Harvey, Irma, Jose, Katia, and Maria. Over the past 30 days, some of the strongest hurricanes ever recorded in the Atlantic Ocean have prompted a diluvial flood, a mass migration, and an ongoing humanitarian crisis in Puerto Rico.

It can feel like there’s little precedent for so many hurricanes in so short a time. In fact, there isn’t—at least within living memory.

According to early estimates, September 2017 is already the most energetic month for hurricanes in the Atlantic Ocean ever recorded, even though the month is not yet over. The month’s accumulated cyclone-energy index of 155.4 breaks the previous record of 155.0, set in September 2004.

“September 2017 has also set the monthly record for major hurricane days and looks to probably set the record for hurricane days (and possibly named storm days) as well,” says Phil Klotzbach, a meteorologist at Colorado State University who researches the history of Atlantic hurricanes.

The month was so powerful that it has assured certain records for the 2017 Atlantic hurricane season as a whole, which is already among the 10 strongest seasons ever measured. Two Category 5 storms (with wind speeds exceeding 157 miles per hour) formed this year. Only a handful of years before this one have seen multiple Category 5 storms take shape: 1932, 1933, 1961, 2005, and 2007.

And only once before in history, in 2007, did multiple Category 5 storms make landfall in North America.

The new record verifies that something extremely unusual is happening in the Atlantic Ocean this year. It also sets the coming autumn in perspective: With a La Niña current budding in the Pacific, the final two months of the Atlantic season will likely see a higher-than-normal number of tropical cyclones. (Hurricanes normally stop forming around November 30.)

But the record also provides a glance into the messy nature of the Atlantic hurricane record. Over the past decade, our understanding of that record has changed—as has what researchers can safely say about the effects of human-caused climate change.

Accumulated cyclone energy, abbreviated ACE, measures the strength of a single hurricane over its lifetime. Because it’s a durational measure, it can also gauge the intensity of an individual month or an entire hurricane season. Meteorologists calculate ACE by taking a hurricane’s maximum wind speed every six hours, then summing the squares of those figures. They divide the final product by 10,000 to make it easier to use.

Hurricane records in some form go back more than a century. The U.S. Signal Service, a predecessor to the National Weather Service, began tracking Atlantic hurricanes in 1878. In that context, September 2017’s record can seem extraordinary: the strongest hurricane month in 139 years.

But climate scientists say that there are issues extrapolating hurricane history that far back, especially when measuring storms with ACE. Accumulated cyclone energy requires that meteorologists correctly observe the strongest winds in a cyclone every day. The National Weather Service does a much better job of that now, they say, than they did even a couple decades ago.

“The probability of us observing the true maximum of a storm now is much higher than it was 50 years, or 100 years, or even 20 years ago, honestly,” says Gabriel Vecchi, a professor of geoscience at Princeton University and a former climate modeler for the National Oceanic and Atmospheric Administration.

Why? In part, weather satellites have gotten better and more accurate. But meteorologists didn’t always have access to weather satellites. In fact, this exact problem—comparing hurricane intensity across 100-year time spans—is a microcosm of the current challenge facing climate scientists studying hurricanes. Over the past decade, it’s become clear that long-term hurricane data does not mesh together as seamlessly as some researchers had once hoped.

Before satellites, the National Weather Service (and its predecessors) pieced hurricanes together from a few types of data. The most significant of these were shipping records: observations taken by commercial or passenger vessels caught up in a storm. By slicing together multiple observations of the same storm, hurricanes emerge from the data.

The National Weather Service also monitors tropical cyclones through research flights and historical landfall accounts. It merges all of these different data—from ships, satellites, and land-bound weather stations—into a common set called HURDAT.

For many years, researchers knew HURDAT to have problems, especially further back in time. The National Hurricane Center suspected that the earliest part of the record missed between zero and five storms per year. But about 10 years ago, Vecchi and his colleagues at NOAA sought to model exactly how many tropical hurricanes it might be missing in the early 20th century.

