Impact of Assimilating WSR-88D Radar Observations on Snowband Prediction within the Comma Head of Northeast U.S. Cyclone

Snowbands embedded within cool-season extratropical cyclones over the Northeast United States represent a significant challenge in numerical weather prediction even at short time scales. In this presentation we test the impact of radar data assimilation (DA) on a convection-allowing ensemble and resulting short-term forecast of a snowband event on 9 February 2013 over Long Island, NY. This event featured a cyclone rapidly deepening and a series of multi-bands that formed just before 2000 UTC 8 Feb and eventually merged into a primary band that persisted until 0230 UTC 9 Feb. The complex evolution of this banding event makes it a good candidate for examining the impact of radar DA.

The Ensemble Kalman Filter (EnKF) implemented within the Data Assimilation Research Testbed (DART) package was employed to run a 40-member WRF continuously cycled data assimilation (CCDA) system. This system assimilated WSR-88 radar reflectivities and radial velocities from three coastal radars: KOKX, KBOX, KDIX along with conventional observations (radiosondes, satellite winds, ACARS, and surface observations) to a 40 member WRF mesoscale ensemble. This CCDA run was done in two phases. First, a run at 9-km grid spacing with a domain covering the eastern half the U.S. where only conventional observations (radiosondes, satellite winds, ACARS, and surface observations) were assimilated every 6 hours for a 5 day period between 0000 UTC on 04 Feb to 1800 UTC on 8 Feb 2013. The resulting 40-member 9-km ensemble with 39 vertical levels then was used as the initial conditions for the second phase. This second phase consisted of assimilating, in addition to conventional observations, radar observations every 15 minutes for a 6 hour period (1800 UTC Feb 8 - 0000 UTC on 9 Feb) on a downscaled 3-km grid which expands from coastal NJ to east of Cape Cod, MA.

This radar DA ensemble was run for three experiments in which the radar observations assimilated were varied. One run used only reflectivities, while another used only velocities, and a third included both reflectivities and velocities. As a final step, starting at 0000 UTC 9 Feb, the results from the three radar DA ensemble are used to initialize 20 member, 6 hour forecast (free-run) ensembles at 3-km grid spacing for each experiment.

This presentation will compare the forecast ensemble initialized from these three radar-DA ensembles against an identical setup that did not include the assimilation of radar. Results from this comparison suggest that radar reflectivity DA improves the structure and intensity of the primary snowband, but the combined reflectivity and velocity run had a larger positive impact. These DA benefits were shown to become negligible after only ~1 hour of the free run. However, the impacts are shown to extend longer into the forecast when combined with the use of a digital filter, which assigns a latent heating profile representative of the simulated radar signature.