In the days before a hurricane makes landfall, we all become storm experts. We nod our heads with our favorite TV forecaster. We posture about spaghetti models. We hyper-focus on one model and say, surely, this is the one to follow.
It’s the GFS this time. No, maybe the European is the one to trust. Or how about that one random squiggle that has the storm missing Florida entirely and steaming to the Flemish Cap?
Hurricane Ian, which clobbered Southwest Florida in late September, showed how frustratingly disparate the models can be for the public — particularly several days before landfall.
Many of the models had Ian headed for the Panhandle. Then a good bunch had it aimed straight for Tampa Bay. Then some predicted a landfall south of Tampa Bay, which turned out to be prescient.
But what exactly are these models and where did they come from? And how can they be so different from one another? And is there one we can trust more than the others?
Fact is, none of the models are perfect. And Floridians shouldn’t pick a single favorite.
Like neighbors helping one another prepare for a storm or digging out afterward, they work best when they work together.
What is a hurricane model?
Very few models are used only for hurricanes. Government agencies use them every day to forecast the mundane: Will it be nice this weekend? Will I need an umbrella when I head to work Monday morning? When will a cold front finally break our oven-like summer?
Super computers run these models, calculating complex formulas to understand what’s going on with the Earth’s atmosphere, said David Nolan, a professor in the Department of Atmospheric Science at the University of Miami.
But come hurricane season, these models can be just as useful telling us where that Category 4 monster is likely to land.
The models split the atmosphere into a grid, then take a snapshot of a piece of the atmosphere. The more snapshots, the better understanding forecasters have of the atmosphere and the better these models will perform.
Think of it like pixels in a photo: The more pixels, the better the quality of the image.
“The Earth’s atmosphere is broken up into little squares, and then every square basically says to itself: Alright, I know my temperature and my pressure and my wind speed and wind direction, and whether or not it’s raining, and so on,” Nolan said.
Based on that square, and what’s happening in the squares next to it, the models predict how that’s going to change in the next minute.
“So all of them step forward in time, basically using information based on themselves,” Nolan said.
The models make errors. And those errors swell the further out the forecast goes. So the forecast for five days from now will be less accurate than the one for two days from now.
There are sizable areas where scientists can’t collect data, like in the middle of the Atlantic Ocean, where there are no weather stations or weather balloons.
However, satellites and commercial planes, which also collect data, can help to fill these gaps, said Jeff Masters, a meteorologist for Yale Climate Connections.
That’s why you need big computers. There are billions of these images across the planet, Nolan said, and the models march forward in time for weeks. The computers do billions of calculations just to get our normal weather forecast.
“Nonetheless, you know, it’s not the real atmosphere, it’s not perfect,” Nolan said. “I would say it’s amazingly good what weather forecasts can do.”
Why so many models?
There are probably more weather models than you think, said Richard Rood, a professor in climate and space sciences and engineering at the University of Michigan.
The National Hurricane Center breaks some of these models into categories: dynamical models, statistical models, and ensemble and consensus models.
In all, the Hurricane Center lists 27.
So yes, there are a lot.
The Hurricane Center uses what it considers to be the top models for forecasting hurricanes. The models come from different countries with their own weather services, and some come from the United States, which has multiple agencies that have their own weather models for their own purposes: the National Oceanic and Atmospheric Administration, the National Weather Service, the Navy, NASA and more.
But the two we hear about most are an American model known as the Global Forecast System, or GFS; and the European model or the ECMWF, which is the European Center for Medium-Range Weather Forecasts. The two global models are considered the best in the world, and are often pitted against one another.
Experts generally agree the European model has an edge over the GFS, Rood said.
Both are modeled from equations performed by high-powered computers. The GFS and the European both are medium-range forecasting models and can forecast fairly far into the future. The Euro can forecast 10 days out, while the GFS can model 16 days out.
These two models, along with all the others, have the opportunity to use the same shared data in their model runs. Yet, they result in different outcomes.
