The United States’ Global Forecast System (GFS) has received a major upgrade to its modeling capabilities. The new dynamical core that has been added to the GFS – its first new dynamical core in nearly 40 years – will dramatically improve its forecasting abilities. The National Oceanic and Atmospheric Administration (NOAA) also hopes it will help to reestablish U.S. leadership in the international weather modeling community.
After years of rain-checked upgrades, U.S. weather modeling was definitively snowed under in 2012, when the GFS failed to accurately predict Hurricane Sandy’s landfall. The ensuing years saw the National Weather Service (NWS) showered with improvements – its total supercomputing power increased from 776 Linpack teraflops in 2012 to 8.4 Linpack petaflops in 2018, and its storage capabilities were increased in similarly dramatic fashion.
All of this was in preparation for a massive upgrade to the GFS model itself – and, at long last, that upgrade has arrived.
The Finite-Volume Cubed-Sphere (FV3) dynamical core is responsible for computing wind and air pressure in weather prediction modeling. The model divides the atmosphere into cells, simulating how air particles move through each cell around the globe. After selecting FV3 from a variety of new dynamical cores, NOAA worked for three years to integrate the FV3 into the heart of the GFS.
The upgrade was run through a battery of tests. After rigorous analysis – several simulated years of side-by-side comparisons with the previous GFS, conducted over the course of a year by more than a hundred experts nationwide — NOAA concluded that the FV3-based GFS showed clear improvements over the previous GFS in a number of areas. It outperforms the old GFS in predicting winter storms and tropical cyclones – in terms of both track accuracy and intensity – and promises to generally improve both 1-2 and 3-7 day forecasts, specifically with regards to rain and snow.
“The weather forecasts that you hear are provided by NWS forecasts or folks that are using the numerical guidance that our numerical models provide to them,” said Brian Gross, Ph.D., director of NOAA’s Environmental Modeling Center. “So we’re really looking at improvements in this foundational information that the individuals that forecast the weather use on a day-to-day basis.”
“The GFS with the FV3 dynamical core brings together the superior dynamics of global climate modeling with the day-to-day reliability and speed of operational numerical weather prediction,” Gross added.
The FV3 GFS has also been ported to commercial cloud technology – a capability that is aimed at giving researchers and universities the ability to run global model forecasts. This is a stark change from the previous GFS, which was designed to run specifically on NOAA hardware.
The FV3 is an extremely promising start to a new era in U.S. weather forecasting. “However,” says Louis W. Uccellini, director of the National Weather Service, “Our job doesn’t end there – we also have to improve the physics as well as the data assimilation system used to ingest data and initialize the model.”
NOAA and the National Center for Atmospheric Research (NCAR) are working on a common infrastructure that will bridge the gap between operational weather forecasting and the research-oriented modeling community. “This new dynamical core and our work with NCAR will accelerate the transition of research advances into operations to produce even more accurate forecasts in the future,” said Uccellini.
The newly-upgraded GFS is just one element of NOAA’s long-term weather modeling trajectory. “Implementing the FV3 GFS is the first step in our work to deliver the Next-Generation Global Prediction System – or NGGPS – which is an ongoing effort and will include a series of future upgrades,” said Neil Jacobs, Ph.D., acting NOAA administrator.
As for the previous GFS? It will run in parallel with the new GFS until the end of September 2019 to allow for further comparisons between the two models – then, it will be taken out of service.