A rain-wrapped tornado struck the small town of Elk City, Okla., on 16 May and damaged more than 20 homes and businesses. Thanks to an experimental storm prediction model from the National Severe Storms Laboratory (NSSL) in Norman, Okla., residents received more than twice the usual warning time to seek shelter.
Pamela Heinselman, project manager of the prediction model called Warn-on-Forecast, told Eos that the National Weather Service (NWS) forecasters in Norman saw the test as “significant in terms of being able to put out that advisory and then have confidence to issue a warning…earlier than they would have been able to do otherwise.”
Warn-on-Forecast predicted the 16 May supercell, a type of strong, rotating thunderstorm that commonly precedes tornadoes, hours before it formed in the Texas panhandle. As the program processed real-time weather data, it reported a 90% probability that extreme wind shear and updrafts would form across a swath of western Oklahoma and reach a peak over Elk City.
NWS’s parent agency, the National Oceanic and Atmospheric Administration (NOAA), publicly announced the test of the new model on 13 July.
Turning Predictions into Tornado Warnings
Warn-on-Forecast combines a large number of high-resolution atmospheric convection models with real-time radar observations to generate probability-based forecasting maps. The comparison of modeling results with the ever-changing radar information yields maps that predict the formation and path of severe weather more precisely and with more lead time than current forecasting models allow.
Before the Elk City tornado, the software “indicated the possibility of intense supercell thunderstorms capable of producing tornadoes moving from the Texas panhandle into western Oklahoma over the course of [a] 3-hour period,” said Patrick Skinner, a research scientist with NOAA and the University of Oklahoma Cooperative Institute for Mesoscale Meteorological Studies in Norman.
Researchers from NOAA’s NWS and NSSL working out of the NWS Forecast Office in Norman issued a tornado watch for 33 at-risk counties 90 minutes before the storm. A tornado watch means that tornadoes are possible but have not yet been spotted. Local emergency management then activated outdoor warning sirens 30 minutes before the tornado touched down in Elk City.
Currently, NWS issues a severe weather warning when meteorologists first see a supercell on radar or receive reports of local sightings. This “warn on detection” approach gives the public an average of 13 minutes of advance notice before a twister hits.
Local news outlets later reported that the tornado damaged between 20 and 30 homes and businesses in Elk City, injured multiple people, and caused one fatality. NWS categorized the tornado as an EF2 on the Enhanced Fujita Scale, with wind gusts faster than 179 kilometers per hour (111 miles per hour).
Although Warn-on-Forecast’s current 30-minute warning time is a significant improvement, Heinselman explained that her team aims to increase this time to an hour to reduce fatalities and injuries and give high-density locations like hospitals and stadiums enough warning to evacuate.
This successful test of Warn-on-Forecast was part of the 2017 Spring Forecasting Experiment at NOAA’s Hazardous Weather Testbed. Warn-on-Forecast was a first-time participant in this annual NOAA event, a modeling-focused experiment coordinated between the Storm Prediction Center and NSSL.
According to Heinselman, experiments like these are critical for understanding the strengths and weaknesses of the forecasting program and for improving the current version of the model. Warn-on-Forecast is still in the experimental stages of development and is not expected to be fully operational for at least 6 to 8 years, she said.
Because the Elk City forecast was the first attempt by NWS forecasters to use Warn-on-Forecast for real-world predictions and storm warnings, the team does not yet know whether the accuracy of the prediction typifies the software’s capabilities or was just a one-time success. The researchers expect to continue testing with NWS. These tests should enable the researchers to “collect more cases with the model, which will allow us to more thoroughly evaluate its performance,” added Heinselman.
Warn-on-Forecast and a related program, Forecasting a Continuum of Environmental Threats, aim to deliver detailed storm predictions across a “threat grid” that updates on the basis of real-time data, Henselman said. Other currently in use predictive models issue a “yes or no” storm prediction for large areas and can have a high level of uncertainty. “From probabilistic-type guidance you can give a more specific area of impact associated with different threats,” she explained. “The goal is to be able to reduce the [geographic] size of that warning based on, in part, model forecasts.”
—Kimberly M. S. Cartier (@AstroKimCartier), News Writing and Production Intern