Meteorologists running an experimental forecasting model, powered by the computing muscle of the Pittsburgh Supercomputing Center, this spring were able to predict thunderstorms in unprecedented detail 24 hours in advance.

	Martha Rial, Post-Gazette Dark clouds hover over Downtown yesterday, as a fast-moving line of thunderstorms passed through the region. Click photo for larger image. Related coverage Strong storms flood streets, cut electricity Hurricane season off to busy start Get updated forecasts on post-gazette.com's Weather page

In some cases, the forecast model would predict storms to within 20 miles of their actual location and to within 30 minutes of their actual time.

Just as important, the computerized forecasts produced images "that looked very similar to what we see on radar," said Steven Weiss, science operations officer at the National Weather Service's Storm Prediction Center in Norman, Okla.

Seeing the structure of the predicted storm, he explained, is important in determining whether a storm is likely to produce tornadoes, hail or dangerous winds.

"It was an eye opener in many respects," he said of the experiment.

The forecasts were produced by the Center for Analysis and Prediction of Storms at the University of Oklahoma, which has been working with the Pittsburgh Supercomputing Center for almost two decades.

The modeling is not yet perfect and the experiment showed more work is needed to increase the amount of data gathered regarding atmospheric conditions, Weiss said.

"But we've raised the bar pretty darn high," he added. "Before long, people will say, 'Don't just tell me it will rain somewhere in Western Pennsylvania this afternoon, tell me which county and what time.' "

The experiment ran from mid-April through mid-June and covered two-thirds of the continental United States -- from the Rockies east to the Appalachians. It was part of the so-called Spring Program, run each year by the National Oceanic and Atmospheric Administration to coincide with tornado season.

This was an unusual tornado season. For the first time since the weather service began keeping records in 1950, no one was killed by a tornado from April through June.

"We didn't have any storm systems that were prolific tornado producers," Weiss said, saying the weather, not improved forecasts, played the biggest role in the lack of tornado deaths. "But we had ample severe storm activity," he added. "We were still able to see a great number of [forecast] successes."

Kelvin Droegemeier, director of the Center for Analysis and Prediction of Storms, said the LeMieux supercomputer in Pittsburgh allowed his team to run the forecast model at a level of detail five times greater than the National Weather Service's operational forecast model. That level of detail, in turn, required 300 times more computing power than needed to run the operational model.

Each evening, the researchers in Norman would transmit new atmospheric conditions to the LeMieux computer, housed in the Westinghouse Energy Center in Monroeville. By next morning, the computer would produce a 30-hour forecast, which was then sent to the Storm Prediction Center and the National Severe Storms Laboratory, also in Norman.

The numerical storms that grew inside the computer didn't always correspond exactly to the storms that developed in nature 24 hours later, Droegemeier said. But they were close enough -- and reflected enough of the actual structure of the real storms -- that the experiment may have changed some minds about how useful the advanced forecast models can be.

Droegemeier's group, as well as groups at the NOAA Environmental Modeling Center in Camp Springs, Md., and the National Center for Atmospheric Research in Boulder, Colo., all used the Weather Research and Forecasting Model, an advanced model that is being developed for both operational and research use.

The model can be run at various levels of detail, with predictions made for weather conditions at points on a grid with spacing varying from 10 kilometers to 1 kilometer. While the other groups operated with grid spacings of 4 to 4.5 kilometers, Droegemeier's group produced forecasts based on grid spacings of 2 kilometers, or every 1.2 miles.

That level of detail places great demands on computing resources, Weiss said, but also strains the weather service's ability to provide sufficiently detailed environmental conditions to plug into the computer programs.

For instance, weather balloons are launched twice a day from Pittsburgh to gather atmospheric information. But other weather balloons are launched no closer to Pittsburgh than Buffalo, N.Y., Wilmington, Ohio, and Washington, D.C., Weiss said. Weather satellites can provide some atmospheric information, "but there are large areas in between without sampling."

The weather service still must determine whether the advantage of more detailed storm modeling is worth the expense of higher-powered computers and more extensive gathering of atmospheric data, Weiss said.

LeMieux, the name given the supercomputing center's workhorse, has 3,000 processors and is capable of up to 6 trillion calculations each second; 1,228 processors were used each night for the forecasting work. The center is now installing a new Cray supercomputer, boasting 2,000 processors capable of up to 10 trillion calculations each second.