The Colorado River is a lifeline. It flows from the central Rocky Mountains of Colorado and Wyoming through five other U.S. states, 11 U.S. national parks and two Mexican states on a 1,450-mile-long journey that supplies water to 40 million people, supports uncounted wildlife and irrigates more than five million acres of farmland. And the river is endangered.

Not only are humans drawing unsustainable amounts of water from this source, but abnormally low precipitation and hot, dry conditions have been shrinking it for years—an alarming trend that is likely to worsen as climate change takes its toll. “To the extent that water is life, the idea that we lose the Colorado River—or even that it is diminished—has an outsize impact on this region,” says Jennifer Pitt, Colorado River Program director at the National Audubon Society. Yet despite the river’s importance, scientists have had a hard time pinning down how much its flow may decline as the world warms. To Chris Milly, a senior research scientist at the U.S. Geological Survey, the question is both disconcerting and fascinating. “I was pulled into the mystery of what was really going on in the river basin,” he says. “My interest bordered on obsession.”

That obsession turned into a year-long immersion in data. The results, published Thursday in Science, suggest that by midcentury, the river could lose 14 to 31 percent of its historical flow from the period of 1913–2017.

Milly and his colleague Krista Dunne, also at USGS, created an extremely detailed computer model that analyzed how water moves in and out of the Colorado River basin via precipitation, melting snowpack, evaporation and other key processes. But because there are several physical parameters with values that are difficult to measure (such as the maximum amount of water the soil can hold at any given location in the basin), the researchers ran the model a whopping 500,000 times—tweaking those unknown parameters in every instance—until they found 171 versions that reproduced historical records remarkably well. They then projected their new and improved model into the decades ahead in order to estimate how the river might shift if the basin’s temperature increases by one degree Celsius (1.8 degrees Fahrenheit). The answer was grim: its flow would drop by 9.3 percent. Based on a range of climate scenarios, Milly and Dunne went on to predict that by midcentury, the Colorado River’s flow will likely decrease by as much as 31 percent, compared with historical values.

The study indicates the Colorado’s future hinges on snowpack, which is a major source of its water, because as the snow gradually melts in the spring and summer, the resulting water trickles into the ground, the river and its tributaries. “We discovered that snow cover behaves as a protective shield,” Milly says. Its high surface reflectivity, or albedo, throws back incoming solar radiation and keeps the ground beneath it relatively cool. But climate change is reducing the extent of that shield, allowing more solar radiation to penetrate the surface and thus creating a number of cascading effects. A large amount of moisture within the soil and trees will likely evaporate. Much of the remaining snowpack and groundwater will do so as well, leaving little water to run into the river.

Brad Udall, a senior scientist at Colorado State University, who was not involved in the new paper, calls its findings—particularly the 9.3 percent drop in flow—“eye-popping.” Udall co-authored a 2017 study that suggested the flow would decrease by 3 to 10 percent per 1.8 degrees Fahrenheit of warming, so the fact that Milly and Dunne’s number sits at the high end of that range grabbed his attention. But he does not doubt the researchers’ results, which, he says, went into much greater detail than previous efforts. “I would argue that they did it more elegantly and more rigorously,” he says. “And you have to take this result pretty seriously.”

Udall thinks the findings will have major ramifications for water managers and users alike. “Every drop in that river is being used. And any reduction like that is going to cause serious pain,” he says. But he is hopeful that conservation managers will find the best route forward. “I like to say, ‘Hey, if we’ve got 20 percent less, that still means the glass is 80 percent full,’” he says. “Let’s get smart and savvy and figure out how to use what we’ve got.” Meanwhile Pitt, who was also not involved in the new study, is similarly inspired by a resolution reached last year when the seven U.S. states that host the river agreed to voluntarily cut their water use.

Still, Pitt worries that the Colorado River will continue to change—and in unpredictable ways. Although scientists have made significant strides in forecasting the impacts of rising global temperatures, those projections cannot include the inherent variability of water flow in the river. The historical record, for example, shows it might drop to roughly four million acre-feet in one year and climb to about 24 million acre-feet in another—all because of a varying snowpack. (An acre-foot is the volume of one foot of water over an area of one acre, or roughly 326,000 gallons.)* In addition, these studies cannot take into account the many broader changes that the decreasing snowpack will manifest in the Southwest. Not only does the early snowmelt create a darker, more absorptive earth, it also bumps summer—and fire season—earlier. That process will further dry the region and reduce the flow of water into the Colorado River.

“Water is sort of a silent utility,” Pitt says. “People don’t tend to think about it or worry about it. But the potential impacts are enormous.”

*Editor’s Note (2/2/20): This sentence was edited after posting. It originally described an acre-foot as the volume of an area of one foot of water over a depth of one acre.