SILVER SPRING - In an era when computers are supposed to be introducing greater accuracy and certainty into our lives, weather forecasters are moving to use their computers to introduce more uncertainty into their winter storm predictions.
Still smarting from two "busted" forecasts a couple of years ago, one a surprise foot of snow and the other a no-show, meteorologists are developing a system that would deliver storm predictions in terms of probabilities instead of expected accumulations.
The idea was aired Friday at a conference called by the Washington Chapter of the American Meteorological Society to discuss the difficulties of forecasting winter storms in the mid-Atlantic states.
This region is one of the toughest for such forecasting, said Louis W. Uccellini, director of the National Center for Environmental Prediction. Major snowstorms here typically occur when low-pressure systems form in the Gulf of Mexico and strengthen as they move up the Atlantic coast.
"Given where the ocean is and where the mountains are, you have a rain-snow line in every storm," Uccellini said. "This is a major forecasting challenge in itself."
The forecasts matter here because people act on predictions of snow, often days in advance. They mob stores in search of bread and toilet paper, and cancel many scheduled events.
The weather models and supercomputers don't always work so well in predicting how much snow or ice will accumulate.
They scored a triumph when they successfully predicted the "superstorm" of 1993 and the Blizzard of 1996.
Two forecasts go awry
But then forecasters, seduced by their successes, followed the models over a cliff. Twice.
In 2000, the computers led them to predict the Jan. 25 "flurries" that blew up into an 11-inch storm in Baltimore. Then, a predicted Dec. 30 "storm" fizzled here and buried Albany, N.Y.
"The forecasts for two to seven days out have become so good that people start thinking that a storm is going to form as the models say it will," Uccellini said. "We set ourselves up."
"Now, when we say it's going to snow, people believe it's going to snow," he said. "People [television weather broadcasters] get on the air, and they're talking about the forecast as if we're absolutely certain that's the way it's going to happen. And we don't have that certainty."
No model by itself can account for all of the variables that influence the atmosphere. As a result, the National Weather Service is developing an experimental forecasting system designed to restore a healthy uncertainty to its winter storm forecasts.
Called "ensemble" forecasting, it will combine the results of dozens of computerized weather models and arrive at sort of a consensus about what is likely to occur. It's a bit like relying on two dozen arrows fired at a target to suggest where the bull's-eye might be, rather than one or two that claim to hit dead-center. It's less likely to go badly awry.
The results would be expressed as probabilities rather than certainties. They might be displayed on a map much like today's hurricane track forecasts. Places where heavy snow is most likely would be highlighted and hedged with a probability factor of say, 50 percent or 90 percent.
"If we're not certain about an event, the product should show that," Uccellini said.
Some forecasters wary
Not every meteorologist is comfortable with the idea of introducing uncertainties into the forecast. Some expressed concern that the public would perceive the change as a scientific step backward.
Ken Reeves of AccuWeather said the public "is still going to want to know: Is it going to rain or is it going to snow?"
Forecast "consumers" such as energy traders and road crews, he said, have expensive decisions to make and need some certainty as a basis for making such decisions. "If we miss that viewpoint, they'll say, 'That isn't going to do me any good,'" he said.
In January 2000, forecasters looked at computer models and believed them when they indicated that the storm would pass off the Carolina coast, bringing only flurries or light snow to Baltimore and Washington.
"All the models were focusing on creating this trough and taking it out to sea," Uccellini said. "We're only talking about a difference of 100 or 200 kilometers ... , and yet that made all the difference in the world."
When forecasters realized that the storm had turned toward Baltimore, they began increasing their forecast of snow totals, but many residents had gone to bed and missed the most dire late forecasts.
"We took a lot of heat for this storm," Uccellini said. "I had to defend the National Weather Service while we were being buried under 10 to 15 inches of snow. I think there is a public perception that we've taken a step back. And the media, they have the expectation that we can do this, and that has been a crusher for us. People are angry."
Snow forecasting is difficult, and at least in the mid-Atlantic states, heavy snowstorms are relatively rare. In the past 50 years, only 14 storms have dumped 10 inches or more on Baltimore.
Forecasters, who haven't had much snow to practice on the past two winters, could be busier this winter.
The Weather Service has forecast an increase in the number of storms cruising up the East Coast, a likely result of the moderate El Nino conditions that developed in the spring in the eastern tropical Pacific Ocean.
The weather service has said nothing about whether those storms would bring snow or rain, but there might be signs that El Nino and other climate factors are aligning to increase the chances for snow.
Joseph D'Aleo, chief meteorologist for WSI Corp., a Massachusetts-based weather data and services company, said that in addition to El Nino, climatic factors elsewhere in the Pacific and in the North Atlantic have also shifted in ways that tend to bring lower temperatures and more low-pressure systems to the mid-Atlantic states.
"It suggests it's going to be a snowier winter," he said.