First, standard radiosonde, aircraft, and surface observations from 31 May 2013 are assimilated onto a 15-km grid every hour from 0000 UTC to 1600 UTC using an Ensemble Kalman filter (EnKF) method within the Weather Research and Forecasting (WRF) Data Assimilation and Research Testbed (DART) system. The three targeted radiosonde observations are then assimilated in half-hourly periods from 1600 UTC to 2000 UTC. A convection-permitting (3-km) grid is embedded at 2000 UTC, and a 36-member ensemble of forecasts are run out to 9 h. This experiment is compared to a control run that excludes the three targeted soundings, consistent with the resources available in an operational setting. Additional experiments vary the localization radius and initial condition observation error in the EnKF assimilation, and exclude the 1800 UTC National Weather Service soundings from the control experiment in order to test the sensitivity of these results. The impact of the targeted radiosonde observations are evaluated by quantifying the accuracy of the timing and location of convective initiation, the tornadic supercell near El Reno, and the transition to a heavy-rain/high-wind producing Mesoscale Convective System (MCS) among the set of 0-9 h forecasts.