Will it snow or not? It seems like a simple question that weather forecasters ought to be able to readily answer. But the degree to which they can be sure depends on a lot of things, including what sort of technological crystal ball they are using and how well they know their local weather.
“We have a number of models,” said Bruce Sullivan, senior forecaster at the National Weather Service's Weather Prediction Center in College Park, Maryland. “There is the GFS (Global Forecast System) model, NAM (North American Mesoscale Forecast System); we have an ensemble forecast model called SREF (Short Range Ensemble Forecast), the GEFS (Global Ensemble Forecast System).” Plus many others both in the U.S. and abroad. But how are these used?
“In ensembles you take a model and use different initial conditions,” Sullivan explained. Those conditions come from radar, satellites, weather balloons -- the real world, in other words. Then you perturb those conditions in several possible ways and see where each perturbation takes the model. “If at 36 hours they all agree, that gives us some confidence. But if they diverge you get less certain forecast.”
As in all computer modeling, the old adage applies: garbage in, garbage out. So it is extremely important to get the initial conditions right, Sullivan added.
“If you don't have those initial conditions right the error increases with time,” he said.
Then there is the matter of tailoring the forecast to a specific area. That's even more of an art, explained Dan Satterfield, chief meteorologist for the CBS affiliate WBOC TV in Salisbury, Md.
“A model that is going back and forth or showing unrealistic temperature forecasts or has been verifying badly is not likely be given a lot of credibility,” said Satterfield. “A particular fave of mine is the WSI Rapid Precision 4-km resolution model. This model has great physics and has done very well here on Delmarva for snow.”
The same ensemble process is used whether it's summer or winter, said Sullivan, but there are some advantages to certain kinds of models in the winter, according to Satterfield.
“The models with the higher resolution will do better with snow than the global models because they have better physics and also usually a better vertical resolution,” said Satterfield. “A good forecaster develops a list of rules in his or her mind as well, and one for me is that if the upper level trough is negatively tilted then the storm will be as strong or stronger than the models show. This was the case in this snow event as well.”
This kind of insight, or failures of models, is why the models themselves are always being improved.
“Universities are always looking at new models and we're always tweaking them,” Sullivan said. He recalls in 2000 when there was a big storm that the models all got wrong. It was very instructive, he said.
“That was the biggest bust up we've had in 14 years,” said Sullivan. And meteorologists learn from their mistakes.