Hurricane experts predict better forecasts.

When Hurricane Emily approached the United States last August, most of the forecasting models used by the National Hurricane Center in Coral Gables, Fla., predicted the storm would plow squarely into the Carolinas. But an experimental forecasting model run in Princeton, N.J., projected that Emily would turn before reaching the mainland, striking only Cape Hatteras. That information helped hurricane center meteorologists accurately predict Emily's curving path.

With hurricane season officially open on June 1, the National Weather Service is working to incorporate that successful research model into its regular forecasting routine. Along with a string of other new tools, the experimental model promises more accurate forecasts of the greatest storms on Earth, says Robert C. Sheets, director of the hurricane center.

The National Weather Service typically runs a suite of different forecasting models on supercomputers at its National Meteorological Center in Suitland, Md. But the new model, developed at the National Oceanic and Atmospheric Administration's Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, N.J., most realistically represents atmospheric physics.

Last year, results from the GFDL model arrived late at the hurricane center because it took researchers 6 hours to run each simulation on their supercomputer in Princeton. At the National Meteorological Center, an improved version of the model is now running on a Cray C90 supercomputer that the center recently acquired.

"Using the faster computer and a more efficient model, we can now run it in 20 minutes," says GDFL's Robert E. Tuleya, who developed the model along with Morris Bender and Yoshio Kurihara.

As a measure of the model's forecasting ability, Tuleya compares how accurately it predicted the storm's track a day in advance. For Hurricane Emily, the GFDL model had an average error of 48 miles. The most accurate model in operation at the time missed Emily by 82 miles on average.

Hurricane forecasters also look to the GDFL model for help in anticipating the intensity of winds and the size of storms--two critical features that meteorologists cannot currently predict with skill.

Unlike other models, the GFDL model forecasts wind speeds around the storm, providing some measure of its size. "It's showing some skill. What we've seen is encouraging," says Jerry Jarrell, deputy director of the hurricane center.

Aside from modeling and computer improvements, hurricane forecasters have several new tools at their disposal. In April, the United States launched a much-needed geostationary weather satellite that should start providing routine images by late summer. The satellite will greatly enhance meteorologists' ability to track storms from their birth off the coast of Africa. The weather service is also updating its antiquated radar network system with new Doppler weather radars.

"In 2 years we think our 36-hour forecast will be as accurate as our 24-hour forecast was last year. So that's a marked improvement," says Sheets.

"Unfortunately," he adds, "population is increasing along the coast at a faster rate than our ability to forecast where the storm will go. We're fighting somewhat of a losing battle in that it's taking longer and longer for people to respond."