WEATHER NEWS: How Hurricane Sandy sprung weather models into the mainstream
Ten years have passed since Hurricane Sandy walloped New Jersey and New York, becoming one of the deadliest and most destructive cyclones in modern U.S. history. In the days leading up to the storm, computer models took center stage as they showed the massive storm making an improbable turn toward the nation’s most populous coastline.
No model gained more publicity than “the European,” which was the first to project that unlikely turn. The news media heralded its performance and asked why “the American model,” run by the National Weather Service, wasn’t as quick to accurately simulate the storm’s path. Congress took notice and authorized the first of several rounds of funding to improve U.S. computer modeling.
Scientists acknowledge that Sandy raised awareness of computer models and paved the way for improvements. Since Sandy, models have been run on ever-faster supercomputers and become more accurate.
But some scientists also say the public’s and media’s focus on individual computer models has — at times — become a distraction. They say a misplaced emphasis on what the models show has diverted public attention from the forecasts produced by experts.
The European and American models, and Sandy
Weather is created by air sloshing around in the atmosphere, with movement largely governed by a small number of very complicated equations. Given sufficient computing power, it is possible for computer models to solve these equations and simulate the atmosphere’s erratic motions.
Because of the immense computing resources required to operate these models, they are almost all hosted by governments. The United States has a small handful, with the American model or Global Forecast System (GFS) the most well-known. Britain, Canada and Japan each have their own forecast models, while the European model — operated by the European Center for Medium-Range Weather Forecasts (ECMWF) — is supported by 35 member states.
There is a great deal of international cooperation in building forecast models, according to Louis Uccellini, the previous director of the National Weather Service.
“This is one of the more astounding collaborative relations we have as a country, built off a long history of international collaboration in the meteorological community,” he wrote in an email. Still, despite the note-sharing, various forecast models obtain data and compute equations slightly differently. In a system as complex as the atmosphere, this can yield big differences in their forecasts.
On Oct. 21, 2012, the European model projected that a young tropical storm south of Jamaica would merge with a cold front and slam the northern Mid-Atlantic eight days later. The American one, meanwhile, showed Sandy veering well out to sea. Not for three to four more days would the American model begin projecting a track similar to the European, depicting Sandy’s infamous left hook into the New Jersey shoreline.
The European model’s noteworthy early success in simulating such a high-impact and complex storm led many meteorologists, the media, politicians and members of the public to declare that “the Euro nailed Sandy.” That saying has since become a popular meteorological meme.
“Sandy was when [models] entered mainstream popular culture,” said Neil Jacobs, former acting administrator of the National Oceanic and Atmospheric Administration and chief science adviser for the agency’s next generation modeling effort, in an interview. For the first time, people who weren’t meteorologists saw a “competition” between the American and European models, Jacobs said. That notion has proved influential and difficult to shake in the decade since.
Experts: Forecasters ‘nailed’ Sandy, not computer models
According to some experts in forecasting and computer modeling, the European model should only partially be credited for the successful Sandy forecast.
Richard Rood, a professor at the University of Michigan who has studied biases and successes of atmospheric models, said that the European model’s long-range simulation for Sandy didn’t result in an accurate forecast by itself. Rather, it helped meteorologists understand one edge of an uncertainty envelope that facilitated a skillful forecast. Such a holistic analysis of model data is how experts at the National Hurricane Center more accurately forecast hurricane tracks and intensities than any individual model.
Uccellini wrote in an email that “ultimately, it was the forecasters who nailed the details of Sandy’s track.” While the European model did indicate Sandy’s potential East Coast impacts before the others, forecasters who analyzed the entire suite of models and their respective ensemble members (dozens of alternative simulations with slightly different input data) helped assure that the predictions were steady and reliable, Uccellini said.
While crediting the European model for nailing Sandy may not paint the full picture, the model does, in fact, have a well-documented edge over the American one. Evaluations of model accuracy show that the European model has produced somewhat superior simulations of the global atmosphere since the 1980s. But these benefits are readily washed out by the nuances of individual storms or even groups of storms.
The European model doesn’t outperform the American in every storm. For example, the American model produced better forecasts than the European on average across the 2021 hurricane season. Scientists say this is why it is essential to look across all models, and understand their biases, before making a forecast.
Sandy’s modeling legacy
The European model’s successful long-range simulation of Hurricane Sandy’s track was a watershed moment in creating public discourse around individual model forecasts with both positive and negative consequences, scientists say.
Rood described a number of “upgrades and documented improvements” to the American model over the past decade, with “the performance in Sandy [being] a major motivator.” Jacobs agreed, saying that Sandy had “almost everything to do” with the political push to improve the American model.
“Sandy was a big eye-opener to Congress about what needed to be funded,” Jacobs said. He noted that Sandy was a motivation for the Next Generation Global Prediction System, the Weather Service’s current effort to increase the accuracy of forecasts out to 30 days.
Still, Uccellini stressed that a focus on individual models can become a distraction when it interferes with official forecast messages. These models sometimes show extreme scenarios or waver from simulation to simulation. If the public or decision-makers anchor to a particular model simulation, they can be misled because a given simulation at a given time doesn’t capture the full range of forecast possibilities.
Uccellini urged the public to “spend more time listening to the [professional] forecasts” than “comparing individual model runs during an extreme event.” Forecasters, he added, “are accessing all models and related post processing that assist in the selection of the most likely forecast.”
Although the American model still lags the European in overall performance, efforts to improve it that began after Sandy continue. In June, NOAA inaugurated new weather and climate supercomputers with more computing and storage capacity for running the American. 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.