Youngsun Jung, Ming Xue, Yunheng Wang, Yujie Pan, and Kefeng Zhu
The ensemble Kalman method has shown potential for storm-scale data assimilation and ensemble forecasts. However, its computational cost is still a major issue for operational use of the EnKF. Recently, CAPS has developed an efficient parallel EnKF system that is capable of assimilating multi-scale observations. We applied this EnKF system to the May 10, 2010 Oklahoma-Kansas tornado outbreak case that spawned over 40 tornadoes to test the storm-scale, cycled EnKF DA, and ensemble forecasts. To include both mesoscale and storm-scale features important on this day, the storm-scale ensemble with 4-km horizontal grid spacing was nested inside the regional WRF ensemble analyses at a 40-km. The storm-scale domain was 1750 × 1920 × 21 km3 and is the same as the CAPS VORTEX2 realtime forecasting domain.
The 40-menber regional ensemble was interpolated to form a storm-scale ensemble at 1500 UTC on 10 May 2010. The boundary conditions for each ensemble member were provided by the corresponding regional ensemble member. In the storm-scale ensemble, the ARPS system was used in both simulation and analysis. Using the MPI-OpenMP hybrid ARPS EnSRF, conventional (surface, sounding, profiler) and radar data were assimilated every hour. Additionally, surface and radar data were assimilated every ten minutes during the last one-hour period before the forecast was launched. Finally, forward ensemble and deterministic forecasts were launched every three hours starting from 1800 UTC for 6 hours.
The preliminary results showed the parallel EnKF algorithm exhibited good scalability for dense radar observations. The analyzed base reflectivity mosaic at the end of each assimilation window exhibited a good fit with the observations in shape, structure, and intensity. Also, a line of strong, isolated storms in the central Kansas and Oklahoma was captured reasonably well by the ensemble forecasts throughout the forecast period.
