In a potentially historic marriage of supercomputing and big data, IBM goes live Thursday with a global weather model that it says can provide far more accurate forecasts for the entire world.
Called GRAF — Global High-Resolution Atmospheric Forecasting — the new model offers high resolution weather forecasts globally with a detail for areas as small as 2 miles wide, compared with 6 to 9 miles for weather models covering parts of the world outside such advanced regions as Europe, the U.S. and Japan.
IBM says its new supercomputer, DYEUS, built just to run the model, will issue 12 trillion pieces of weather data every day and process forecasts every hour, while many global weather models update only every six to 12 hours.
"We're actually getting down to the cellular level of the thunderstorms now, where we weren't able to do that before. And with that information, we can provide better support to critical decision-makers,'' said Kevin Petty, head of science and forecasting at The Weather Company.
IBM became a leading player in meteorology in a few years ago when it purchased The Weather Company in 2016, retaining the data gathering and forecasting units while splitting off the Weather Channel.
While Americans are accustomed to turning to their phones for storm warnings and accurate, hourly forecasts, billions of people around the globe lack even basic forecasts, making the world a much more chaotic place. For companies like American Airlines, better international weather forecasts can mean more comfortable flights for passengers who can be routed around turbulence. If dispatchers have more confidence in storm locations and strengths, they can even shorten flights.
"If we can get a better accurate resolution, a better picture of that weather event, we are able to much more confidently operate the flight as planned,'' said Steve Abelman, manager of weather technology for American Airlines. which is currently a client of IBM's weather services. "If we can do a more fuel efficient flight, we can certainly save money on fuel."
But Abelman said the model still has to prove itself and American will wait perhaps months before relying on the data. And even then it will use the data together with other sources. Any weather model, no matter how advanced the underlying technology, can only prove its accuracy over time. But IBM sees a huge need in the market for its new product.
"Any business that doesn't have a weather strategy is missing out on returns,'' said Cameron Clayton, general manager of IBM's Watson Media, Weather and special Red Hat project. Weather "impacts everything. It impacts supply chains, the kind of purchasing decisions we make, when you get up in the morning, how you're going to dress, what you're going to eat. And being able to predict that in advance for retailers, energy companies, travel and transportation businesses is all we do.''
Individuals can access the data thru the The Weather Channel app on smart phones and in other weather sources that use IBM data. IBM says the key to the model is in the supercomputer DYEUS, named for the ancient god that ruled the daytime sky.
DYEUS uses IBM's Power 9 processing chips, which are also used in Summit, the world's fastest supercomputer. In addition, DYEUS employs graphics processing units that are used in video gaming, enabling much faster processing of visual output. GRAF will provide forecasts of up to 15 hours for 26 million locations around the world.
One of the limitations of global weather forecasts is a lack of data, but IBM says GRAF has the ability to pull in data from airplane sensors and even from smartphones, which can provide temperature and pressure data. IBM declined to give revenue projections for the GRAF project or disclose its costs. The model was developed in a public-private partnership with the National Center for Atmospheric Research.
Clarification: Story updated to reflect IBM announced it was buying The Weather Company in 2015 and completed the purchase in 2016.