Supercomputers, Tornadoes, and the Biggest Unsolved Mystery in Weather Technology

Even the best forecasting tools can't predict monster twisters quickly enough.

Predicting tornadoes is so difficult that many people still rely on natural warning signs and folklore. Sky the color of pea soup, eerie stillness, and thundering hailstones are all ominous. But none of these conditions, alone or together, guarantees a tornado is imminent. Not even the best weather-forecasting technology can determine more than a few minutes ahead of time whether a tornado will strike.

By the time a tornado has touched down, it’s often too late for people to reach safety. John Kass, a Chicago Tribune columnist who survived a devastating tornado in 1967, wrote that it sounded like a “freight train of demons,” and remembers the feeling of “nature as evil, nature with a mind, predatory, nature intent on hunting us down.”

Even as forecasting technology has made remarkable advances in recent decades, tornadoes are an outlier. They remain dangerously difficult to see coming.

There are a few reasons for this: Tornadoes are extremely localized weather events and they don’t last long. Even the most destructive tornadoes are over soon after they begin. Most last less than ten minutes. There are also a wide variety of conditions that tornadoes favor, which means that most storms that could produce a tornado don’t. And the ones that don’t? They seem to look just like the ones that do.

Researchers understand the kinds of storms that can create tornadoes, but those storm systems can cover enormous swaths of land, well into the hundreds of miles. It’s next to impossible for forecasters to say with any precision which storms will produce the big tornadoes. And forecasters have “very little skill” predicting when in a storm’s lifetime a big tornado will emerge, as Joshua Wurman, the president of the Center for Severe Weather Research, told NBC News in 2014.

That’s not to say there haven’t been improvements. “Back in the day, you would basically be weather observers, more nowcasting as opposed to forecasting,” said Christopher Vaccaro, a spokesman for the National Weather Service. “As conditions were unfolding, people were sharing that by word of mouth. A few decades ago, a ‘tornado warning’ would basically be a nowcast.”

Today, officials issue a tornado watch when weather conditions mean a tornado is possible, and they issue a tornado warning when a tornado has actually been spotted. The difference, as the National Weather Service puts it, is that a watch means you should be prepared to take action, and a warning means you should take action immediately. Even if someone is alerted to a tornado warning the moment it’s issued, there’s little time to seek shelter.

Officials put out a warning 16 minutes before the tornado that decimated Moore, Oklahoma, in 2013, touched down and killed 24 people. Two years before that, in Joplin, Missouri, sirens sounded 20 minutes before the catastrophic twister touched down and killed 158 people.

In the 1980s, the vast majority of tornadoes had no warning at all. If there was a warning, it would occur only maybe two minutes before a tornado hit. (And there weren’t emergency text-messages in those days.) Today, forecasters are able to give closer to 14 minutes notice before a tornado touches down—but not in every case. About one-fifth of the time, still, there is little to no warning. “We’re able to issue warnings based on our radar,” Vaccaro said. “Certainly the visible confirmation of tornadoes through spotters is still very necessary.”

There are some technologies that can help. A specific radar signature called a “debris ball,” for example, suggests that there is debris in the air from a tornado having wrecked something. Some researchers believe improvements in radar could eventually mean reliable tornado warnings up to an hour ahead of time. Vaccaro says he can imagine even longer lead-time, given enough observational data and the processing power from supercomputers to analyze it.

Computers already process trillions of weather-related data points per second, which is a large part of why long-term forecasting accuracy, in general, is so much better today than it was even a decade ago. “There might be a time where we could warn of a tornado before we could even detect it on radar,” Vaccaro said. “It’s about getting an even more complete and accurate view of the atmosphere, and then the supercomputing to build models, pushing that boundary of science.”

Meteorologists aren’t there yet, but there’s perhaps some reassurance in the fact that other forecasting feats that would have seemed impossible in the past are now routine. Hurricanes and nor’easters, for example, are no longer surprises.

“If you were to go back and tell meteorologists in the 1970s we would be able to forecast these big events three to five days in advance,” Vaccaro said, “They would think you were experimenting in witchcraft.”