Predicting the whirlwind
Across the wide open plains of the central US and inside air-conditioned computer laboratories, scientists of different stripes are probing one of nature’s most devastating phenomena: tornadoes. Stephen Ornes offers a snapshot of their work
In more than 20 years as a meteorologist, Joshua Wurman had seen – and chased – more than 150 tornadoes. But the one that hit his pickup truck caught him by surprise.
On a warm spring evening in 2012, Wurman was driving down a two-lane highway in the small plains town of Russell, Kansas. In Russell, as in hundreds of other communities across the central and eastern US, late spring is tornado season, and Wurman – who became something of a TV celebrity in 2008 after a string of appearances on the Discovery Channel show Storm Chasers – knew that their chances of seeing a twister were good. Earlier in the day, radar had shown a giant storm brewing over central Kansas. When he and his team hit the road, their aim was to collect real-time data on the wind and weather using mobile radar units that can scan and image the structure of storms. Wurman had been trying to put instruments in the twister’s path. About a mile away, collaborator and atmospheric scientist Karen Kosiba was directing the mission from one of those mobile units.
The tornado chaser’s gamble is to get close enough for good data, but not close enough to endanger your life. That night in Russell, Wurman – a father of four – came close to losing the bet. An hour after sunset, a tornado formed unexpectedly just south-west of where he’d been driving. As it blazed north-east, it grazed the side of the truck. Wurman kept driving, but the tornado turned, striking the truck a second time. As it continued north, roughly following the highway, the twister picked up a house with its occupant inside and threw it 30 m through the air (figure 1).
“That wasn’t our experimental design,” Wurman deadpans as he reflects on that night from the safety of his office at the Center for Severe Weather Research in Boulder, Colorado. “In many ways, it was a botched mission.” The good news, he adds, is that nobody was seriously hurt. He and Kosiba were shaken but fine, and the disoriented homeowner was found sitting on a pile of rubble that had formerly been her house. Members of the team who raced to help were relieved to discover that her injuries were limited to scratches and a broken collarbone. The only other casualty was one of the team’s equipment vehicles.
Nevertheless, the incident was an unwelcome reminder that chasing storms is an unpredictable and hazardous occupation. These violent, swirling storms don’t follow obvious tracks. They’re often small and short-lived, although big ones can linger for hours and cause widespread destruction. Their interiors are too dangerous for humans to explore, and flying debris can wreak havoc with even the most robust instruments. And as Wurman discovered in Russell, they can be almost impossible to predict: the best forecasts give about 13 minutes’ advance notice, max, and most warnings turn out to be false positives. “How do you study a phenomenon that likely won’t happen, only lasts a short amount of time, and if you get too close it will kill you?” he asks.
Winds of creation
Some of the science behind tornadoes is well understood. Physicists have long observed that fluids form vortices when they become unstable. Such vortices efficiently transfer liquid or gas from one place to another; we see them when we drain the bathtub or flush the toilet. So essentially, a tornado is a real-world fluid dynamics experiment gone out of control, quickly shuttling air upward along a spinning column of wind that connects the ground to a storm cloud.
The biggest, most destructive tornadoes spawn from supercells: gargantuan, long-lived storms that churn like engines in the sky. A tilted, rotating updraft, often 10 miles or more in diameter, poaches warm moist air from lower climes to feed the storm. This mega vortex in the sky is called a mesocyclone, and it also thrives on a phenomenon known as vertical shear, where winds at different heights blow in different directions, and at different speeds. Tornadoes often seem to drop out of the bottom of these mesocyclones like the tiny, destructive tails of giant beasts. But they don’t really drop: the wind near the surface is already rotating, and when a tornado materializes the spinning wind becomes more intense and concentrated.
The problem is that although atmospheric scientists know the ingredients that make a tornado, they still don’t know the recipe. Some suspect that tornadoes might be too complicated to fully comprehend, beyond the reach of nonlinear physics. The chaos that drives tornadoes might prevent physicists from ever finding all sets of favourable initial conditions to plug into an equation. The unpredictability of the “butterfly effect” – the idea that a small perturbation can hugely change an outcome – further hampers their efforts. But other researchers continue to hope that some marker, some Achilles heel, may yet be found that would let researchers predict, with a probability better than today’s warning systems, whether or not a supercell will spawn a tornado. “We’re still looking for that holy grail,” Wurman says, although he adds that scientists may need to find and understand several different markers, not just one, before they can reach that goal.
