Of strong public interest, the New York City (NYC) transit system was shut down in preparation for the storm. However, observed snowfall accumulations in NYC were significantly lower than expected: Central Park reported less than 10 inches (of snow); LaGuardia and John F. Kennedy international airports, which canceled flights in anticipation of the storm, reported less than 12 inches; and areas north of NYC (White Plains and Katonah) reported under 6 inches. Decisions by government agencies were heavily scrutinized in the aftermath of the storm as the total cost to business, citizens, and the government of NYC was estimated at USD $1.25 billion. However, parts of Long Island, Massachusetts, and Eastern Connecticut received over 25 inches, consistent with most forecasts.
The over-prediction of snowfall in NYC was generally attributed to the storm tracking 90 miles further east than expected. We use this event as a case study to explore methods for improving regional forecasts of severe winter storms. We perform an ensemble of hindcasts using IBM Research's Deep Thunder, a state-of-the-art high spatial- and temporal-resolution forecasting system, customizable to meet the needs of specific weather-sensitive business decisions. It is based, in part, on the ARW core of the Weather Research and Forecasting (WRF) model. It has been running operationally out to 84-hours at 2-km resolution for the greater NYC metropolitan area since 2008 and successfully predicted the intensity and location of the January snowstorm. We use different combinations of initial and boundary conditions and also examine the impact of 3-D variational data assimilation of available observations. Preliminary results indicate that the Deep Thunder operational forecast produced the best representation of snowfall, with greater accumulation over parts of Long Island and New England and less across NYC and southern Connecticut. We also show the assumed correlation between storm center position and snow accumulations in NYC is not as robust as generally assumed. This suggests added complexity in the publicly available forecasts, possibly due to banding or synoptic scale differences in the model initialization.