“Just because they have the opportunity doesn’t mean they are all starting with ... the same data,” Rood said. “The other thing is, they all use the data in a little bit of a different way, so they have different algorithms.”
The main difference in the models is “data assimilation,” which essentially means how the models start. Scientists use different methods to begin these assimilations, which gives them different results. Rood said the European model generally has a better assimilation process than the GFS, which can at times lead to more accurate models.
The same models can also be run about 30 to 50 times to make what’s called an ensemble. Slight tweaks to the information used in the same model can result in dozens of different tracks. While it may look overwhelming to the untrained eye, these ensembles are helpful for forecasters.
Outside of the models that are used for everyday weather, there are exclusively hurricane models that focus on a much smaller area where a storm is located. These models are good at gathering more details like wind speeds and have become another piece in a forecaster’s tool belt.
“From a forecaster’s point of view, there is an advantage of having multiple models, because it helps them to better understand the uncertainty,” Rood said.
The Great Model Debate
After a storm has passed and made its landfall, forecasters can go back and see which model performed best.
With Hurricane Ian, the individual models that performed best depend on the time frame, but overall, the UKMet model, the United Kingdom Met Office’s global forecast model, and the COAMPS, a hurricane specific model, were best, Masters said. Both models are publicly available, according to the Hurricane Center’s list of models.
Weather models are always being improved across the globe. The GFS rolled out advancements in 2021 to the model’s resolution and data assimilation, among other things.
The worst individual model for predicting Ian was the GFS, which was actually the best performing model on average in 2021. The GFS model had Ian moving into the Panhandle — a nearly 400-mile miss.
The European model fared better than the GFS, but still skewed slightly to the northwest of Ian’s ultimate track, the Washington Post reported. In the days leading up to Ian, the Euro was concentrated on the Tampa Bay area while the GFS was focused on the Panhandle. Both models slowly tracked to the east ultimately deciding on a landfall near Fort Myers not long before the storm made it to shore.
There are a number of reasons the GFS lags behind the European model. Outside of some technical reasons, there are some systematic reasons, Rood said.
The United States has multiple models run by multiple agencies, which splinters resources. The European model is the only model used by the European Center, so it doesn’t have to split money or resources.
The debate over the best model — the GFS or the European — isn’t particularly helpful. The nation’s top forecasters rely on multiple models, not just one.
The Hurricane Center released a report early this year that reviewed the past year’s hurricane models and tracks, and determined which are most reliable. Despite the American GFS model outperforming other individual models, the Hurricane Center’s forecast still was the most reliable in predicting where a storm would land, Masters said.
The Hurricane Center’s forecast generally takes the average of five or six of the world’s most accurate weather models to create its forecast. Masters said the Hurricane Center sometimes will make adjustments based on what the perceived biases are, or what model is performing best.
In an article summarizing the Hurricane Center’s findings, Masters wrote that the Hurricane Center’s track forecast had “accuracies near or better than the five-year average, with two-day and three-day track forecasts setting new records for accuracy.”
Masters said forecasting Ian was more difficult than normal because the models split into two camps: one to the north near the Panhandle, and one to the south near Fort Myers.
The Hurricane Center averaged this split, Masters said, causing the center of Ian’s cone to land over Tampa Bay for a few days until the Fort Myers track became more likely.
“If you look at the top six models, pretty much any of those things could come true,” Masters said. “Don’t put your trust in any particular one. The average of them is usually going to be the best forecast.”
• • •
Tampa Bay Times Hurricane Ian coverage
FEMA: Floridians hurt by Ian can now apply for FEMA assistance. Here’s how.
THE STORM HAS PASSED: Now what? Safety tips for returning home.
POST-STORM QUESTIONS: After Hurricane Ian, how to get help with fallen trees, food, damaged shelter.
WEATHER EFFECTS: Hurricane Ian was supposed to slam Tampa Bay head on. What happened?
MORE STORM COVERAGE: Get ready and stay informed at tampabay.com/hurricane.