Jeff Masters Weather Blog

Inside the quest to develop long-range tornado forecasts » Yale Climate Connections

Nearly three decades ago, William Gray, a hurricane researcher at Colorado State University, embarked on the first sustained effort at seasonal hurricane prediction. His initial outlooks each year were issued at the start of the Atlantic hurricane season, which lasts from June 1 to November 30, giving an idea months in advance of how active the full season might end up being.

For years, Gray labored alone, relying largely on sea-surface temperatures and on links between hurricane activity in the Atlantic and the El Niño–Southern Oscillation, which drives the periodic warming and cooling of the eastern tropical Pacific known as El Niño and La Niña. There were plenty of both hits and misses. Over time, Gray and protégés Phil Klotzbach and Michael Bell gradually increased the accuracy of their predictions, and others joined the fray.

As of 2022, more than two dozen entities in the U.S. and Europe — including NOAA — were issuing some type of seasonal outlooks of tropical cyclone activity in the Atlantic. Many insurers and other firms now consider such products as they make business plans for hurricane season. The outlooks also serve as public reminders of the threats posed by landfalling storms.

Tornadoes don’t inflict the same scale of havoc as hurricanes. Yet a bad U.S. tornado year can still take hundreds of lives, terrorize millions of people, and inflict billions of dollars in damage. With this in mind, a handful of scientists scattered across government, academia, and the private sector are now exploring the potential benefits and pitfalls of longer-range tornado outlooks. The researchers are looking at time periods ranging from two weeks in advance to an entire season. They’re motivated in part by sheer scientific interest but also by hopes that their work might ultimately raise awareness and save lives.

Figure 1. Damage in Dawson Springs, Kentucky, from an EF4 tornado that struck Dec. 10, 2021. Spawned by a rotating supercell that traversed parts of four states, this tornado was on the ground for 165.7 continuous miles. (Image credit: Scott Olson/Getty Images)

In this deep dive, we’ll look first at efforts to craft full-season outlooks, then explore predictions in the shorter multi-week range.

The full-season challenge

Tropical phenomena with wide-ranging effects, such as the El Niño–Southern Oscillation and another known as the Madden-Julian Oscillation, do show some correlations with U.S. tornado activity. But the connections to date haven’t been strong enough to justify official seasonal outlooks for the public from the National Weather Service. Plus, there are knotty questions about the usefulness of a three-month outlook for the kind of extreme weather that plays out over minutes to hours.

“You can’t head for the storm cellar based on a seasonal tornado forecast,” said Michael Tippett, a researcher at Columbia University.

Tippett was among the first to explore monthly-scale prediction of tornado-favorable weather patterns. In 2017, he and John Allen of Central Michigan University joined an project led by Columbia colleague Chiara Lepore, now at Gro Intelligence, to issue experimental springtime outlooks each March. These outlooks span the period March through May and are based on the state of the El Niño–Southern Oscillation in the previous winter.

On March 9, NOAA issued its final advisory on the unusually prolonged La Niña event that held sway for most of the time since late 2020. We’re now in what’s called a Trans-Niño period, as the tropical Pacific moves into a neutral state and perhaps into El Niño conditions as soon as this summer. Research has shown that positive-phase Trans-Niño springs — those going from La Niña to neutral, as we are now — have a tendency to be more tornado-favorable, especially in early spring.

Along these lines, the Columbia outlook for spring 2023 shows a modest enhancement of tornado and hail risk for most areas east of the Rockies. Much like NOAA’s seasonal outlooks for temperature and precipitation, the Columbia outlook shows where a standard division of odds – 33% each for average, above-average, and below-average activity at each location – has been nudged this spring toward greater or lesser activity.

Figure 2. Seasonal tornado outlook issued in early March 2023 by researchers at Columbia University. The map shows predicted shifts in probabilities for tornado and hail across the contiguous United States during March-May 2023, as compared to an average spring (e.g., the 0.4-0.5 range from Texas to Minnesota denotes a 40-50% chance of an above-normal season for tornadoes and hail, as opposed to the standard 33% chance). Values from the Rockies westward may be spurious because of the low level of springtime tornado and hail activity in those areas. (Image credit: Michael Tippett)

Tippett hastened to add caveats, noting that the skill of such forecasts — in other words, their ability to perform better than automatically calling for a near-normal season — is still quite limited. “Overall, I don’t think there’s a scientific reason to expect a seasonal tornado forecast to have more skill than those of near-surface temperature and precipitation, and their skill is pretty modest.”

Victor Gensini of Northern Illinois University, another pioneer in extended-range severe weather outlooks, is similarly cautious. “We do not have many of these fading-Niña samples,” he said, “although the ones we do have tend to be above average [in severe weather] when the season is said and done.”

