On Feb. 19, the National Weather Service's four- to five-day outlook called for storms over Louisiana that weekend. But three days of intensive research flights over the Pacific indicated that the forecast models were underestimating the strength of the jet stream, a high-altitude, high-speed river of air that steers storm systems.
When the new information was fed into computers, they generated a track for the storm that took it deeper into the Gulf of Mexico, where it picked up energy. And the projection placed the storm that weekend not over Louisiana, but over central Florida.
"To me, that's a miracle," says Melvyn Shapiro, a senior research meteorologist at the National Weather Service Environmental Technology Laboratory in Boulder, Colo., who took part in the flights. The results helped NWS forecasters focus more quickly on a system that would generate seven destructive tornadoes in the early hours of Feb. 23.
"There has been no more active period of change [in research] than the one we see now," acknowledges Frederick Zbar, a meteorologist in the office of systems development at National Weather Service headquarters in Silver Spring, Md.
For nearly five weeks prior to the outbreak of tornadoes in Florida, in January and February, a pair of US Air Force C-130s and a small jet attached to the National Oceanic and Atmospheric Administration (NOAA) had been dropping weather probes into storm systems wheeling across the North Pacific.
The goal of the six-week test was to see if the extra information would help the Naval Research Laboratory in Monterey, Calif., and the National Centers for Environmental Prediction in Camp Springs, Md., improve the accuracy of their computer-generated weather outlooks.
The flights over the Pacific, which ended Feb. 28, were part of a larger effort to design the continent's next-generation forecasting system, one that would take full advantage of rapid changes in technology - from ground- and space-based sensors that monitor current weather conditions to the computer programs designed to turn those observations into forecasts.
Field-research projects in recent years sound like folders drawn from a NWS version of the "X Files": NORPEX, VORTEX, FASTEX, and over the next two years, TIMEX.
Some, like NORPEX, which involved the research flights over the North Pacific, are experiments to see how operational forecasts can be improved. Others, like TIMEX, which will focus on the atmospheric processes that lead to the formation of thunderstorms, are expected to improve the models used for forecasting severe weather conditions and sharpen forecasters' ability to interpret the models.
AS supercomputers have improved, so has the output from forecast models, which programmers are designing to forecast smaller-scale changes in atmospheric conditions over shorter time periods.
Improved sensor technologies are feeding the models' voracious appetites for information. Twice-daily weather-balloon launches from 70 locations around the country are being supplemented by 40,000 readings a day taken by commercial aircraft nationwide as they take off, cruise, and land.
Researchers are finding ways to tease additional information from satellite and radar images, which can be rapidly sampled.
But for all the severe-weather forecasting successes, such as Florida's tornado outbreak, the service still trips up over such seemingly mundane features as rainfall rates.
"Precipitation has been a hard one for us," says Alexander MacDonald, director of the NWS Forecast Systems Laboratory in Boulder. "It's smaller in scale, unstable, and so much harder for forecasters to hit. Essentially it's like the small fish that escapes the net with a wide mesh."
And as the lead time on severe-weather forecasts improves, some are concerned about the public's response to watches and warnings.
As events progressed on the Friday and Saturday before central Florida's tornado outbreak, forecasters in Florida and at the Storm Prediction Center in Norman, Okla., could see that by Sunday, conditions would be ripe for violent, and possibly tornadic, thunderstorms.
"We issued our first tornado watch on Sunday at 2:15 p.m. to last until 9 p.m.," recalls David Ivy, chief of the center's operations branch. When 9 p.m. rolled around, "the forecaster on duty here was very concerned about the fact that not much had happened.
"There was some discussion about extending the watch, or canceling it and waiting to see what happened next."
If the watch was canceled, forecasters were worried that people would go to bed thinking the danger was over, when it wasn't. But based on the day's lack of tornadoes, people might just ignore the extended watch.
In the end, Mr. Ivy says, forecasters extended the watch until 3 a.m. the following morning. The first tornado warning was issued by the Tampa forecast office at 9:12 p.m.