The National Oceanic and Atmospheric Administration announced Monday that a major upgrade has been applied to the American Global Forecast System model, one of the main computer models used to predict weather across North America and the world.
The upgrade focuses on addressing the underlying physics of the model and how it handles various features of the atmosphere. It’s known as version 16.0 of the model frequently referred to by forecasters as simply the GFS or the “American” model.
The upgrade piggybacks off the launch of the GFS FV3 model, or Finite-Volume Cubed-Sphere Dynamical Core, a souped-up version of the previous GFS model that debuted in summer of 2019. Its release was delayed while model biases were addressed, including a tendency for model depictions to skew too cold and snowy. After changes, the FV3 was released, fully replacing the legacy GFS model in September.
The latest upgrade focuses on addressing some additional biases. The upgrade also adjusts how initial conditions, or current weather information, are ingested into and processed by the model, while integrating more sources of data from weather satellites and ordinary aircraft.
Furthermore, the model’s resolution in the vertical will nearly double. The atmosphere will now be simulated as having 127 vertical slices, rather than just 64.
“When we announced our upgrade to the GFS in 2019, we described it as replacing the engine of a car,” Louis Uccellini, director of NOAA’s National Weather Service, said in a phone conference Monday. “With today’s upgrade, we’re adding more horsepower and more upgrades to the entire car as we move forward.”
The upgrade “brings together the day-to-day reliability and speed required” in an operational weather model, Uccellini said. Software engineers reconfigured existing NOAA infrastructure to allow for more computationally demanding model simulations to be run. The physical computers themselves are located in Reston, Va., and Orlando.
Version 16.0 of the GFS also absorbs a global ocean wave model known as Wavewatch III, which should allow for better marine forecasts, particularly with regard to water waves driven by wind. Wave forecasts will now stretch to 16 days out rather than just over one week into the future.
“This implementation is the first time it allowed us to couple the GFS to the global wave model,” said Vijay Tallapragada, chief of the Modeling and Data Assimilation Branch at NOAA’s Environmental Modeling Center.
Tallapragada explained that the highest altitude simulated by the new GFS will jump to 80 kilometers (50 miles) up from 55 kilometers (34 miles), effectively raising the ceiling of the model. The additional layers added to the model will allow for improvement in two key areas — the near-surface “boundary layer,” and the stratosphere, the second layer of earth’s atmosphere.
Most weather occurs in the troposphere, or the part of the atmosphere in contact with the ground. In the stratosphere, temperature increases with height due to the absorption of ultraviolet solar radiation.
Increased resolution in the stratosphere will allow for better prediction of sudden stratospheric warming events, which are known to have major implications on weather systems closer to the surface. In early January, a sudden stratospheric warming event spurred the disruption of the polar vortex, which, through a chain reaction of events, unleashed an outbreak of bitter Arctic air that wrought havoc in Texas in mid-February.
Tallapragada said that users can expect significant improvements in forecasting high-end events such as heavy precipitation or tropical cyclones.
He said that Version 16, when run in parallel with the previous iteration of the GFS model, resulted in “more well-accurate timing and magnitude of the snow in Colorado” that struck early last week.
NOAA has been experimenting with aspects of Version 16, running it internally since 2018. When it comes to tropical storms and hurricanes, they say the results are promising.
“We found about a 10 to 15 percent improvement in the track and intensity forecasts in the Atlantic Basin, especially at longer lead times,” Tallapragada said.
Programmers and meteorologists also noted that the tweaked GFS can signal trouble areas at risk for brewing a tropical storm or hurricane about 36 hours further in advance.
Despite what NOAA touts as an impressive step in the future of weather forecasting, some meteorologists still believe the United States lags behind Europe in its ability to produce a good model. The European Center for Medium Range Weather Forecasting, or ECMWF, has a model that is generally considered better than the GFS.
“We have been running the GFS v16 maps in parallel on our models page for the past few months, and I’ve honestly not seen much to impress me,” wrote Matt Rogers, a meteorologist at Commodity Weather Group. “The model is still quite volatile from run-to-run with significant changes that lack consistency. While it may have a decimal improvement in skill score, it will likely continue to verify as a weaker model against the European and all the various ensemble guidance.”