American infrastructure is designed around a simple idea: We can predict how often the worst storms will come. Take the benchmark that undergirds the $1.2 trillion National Flood Insurance Program: the 100-year flood. That’s a flood that’s supposed to occur once a century, on average. A once-in-a-lifetime event. Similar estimates govern basic stormwater infrastructure: Sewers might be designed for a two-year rainstorm (likely to occur once every two years) or culverts for a 25-year rainstorm (once every quarter-century).
These probabilities are also used in modeling the absolute worst cases, such as California’s ARkStorm scenario—a 2010 U.S. Geological Survey project that imagined how a month of constant rain might turn the Central Valley into a giant lake, flooding 1 in 4 of the state’s buildings, forcing the evacuation of 1.5 million people, and causing more than three times as much damage as the better-known nightmare earthquake popularly known as “The Big One.” In addition to offering a potent biblical allusion, ARk stands for “atmospheric river 1,000,” because scientists originally thought those levels of precipitation would occur once every 500 to 1,000 years.
But other estimates suggest that such a storm might occur once every 100 to 200 years. And a more recent analysis concludes that climate change has caused the likely “recurrence interval” of an ARkStorm to creep closer to once each century—an ARcStorm. The last one was in 1862. “Climate change is dramatically upping the odds of a very high magnitude flood event happening in our lifetimes, specifically,” the authors wrote last August. This month’s record-breaking California rain, in other words, is a warmup. How did they come to that conclusion? By plugging more recent weather history into their simulation.
It’s older data, however, that determines much about the way we plan cities, price flood insurance, and build infrastructure, because historical records determine the crucial benchmark of the “100-year” storm and all its derivatives. The idea of evaluating storms and floods by their likely interval of recurrence was adopted in the 1960s to help administer the newly created National Flood Insurance Program, and has been complained about ever since.
The 100-year flood metric, wrote the geographer Rutherford Platt, is “unique in the annals of resource management policy for its durability over several decades … despite chronic griping and hand-wringing by three generations of floodplain managers.” In theory, Americans ought to want to prepare for the 100-year flood, which sounds frightening. But as Platt observed, if you were more likely to die than see one, why bother? Probabilities like this govern the design of everything from storm drains to levees to flood insurance, but they are probabilities from last century—and the storms are different now.
When Hurricane Harvey struck Houston in 2017, it was the city’s third 500-year flood in three years. Over four days, Harvey’s 30 to 40 inches of rain across the region made it a storm likely to occur just once every 3,000 to 20,000 years, at least according to the established metrics. Last summer, there were five “1,000-year” rain events in five weeks, in Dallas, eastern Kentucky, eastern Illinois, St. Louis, and Death Valley. California just broke a bunch of rainfall records this month, including in well-measured locations like Los Angeles and San Francisco.
The long-term trend is beyond dispute: Between 1958 and 2012, according to the 2014 National Climate Assessment, the amount of rain falling in the biggest storms grew by 37 percent in the Midwest, 27 percent in the South, and 71 percent in the Northeast. Do last century’s weather records still hold water? Some researchers have concluded they don’t. One study concluded that a 7.4-foot storm surge flood in New York City has progressed from a 500-year event before the industrial revolution to a 25-year event today, and will be a five-year occurrence in a few decades. Something similar is happening with heavy rain events.
Unfortunately, yesterday’s numbers are currently being used to plan tomorrow’s stormwater infrastructure, a burst of which will be funded by the Bipartisan Infrastructure Law that President Joe Biden signed in 2021. As NPR’s Lauren Sommer put it this month, “many cities aren’t constructing infrastructure to handle increasing amounts of water, because the rainfall records they use to design it are decades-old in most states.” Lacking money, expertise, or political will, they are preparing for a bygone world of weather.
The National Oceanic and Atmospheric Administration has been authorized to develop a more modern rainfall atlas, as well as one that takes climate change into account, but that work won’t be done in time for current projects. “There’s a timing disconnect for sure,” said Chad Berginnis, executive director of the Association of State Floodplain Managers. “No doubt this infrastructure is not going to have as much benefit because it’s authorized at the same time as this NOAA update. But I’m not going to let the perfect be the enemy of the good.” Better to build something, in other words.
Some cities are going a step further. After Harvey, Houston officials amended the land use code to apply identical standards to the 100-year and 500-year flood plains—even though one is supposedly five times less likely to flood. In Seattle, planners looked at a climate impact tool and decided to expand a new stormwater tunnel from 14 feet in diameter to 18 feet. After Madison, Wisconsin, flooded in 2018, the city decided to raise its design standards for new development—a culvert under a road, for example, now must be built for the 100-year event, instead of the 25-year event. But many smaller jurisdictions do not have the expertise, money, or political will to go beyond what the federal government is recommending.
