The new version of the model is known as the GFS-FV3 (FV3 stands for Finite Volume Cubed-Sphere dynamical core), and it contains what is known as a “dynamic core.” The Weather Service says this dynamic core will make the model run more efficiently and modernizes the agency’s approach to forecasting.
The Weather Service had planned to launch the FV3 model in March but paused implementation after users of the test version of the model reported that its forecasts were unrealistically cold and snowy.
Since February, the Weather Service has worked to fix these problems. “We took this time to correct some of the deficiencies,” said Brian Gross, director of the Weather Service’s Environmental Modeling Center in an interview. “It’s better than where we were. We’re pretty confident in its performance.”
In a memo released Monday, the Weather Service explained that in previous test versions of the FV3, “snow was not adequately melting under warm conditions” and that its calculations for the sun’s intensity were incorrect. Together, these two factors meant the model’s forecasts were too cold and too snowy.
In recent months, the Weather Service made modifications to address these problems and, after some initial testing, concluded that “the results are promising and give NWS confidence to proceed.”
However, the memo noted that some cold bias remains in the model, which may still inflate snow amounts, mainly in forecasts three or more days into the future.
“The cold bias has been substantially reduced, but it is still there,” Gross said. “You get a cold bias, you can still get too much snow. That’s what we’re working on now and figuring out what to do.”
Assuming the Weather Service determines that the FV3 is ready to go after its upcoming 30-day test and launches it mid-June, it will continue to run the old version of the GFS until Sept. 30 so users can compare and contrast forecasts.
Gross said user feedback on the experimental version of the FV3 thus far has been “enormously” helpful.
“One of our key messages to our partners is to keep it [feedback] coming,” he said. “The more eyeballs we have, the better it’s going to be.”