For most people, snowfall forecasts are simple—your weather app tells you how much snow is going to fall and when, and the school decides whether or not to close because of it. However, behind that forecast are weeks of pattern watching, forecast models, and forecaster input. It all comes back to the exact placement of ridges (areas of high pressure) and troughs (areas of low pressure) around North America.
First comes the pattern watching. Whether the El Niño Southern Oscillation (ENSO), which tracks sea surface temperatures in the South Pacific, is positive or negative determines the base state of the pattern over the continental 48 states. Meteorologists can forecast several months out with ENSO. After that, there are various teleconnections, which are metrics that track whether there is high or low pressure in the atmosphere in a certain area and can usually forecast with some accuracy 10 to 15 days out. For example, the Pacific/North American Oscillation (PNA) and the North Atlantic Oscillation (NAO) are two of the most important teleconnections for storms in the northeastern US. If the PNA is positive it means there is a ridge, or high-pressure system, in the west which allows a trough to form in the east, where snowstorms can come up the coast. The NAO determines whether there is a ridge or a trough south of Greenland. If there is a ridge, it can help prevent storms from cutting to the west, which would bring warm air and rain for southern New England, instead of forcing a colder, offshore track.
Within seven to ten days of a potential storm that the general pattern would indicate, forecasters turn to ensemble forecast models. These are many simulations of the upcoming two weeks that all begin slightly differently, to account for errors in how the current global environment is measured. If many members on the ensemble models agree on a storm threat, it is likely to happen in some form, although the details of where the rain or snow line sets up, how much precipitation there will be, and how fast the storm moves will not be set in stone until a few days before the event. Forecasters look at ensemble models to see where they project the snow will fall, and if there is a possibility that warm air intrudes aloft above cold air, which would lead to sleet or freezing rain even if it is below freezing at the surface. Forecaster input is involved at this stage to take out clear outlier solutions and outcomes that are highly improbable due to their pattern or uncommonness. For example, if one ensemble member shows a very strong low-pressure system hitting Buffalo while the other 49 members of that model show a weak wave of low pressure passing somewhere south of Nantucket, it would be easy for a human forecaster to discount that one member as an outlier, leading to a more confident and accurate forecast.
As we get within about three days of an event, ensemble and models will come into better agreement, but since they inherently have slightly different starting environments, there will still be a spread among them. At that point, it is good to use mesoscale models, which with higher resolution can resolve finer details in the storm, such as where heavy snow banding may set up. This is when forecasters predict exact snowfall amounts for specific areas, where is likely to see all snow and where it might change to rain, and when the snowfall will begin and end. There are several different high-resolution models. Two are produced by a branch of the National Oceanic and Atmospheric Administration called the National Centers for Environmental Prediction (NCEP), one is produced by the Canadian Meteorological Centre, and one is produced by the European Centre for Medium-Range Weather Forecasts. These models vary slightly between one another but, generally, when one trends in one direction the rest follow, allowing forecasters to tweak snowfall forecasts if models consistently move north or south, or raise/lower the forecasted precipitation amounts uniformly. Forecaster input is heavily involved in this step. For example, frontogenesis is a measure of how rapidly horizontal temperature gradients increase over time. Models often depict heavy snow bands where frontogenesis is expected to be the strongest, but, in reality, the heaviest bands often set up on the northwest edge of that frontogenesis, leading to the highest total snowfall being northwest of where it was expected to be. Forecasters include that known model bias into their forecasts.
The final step of fine-tuning a forecast is colloquially referred to as “now-casting,” and it occurs within 24 hours of a storm. Now-casting involves looking at surface analysis, satellite imagery, and radar, and comparing that to what models show. If a low-pressure system is stronger, farther north, or wetter than forecast, or vice versa, forecasters will slightly adjust their snowfall totals north, south, or increase totals, accordingly.
Finally, the storm arrives after scientists following the threat for several weeks, from a favorable pattern to a possible storm on ensemble models to a full-blown snowstorm. Forecasters can relax and enjoy the storm as everyone else does until it is already time to follow the next threat of a snowstorm. Next time when you are hoping for a snow day, take a minute to think about all the science that goes into that crucial forecast!