Analysis of Accurate and Inaccurate HRRR Forecasts for Heavy Banded Snowfall in the Central United States

The HRRR model, as well as other Convective Allowing Models (CAMs), have been analyzed for their accuracy in forecasting thunderstorms and other summertime phenomena. Less analysis has been performed on the ability of CAMs to predict wintertime phenomena. The most prolific snowfall totals associated with cyclones occur where moisture and instability are collocated with frontogenesis, resulting in banded snowfall. These localized areas of heavy snow can result in areas receiving >12” of snowfall, while an area less than 50 miles away may only see an inch or less. This study analyzes two separate cases in which 4 km grid spacing is necessary to resolve mesoscale banded snowfall. Both cases involve modest Midwest cyclones (990-1000 mb lowest central pressure) which produced swaths of snow >12”. In one case, the HRRR produced an accurate depiction of initiation, strength, location, and timing of banded snow that occurred between 24 Jan 2017 – 26 Jan 2017 in South Dakota, Nebraska, Iowa, and Minnesota. In another event over 01 Jan 2017 – 03 Jan 2017, the HRRR struggled to forecast the timing, intensity, and location of banded snowfall primarily in North Dakota and Minnesota. These two cases were selected after subjective comparison of HRRR forecast reflectivities with observed radar over 27 snowfall cases exceeding 4” in the Central United States over the 2016-2017 winter.