In this respect, the formation of tornadoes may be similar to the genesis of cancer in the human body. In decades of studies, oncologists have identified a raft of cancer-causing genes, chemicals and environmental exposures that spur cells to propagate out of control in the body. Individually, a single mutation or exposure may not be influential enough to cause disease, but when many of them occur simultaneously, through a combination of bad luck and large numbers, they produce a biological “perfect storm” for cancer. Similarly, simulations based on real-world tornado data suggest that these meteorological storms require unlikely but not impossible combinations of wind shear, moisture, instability and other factors, in just the right order, to form.
Danger at the boundary
Wurman is one of the pioneers at getting this real-world data. Back in 1994 he designed a Doppler radar instrument that can quickly collect information about wind speed and direction. Such devices can be mounted on trucks and driven to storm sites; with their enormous antenna dishes, these “Doppler on Wheels” units, or DOWs, look like they could be listening for aliens as they race around Tornado Alley. Other specialized tools have also been tried. In the 1980s US government scientists developed a package of instruments to measure wind speed, pressure and humidity that could fit in an aluminium barrel and be placed in the path of a tornado. That device – named TOTO after Dorothy’s dog in The Wizard of Oz – never made it into an actual twister, but in 2003 meteorologist Tim Samaras successfully used small, rugged devices with barometers built in to measure the intense pressure drop inside an active tornado. Other researchers have launched weather balloons outfitted with anemometers into supercells and used lightning arrays to search for connections between flashes of light and tornado intensity.
Perhaps the most important real-world measurements focus on the winds that whip around closest to the ground, where conditions are most likely to be hazardous to people and buildings. Atmospheric scientists call this the tornado boundary layer, and because of the violence of the winds, it is notoriously difficult to study. But after the Russell tornado battered his truck and equipment, Wurman realized that there was a silver lining: the DOW on the damaged truck had continued to scan the entire tornado every seven seconds at different heights above the ground before it was knocked offline. Those measurements provided unprecedented data on near-ground wind conditions.
Together with Kosiba, Wurman found that the strongest winds occur low to the ground, with the peak intensity only about 5 m off the ground. That’s surprising. After all, the top of a tree or building is usually windier than the bottom. Not so in a tornado: the most dangerous area may, in fact, be at the level where we spend almost our entire lives. Wurman says that many scientists used to believe that most tornadoes were too weak to be dangerous, but recent studies suggest that they’re stronger than previously believed – and capable of inflicting substantial damage. Kosiba and Wurman reported their findings in the December 2013 issue of the journal Weather and Forecasting (10.1175/WAF-D-13-00070.1).
But even though scientists are getting better at finding storms and collecting data, tornado research still relies heavily on individual case studies. It isn’t like particle physics, where researchers can gain confidence in their findings by running an experiment multiple times; as Wurman notes, just because the Russell tornado had its strongest winds near ground level, that doesn’t mean that every other tornado will behave the same way. “We’re still at the stage where we’re not even asking the smartest questions,” he says.
One thing physicists can do, though, is to funnel what tornado data they do have into computer models. And the models are getting better: “They’re doing simulations that 10, 20 years ago you couldn’t have conceived,” Wurman says. But he also notes that atmospheric scientists need better observations to feed their simulations, and better simulations to predict phenomena that can be confirmed by observation. This virtuous loop can’t continue indefinitely. “At some level,” Wurman says, “tornadoes are going to be unobservable. There are things we can’t directly observe because of the violence.”
On 31 May 2013, just over a year after Wurman’s close call in Russell, that violence claimed high-profile victims within the chaser community. A trio of scientists in El Reno, Oklahoma, were trapped and killed by a tornado that formed beneath an intense, mile-wide supercell with multiple vortices that followed unpredictable paths. The group, which included Samaras and his 24-year-old son, had been trying to deposit “pods” that collect wind data as they’re pummelled by tornado. Samaras had decades of experience studying tornadoes and was known for his conservative approach to safety. But the lashing wind and rain, together with the tornado’s tortuous trajectory, likely made it impossible for him to see a safe exit.
