On THE morning of December 26, 2004, villagers from Bang Koey in Thailand noticed something strange. Buffalo grazing on the beach lifted their heads, pricked their ears and looked out to sea, then stampeded to the top of a nearby hill. For the baffled villagers who chose to follow them, it was a live-saving move. Minutes later, the tsunami struck.

Since then, there have been hundreds of reports of animals seemingly foretelling catastrophe - not just minutes, but sometimes hours and even days before it occurred. These include tales of bizarre behaviour by wild beasts including elephants, antelopes, bats, rats and flamingos, plus stories of dogs refusing to go for their usual morning walk.

Could these creatures have sensed the massive earthquake that triggered the 2004 tsunami? It is an outlandish assertion, given that seismologists have so far failed to come up with any sign that a quake is imminent. Yet, for the same reason, the possibility animals might hold the answer cannot be ignored. After all, an advance warning system could save thousands or even millions of human lives.

The idea that animals can predict earthquakes has ancient origins. In 373BC, the Greek historian Thucydides recorded that rats, dogs, snakes and weasels deserted the city of Helice in droves a few days before a catastrophic earthquake.

It was the first in a long line of such anecdotes. There is also no shortage of theories about what might be going on.

What has been lacking, however, is real scientific data linking strange animal behaviour with earthquakes. Now at last we have some, and from an unusual quarter. Last November, a psychologist looking for depression in dogs announced he had stumbled upon the elusive evidence.

In late 2000, Stanley Coren from the University of British Columbia started a study into whether dogs, like some humans, suffer from "seasonal affective disorder", or the winter blues. Twice a week, he would email 200 dog owners in Vancouver, asking them to rate their pets' activity and anxiety. The results were disappointing.

In general, there was little daily variability, and Coren's initial analysis of many months' information strongly refuted his suspicion that dogs can become depressed in winter.

So it was some time before Coren realised what he had unknowingly recorded. "When I finally did go through the figures in detail a couple of years later, I noticed this strange point of data just floating there."

The anomaly occurred on February 27, 2001. Of the 193 dogs whose behaviour was recorded that day, 47 per cent were much more active than usual and 49 per cent - mostly the same dogs - were much more anxious than usual. Both measures were statistically significant, and the likelihood that such a big difference was coincidental was less than 1 in 1000.

At first Coren thought severe weather, perhaps a thunder storm, might have driven the dogs berserk, but newspaper archives proved him wrong. Then he realised there had been an earthquake. On February 28, a 6.8 magnitude quake shook the north-west Pacific, with an epicentre at Nisqually, Washington, about 240 kilometres south of Vancouver.

Coren started to wonder what the dogs could have sensed. There were few clues in the standard literature. Seismologists looking for signals foretelling a big earthquake have failed miserably to find any. Even the famed Parkfield experiment in California, the most comprehensive long-term earthquake research project in the world, has so far drawn a blank. Since 1985, researchers from the US Geological Survey and the state of California have exhaustively monitored a section of the San Andreas fault near the town of Parkfield. To date, they have found nothing that would reliably indicate a quake is imminent.

Despite this, the mass of anecdotal evidence about animals predicting earthquakes has attracted much speculation about what they might be sensing. One idea is that some animals detect changes in the Earth's electrical field, although the existence of electromagnetic changes associated with earthquakes is disputed.

Another theory is that animals are responding to subterranean gases such as radon and hydrogen released from rocks before a quake, despite the fact that few experts accept such gases are produced.

A more exotic explanation comes from Rupert Sheldrake, a biologist and director of the Perrott-Warrick Project in London, which researches unexplained human and animal abilities. "I think animals are picking up on something that we are not. It could be electromagnetic, or perhaps even a sixth sense," Sheldrake says.

Coren's attention, however, was drawn to a more down-to-earth idea. He suspected the dogs in his study might simply be hearing vibrations. So back he went to his data to look for evidence of the idea. Sure enough, of the 14 dogs in the study with hearing impairments, only one was significantly more anxious that day, and it lived with a hearing dog that had also become anxious. Encouraged by the finding, Coren went on to look for factors that might explain why some dogs became agitated on February 27 and others did not.

He found that dogs with floppy ears showed only half the change in activity and one-third the change in anxiety of dogs with pricked ears. Not only would an ear flap reduce the sound reaching the inner ear, Coren also realised it would reduce high-frequency sounds more than low-frequency ones.

