Animals appear to predict earthquakes by sensing electricity in the air, the first study to find reliable evidence of the phenomenon has shown.
Camera traps revealed an "amazing" drop in the number of animals spotted up to 23 days before a major quake hit their rainforest home at Yanachaga National Park in Peru.
For five of the seven days leading up to the magnitude seven Contamana earthquake in 2011 no animal movements were recorded at all. This was totally unexpected in a region normally teeming with wildlife.
Lead scientist Dr Rachel Grant, from Anglia Ruskin University, who searched camera trap records for evidence of altered animal activity before earthquakes, said: "The park was 320 kilometres from the epicentre, and I thought, there was not much going to be happening. But when I saw the results I was totally shocked. It was amazing. The analysis showed that just before the earthquake animal activity dropped right down."
Other evidence suggested that prior to the earthquake the air around the high mountain site was filled with positive ions - electrically charged molecules that can be generated when rocks are placed under stress.
Scientists believe the animals, mainly rodents and other creatures living close to the ground, were made to feel uncomfortable by the positive ions leading them to avoid the area.
They are thought to have escaped to lower ground where the air was less ionised.
Positive ions - unlike "feel good" negative ions - have been known to trigger headaches and other ill effects such as agitation, hyperactivity and confusion in humans as well as animals.
The findings may help experts develop better short-term seismic forecasts, the researchers claim.
On a typical day the motion-triggered cameras placed around Yanachaga National Park record between five and 15 animals.
But in the 23 days before the Contamana earthquake the number of sightings dropped to five or fewer per day. No animals were photographed at all on five of the seven days immediately preceding the quake.
A comparison study of camera trap sightings showed that animal activity remained normal in the park over a different period when seismic activity was low.
Other records revealed disturbances to the ionosphere - an electrically charged layer of the atmosphere - which began two weeks before the earthquake and could have been linked to positive ions.
An unusually large effect was detected eight days before the earthquake, coinciding with the second significant decrease in animal activity.
Dr Grant said: " We believe that both of these anomalies arise from a single cause: seismic activity causing stress build-up in the Earth's crust, leading - among other things - to massive ionisation."
Previous research has demonstrated that rocks placed under stress generate positive ions in the air around them.
There are numerous anecdotal reports of animals seeming to predict earthquakes, but none have been regarded as reliable.
The new study, published in the journal Physics And Chemistry Of The Earth, did not depend on witness recollections that might be subject to psychological bias.
Although humans could be affected by positive ions, Dr Grant did not think we share an ability to sense approaching earthquakes with animals.
"Unfortunately we humans have lost a lot of our ability to sense things," she said. "We've created our own environment, cocooned in our houses, so we don't have to. Also humans are very generalist in what they respond to.
"The animals most sensitive to positive ions are those that live in close proximity to the ground. They are much more likely to feel something. Those we found to be most affected were small rodents - they had totally gone eight days before the earthquake."
She added: "I don't think animals have evolved an ability to predict earthquakes. What's more plausible is that animals in general will tend to move away from unpleasant stimuli. It's a simple avoidance from a chemical they don't like."
Co-author Professor Friedemann Freund, from the Seti Institute at the American space agency Nasa's Ames Research Center in California, said: "The camera traps were located on a ridge at an altitude of 900 metres. If air ionisation occurred, it is likely that it was particularly strong along such a ridge. Hence, the animals would have escaped to the valley below, where they were exposed to fewer positive airborne ions.
"With their acute ability to sense their environment, animals can help us understand subtle changes that occur before major earthquakes. These changes, that we are now able to measure, express themselves in many different ways at the earth's surface and above."