For this case, ensemble Kalman filter (EnKF) data assimilation (DA) is performed on a 500 m grid, assimilating data from multiple Doppler radars and from the conventional networks including the Oklahoma Mesonet. The EnKF DA cycles of 5-minute intervals are run over a one-hour period, arriving at the final ensemble analyses at 1930 UTC, about 26 minutes prior to the start of the observed Newcastle-Moore tornado. Ninety minute ensemble forecasts are then run for 90 minutes, covering the tornadic storm life cycle. The 500-m ensemble grid is nested within continental U.S. domain 4-km ensemble forecasts initialized at 000 UTC, 20 May.
Most members of the forecast ensemble on the 500 m grid successfully reproduced the observed tornadic supercell, with strong low-level mesocyclone with prominent low-level rotation that tracks closely to the observed tornado track. The predictability of tornado is first assessed, using the low-level mesocyclone as a proxy, based on the ensemble forecasts and additional forecasts with perturbations of different magnitudes and structures added to the initial condition of the best forecast member.
To further assess the predictability of tornado within tornado-resolving simulations, an ensemble of forecasts is produced on a nested 50-m grid. All ensemble members predict tornadic vortices with surface winds of 30 m s-1 or greater, and some members predict intense tornadoes with wind speeds meeting EF5 criteria (90 m s-1). There are substantial variations among members in the duration and track of predicted tornadic vortices.
Probabilistic forecasts are produced for the tornado, using several combinations of criteria, including near-surface wind speed and vertical vorticity. These probabilistic forecasts are verified against the observed tornado, and compared against probabilistic forecasts produced for commonly-used proxy variables, including updraft helicity. Based on the ensemble of explicit tornado forecasts as well as a set of forecasts with small initial condition perturbations, and predictability of the tornado at tornado-resolving resolutions is discussed.