News of the WeekWEATHER

Forecasters Learning to Read a Hurricane's Mind

See allHide authors and affiliations

Science  23 Apr 1999:
Vol. 284, Issue 5414, pp. 563-565
DOI: 10.1126/science.284.5414.563

Hurricane forecasting has come a long way since one sneaked up unannounced on Galveston Island, Texas, in 1900 and killed 8000 people. Nowadays, meteorologists know when a storm is on its way, but predicting just where it will hit land still isn't easy. For most of the past half-century, forecasters have struggled to narrow their predictions of a hurricane's next move, but as recently as the 1970s, guesses of a hurricane's position 24 hours ahead of time were off by an average of more than 200 kilometers. Now hurricane researchers finally have something to celebrate.

“It's been a pretty exciting 5 years,” says hurricane specialist Russell Elsberry of the Naval Postgraduate School in Monterey, California. Better observations of the streams of winds that carry hurricanes toward land are feeding new computer models for predicting how those winds will shift. And, as recent analyses—including one in last month's Bulletin of the American Meteorological Society—show, these new tools are getting results. “It's quite clear that the [U.S.] National Hurricane Center has been making much improved track forecasts” of future storm movement, says Elsberry. The new forecasting skill means that crowded coasts will have more time to prepare for storms, and warnings can be limited to smaller sections of coast, saving millions of dollars on unnecessary evacuations.

Hurricane forecasting has spent a long time in the doldrums. In the 35 years after record keeping was begun in 1954, forecasts of a storm's position 24 hours in the future improved by only about 1 kilometer per year, even after satellite images made it easier to track the position, winds, and extent of a hurricane. One problem was that neither satellite images nor the scattered data from weather buoys and ships offered many clues about the stream of air surrounding a storm, which determines its speed and direction.

“There is no substitute for in situ observations,” says meteorologist Kerry Emanuel of the Massachusetts Institute of Technology. For 15 years, researchers had been collecting those observations by flying aircraft near the storms and releasing instrumented packages called dropwindsondes—a sort of weather balloon in reverse that radios back wind speed and direction, temperature, pressure, and humidity as it falls. But those efforts were sporadic until 1997, when the National Weather Service (NWS) made such observations routine and introduced a new dropwindsonde that tracks itself using the satellite-based Global Positioning System, allowing more precise wind mapping. The NWS also acquired a Gulfstream-IV jet, which could fly higher and faster around storms than the traditional hurricane-hunter aircraft, probing more of the nearby atmosphere.

In the March Bulletin of the American Meteorological Society, Sim Aberson and James Franklin of the National Oceanic and Atmospheric Administration's (NOAA's) Hurricane Research Division in Miami, Florida, describe the payoff: The 1997 dropwindsonde observations improved storm-track forecasts by 31% 24 hours ahead, by 32% at 36 hours, and by 12% at 48 hours, they report, compared to computer forecasts made without the observations. The tropics were relatively quiet in 1997, prompting just five missions by the Gulfstream-IV, so “you don't want to make too much of the numbers,” says Franklin. Still, he says, “we're fairly confident '98 will be like '97.”

Along with better data, forecasters have better tools for interpreting the information. Their primary aid is computer modeling that incorporates the latest observations to create a picture of the storm and its surroundings and calculates how the storm will move and develop. “There has been a quantum increase in the skill of the models,” says Stephen Lord, a deputy director at the NWS's National Centers for Environmental Prediction in Camp Springs, Maryland.

The prime example has been the hurricane model developed by Yoshio Kurihara, Morris Bender, and Robert Tuleya of NOAA's Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, New Jersey. The GFDL model works on two scales. Like standard global atmospheric models, it simulates the atmosphere in broad strokes to capture the river of air, thousands of kilometers across, that sets the hurricane's overall course. But it also zooms in on the hurricane's vortex, using the latest satellite and in situ data to model the storm and the way it interacts with its surroundings in fine detail.

In tests prior to becoming operational at the National Hurricane Center (NHC) in 1995, the GFDL model outperformed its predecessor, logging average track errors that were about 12%, 24%, and 28% better at 24, 48, and 72 hours, respectively. Since then, “it's been the best performer” of the half-dozen models that NHC forecasters consult before issuing an official forecast, according to James Gross of the NHC.

Even so, it can be hard to tell whether better data and models are actually improving the official forecasts, because the improved tools are new and forecasters have always had good seasons and bad, depending on the nature of the storms. But meteorologist Colin McAdie of the NHC thinks track forecasts are improving at an accelerating pace. His recent analysis shows that at all forecast times, the predictions improved twice as fast during 1992 to '96, the period when the GFDL model debuted, as they had during the previous 2 decades. The routine dropwindsonde observations that began in 1997 seem to have helped sustain that progress.

Such improvements should allow the NWS to target its hurricane warnings more precisely. When the weather service issues a hurricane warning, prompting an evacuation, it generally includes a stretch of coast three times longer than the section that eventually suffers high winds, just to be sure—which means that hundreds of kilometers are cleared but suffer little damage. With costs averaging half a million dollars per kilometer of evacuated coast, according to the NWS, not to mention a toll in public goodwill, that's an expensive insurance policy. If the improvements of the '90s can be continued, averting hurricane disasters should be cheaper and less disruptive.

Navigate This Article