Here, a 100 member ensemble initialized from an Ensemble Kalman Filter (EnKF) is applied to two powerful winter storms along the East Coast of the United States in 2010. The differences in the operational forecast skill of these two events has been documented but has focused primarily on the synoptic-scale predictability rather than the mesoscale details. The spread in the rain-snow transition line along the East Coast, as inferred from model-derived precipitation fields as well as typical operational rules-of-thumb, is examined as a mesoscale feature, while the track and strength of the cyclone are examined as synoptic features. It is found that the event with increased synoptic predictability showed large differences among the ensemble members in the amounts and distribution of snowfall at forecast lead times of only 36 hours. Additionally, the ensemble perturbation kinetic energy does not seem to show an appreciable upscale propagation of error. Instead, perturbations from the cycling EnKF are maximized at large scales and continue to amplify at large scales throughout the forecast period without waiting for upscale propagation. This suggests that relatively small errors in the synoptic scale initialization of a forecast may have more importance in limiting predictability than an inverse energy cascade resulting from small-scale processes.