic supercell both in good timing and location, although it is weaker and less organized than the observations.

However, when additional high-temporal-frequency (5-minute interval) Mesonet surface observations are assimilated prior and together with radar radial velocity, the supercell near Moore disappeared. Further comparisons with the observations show much lower surface 2-m temperature as well as higher 2-m moisture. In one hand, this deficiency is resulted from a systematic bias within the EnKF system, and can be reduced when potential temperature and dew point are assimilated instead of kelvin temperature and water vapor mixing ratio for surface observations as well as updating soil temperature and moisture. In the other hand, there are large areas of spurious precipitation that produces a cold and wet surface, thus both surface temperature and moisture becomes much closer to the observations when these precipitations are removed using reflectivity control.

These experiments again prove the sensitivity of supercell simulation to the initial conditions, especially within the PBL. Further experiments of cycling EnKF in a LES domain are planned.