New weather supercomputers faster than a petaflop
Weather service upgrade will boost forecasts and improve models.
The Earth’s atmosphere is under an increasingly fine-tuned microscope, analyzed with bygone weather balloons still launched twice daily to pioneering satellites that capture the dazzling dances of frosted cloud tops.
It’s all in an effort to predict Mother Nature’s next move — whether tropical cyclone or Alberta clipper — but can fall short if the cornucopia of data feeds into a dullard computer.
This month, two brawny supercomputers the size of school buses will be added to the weather forecasting capacity of the U.S. They are the cherries on top of a multiyear, multimillion-dollar improvement to the National Weather Service’s forecast muscle and will process mountains of calculations for a souped-up Global Forecast System (GFS) weather model.
“The supercomputer enables us to upgrade the model without suffering any lag time in getting information back,” said Robert Molleda, the warning coordination meteorologist for the National Weather Service office in Miami. “We can capture more data and make the calculations even faster.”
The computers’ lickety-split speed is expressed in petaflops. One petaflop equals one quadrillion calculations per second. The addition of the new supercomputers boosts the overall braininess of the league of weather computer heavyweights to 8.4 petaflops.
That means if everyone on Earth was given a calculator and asked to keep up with the supercomputers they would have to perform
1.2 million calculations per second, according to Dave Michaud, director of central processing for the National Weather Service.
“For us, the important thing is the outcomes in terms of our mission to protect life and property,” Michaud said. “We want to give the decision makers the best information possible to help people in high-impact situations.”
It’s a multistep process. Atmospheric readings are gathered worldwide by weather balloons, ships, ocean buoys, airplanes and ground stations. Those readings are fed into the computers, which put them into calculations weather models can use to create a forecast.
Advanced satellites, such as the GOES East, which was launched in the fall of 2016, and the enhanced Joint Polar Satellite System-1, which launched in November, also add to the data going into the computer.
This year’s computing upgrade is part of a 10-year National Oceanic and Atmospheric Administration contract awarded in 2012 and capped at $502 million. About $44.5 million was added to the budget after Hurricane Sandy devastated the northeast in October 2012, and limitations of the GFS, or American forecast model, were exposed.
The GFS underperformed compared with the model from the European Center for Medium-range Weather Forecasts, or European model, which correctly forecast Sandy’s track toward the East Coast.
The GFS was outdone again during 2015’s Category 5 Hurricane Joaquin. The GFS and European model disagreed on the forecast track, with the GFS showing the dangerous cyclone making landfall in the mid-Atlantic region, while the European had it going out to sea. The European model won. Thirty-three crew members aboard the cargo ship El Faro died when it sank in Hurricane Joaquin. The U.S. Coast Guard ultimately determined the captain misjudged Joaquin’s path and “failed to understand the severity of the situation.”
“As other computing centers around the world step up their game, it is vital that the U.S. atmospheric scientists also have access to the computing power they need for continual improvement,” said Bob Henson, a meteorologist and blogger for Weather Underground.
Of the 10 hurricanes and seven tropical storms that formed during the hyperactive 2017 hurricane season, four reports evaluating forecast accuracy have been released by the National Hurricane Center. Those include for tropical storms Arlene, Don, Katia and Rina.
“All mention poor forecasts of formation of the tropical cyclone,” noted Rick Knabb, former NHC director who is now at The Weather Channel, in a tweet this month. “Exemplifies how tropical cyclogenesis remains a scientific and operational challenge, and why forecasts are probabilistic.”
The new GFS model, which will be used experimentally during the 2018 hurricane season with the new supercomputing capacity, will have increased resolution in more levels of the atmosphere and out for 16 days, versus the current 10 days.
“It’s really about bringing the big picture information and large-scale weather systems into better resolution and applying the information as much as possible to local weather,” Molleda said. “It will help us provide more accurate information sooner.”