NOAA Bets Forecasting Success on Supercomputers

George Putic
August 30,2013

COLLEGE PARK, MARYLAND — Early fall, as hurricane season begins, is the time when many Americans start to closely follow weather reports.

Because predicting the strength and movement of huge storm systems that bring destruction and death is of crucial importance, meteorologists at the National Oceanic and Atmospheric Administration (NOAA) have invested in supercomputers that allow for better predictions up to as six days in advance of a storm's arrival.

Hurricane Sandy, which hit the U.S. East Coast in October last year, was the second costliest storm in U.S. history — causing widespread damage and killing 285 people. At the time, some blamed meteorologists for not accurately predicting the path of the storm.

But according to Ben Kyger, Director of Central Operations at NOAA's College Park headquarters, weather forecasting is extremely difficult.

"You’ve got major patterns in the atmosphere, like the jet stream, but you’ve also got little eddies, little currents, little things happening all over the place," he said. "All these little changes are interacting with each other, continuously, all day long. So if you look at it from above, from a satellite, you see the atmosphere moving and churning in big ways and little ways.

Oceans, he says, are another factor because they closely interact with the atmosphere and have a huge effect on storms. In order to improve reliability of its forecasts, NOAA spent about $20 million on two new supercomputers that started building models of the weather patterns on July 25.

"These computers generate the initial model guidance that the whole forecast process depends on, for all the weather information that you see, with snowstorms, tornadoes, hurricanes, how hot it’s going to be today — all of your weather forecasts start with what comes off of these supercomputers," he said.

While the computational power needed to analyze data from weather satellites, ground stations and other sources is enormous, human brainpower and experience remains crucial to accurately predicting temperature, air pressure, humidity and wind speed.

According to Kyger, meteorologists at the National Centers for Environmental Prediction scan the same data the supercomputers get before issuing a forecast.

"They look at lots of different models, assess that and create the five-day forecast," he said. "They have a lot of scientific and subjective knowledge from doing it year after year. They know where the models are strong, where they’re weak and they give us significantly better forecasts than the models would do by themselves.

NOAA issues worldwide forecasts every six hours each day of the year. Access is free — a benefit to countries that cannot afford their own weather service. NOAA has planned upgrades for its weather-predicting supercomputers set for 2015.