On the evening of July 3rd, as the sun went down over Kerrville, Texas, a small city of some twenty-five thousand people and the seat of Kerr County, the water in the Guadalupe River was just four inches deep, according to the U.S. Geological Survey’s stream gauge there. The area hadn’t had any rain since mid-June. R.V. campers in the HTR TX Hill Country campground in Kerrville, some of whom had arrived at the sixty-five-acre facility only hours before, could barely hear the river, even though they were parked in premium spots next to it.

At 3:30 A.M. on July 4th, Dalton Rice, the city manager, went out for an early-morning jog along the sluggish waterway. The river had risen to 1.71 feet, around the average depth. Kerrville’s much anticipated “Fourth on the River” celebration at the riverside Louise Hays Park was scheduled for that afternoon, and Rice saw “not a drop of rain” during his run, he later told Dan Patrick, the lieutenant governor of Texas. By 4 A.M., when Rice went home, “there was very light rain,” he said that day. “We did not see any signs of the river rising at that time.”

Rice was apparently unaware that a few hours earlier, at 1:14 A.M., the Austin-San Antonio office of the National Weather Service had sent a flash-flood warning for south-central Kerr County. The area it covered included the town of Hunt, about twelve miles upstream from Kerrville, where Camp Mystic, a girls’ sleepaway camp, was situated, at the confluence of Cypress Creek and the south fork of the Guadalupe. At 4:03 A.M., the warning was upgraded to an emergency: “This is a PARTICULARLY DANGEROUS SITUATION. SEEK HIGHER GROUND NOW!” The stream gauge at Hunt showed the river nearing twenty-two feet, twelve feet higher than its banks. In the next hour, it would rise to thirty-seven feet, at which point the gauge stopped transmitting.

Network-news footage of flood-related disasters has traditionally shown low-lying coastal towns swamped by storm surges from named weather systems, with residents and rescue workers navigating the streets in hip waders, canoes, and powerboats, and the water level sometimes taking days to subside. Houston, New Orleans, Tampa, Charleston, and New York are among the cities that have been inundated in the past three decades. Yet storm-surge-driven inundation flooding is not the deadliest kind of flood. The disaster often unfolds slowly enough that there’s time for many people to get to higher ground—and for weather reporters to get in place. The water isn’t moving anywhere near as fast as that of a flooded river. The flowing water in a mad river reshapes the entire landscape, a process known as fluvial erosion.

The flood barrelling down on Kerr County more resembled a tornado or a wildfire, a volatile, rapidly changing hazard, with a narrow window within which to act before the deadly force of the raging river arrived at your door. The September, 2024, floods that killed a hundred and eight people in the high country of western North Carolina, when Hurricane Helene passed over, were this kind of event, as were the floods in Valencia, Spain, in late October, 2024, which killed two hundred and thirty-two people. The recent flooding in Ruidoso, New Mexico, which claimed three lives, was, too. In my part-time home of Vermont, storms flooded much of the state, including the capital, Montpelier, in July, 2023, and flooding caused by the remnants of Hurricane Beryl hit the town of Plainfield, among other places, in July, 2024. Two people died in each event, and the damage from erosion was severe. The 2023 disaster wasn’t even caused by a named storm. Rain-soaked “thunder boomers” were enough to do the damage.

Historically, this kind of flooding, which often occurs in hilly and mountainous regions, has received far less attention than storm-surge flooding, and local municipalities tend to be less prepared than coastal towns to deal with it. People who have moved away from the beach to escape hurricanes and rising sea levels, and have settled in supposed “climate havens” like Asheville, North Carolina, or Plymouth, Vermont, are reluctant to accept that they have merely traded a devil they know for one they don’t. Francis J. Magilligan, a fluvial geomorphologist at Dartmouth College, told me, “We are actually getting pretty good at figuring out inundation risk. I would say, comparatively, we’re at the starting point of thinking about erosional risk. And we don’t have a good set of tools, FEMA or otherwise, to understand that risk.”

The floodplain maps that the Federal Emergency Management Agency uses to designate a Special Flood Hazard Area, or S.F.H.A., are based on historical climate and geological data. The maps don’t show the actual boundaries of past floods; rather, they illustrate statistical constructs based on the probability that a flood of a certain magnitude will occur there in a certain time range. The National Flood Insurance Program uses the maps to calculate risk. People who live in a hundred-year floodplain, for example, have a one-per-cent chance of being flooded in any given year. Anyone who builds or buys property in a mapped floodplain must take out federal flood insurance in order to qualify for a mortgage. The maps cover the coastal areas and the borders of major rivers, like the Mississippi, that flood frequently. They often don’t cover mountainous communities with smaller rivers and streams, where flooding can affect not only people living down by a river but also those living on banks high above it, whose homes may be lost to erosion. Moreover, most of the data FEMA uses to designate S.F.H.A.s are based on readings from the nineteen-sixties and earlier—data that climate change has largely rendered obsolete. The language, too, engenders a false sense of confidence. It makes it sound as though a bad flood will occur only once every hundred years, and that’s not the way the statistic is supposed to work.