They split HURDAT in two parts—to see if the technological past could detect the climatic present. First, they took the 40 years of hurricanes observed by satellites and made small, random changes to their paths, essentially creating a library of 650 years containing storms that did exist and storms that could have existed. Second, the team ran these 650 simulated years through each year of the observational network. They essentially asked: Would the ships and weather stations of 1878 detect storms across these 650 years? Would the ships and weather stations of 1879 detect storms across these years? And so on for all the years in the pre-satellite record.

They found that the pre-satellite record might have missed several storms per year, about as many as the National Hurricane Center estimated. It was most likely to miss storms before 1914, because—before the Panama Canal was opened—most ships stayed in the North Atlantic near Europe, and few traveled to the tropical Atlantic, where most hurricanes form. The record was also more likely to miss hurricanes during the two world wars.

A graphic from Vecchi and colleagues’s 2008 paper. Figure A shows every known tropical storm path from 1966 to 2006. Figures B, C, and D show the density of ship routes through the North Atlantic Ocean at different periods of the 20th century. Notice how many ships avoided the tropical Atlantic before the opening of the Panama Canal in 1914. (Vecchi, et al. / Journal of Climate)

Over the next few years, the NOAA team was able to show how disjointedly the data sets fit together. Vecchi and his colleagues showed that most of the rise in hurricanes over the past century was due to the rise of small, one- or two-day storms. These hurricanes were mostly observed during the satellite era. Once they were accounted for, then the century-scale rise in hurricane intensity went away.

This isn’t to say that there hasn’t been an increase in storms over the last century, only that you can’t find it in the data. At the same time, the satellite record shows a clear increase in hurricane activity since 1980.

“I think it’s unambiguous that over the last 30 years, hurricane activity in the Atlantic has increased. I don’t think there’s reason to doubt that,” Vecchi told me. “The question then becomes, how does this increase fit into the larger century-scale record? And then there’s a lot more room for reasonable people to disagree there.”

The problems in long-term hurricane data have left hurricane researchers with a dilemma.

“Our theory, our understanding of how the atmosphere works—which is based on decades of research, thousands of studies by hundreds of people—indicates to us that as the planet continues to warm from greenhouse gases, we expect that the strongest storms on the planet should get stronger,” Vecchi told me. “That is what our theoretical expectation is, and it’s rooted in a lot of research and experience.”

“At the same time, looking back, we know our data is problematic, and that means that we can’t see that signal in the data. And that’s a troublesome situation to be in,” he said.

How long will it take to see a signal in the data? Vecchi and his colleagues have estimated that a human-caused climate-change signal should appear in the observational data by the middle of the century. And of course, the theory tells them what they should already expect.

All of these data problems point to why it’s difficult to compare hurricane records—especially durational records like ACE—from the early 1900s to those that follow the Carter administration. While the issues of merging disparate data sets isn’t limited to hurricane data, it is more acute there, because tropical cyclones are singular events that form and often die in the open ocean. Temperature records, for instance, are considered much more robust.

While September 2017 has broken records, it has not shot completely off the chart, only slightly exceeding the last strongest-ever month. But there are certain ways that the 2017 season really is anomalous, Klotzbach said. As late as the first of the month, 2017 seemed like an average hurricane season. But hurricane activity rapidly crescendoed after that, diverging from the historical norm.

2017 Hurricane Season ACE for the North Atlantic Ocean, Compared With the 1981-2010 Average

Phil Klotzbach

Vecchi said that the problems facing hurricane climatology do not make its findings any less stark.

“People say, the data in the past isn’t good, so we don’t know anything,” he told me. “That’s not true. We have our theory. We know the physics. We make forecasts of storms that we’ve never seen before—the forecast of Hurricane Sandy, for instance, was accurate many days out. So we can’t discount the theory going forward just because we know the data going backward are problematic.”

“It’s just that now we have to make decisions in this strange world,” he said.