Hibernating Black Bears Sleep Twice as Long in Winter as in Summer, AI Helps Confirm

Black bears vanish into their dens each winter and emerge months later, and science has long assumed they spend most of that stretch asleep. A new study tracking brain activity in American black bears during hibernation confirms just how much: these animals sleep roughly twice as long during hibernation as they do in summer, the first time automated brain-wave analysis has established the pattern across both seasons in bears. The tool that made it possible? Artificial intelligence trained to score sleep stages from brain-wave and related signals.

For years, scientists have wanted to know what’s really happening in a bear’s brain during those long winter months. Proving it required recording brain activity around the clock for years on end, generating so much raw data that no human team could reasonably analyze it by hand. Researchers from the University of Alaska Fairbanks, Stanford University, the University of Oxford, and the University of Melbourne set out to solve that problem by testing whether AI-powered sleep-scoring software could do the job accurately, even in an animal whose body temperature, and therefore brain activity, shifts during hibernation.

What they found, published in the journal PLOS One, opens a window into one of nature’s most fascinating survival strategies, and it could help researchers better understand how sleep fits into the deep metabolic slowdown of hibernation, a process with possible relevance to human medicine.

A Mountain of Data Too Big to Read by Hand

Researchers implanted 16 captive American black bears with small devices that wirelessly transmitted brain-wave readings, eye-movement signals, and muscle-activity data throughout the year. These are all standard measures used to identify sleep stages. The bears were kept in outdoor enclosures near Fairbanks, Alaska, in conditions meant to mirror their natural environment. Over the course of the study, the team accumulated more than 3,500 days’ worth of continuous recordings.

Sorting through that volume of data manually to label each moment as non-REM sleep, REM sleep (the dream stage), or wakefulness would have been practically impossible. So the researchers turned to two AI programs designed to automatically classify sleep stages: one called Somnotate and another called Somnivore. Both had previously been tested in humans and common lab animals such as mice and rats, but neither had ever been applied to a hibernating animal.

Why Bear Brain Waves Are Tricky for AI to Read

Hibernation posed a unique challenge. A bear’s core body temperature drops during hibernation, though not as dramatically as in smaller hibernators like ground squirrels. Black bears cool from about 38°C in summer down to roughly 30 to 35°C during winter, and that temperature doesn’t stay steady. It cycles up and down over multi-day periods throughout the hibernation season.

Brain wave patterns shift with brain temperature. Cooler brains produce electrical signals at slightly different frequencies, and since AI sleep-scoring tools learn to recognize patterns in those signals, the researchers worried that temperature-driven changes could confuse the software. To find out, they carefully selected three 24-hour recordings from each of six bears: one during hibernation when body temperature was near its highest point in the cycle, one near its lowest point, and one during summer when the bears were fully active.

Three trained human scorers independently scored those reference recordings by hand. Scores where at least two of the three agreed were used as the gold standard against which the AI tools were measured.

AI Approached the Experts, With One Important Condition

Both AI tools performed well once trained on bear-specific data. Accuracy scores landed mostly in the range of 0.90 to 0.98 for both programs. For context, the human scorers themselves agreed with the consensus roughly 90.5% to 96.5% of the time, putting the AI in a remarkably similar range.

There was a catch. Each AI program needed to be trained specifically on bear data, and, in the case of one program, separately for hibernating versus non-hibernating periods. Applying a model trained on a different bear entirely caused accuracy to drop substantially. The researchers did find, however, that a single training model worked well across the full range of body temperatures observed during hibernation, meaning the AI did not need to be retrained whenever a bear’s temperature cycled up or down. That matters in practice, because temperature shifts gradually over weeks, and it would be nearly impossible to determine exactly which training model should apply at any given moment.

During summer, the AI tools slightly overestimated sleep time, likely because bears moved in and out of light drowsiness more frequently, which can blur the line between sleep and waking for any scorer, human or machine.

Bears in Hibernation Sleep About Twice as Long

With reliable automated scoring in place, the team compared sleep patterns across seasons. During summer, bears spent about 32% of their time asleep. During hibernation, that figure rose to roughly 65%. Bears spent nearly twice as much time in non-REM sleep and more than twice as much time in REM sleep during winter compared to summer. Waking hours dropped by roughly half.

Both AI programs confirmed these findings, and their results during hibernation did not differ significantly from those recorded by the human scorers. As the authors write in the paper, “the total distribution of vigilance states based on automated scores of the reference data set show for the first time evidence that bears sleep about 2x as long during hibernation compared to summer.”

Bears matter to sleep researchers for more than just sleep. During hibernation, a black bear’s metabolism drops to about 25% of its normal rate, yet its body temperature falls only about five degrees. That combination is unusual, and understanding how sleep fits into it could inform research on medically induced states of reduced metabolism in humans.

This analysis covers only a snapshot, three 24-hour recordings per bear (one near the peak and one near the trough of the mid-hibernation temperature cycle, plus one summer recording), drawn from six bears. Because those winter recordings came from the coldest stretch of hibernation, shivering may also have influenced the sleep patterns the software detected. Scoring the full 3,500-day archive across more animals and across the entire hibernation season will be needed to fully understand how sleep patterns shift throughout winter and as bears enter and exit hibernation.

Still, the proof of concept is now established. Reliable, AI-assisted brain-wave analysis of bears works. Scientists finally have the tools to work through one of the most extensive sleep archives ever assembled, and to ask sharper questions about what sleep is actually doing inside a hibernating bear.

---