The storm in the machine
As field scientists search for safer ways of collecting storm data, Leigh Orf at the University of Wisconsin’s Space Science and Engineering Center is developing new ways of putting those data to work. Like Wurman, Orf is an atmospheric scientist, but he’s not a storm chaser. Instead, he uses supercomputers to simulate the complexity of a supercell. He’s pursuing the same holy grail that Wurman mentioned: what initial conditions give rise to the most destructive tornado?
Numerical simulations can reveal dynamic structures of different scales within a tornado, while visualizations of those simulations can help researchers better understand what these phenomena look like (figure 2). But getting there requires “a whole lot of grunt work”, Orf says. “When we walked into this, we had to handle lots of data and do something no-one else had done before, to capture from birth to death one of these nasty tornadoes in a good simulation,” he says. He and his collaborators have been putting in the hours since the 1990s, trying to figure out how to use powerful supercomputers to tame the volumes of data needed to create models of weather systems.
About five years ago, Orf started working on a simulation that uses real-world initial conditions to replicate a real-world disaster. “My ultimate goal was to simulate the most devastating type of tornado,” he says. Tornadoes are rated on the Enhanced Fujita (EF) scale, which estimates tornado wind speeds based on damage surveys. Tornadoes in the most destructive category, EF5, have winds that exceed 300 miles per hour. While most tornadoes are short-lived, high-end behemoths can be devastating. In May 2011 a tornado outbreak in the midwestern and southern US killed more than 150 people and caused $7bn in damage. On 24 May one particularly violent tornado in that outbreak stayed on the ground for more than two hours and carved a path 63 miles long. Coincidentally, it began near El Reno, Oklahoma – the same town where, two years later, Tim Samaras was killed collecting data.
To model this storm, Orf created a numerical simulation using an existing model for atmospheric phenomena called CM1, which he describes as “Navier–Stokes [equations] plus a bunch of physics parameterizations”. He also used data visualization tools to produce images of the simulated tornadoes. For the model’s base state, he plugged in forecasts from another model based on airborne measurements from near El Reno on 24 May 2011 – similar to the wind speed and other observations made by scientists like Wurman on the ground. Then he perturbed the state and let the model take over.
Orf simulated that storm dozens and dozens of times on Blue Waters, a supercomputer at the University of Illinois at Urbana–Champaign that holds the distinction of being the world’s fastest university-based supercomputer. Most of those iterations didn’t produce tornadoes. “Even in this one environment,” he says, “certain things have to happen in the right order.” But after years of trials, he and his collaborators managed to poke the supercell in just the right ways to produce an EF5, long-track tornado like the one that formed near El Reno in 2011. They published their findings in the January 2017 issue of the Bulletin of the American Meteorological Society (10.1175/BAMS-D-15-00073.1).
This work revealed detailed, complex structures within the tornado, including multiple smaller vortices twisting up to form larger ones (figure 3). Those vortices, says Orf, support Doppler radar observations made by storm chasers who scan real tornadoes. Another persistent feature in the simulations was a horizontal tube of air that follows the ground and tilts upward into the updraft – something that has also been documented in the field. “The storm is finding interesting ways to organize vorticity, and tilt and stretch it,” says Orf. “It’s sort of found a way to recycle energy back into itself, and it’s really quite fascinating.”
The ideal model would use a set of initial conditions, sampled from the environment, to simulate what’s about to happen, in less time than it would take for the tornado to form or not form
Orf points out that his intensive simulation covers just one case, making it difficult to generalize. On the other hand, accurate simulations offer a repeatable way to study supercells and vortices – without plunging into the belly of the beast. Encouraged by his results with the El Reno tornado, Orf says he’s now preparing simulations of other recent devastating outbreaks, to help researchers identify commonalities and, ultimately, turn that knowledge into better safeguards for the people at the greatest risk. The ideal model would use a set of initial conditions, sampled from the environment, to simulate what’s about to happen, in less time than it would take for the tornado to form or not form – making it possible to provide accurate warnings without the stress of false alarms or the related hazard of “tornado fatigue”.
“The end game of all this research,” he says, “is better prediction.”