What's more, dogs with smaller heads were almost twice as likely to behave strangely before the earthquake than those with larger heads.

This was particularly interesting, given that dogs with smaller heads tend to be more sensitive to high frequencies.

Taken together, Coren's results present an alluring hypothesis. He suggests that the kind of high-frequency sounds many dogs can hear are emitted before an impending earthquake, perhaps from rocks scraping or breaking underground.

Admittedly, this is only one study. Even if Coren is right about dogs, it is still possible other animals may be able to predict quakes in different ways.

Still, his findings will appeal to anyone interested in putting the phenomenon on a more scientific footing. One of these is Eric Wikramanayake, a conservation scientist with the World Wildlife Fund, who was studying elephants in Sri Lanka when the Asian tsunami struck.

"There were all these reports of elephants fleeing beaches hours and even days before, due to a sense of impending danger, which people attribute to a sixth sense," Wikramanayake says. "Our data just do not show any of this."

Like Coren, Wikramanayake had not set out to investigate earthquake prediction. He was interested in how elephants range over their habitats and had fitted animals in two herds with radio collars to monitor their movements.

What he found was precisely nothing. One herd was only 100 metres from the beach when the tsunami arrived and took cover behind a large sand dune when the wave came into sight. The other herd was safely about five kilometres inland and did not behave unusually before, during or after the tsunami.

That is at least consistent with Coren's idea that dogs can detect high-frequency sounds. Elephant hearing is geared towards low frequencies, and they reacted only when they could see they were in danger.

"As far as I can tell, the evidence from dogs and elephants is strongly suggesting that we need not look any further than the five senses we already know about," Wikramanayake says.

IF COREN's idea is correct, however, that raises the question of why the extensive Parkfield experiment has detected nothing. The answer may lie in finances and logistics.

"We expected to see an earthquake of 6.0 or more at Parkfield within seven years of the project start date, and it didn't happen," says seismologist Peggy Hellweg from the University of California, Berkeley. The expected earthquake eventually struck in 2004, but by that time inaccessible instruments had started to fail, some of the funding had been pulled and some researchers had left the project. In any event, it is debatable whether the Parkfield project would have detected any acoustic early warning signals even if a quake had come sooner.

Hellweg points out that earthquakes are highly variable. "We don't understand earthquakes well enough to say why each quake gives off different signals," she says.

Even if some quakes are preceded by high-frequency vibrations, is it possible that dogs in Vancouver could detect sounds emitted near Nisqually?

Andy Michael, a US Geological Survey seismologist, is sceptical. He points out that the epicentre of the quake was more than 240 kilometres south of Vancouver. "It is physically implausible for seismic waves in the kilohertz range to travel that far and be above the background noise level," he says. "After all, if they have trouble getting through a dog's ear flap, how will they get through hundreds of kilometres of rock?"

Normally, he says, seismic waves at such high frequencies are detectable at most a few tens of metres from their source.

Although Michael is unconvinced by Coren's ideas, he is intrigued by the blip in the data. "I am inclined to believe that this is just an extraordinary coincidence," he says. But he adds a cautionary tale.

"When Alfred Wegener presented his theory of continental drift he proposed mechanisms for continental movement that were easily disproved. His core ideas about tectonics were right but nobody was listening because the mechanisms were wrong."

Besides, if animals really can foretell earthquakes, does it really matter how? Given the potential to save human lives, Sheldrake thinks not. "We know that before at least some earthquakes animals behave strangely. That should be a powerful incentive for us to develop some sort of prototype warning system."

The idea is not so far-fetched. While Western society has been reluctant to use animals as earthquake predictors, China has embraced the idea. At the Nanning earthquake bureau in south-central China, experts use a video link to keep a 24-hour watch on snakes in farms throughout the country. If the snakes begin making desperate and concerted attempts to escape, the observers raise a warning. Other governments may not be willing to go that far yet, but Coren believes they should at least explore the possibility of animal quake prediction.

"A centre that people could call in to when they see unusual things would be useful," he says. "We would be able to collect much more data - and if that proves useful, we could begin considering giving warnings when we suddenly get hundreds of calls from a single area."

It could cost little more than the price of setting up a phone line. Put that way, what have we got to lose?