Flash floods often occur in terrain with steep, narrow valleys that drain into rivers where the water is confined in natural and man-made channels and has nowhere else to go. The trigger is very heavy, concentrated rainfall. Warmer oceans lead to more surface evaporation, and warmer air is capable of carrying more moisture. When the warm air is forced upward by mountains, it cools and loses its ability to retain water, causing a sudden release of rain. With the Valencia floods, a year’s worth of rain fell in a single day. In North Carolina, upward of twelve to fourteen inches of rain fell in several hours. As global temperatures rise, outbursts of extreme precipitation will increase. Carl Renshaw, a hydrologist at Dartmouth, told me, referring to the meteorological conditions that led to the Texas floods, “You’re dumping so much water so quickly because you’ve primed the system with moisture. It’s like loading a gun. The people there had no time.”

Kerr County and its neighbor Kendall County sit in a valley that is known as Flash Flood Alley, because of the speed with which rainfall runs off the cliffs of the Balcones Escarpment and sluices down through streams and rivers to the coastal plain around San Antonio. The hillsides are steep, and the thin, dry topsoil over bedrock retains little water. Major flooding occurs regularly. And yet, even in Flash Flood Alley, which the judge of Kerr County, Rob Kelly, described in a press conference on July 4th as “the most dangerous river valley in the United States,” people continue to build in hundred-year floodplains. When properties are flooded, old-timers and newcomers alike tend to fix things back up. Kimberly Meitzen, a geography professor at Texas State University, in San Marcos, a small city in the Guadalupe River basin, told me, “We have this history of flooding and rebuilding here.” She mentioned five floods within living memory, in 1978, 1987, 1998, 2002, and 2015, in which not just houses but whole neighborhoods were washed away. “They just rebuild,” she said. Moreover, she added, “people continue to request variances to rebuild even closer” to the river. “There’s no really strong regulatory authority to prevent it,” she went on. “And so it just continues to put more people at risk.”

In Vermont, a flood in August, 2011, caused by Hurricane Irene, was so devastating that the U.S. Geological Survey called it a five-hundred-year flood, and at least one analysis deemed it a thousand-year flood. The flood that tore through Plainfield on July 10, 2024, occurred one year to the day after the 2023 flood, which had been a hundred-year one. And, on the same day this July, a violent flood struck Caledonia County, in the north of the state. Although such a tight grouping of floods isn’t statistically impossible, it suggests that the probability curves used by FEMA are out of whack with real climate conditions.

A recent national study commissioned by First Street Foundation, a private risk-assessment firm, concluded that floods previously considered to be hundred-year events have become, on average, sixty-two-year events. (In some places, the estimate is as low as eight years.) The study also found that in Vermont’s Washington County, which includes Montpelier, Plainfield, and the city of Barre, more than forty-eight hundred properties (out of nearly thirty thousand total properties) are at high risk of flooding, yet FEMA includes fewer than fifteen hundred of those properties on its special-flood-hazard maps. Nationally, the study found that 17.7 million properties are at risk of flooding, but only about five million properties are in a FEMA flood-hazard zone. That means that millions of home buyers and owners are making decisions with an incomplete understanding of the true physical and financial risks they face. Jeremy Porter, the head of climate-implications research at First Street, told me, “Ultimately, what you end up doing is systematically underestimating flood risk.”

Vermont feels like the frontier of climate change in the Northeast. Farmers in the bottomlands, who previously planted wheat and barley, are beginning to plant rice, which can be underwater for two days without damage to the crop. The old roads that early Vermont settlers hacked out on hilltops, which lasted for more than two hundred years, are melting back into the forest. Extreme-rain events scour the roads down to bedrock ledges, rendering them impassable, and, because no one then uses them, any blown-down trees don’t get cleared. The next storm brings more blowdowns. A road that I went mountain biking on ten years ago, when it was a distinct pathway with old-growth trees on each side, lined by aged stone walls, is now such a tangle of fallen trees, branches, and rocks that it’s hard to tell a road was ever there.

Vermont is the second least populated state, after Wyoming, with fewer than six hundred and fifty thousand residents; it is also the fourth highest in disaster-relief funding per capita, nearly all of it flood-related. Washington County ranked first nationally in disaster declarations between 2011 and 2024. Annual precipitation in the state has increased six inches since the nineteen-sixties, and heavier-than-normal rain events in the Northeast are expected to increase by as much as fifty-two per cent by 2100. Vermont is a laboratory for the study of intense rainfall in steep terrain, and a proving ground for scientists, policymakers, regulators, and land-use planners who are on the front lines of a recurring catastrophe that traditional methods of prevention—dredging a river’s bottom, armoring its sides, berming its banks—have only made worse.