The forecast for Hurricane Irma, one of the strongest tropical cyclones ever recorded in the Atlantic Ocean, is bleak. According to guidance issued Wednesday morning by the National Hurricane Center, the storm will turn north in a day or two, slicing through the Florida Keys and into the center of that state’s peninsula.

Key to these forecasts are anticipations of the storm’s track: where its eye will move over the next five days. Storm track is so critical to hurricane forecasts that people often call it the hurricane forecast, even though a full hurricane forecast will also project a storm’s future pressure, temperature, wind speed, and storm surge.

Yet a storm’s track—combined with its strength—may be the most important, since it determines whether a certain city or metropolitan area gets evacuated. It’s also where hurricane meteorology has most clearly improved since the 1980s.

There’s an irony, of sorts, in how a tropical cyclone tracks across the ocean. Hurricanes—some of the largest, most fearful storms produced by the Earth system—are blown around the planet by much weaker storms. The standard warm and cold fronts of the middle latitudes—which meteorologists call baroclinic storms or extratropical cyclones—determine where a massive cyclone makes landfall.

Why is that? Baroclinic storms often have hot and cold air at the same altitude. (Readers from Colorado to Connecticut know that after a thunderstorm pushes through, it can sometimes cool off a warm day.) This difference in temperature leads to forward motion, pushing a storm through the atmosphere and driving weather across the planet.

“With a tropical cyclone, its very symmetric. The temperature differences driving it are vertical—there’s warm, wet air below, and cool, dry air up above—so there’s a vertical circulation, but it doesn’t move in a horizontal direction,” said Neal Dorst, a research meteorologist at NOAA’s Hurricane Research Division.

“There’s no impetus driving a tropical cyclone—it will have a tendency to stay where it is,” he said. If a hurricane appeared on a globe with no other weather systems, it would drift northerly only at a couple miles per hour, thanks to the Coriolis effect. It would not move in any other way, he told me.

The National Weather Service runs multiple weather models for each new forecast, which are issued four times a day.

Three different technological breakthroughs are key to the improved storm-track models, according to Kerry Emanuel, a professor of atmospheric science at the Massachusetts Institute of Technology. First, scientists now observe much more of a storm than they used to, through both constant satellite surveillance and data-rich aircraft flights through individual storms.

Second, meteorologists have improved their simulations of atmospheric processes, and they have better computers to resolve the math required by them. Thirty years ago, a high-quality atmospheric model might have had grid points 150 kilometers apart; now, they are five kilometers apart, he told me.

Finally—and most importantly, according to Emanuel—scientists are much better at “initializing” forecast models; that is, models now start their forecasts with a vision of reality that’s much closer to the real world. These initial states often start from a six- or 12-hour-old forecast, which the computer model then brings to a current state by assimilating new observations.

“The fact is that observations are few and far between—we’re certainly not measuring every molecule of the atmosphere,” said Emanuel.

About half of the computing power thrown at any hurricane forecast is spent entirely on describing this initial, current state, he said. The other half is spent projecting the future.

Yet those model runs alone do not determine what shows up in a broadcast or on a news site. (Unless someone has improperly tweeted a spaghetti plot.) Ultimately, the team at the National Hurricane Center might sit down with hundreds of simulated storm-track runs. Major models like the European or the U.S.-operated Global Forecast System also prepare dozens of “ensemble” runs, where they adjust small aspects of the weather’s initial state.

Then an experienced human being sits down and looks at them all.

“The official track is—you have a human being sitting there, weighing runs, usually mentally, with which ones he trusts the most and which ones agree,” Dorst told me. “You always want to have a human being somewhere in the loop when you’re issuing forecasts saying, this is where [the hurricane] will go.”

That researcher will draw the anticipated-track line, based on the model runs and their own experience. Then they add the cone of uncertainty around the anticipated track. This cone doesn’t have anything to do with what the models say: It’s an average of how forecasters have gotten other tropical cyclones wrong over the past five years. “That’s why it spreads over time,” says Dorst. “In 12 hours, they might have a 20-mile error, and at 24 hours, a 40-mile error.”

Those errors have gotten smaller over time when it comes to storm track. But predicting a hurricane’s intensity is a less certain matter.

“Although the track forecasts have steadily improved over the last 30 years, there’s been no tangible improvement in our ability to forecast how strong a storm will be,” Emanuel told me.

“The forecaster’s worst nightmare is: There’s a storm in the Gulf that’s 12 or 24 hours from landfall, and you go to bed at night, and it’s a tropical storm. Then you wake up the next day and it’s a Category 3. By then there’s just no time to get people out of harm’s way,” he said.

That is more or less what happened when Hurricane Audrey, which strengthened from a tropical depression to a Category 3 cyclone over 24 hours, struck northeastern Texas and western Louisiana in 1957. Because it intensified so close to shore, few evacuated from its path. Audrey ultimately killed more than 400 people, making it one of the deadliest hurricanes ever to strike the United States.

“Even though weather forecasting has improved since then, I think even if you talk to a hurricane forecaster today, they’ll tell you this is what keeps them up at night,” says Emanuel.

Earlier this year, he found in a paper in the Bulletin of the American Meteorological Society that climate change will make this kind of rapid intensification more likely. “There’s a lot more cases of rapid intensification [in a warmed climate], and that includes the accident of it rapidly intensifying just before it makes landfall,” he said.

At the same time, it’s “immoral” to order an evacuation for any tropical storm that could strengthen, he said—since people often die in car accidents during mass evacuations.

Hurricane Harvey exemplifies the difficulty of examining both of these symptoms together. The recent hurricane intensified up until the moment it made landfall, setting it apart from all hurricanes in the Gulf of Mexico observed in the modern-day period. But its track was what really made it devastating: Had it not stalled out above Houston for several days, it would have been a major disaster, but not a multibillion-dollar catastrophe.