It’s unusual to see violent tornadoes near Chicago in the late summer, which is why weather watchers were stunned when a high-powered twister tore through Plainfield, Ill., on a hot August afternoon in 1990. Karen Kosiba, who was a child at the time living in nearby Homer Glen, remembers its terrifying force. The tornado killed 29 people, leveled 470 homes and left a path destruction up to a half-mile wide.

According to the National Weather Service, tornadoes as intense as the one that hit Plainfield are exceedingly rare, accounting for only about 1 percent of all twisters. But that reassuringly small figure may be a product of the imperfect way that experts gauge tornado intensity, according to a study co-written by Kosiba, an adjunct professor at the University of Illinois’s Department of Atmospheric Sciences.

The findings suggest that the actual proportion of violent, Plainfield-like twisters is upward of 20 percent. Researchers show that the National Weather Service is underrating tornadoes that move through rural areas, leaving little wreckage behind. The study is part of an ongoing effort to update the tornado rating system.

Because tornadoes are so short-lived, it can be difficult to measure wind speed in real time, so the National Weather Service infers wind speed from the damage tornadoes leave in their wake. It then assigns a rating, from 0 to 5, on the Enhanced Fujita scale of tornado intensity. An EF-0 could uproot a sapling. An EF-3 could tear the roof off a sturdy building. An EF-4 or EF-5 such as the Plainfield tornado could rip a house from its foundation. These are classified as “violent” tornadoes.

The shortcoming of this system is that most tornadoes sweep over rural areas with few buildings, so they inflict little damage. A twister that could level a small town but instead glides over the prairie will leave scant evidence of its destructive potential, which puts the National Weather Service at risk of undercounting the true number of violent tornadoes.

“Just because it doesn’t look like it’s a big, strong tornado, or you don’t have the corroborating damage with it, doesn’t mean that it doesn’t have substantially strong winds,” Kosiba said. “A lot of these tornadoes have more power to cause damage than people think that they do.”

To gauge the accuracy of damage-based tornado ratings, Kosiba and her colleagues needed hard data on tornado wind speeds. That meant radar data. Because a Doppler radar casts a beam that, like a flashlight’s, gets wider and fuzzier at a distance, they needed up-close readings to get an accurate picture. So researchers used data from Doppler on Wheels, a mobile radar system that allows experts to capture high-resolution radar cross-sections of live tornadoes from about a mile away.

“It’s like a camera. If you stand across a parking lot, maybe you can count somebody’s fingers. If you want to see their fingerprints, you’ve got to get within a foot. And we’re trying to see those fingerprints,” said Joshua Wurman, an adjunct professor at the University of Illinois’s Department of Atmospheric Sciences and lead author of the study. Wurman invented Doppler on Wheels in 1995.

For the study, authors analyzed wind speed data gathered by Doppler on Wheels from 120 supercell tornadoes over the Great Plains between 1995 and 2006. The National Weather Service had rated 82 of those tornadoes.

When researchers compared radar data with damage-based ratings, they found that tornadoes were generally larger and more powerful than damage-based ratings indicated. The findings suggest that between 20 and 25 percent of supercell tornadoes boast wind speeds that would cause EF-4 or EF-5 levels of damage, even though only about 1 percent of tornadoes are typically rated EF-4 or EF-5. On average, tornado ratings are 1.2 to 1.5 categories too low, Wurman said. The findings were published in the Proceedings of the National Academy of Sciences.

“If a tornado goes through open fields, it essentially gets rated as a zero or a one,” Wurman said. “It gets rated very low, even though it might have 300-mile-an-hour winds.”

In one particularly egregious example, researchers tracked a slender tornado that tore through rural Nebraska in May 2004. The National Weather Service rated it an EF-0, but with wind speeds nearing 200 mph, it would have clocked in as an EF-4 had it run into more buildings.

Robin Tanamachi, an atmospheric scientist at Purdue University who was not affiliated with the study, praised the paper, which is the first to quantify the bias of damage-based ratings. She said that radar data can help scientists understand how tornadoes inflict damage.

“The core objective of the Fujita method is to infer wind speeds based upon damage. But because all buildings are constructed differently, and all tornadoes are different, no tornado-building encounter is the same as another,” she said in an email. “Did the tornado strike the building directly, or graze the side of it? Did the tornado pass over quickly, or spend more time on top of the building, potentially inflicting more damage? These are questions that radar measurements can help answer.”

The study will be used by a joint committee of the American Meteorological Society and the American Society of Civil Engineers, which is working to improve tornado wind speed estimates made by the National Weather Service. Engineers need to understand the true power of tornadoes to erect buildings that can stand up to them, so weather experts are looking to refine the Enhanced Fujita scale and begin issuing wind-speed estimates using radar and other tools, said Jim LaDue, a National Weather Service meteorologist and co-chair of the joint committee, which includes Kosiba and Wurman.

This new paper adds to a growing body of work showing that damage-based ratings tend to be more accurate when tornadoes run into more trees and sturdy buildings, LaDue said.

“It all fits this narratives that we’re just not capturing the strengths of tornadoes in rural areas,” he said. “We’re seeing multiple lines of evidence about the underrating problem.”

He said that future research should compare different methods of gauging wind speed to produce a more accurate picture of tornado intensity. The results could prove crucial to producing more tornado-resistant homes.

“This is where I think Josh and Karen’s work is really critical,” LaDue said. “They’ve really connected with the engineering community and through their data, the engineering community can help clarify their tornado models to evaluate building resistance and then try to come up with improved designs.”