Investigating the Uncertainties of Forecasting NE Cold Season Precipitation in Numerical Weather Prediction Models

A major winter storm on March 2, 2018, was a big forecast challenge with respect to predicting precipitation type for the northeast U.S including the Hudson Valley of eastern New York. Forecasts from the National Weather Service (NWS) blended the colder solution from the operational North American Mesoscale (NAM) model, and the warmer solution from the operational Global Forecast System (GFS) model, resulting in a mix of rain and snow in the Hudson Valley. However, the observations showed that the major precipitation type was snow even at the lower elevations in the mid and upper part of the Hudson Valley, and the snow depth forecast errors were around 8-12 inches near the Albany area.

The Weather Research and Forecasting (WRF) model was used to run nested simulations of this winter event at 25 km, 5 km and 1 km horizontal grid spacing. Experiments exploring sensitivity to model physics (including microphysics, cumulus, boundary layer, and radiation schemes) and domain configuration were made. The largest sensitivities with respect to snow depth were associated with cumulus and radiation schemes, and the vertical resolution in the middle troposphere was also found to be influential.

The WRF simulated results were compared with the NWS snowfall analysis observations from the Gridded Automated Zonal Precipitation and Complete Hi-res Output (GAZPACHO) using the point observations from public data statements (PNS) with an enhancement scheme that provides a more representative analysis for snowfall amounts in mountainous areas based on terrain slope. Additional measurements from the New York State Mesonet (NYSM) were added to the PNS and also used for the verification of surface temperature, snow depth and snow liquid water equivalent. These comparisons are providing clues regarding the causes of this dramatic forecast failure.