It’s late August, which means it’s the peak of hurricane season. Not surprisingly, weather models have begun to key in on a particularly interesting area of disturbed weather near the Eastern Caribbean islands that shows potential for growing into a tropical storm or hurricane that may track toward the U.S. coast. If it gets a name, it’ll be called Hermine. (The National Hurricane Center has unfortunately turned down a request to rename it Harambe.)
Behind the scenes, meteorologists have sprung into action, downing extra cups of coffee, sending cryptic tweets and encouraging caution. The National Oceanic and Atmospheric Administration, the parent organization of the National Weather Service, has dispatched an Air Force Reserve “hurricane hunter” aircraft to fly through the storm and gather additional data.
But this area of disturbed weather — officially designated “invest 99L,” as in, “we should investigate it further” — is at the edge of predictability when it comes to forecasting what its location and strength will be in about six days, when some weather models show that it could be approaching Florida. (Errors for predicted five-day paths issued by the National Hurricane Center, which uses weather models as guidance, averaged 222 miles in 2011-15.)
So, how do you responsibly talk about a hurricane that could be potentially devastating when the probabilities are still unclear, particularly when it might be the first to make landfall in Florida in a decade?
During times such as this, meteorologists have increasingly relied on an approach not unlike that of FiveThirtyEight’s election forecasts: Take all available data, weight it by what has historically been the most reliable and rely on the average to guide you. Meteorologists call this approach “ensemble forecasting.”
Problem is, when you do that on a storm that’s still about six days from a potential U.S. landfall, this is what you get:
Yep, a big mess.
Since meteorologists realize ours is an imperfect science, these so-called “spaghetti plots” are designed to quantify the uncertainty of a forecast at any given moment. They’re produced by running the same models multiple times with the initial conditions slightly changed, which tests the sensitivity of the forecast to the inherent error in how much we know about what the atmosphere is doing right now. The above plot shows the range of possible future paths of the storm, but there are similar uncertainty plots for the storm’s strength forecast, too.
A single weather model showing a strong hurricane hitting Florida and then the Gulf Coast, just like a national poll showing Donald Trump in the lead, will get your attention — and maybe even make the news. For the past three model cycles, since Monday afternoon, the European Center for Medium-Range Weather Forecasts — which many meteorologists consider the most reliable weather model for hurricane forecasting, had a result like this for 99L. But until other models start trending in that direction — and more importantly, until those spaghetti strands start bunching together — we have very low confidence in an extreme forecast, even one from the ECMWF.
Sure, all the ingredients are here for this area of disturbed weather to become a memorable storm. Water temperatures on its projected path are near record highs, a potential fuel source that could aid in rapid intensification. There’s also a strong blocking ridge of high pressure that is forecast to set up over the mid-Atlantic region and could help steer this storm, if it forms, toward the U.S. Those are important details that meteorologists can add to the messy weather model guidance. But ensemble forecasting helps meteorologists avoid focusing too intensely on the extreme possibilities and take a step back to gain perspective.
This uncertainty is difficult to navigate because lives are on the line, of course. On the one hand, you don’t want to draw too much attention to an extreme result for fear of crying wolf, and on the other hand, you don’t want to miss an opportunity to give residents valuable days of preparation time.
I always try to err on the side of caution by describing “reasonable worst-case scenarios.” Right now, for this storm, a hurricane striking the mainland U.S. (or any land) is a reasonable worst-case scenario. But most likely, this storm may not get its act together at all. That’s the sort of information you can gather by looking at many models (or polls) at once, not just focusing on the scariest ones.
Some additional context for this specific storm is that no matter what the current forecasts show, there’s a looming concern when you consider more than a century’s worth of weather records in Florida. Between the 1950s and Florida’s last hurricane strike in 2005, the state averaged one landfall about every other year. (That’s not to say that the atmosphere subscribes to a set schedule. Hurricanes can come in bunches, or with long breaks in between.)
Amid what by some measures has been a historically inactive stretch for hurricane landfalls in the U.S., the mere threat of a potential hurricane heading toward Florida is relatively big news in the weather world. In May, meteorologist and hurricane specialist Philip Klotzbach tried to quantify just how unusual Florida’s current lucky streak is:
Since that tweet, another hurricane — Earl — came and went. That’s now 67 in a row that have missed Florida.
As Matt Lanza wrote for FiveThirtyEight last year, there’s been a huge increase in vulnerability on our coastlines over the past decade. According to U.S. census data, more than 2 million more people live in Florida today than during the last strike: Hurricane Wilma on Oct. 24, 2005. More than half a million of those moved to the Miami area. Throw in the steady unrelenting pressure of sea level rise, and the Miami area could be facing real trouble.
A major hurricane hitting the U.S. could produce a lasting catastrophe. When predicting such a thing, you want to make sure you’re very likely right.