Satellite image of the Mississippi River surrounded by a green landscape
Deep learning techniques revealed that rivers across the United States (such as the Mississippi) and central Europe are warming up and losing oxygen even more rapidly than oceans. Credit: Lauren Dauphin, U.S. Geological Survey

A deficiency or absence of oxygen in Earth’s bodies of water could increase greenhouse gas emissions, mobilize metal toxins, and suffocate oxygen-breathing aquatic life. Although deoxygenation is relatively common in bodies like lakes and oceans, new research in Nature Climate Change reports that rivers across the United States and central Europe are warming up and losing oxygen even more rapidly than oceans.

Scientists trained a deep learning model to fill in the gaps of 4 decades of water quality and temperature records collected from 796 rivers. This model allowed them to reconstruct trends that would otherwise be impossible to pick out in the hodgepodge of historical data. Running their model into the future, the researchers predict that oxygen levels will sink between 1.5 and 2.5 times faster than they have over the past 40 years.

“The rate is small, but you’re changing your baseline in a way that can cause extremes to become more frequent,” said aquatic ecosystem ecologist Joanna Blaszczak of the University of Nevada, Reno, who was not involved in the study. “And those extremes have many implications—both biogeochemical and for aquatic communities.”

A bit like the weather, dissolved oxygen levels vary from day to day. But similar to how higher average temperatures mean more frequent and extreme hot days, sustained low baseline dissolved oxygen means more frequent, more intense lows, which can lead to “dead zones” for aquatic life that needs oxygen to breathe. It can also spur greenhouse gas emissions because anoxia nudges bacteria to switch from respiring oxygen to another metabolism that produces nitrous oxide, a potent greenhouse gas. Many toxic metals, such as arsenic, are also sensitive to oxygen and move more easily through rivers in periods of extreme anoxia.

Patchy Data Stymie Efforts

But rivers are not flasks in a lab. In the real world, things are more complicated.

At first glance, the link between temperature and dissolved oxygen in water is straightforward: Oxygen dissolves more easily in cold water than it does in warm water. This relationship is one reason why climate change is expected to drive down dissolved oxygen levels in some waterways.

But rivers are not flasks in a lab. In the real world, things are more complicated.

For example, rivers are aerated by their flow. And biological activity in rivers can either increase oxygen, which happens during photosynthesis, or use it up, which happens when animals and bacteria respire. The landscapes that rivers traverse can substantially affect dissolved oxygen, too.

So scientists can’t simply assume that climate change will drive anoxia in rivers everywhere. They need real data. Luckily, agencies like the U.S. Geological Survey have decades’ worth of it. There’s just one catch.

“The data tend to be, in general, relatively sparse,” said environmental scientist Wei Zhi of Hohai University in Nanjing, China.

As might be expected for data collected across multiple decades, agencies, and countries, river quality data are “patchy,” Blaszczak agreed. And this patchiness stymies efforts to reconstruct reliable large-scale historical trends in things like river temperature and dissolved oxygen.

Deep Learning Uncovers Trends

Zhi and his colleagues thought deep learning might be able to fill in the gaps in the historical record. So using data on water quality, temperature, topography, land use, and weather, they trained a type of neural network called a long short-term memory model to predict the temperatures and dissolved oxygen levels in U.S. and central European rivers. Long short-term memory is a classic model for filling in time series, Zhi said.

“They’re overcoming sparse data limitations.”

“They were able to use this long short-term memory model, so deep learning, to reconstruct past historical time series and then evaluate trends,” Blaszczak said. “They’re overcoming sparse data limitations.”

The trends revealed that 87% of rivers in the study warmed between 1981 and 2019, with median warming rates of 0.16°C and 0.27°C per decade for the United States and central Europe, respectively. About 70% of rivers deoxygenated, with the decadal decrease in average dissolved oxygen concentrations reaching as high as 1%–1.5% in some rivers.

Because of the many factors affecting real-world rivers, the straightforward link between higher temperature and deoxygenation didn’t entirely hold: Rural rivers warmed the least but deoxygenated the most rapidly.

“Agricultural sites usually have a lot of nutrients,” said environmental scientist Li Li of the Pennsylvania State University, one of Zhi’s colleagues. “And that could stimulate a lot of biological activity like respiration,” which depletes oxygen.

Also, rivers are “losing oxygen faster than the oceans,” said Li. That’s counterintuitive because rivers are shallow, well aerated, and illuminated by sunlight, which powers oxygen-producing photosynthesis.

How, exactly, any given river will change in the future will depend on local factors. “We would love to do global-scale study,” said Li, but for now, she and Zhi could consider only North American and European rivers because of data limitations. Even so, the detailed nature of their findings revealed substantial regional variations that wouldn’t be detectable using other methods—both between the United States and central Europe (which is warming and deoxygenating more quickly) and within smaller local areas.

“And when we’re thinking about prioritization of how we can manage rivers to sort of preemptively try to slow rates of deoxygenation,” said Blaszczak, “I feel like that’s an important conclusion.”

—Elise Cutts (@elisecutts), Science Writer

Citation: Cutts, E. (2023), Rivers are warming up and losing oxygen, Eos, 104, Published on 3 November 2023.
Text © 2023. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.