The assimilation system was designed similarly to previous convective-scale data assimilation studies that employed the ensemble Kalman filter. Conventional observations (e.g. radiosondes, surface data) were assimilated at hourly intervals in a cycled, 36-member, WRF-DART based, EnKF analysis system beginning at 00 UTC 17 November 2013. The model was configured with two domains, a CONUS-wide domain with 15 km grid spacing and a nested domain with 3 km grid spacing centered over Illinois. Between 15 UTC 17 UTC, the assimilation interval was decreased to 5-min to assimilate WSR-88D and surface observation datasets. This two-hour period captured the initial development of convection and ended immediately prior to the development of the Washington tornado. During this two-hour period, additive noise was used each cycle to accelerate convective spin-up and promote ensemble spread. In a separate experiment, observation innovations were used to determine the locations where the additive noise procedure was performed.
Ensemble forecasts were initialized every 15 minutes between 16 UTC and 17 UTC from the nested 3-km posterior ensemble analyses, producing forecasts with lead-times of 0, 15, 30, 45, and 60 minutes prior to the development of the Washington tornado. Updraft helicity (UH) in the near-surface layer (0-1 km AGL) was computed as a proxy for tornadic potential and was used to assess differences between the forecasts. This presentation will focus on characteristics of the forecasts related to the storm that produced the Washington tornado, and compare these to other non-tornadic storms in the domain. This includes the impact of forecast lead-time, the use of near-surface UH in discriminating tornadic potential, and the benefit of using innovations to direct the additive noise procedure in the analysis system.