That struggle to pinpoint Ian’s path underscores the challenges of formulating and communicating forecasts for such a complex, unpredictable and severe storm. The American model, known as the Global Forecast System, or GFS, is among a host of supercomputer-based tools meteorologists use to combine current weather observations with knowledge of atmospheric science to predict where a storm such as Ian will go.
And while those weather models quite accurately predicted that Ian would rapidly intensify as it moved into the Gulf of Mexico, they gave mixed signals on what path the storm would take — contributing to surprise for many on the ground in Southwest Florida when Ian made landfall there.
As Florida assessed Ian’s damage, Gov. Ron DeSantis (R) stressed the challenge of interpreting the models when a reporter asked whether more should have been done to warn residents about the storm’s severity.
“I followed not just the NHC [National Hurricane Center] track, the Euro [European] model, the ICON model, the GFS — most of you probably don’t even know what those are,” he said, the Miami Herald reported Monday. He said that while it’s important to “look to see what you can do better,” he did not elaborate and said he was focused on recovery.
So what are those models? And what were they telling DeSantis and other Florida leaders as the storm approached?
Here is a look at how they performed, based on a preliminary NOAA review scientists presented via webinar to meteorologists and media Thursday.
When pressed on preparations, DeSantis stressed that many media reports days ahead of the storm had focused on weather models’ early predictions of landfall somewhere around Tampa Bay, about 100 miles to the north of the parts of southwest Florida that ended up hardest hit.
It’s true — key models’ predictions of the storm track were almost entirely too far to the northwest, though some were farther off than others.
The GFS had the largest error of the major weather models when it came to predictions of Ian’s track, according to NOAA researchers. For example, the GFS was still suggesting potential landfall as far north as the Florida Panhandle at the same time models run by the European Centre for Medium-Range Weather Forecasts, commonly known as the European model, and the United Kingdom’s Met Office, known as the UKMet model, were predicting that the storm would hit south of Tampa Bay.
Even the European model, considered to be the most accurate forecaster of global weather patterns, skewed slightly northwest of Ian’s eventual track, gradually shifting its predictions toward the southeast as the storm approached.
The uncertainty in track predictions was largely linked to how Ian interacted with a frontal boundary that stretched across the Southeast, separating an unseasonably cool and dry air mass over the eastern United States from the tropical moisture fueling Ian, said Geoffrey Manikin, a NOAA researcher, in a presentation Thursday.
The UKMET model was the best predictor of the landfall in Southwest Florida, and it also most closely called Ian’s path into the Atlantic and then the South Carolina coast, the NOAA researchers said. The European model, too, had a narrower spread of track predictions than the American model, operated by NOAA.
But that model and others NOAA uses to analyze hurricanes, on the other hand, reliably predicted the rapid intensification that made Ian so devastating.
A pair of high-resolution hurricane models run by NOAA are designed to detail wind speeds and storm structure in ways large-scale models like the European and GFS cannot. The models began running on a new version of the agency’s supercomputing system ahead of this year’s hurricane season, something NOAA said would allow for “uninterrupted” tropical cyclone forecasting.
These hurricane-focused models repeatedly called for the rapid strengthening that would occur as Ian entered warmer-than-normal Gulf of Mexico waters and predicted wind speeds as high as 168 mph, slightly higher than the 155 mph winds ultimately observed at the storm’s strongest.
Brian Tang, an associate professor of atmospheric and environmental science at the University of Albany, called that level of error “really low, especially for a storm that underwent rapid intensification.”
What did Hurricane Ian reveal about the models?
The variance in model forecasts of Ian underscores the difficulty of summing up so much uncertainty in a single forecast.
Each of the Hurricane Center’s periodic updates to Ian forecasts, after all, represented the full spread of those predictions, accounting for “the uncertainty inherent in all numerical prediction models,” said Louis Uccellini, former director of the National Weather Service and now a visiting professor at the University of Maryland at College Park.
In the case of Ian, the high degree of uncertainty meant forecasters had to proceed with caution when adjusting and communicating forecasts. As models predicted landfall farther and farther south and east, the Hurricane Center was slow to adjust its forecast cone and continued to emphasize that a wide stretch of coastline should prepare for the worst.
“What the Hurricane Center wants to avoid is a windshield-wiper effect,” when the storm track is adjusted one way only to quickly move in the opposite direction, Tang said. “You don’t want to go too far to the extreme, because who’s to say it won’t correct to the west?”
Whether the center’s meteorologists made the right decisions in interpreting the forecast will be reviewed in a year-end analysis of the hurricane season, Uccellini said. That “could reveal deficiencies, but I could not tell what these might be,” he said.
In the meantime, one storm doesn’t dictate the value or accuracy of any given model. Though forecasts from the GFS “were not the best” for Ian, according to Alicia Bentley, an atmospheric scientist at NOAA’s National Centers for Environmental Prediction, the American model was the most accurate overall for the 2021 hurricane season and even for some storms this year, including Hurricane Danielle, which never posed a threat to land.
The GFS, in terms of overall accuracy averaged over long periods, still ranks in third place among global computer modeling systems, behind the European and UK Met models.
Despite major investments, the GFS has lagged behind the European models since the 1980s. In June, NOAA inaugurated new weather and climate supercomputers with more computing and storage capacity for running the GFS. Meanwhile, the ECMWF has continued to invest in the European model. It opened a new computing facility in Bologna, Italy, that it says will boost the system’s performance fivefold.