Forecasting snow squalls using mesoscale models

Snow squalls are an infrequent, but potentially dangerous winter phenomenon, that are usually poorly forecast. They occur when the lower section of the atmosphere is destabilized by a combination of surface heating and strong cold advection aloft. Snow squalls are usually oriented in a band, or multiple bands, last only a few minutes, and generally result in small accumulations of snow. The danger of snow squalls is due to the sudden onset, very poor visibility, and resulting slippery road conditions. Serious traffic accidents can lead to the shutdown of highways, and in some instances, intense fires from spilled fuel that leave the road surface badly damaged. Even the least snow squall can cause enough minor traffic accidents to create a morning commuter nightmare. In Boston, a December, 2003 snow squall event turned 30 minute commutes into 4 hour crawls.

Operational forecasting models, especially at the resolution routinely provided over the internet, often miss the occurrence of snow squalls completely. There is some evidence that the National Weather Service program BUFKIT has the potential to provide forecasting help for snow squalls, even if the operational models themselves are unable to directly simulate these storms.

We have used version 5 of the Penn State University/NCAR Mesoscale Model (MM5) to try to forecast snow squalls directly. From the winter of 2003-2004, we have simulated 5 cases of snow squall occurrence. In some cases, a snow squall line is produced that is remarkably similar to what actually occurred. In other cases, the model is unable to produce anything that resembles the squalls that took place. During the current winter of 2004-2005, we have run the MM5 daily, to look for patterns that distinguish snow squall days from non-squall days, although there have been relatively few snow squalls this winter in New England.

We are running the MM5 with 3 nests, at 36km , 12km and 4km horizontal resolution, and 33 levels in the vertical. The model is running with simple ice physics and Grell cumulus parameterization on the outer two grids with none on the inner grid. Experiments with the Reisner cloud physics package produced poorer results. Model output is imaged with GrADS.

We will show results from both seasons, and discuss the forecasting implications based on both the case studies as well as the more recent daily simulations. It is clear from the outset that higher resolution modeling is better able to simulate snow squalls, but the timing and exact location of the resulting shallow convection continues to be very difficult to pin down.