“I don’t see how you can look at the information we have available now and not be asking those questions,” said Bill Haneberg, the state geologist of Kentucky. “We have the tools to start making realistic assessments of what the future is going to be like if we continue on our current trajectory of fossil fuel emissions.”
Even before carbon dioxide had soared above 400 parts per million in the atmosphere, there were issues with the 100-year-flood standard. How do you project a once-in-a-century calamity from a few decades of data? Haneberg made an analogy to the Super Bowl coin toss: Over the history of the event, it’s 26 heads and 29 tails. But there was one five-year stretch of heads, and there have been three four-year stretches of tails. If you wind up with even a normal statistical anomaly in your sample, you’re going to get some inaccurate results when you try to imagine 10,000 years of water.
“Right after the big floods in eastern Kentucky, I got the records from the U.S. Geological Survey,” Haneberg said. “People were talking about thousand-year floods; well I got those records from one stream gauge location, Whitesburg, and they had data only to 1957. So the question is: How do you determine that?”
Meteorologists counsel against looking too closely at such headlines as evidence that the climate has already gone haywire. For one thing, it’s a big country—there’s always a huge storm somewhere. For another, the number of broken records is a poor judge of a changing climate. For a third, we have more data than ever even across small geographic areas. “We weren’t slicing and dicing things so finely until recently,” cautions Bob Henson, a meteorologist and journalist with Yale Climate Connections. “It’s analogous to getting ever-more fine MRI scans and detecting phenomena you wouldn’t see before.” In Houston, for example, a 500-year flood on Cypress Creek in 2016 was just a 50-year flood a few miles downstream. A hundred years ago we might have had just one gauge there, if that.
But the consequences of predicting the future on a flawed set of records are not abstract. Last year marked the 100th anniversary of one of America’s most consequential forecasting mistakes. The Colorado River Compact of 1922, which set the course for a century of infrastructure, agriculture, and development in the western United States, predicted how much water the mighty river would send toward the Gulf of California based on a small, and very wet, sample of years. The river has since receded to the historical mean and then some, as the western United States has faced repeated severe droughts. States are bickering over their river-water allocations. Hydroelectric dams are running low on fuel.
“It’s kind of a mess, even in the absence of climate change,” said Daniel Wright, an associate professor at the University of Wisconsin. Wright helped Madison adopt climate-conscious design guidelines and works with the magically named U.S. Office of Water Prediction. “Almost all of [the models] assume that data varies from year to year, but underlying drivers are not changing over time. Those assumptions just don’t hold.” These days, the 100-year-flood problem is the opposite of the one that haunts the Colorado River. In most of the United States, the warmer atmosphere holds more water vapor, leading to heavier and more frequent precipitation.
But figuring out how much rain falls from the sky is only half the battle. You also have to figure out where it goes once it lands. Here, climate change brings additional problems that toy with our probability tables. Very wet soil, very dry soil, burn scars, or erosion can change the way a fixed quantity of rain hits the ground. Urbanization has also created new flood zones, shaped not by hills and valleys but by parking lots, undersized sewer lines, blocked storm drains, and broken pumps. In urban areas, runoff is increasing faster than rainfall, and urban flooding may be a harbinger of what stronger storms will bring to greener areas in the decades to come.
When the First Street Foundation mapped flood risk accounting for rainfall flooding and changing storm expectations, it found almost twice as many parcels in the 100-year flood zone as FEMA has in its maps—1 in 10 U.S. properties. But even the idea of a 100-year flood zone doesn’t sit right with some scientists. You’re not in or out; danger comes on a spectrum. Moreover, some flood zones—a few feet from a crashing ocean, or in a steep and isolated river valley—are more dangerous than others.
All of it adds up to a sense that what happened in the last 100 years may not be the best predictor of the weather today—let alone tomorrow. How to adjust for changing rainfall and flooding expectations is complicated. Infrastructure, most people agree, should be designed to serve its purpose for many decades. (The Dutch have built the Maeslant Barrier for a 10,000-year storm.) Insurance, most people agree, should reflect current risk, not future risk. But how do you account for future flood risk in land use or building regulations? Often our expanded sense of risk winds up penalizing lower-income Americans on low-lying land with outdated infrastructure. You can choose to spare them the burden of adaptation, but is that fair to prospective buyers, or to their neighbors who will bail them out?
Ultimately, said Haneberg, the Kentucky geologist, the questions here are political. “What’s the point?” he asked. “Are we trying to help people rebuild or prevent flood damage?” When it comes to rainstorms, a hundred years ain’t what it used to be.