Note for future hurricane prep: Pay attention to probabilities.
Tropical Storm Erika, the storm that churned across the Atlantic for five days last week vexing forecasters and leading to a state of emergency in Florida, was a classic example of probability theory for weather scientists. Under one set of conditions, the storm would likely do one thing. Under another, something entirely different. But nailing down those conditions — wind shear, topography, sea temperatures, atmospheric conditions, even the movement of thunderstorms within Erika — proved difficult and ultimately produced a cone of uncertainty that was as uncertain as the storm itself.
“Both track and intensity had higher errors than normal, so the forecasts were not as accurate as we would have liked,” James Franklin, who oversees forecasters at the National Hurricane Center, said Tuesday.
Blame computer modeling. The massive data-crunching computers have become increasingly accurate at forecasting a storm’s track, but not its intensity. So if a storm — like Erika — is weak and not tall enough to be steered by atmospheric conditions calculated in global models, then heads into warm, wet conditions that could strengthen it, teasing out a forecast that accurately calculates both track and intensity can be difficult.
“They needed better data getting into [the models]. If you don’t have much data to analyze in your model, you’re not going to have good forecasts,” said Weather Underground meteorologist Jeff Masters.
In 2009, the National Oceanic and Atmospheric Administration launched a project designed to cut errors in half over 10 years, Franklin said. Last year, forecasters had their best year ever. But after five years of steady funding and improvements, the project’s budget was cut this year by two-thirds to $5 million, he said, which could potentially slow progress on improving intensity projections.
When Erika first appeared late Aug. 24 nearly a thousand miles east of the Leeward Islands, forecasters warned that predicting the messy storm more than two days out would be tricky. The fast-moving tropical storm was becoming better defined, and moving west into an area with warm water, humid air and low wind shear, hinting it could power up.
To find out where it was going, forecasters turned to their dynamical models that look at conditions around the globe. Since the 1970s, the tracks have become more and more accurate, improving exponentially in recent years as more satellites have allowed forecasters to load more data into increasingly powerful computers. A former boss at the hurricane center compared forecasting track to following a cork in a stream, Franklin said. Follow the stream and you follow the cork.
“Although it may not seem that way after Erika, track is a relatively simple process,” Franklin said. “Mostly, the storm is being pushed around by surrounding air flow.”
But when it comes to intensity, Franklin said the hurricane center still relies in part on old statistical models that look at past records and then base models on how similar storms behaved in the past.
“There’s no physics, no equations that govern how the atmosphere works. It’s entirely empirical. You just look at what’s happened before and use that as a basis for prediction,” he said.
When scientists looked at past El Niño years — this year a powerful El Niño in the Pacific is fueling strong hurricane-slicing upper winds — they concluded that wind shear would keep Erika in check. And they said so in lengthy discussions attached to advisories, dense with references to subtropical ridges and marginal humidity. While they felt confident in their two-day forecasts, forecasters repeatedly warned that three-to-five-day forecasts can be off by as much as 240 miles.
But because they prefer to err on the side of safety, forecasters included projections for a hurricane near the Turks and Caicos Islands early Saturday and over the Bahamas on Sunday in their 5 a.m. forecast on Aug. 25, day two of the storm.
Twelve hours later, with the storm 500 miles east of Antigua, forecasters redrew the five-day cone to include a possible hurricane strike in Florida, although they continued to warn that the forecast had “considerable uncertainty” based on two possibilities: If Erika remained weak, it would tack to the south and west. If it intensified, it would head north and east.
Probability calculations, however, offered a different way of looking at Erika, giving Florida a mere five to 20 percent chance of having tropical storm force winds.
Even so, by Thursday, a day after heavy rains left 20 people dead on the small Caribbean island of Dominica, Miami-Dade County officials were meeting and preparing for a storm and a press conference to warn residents. On Friday morning, after models pointed the storm back toward Florida, Gov. Rick Scott declared a state of emergency. Forecasters continued to warn about uncertainty: They were still struggling to project Erika’s strength until they knew how the storm might recover after its pass over Hispaniola. Wind shear would also snuff out a weakened storm. They shifted the cone of uncertainty squarely over Florida, with the storm’s middle track marching up the center of the state. The probability map now put the odds of tropical storm force winds from Central Florida south at 30 to 50 percent.
Through the day, as the storm passed over Hispaniola, the cone continued to shift further west, changing its target to the west coast. By 9:30 a.m. Saturday, forecasters officially declared the storm kaput.
In the entire five-day life of Erika, the probability map that forecast hurricane strength winds never included Florida. While the cone forecast only became more accurate the closer the storm got, the probability map consistently reflected low risk. The highest threat came at 2 p.m Friday, when the odds of tropical storm force winds reaching Miami-Dade County peaked at just over 50 percent, then receded.
The lesson from Erika?
“We know weak storms often pose challenges and result in higher errors,” the NHC’s Franklin said. “And we did try from the very beginning to point out how uncertain day three, four and five forecasts were because of the wide range of outcomes. But when folks only look at the cone graphic and don’t look at some of the others things we’re putting out, then you can come away with the impression we didn’t do well with the storm.”