Several other scientists are experimenting with full-season severe-weather outlooks.

  • Sang-Ki Lee of NOAA’s Atlantic Oceanographic and Meteorological Laboratory published a landmark paper with colleagues in 2013 on the relationship between the El Niño–Southern Oscillation and U.S. tornadoes. He also led a 2021 paper on a seasonal probabilistic tornado outlook system based on NOAA’s CFSv2 seasonal forecast model. Despite some progress made during the past decade, Lee said the forecast system cannot yet distinguish the very worst seasons from the merely active ones. With that and other challenges in mind, he said, “there is no strong consensus yet among scientists and forecasters within NOAA to justify issuing a seasonal severe weather outlook.” Although the statistical relationships that he found are too marginal for a definitive outlook, Lee did note that, based on his 2013 paper, “the evolving positive-phase Trans-Niño is linked to favorable atmospheric conditions for an active U.S. severe weather season.”
  • Nathaniel Johnson of the NOAA Geophysical Fluid Dynamics Laboratory worked with postdoctoral researcher Kai-Chih Tseng, now at National Taiwan University, on a dynamical-statistical model, which draws on output from GFDL’s dynamical seasonal prediction system, SPEAR. The model developed by Tseng shows potential for national-scale prediction of U.S. tornado activity up to a year in advance. Now preparing a paper on the system, Tseng presented early findings at the American Meteorological Society’s annual meeting in January 2023 in Denver. “Because we are at a stage where we only focus on a single [U.S.-wide] metric of tornado activity,” said Johnson, “we believe the main utility of such a forecast would be to increase public awareness of the potential for an unusually active tornado season.” Over time, the researchers hope to extend the method to specific regions, and perhaps to seasons beyond spring.
  • Through his consulting firm WeatherDeep (a separate project from his day job), meteorologist Ashton Robinson Cook is providing seasonal guidance on severe-weather risk. In a 2017 paper, Cook and colleagues showed how winter and early-spring U.S. tornado outbreaks were heightened by La Niña through shifts in the favored location and strength of the subtropical jet stream, surface cyclones, and other features. “That paper and other bodies of work suggest that Trans-Niño events [La Niña to neutral] favor the occurrence of tornado outbreaks as we head into spring,” Cook said. More recently, Cook started WeatherDeep as an artificial intelligence (AI) platform to incorporate many additional weather features beyond ENSO to identify and forecast severe weather threats. WeatherDeep’s AI suggests a heightened risk of tornado activity from the Ohio and Tennessee River Valleys into Alabama and Georgia through early April. The southern U.S. is expected to remain active for tornadoes through the end of April, while Kansas, Oklahoma, and Texas should become more active through May. “In general, 2023 should be more active for tornadoes than 2022,” Cook said.  

Promise in the multi-week window

Researchers are also drilling down to shorter outlook windows. Weather model ensembles that now extend out to 16 days or even longer are starting to bridge a “valley of death” between explicit weather forecasts going out a few days—including the closely followed convective outlooks from the NOAA/NWS Storm Prediction Center, which extend out to eight days — and the seasonal outlooks that stretch out to months.

In between those time frames, the newest models are showing tantalizing hints that it might be possible to identify some especially dangerous weather patterns two weeks or more in advance.

In work now under way, Kimberly Hoogewind, a researcher at the NOAA National Severe Storms Laboratory, is taking a close look with colleagues at the twin tornado disasters of December 2021: the catastrophic twisters that took more than 70 lives in and around Kentucky on Dec. 10 and a derecho-borne swarm of more than 100 tornadoes on Dec. 15.

One standout feature of this period was unusually warm water in the Gulf of Mexico, which helped generate ample volumes of moist, unstable air to be drawn northward ahead of both storm systems.

“Preceding these events, at least for the previous 15 days, we did not have a cold front that scoured the Gulf, so it was open for business,” Hoogewind noted in an American Meteorological Society talk.

Figure 3. Damage to a factory in Dawson Springs, Kentucky, from an EF4 tornado on Dec. 10, 2021. (Image credit: Chris Conley, via NWS/Paducah, KY)

Another key factor was a series of intense cyclones in the North Pacific, including Super Typhoon Nyatoh. These events helped generate persistent upper disturbances called Rossby waves that propagated downstream toward the United States. By the start of December ­— a full 10 to 15 days ahead of the tornado outbreaks — forecasts from the GFS model ensemble (GEFS) began to depict a pattern change that would support U.S. severe weather, Hoogewind said.

In an interview, Brooks, Hoogewind, and colleague Adam Clark waxed optimistic about the potential value of severe weather outlooks in the two-week time range. Brooks brought up the horrific “super outbreak” — the most prolific in U.S. records — of late April 2011. A two-week span checkered by major tornado activity segued into a swarm of 360 twisters from April 25 to 28, 2011, that took at least 324 lives and left a record $10.2 billion in damage.

