Weather Service to Add Major Might to Computing Power

By Andrew Freedman

After coming under fire for falling behind the capabilities of other nations, the National Weather Service (NWS) is setting out to make an unprecedented increase in its computing power over the next several years, the agency announced this week. The computing boost will triple a key measure of the agency's main weather model, and could yield major improvements to its weather forecasting and warnings capabilities.

The ECMWF model consistently forecast that Hurricane Sandy would make landfall in the Mid-Atlantic region, starting a week in advance.
Credit: Weatherbell.com.

The program is made possible by recent funding from Congress contained in the Hurricane Sandy relief legislation, which was signed into law in January. The NWS plans to use $25 million of the $48 million provided to it in the Sandy supplemental bill, along with funds that are called for in President Obama’s fiscal year 2014 budget proposal, to bring about “unprecedented” computing upgrades — going from an operational computing capacity of 213 peak teraflops at the end of the current fiscal year, to 1,950 peak teraflops by the end of fiscal year 2015, according to NWS Director Louis Uccellini. A teraflop is a measure of how many calculations a computer can make per second, and indicates that a computer can make one trillion “floating point calculations” in just one second. In other words, the agency doesn’t run weather models on your typical personal computer.

“To go from 213 to 1,950 terraflops is the largest increase in computing capacity that we’ve ever had,” Uccellini told Climate Central in an interview. The NWS expects the European Center for Medium Range Weather Forecasts (ECMWF), which currently has a forecast model that is considered to be the most reliable, especially in the 5- to 9-day timeframe, to have an operational capacity of just 554 teraflops by the end of Fiscal Year 2015.

Uccellini said the high-speed computing improvements will help the agency to improve the accuracy of its Global Forecast System (GFS) model by running it at a higher resolution, which will allow it to better capture small-scale weather features — such as thunderstorms — that can affect the accuracy of a forecast. He also said the upgrades, which will be akin to shifting a supercomputer from first gear into overdrive, will enable the agency to put into operation higher-resolution, short-range computer models and observation systems. Those models could provide more accurate predictions for severe thunderstorms and hurricane forecasts.

Uccellini said that the agency has high-resolution computer models that have been developed and tested already, but they have not been implemented because of current computing limits. “We just didn’t have the computing capacity to put them in,” Uccellini said.

The raw power of a computer model is by no means the only key to an accurate weather forecast, and the NWS still needs to improve the ways its models ingest the massive amounts of weather data coming from surface weather stations, weather balloons, aircraft observations, satellites, and other sources. However, faster computers do allow models to be run at higher resolution, meaning that instead of dividing the world into 55-kilometer grid boxes as the GFS model now does, the same model can be run with a horizontal spacing of 10 kilometers.

The GFS model is the main American forecasting model out to 16 days in advance.
Credit: NOAA.

That is important because many weather phenomena, such as thunderstorms, are small in diameter and aren't captured by coarse-resolution models.

The investments will also put the agency on a trajectory to eclipse other forecasting centers that currently have more technological resources at their disposal, and more accurate computer models.

The NWS had come under fire during the past year for the shortcomings of its GFS model compared to the ECMWF. The European model accurately forecast the path of Hurricane Sandy one week in advance, at a time when the GFS model was still showing that the storm would curve harmlessly out to sea away from the East Coast. Similarly, the European model also offered more accurate forecast guidance at longer lead times for some high-impact northeast snowstorms last winter.

Currently, the GFS model is run at a reduced resolution in the extended period beyond seven days, whereas the European model has a higher resolution out to 16 days in advance. 

“They capture these storm systems further in advance with that high resolution model,” Uccellini said of the ECMWF. He said the boost in the NWS computing power “will significantly improve the reliability of our forecasts and bring us on par with the European Center.”

If the computing upgrades result in more accurate weather forecasts, as the NWS expects, it could reduce economic losses from weather events. A 2011 study found that routine weather variability alone affects the American economy to the tune of approximately $485 billion each year, not including the billions that are lost when major storms strike heavily populated areas. 

Related Content
Obama Budget Boosts Weather, Climate Funding
Sans Polar Satellites, Sandy Forecasts Would Have Suffered
Recent Storms Highlight Flaws in Top U.S. Weather Model
Panel Finds Flaws in NWS Guidance on Sandy's Surge