Skip to main content

Fine-tuning the science of hurricanes

University of Wisconsin scientists are working on a new model to improve forecast accuracy

University of Wisconsin scientists are working on a new model to improve forecast accuracy

When Tropical Storm Ernesto fizzled over Florida, it confounded forecasters and reminded those who had listened up and battened down that the tools for predicting hurricanes remain far from perfect.

Hurricane forecasts six to 12 hours out are very good, says Stefan Becker, a climatologist at the University of Wisconsin, Oshkosh, but beyond 48 hours the forecasts still are pretty iffy. “Last year, the 72-hour forecast was very bad,” says Becker, 41, who with UW-Oshkosh atmospheric scientist Marcus Buker is working on a new predictive model he hopes will improve the accuracy of hurricane and tropical storm forecasts. He expects their work to help reduce the margin of error in the longer three- to five-day forecasts especially.

As Ernesto formed south of Puerto Rico, forecasters thought it would make a U.S. landfall in Texas or Louisiana. The storm became a hurricane briefly in the Caribbean, then petered out as it crossed over western Haiti and Cuba’s mountainous east end. The National Hurricane Center thought it would rev up again between Cuba and Florida,fueled by 250 miles of open, 88-degree water, which energizes storms.

Floridians were advised to prepare for a strong 70-plus-mph tropical storm or an 80- to 85-mph Category 1 hurricane. But sustained winds never reached much more than 45 mph in Florida. After slogging up the state, however, Ernesto turned east into the Atlantic, weakened into a depression, then — as forecast — strengthened over warm Atlantic waters. It came ashore again north of Cape Fear, N.C., with torrential rains and 70-mph winds.

National Hurricane Center director Max Mayfield said in a press conference Aug. 30, the day after Ernesto made landfall in Florida, that he didn’t know why the storm didn’t hit the state with more force, except perhaps that it dithered over Cuba longer than expected and lost momentum. The NHC’s forecasts of storm path and intensity are a consensus of the forecasts of nine different models. Mayfield, who retires as director after this hurricane season, said the NHC’s tools for predicting a storm’s intensity still are pretty blunt. He wants funding to sharpen forecasting 30 percent over the next decade.

Becker believes the University of Wisconsin model can help the NHC accomplish that. He says forecast models usually are of two types: statistical, which forecast a storm’s behavior based on how storms at a particular location and time of year have behaved in the past, and dynamical, which forecasts behavior based on real-time ocean and atmospheric data.

“We’re taking a little bit different approach,” says Becker.

The model would use a computerized “neuro-network” to link the dynamical models — their atmospheric dynamics and storm track and intensity predictions — with a statistical database containing 56 years worth of information about Atlantic hurricanes and tropical storms.

“You’re taking everything you’ve learned in the last 56 years [about storm behavior] and applying it to this new tropical storm,” Becker says.

Becker and Buker completed preliminary trials using maybe 100 variables in their statistical model and were encouraged by the results. Working with the NHC and its computers, as well as scientists in Japan, Hong Kong and Hawaii, they hope to expand the number of variables in the statistical model into the thousands and refine it. The pair could have the model up and running in two years.

“We hope that by next year, we’ll be coming up with some forecasts that are more accurate,” Becker says. In the meantime Floridians, who have seen seven hurricanes in the last two years, aren’t really complaining that forecasters missed the mark this time.