As Brooks put it, “How far in advance could we have said: This upcoming two-week period is going to be unlike anything you’ve seen in your life?”

According to Clark, “We don’t know what the time scale of predictability is for a lot of severe weather, but we think we can make some progress in the extended range, looking beyond week one and into week two. We view it as low-hanging fruit.”

Oddly enough, multi-week tornado outlooks might be the most useful in wintertime – a period when tornadoes are less frequent but more often fast-moving and night-striking, often catching people unaware.

Figure 4. U.S. tornadoes reported each winter (December-February) since 1950-51. A record-high total of 280 tornadoes was observed in 2022-23. (Image credit: NOAA/NCEI, based on SPC data)

Wintertime U.S. tornado reports (see Figure 4) haven’t shown a robust long-term increase beyond the overall rise in reports in the late 20th century that’s been attributed to more careful observing. However, some work has found a significant increase in tornado-supportive air masses during winter in the Southeast. And the high-profile, far-flung winter outbreaks of recent years at least raise the possibility that a warming climate will allow tornadic storms to extend farther north in winter than people are accustomed to. The twisters of Dec. 15, 2022, ranged as far as southern Minnesota, an unprecedented northward extent for December.

Gensini is another who sees promise in extended-range tornado prediction across a multi-week time frame. From 2015 to 2020, when funding ended, Gensini and colleagues put together weekly extended-range tornado activity forecasts for overall U.S. tornado activity (below, near, above, or much-above average) in the first, second, and third weeks after the forecast date.

The team paid close attention to upper-level wave trains in the Pacific. One factor of keen interest is the Madden-Julian Oscillation, or MJO, a phenomenon that produces recurrent, sprawling disturbances that straddle the equator and move slowly eastward around the globe. When a strong MJO disturbance enters the western and central Pacific, it appears to signal an increased chance of tornado-favorable U.S. weather several weeks later. “High-skill forecasts have a stronger MJO signal,” Gensini said in an overview talk at the American Meteorological Society meeting.

A powerful MJO disturbance — one of the strongest on record, in fact — was entering the western Pacific in early March 2023. Along with jump-starting the likely transition to El Niño, this MJO pattern could also enhance the U.S. severe-weather risk toward April.

Just as concerning is the prevalence of very warm sea surface temperatures in and around the Gulf of Mexico after a winter virtually free of cold fronts.

Figure 5. As of March 11, 2013, sea surface temperatures across the Gulf of Mexico were running 1-3 degrees Celsius (2 to 5 degrees Fahrenheit) above the seasonal average. (Image credit:

Bringing long-range severe weather into the operations realm

Gensini hopes to see more funding and attention put into developing operational products for the multiweek time frame, and perhaps a workshop or working group to look at the manifold issues surrounding such outlooks — including who would develop and issue them and how academia could be involved.

The NOAA/NWS Climate Prediction Center now has an experimental online tool that predicts the odds of regional temperature, precipitation, and wind extremes out to two weeks in advance. Localized severe weather is not highlighted, though.

A two-week severe weather outlook would fall between the cracks of current Storm Prediction Center and Climate Prediction Center products, and it’s not yet clear how such a product might emerge within NOAA or where it would be housed.

Figure 6. Outlook for winds exceeding 40 mph on Days 12-14, as projected on March 12, 2023, by NOAA’s experimental Week-2 Global Probabilistic Extremes Forecast Tool. (Image credit: NOAA/NWS/CPC)

Other questions are looming. “What should we be forecasting? Counts of tornadoes? Days with hail? Number of flash floods?” Gensini asked.

Hail, in fact, is of keener interest than tornadoes to insurers. Annual insured losses from hail averaged $8-14 billion from 2000 to 2019, according to Aon. That year-to-year average far exceeds all but the very worst tornado years.

“For the private insurance market, major hurricane events remain the primary concern, given the enormous cost and impact that such events can incur,” said meteorologist Steve Bowen, chief science officer at Gallagher Re, a reinsurance broker. “Seasonal tornado outlooks are of interest, but would be more effective if taken in the context of severe convective storm risk as a whole.” He added that hail remains the dominant driver of severe convective storm losses.

The pioneering hurricane outlooks from Colorado State were issued for years before public and business interest in the outlooks congealed and competitors mushroomed. Said Brooks: “If there was an obvious group of people that had a real interest in these forecasts that were clamoring for us to provide them, that would make it a whole lot easier.”

Jeff Masters contributed to this post.

Website visitors can comment on “Eye on the Storm” posts. Comments are generally open for 30 days from date posted. Sign up to receive email announcements of new postings here. Twitter: @DrJeffMasters and @bhensonweather

Source link