麻豆传媒

THE resort of Acapulco in Mexico has long been known for its attractions: gorgeous mountains, upmarket hotels, crystal clear waters. But on 7 September 2021, something happened that was on nobody鈥檚 wish list 鈥� a magnitude-7.0 earthquake rocked the city鈥檚 sandy beaches and seafront high-rises.

Along with trembling buildings and shaking trees, those caught in the quake also witnessed something substantially more eerie. A barrage of blue lights, like , lit up the night sky, apparently right above the fault line. This strange display was an example of what are known as 鈥渆arthquake lights鈥�, a semi-mythical phenomenon that has cropped up in reports of tremors for centuries.

The idea that these blue flashes are caused by an earthquake is often dismissed by scientists. Indeed, after Acapulco, some suggested the flickering lights may have come from damaged power lines. But a small group of researchers now claim to have evidence for an alternative hypothesis. It says that when tectonic faults rupture, electrical currents are created. And whether these currents produce lights or not, there should be telltale electromagnetic signals produced by them that would be detectable in advance.

If they are right, we could potentially use these signals as a warning of disaster. It is a long shot: the search for ways to predict earthquakes has frustrated us for decades. But new evidence linked to these uncanny, dancing lights in the sky is shaking up the field.

Predicting major tremors is currently just about impossible. Scientists, including those at the United States Geological Survey (USGS), a national agency, compile long-term seismological data that can tell us the chance of an earthquake hitting a given area, but only across a window of time that spans years or decades, rather than anything more precise. Then, there are warning systems like in the US, which uses seismometers to give people alerts of incoming quakes 鈥� but only seconds in advance.

Predicting earthquakes

To do better, we would have to find what is known as an earthquake precursor, a signal that reliably precedes an earthquake much further ahead of time. The trouble is, it isn鈥檛 clear what that could look like. 鈥淭here are some schools of thought that hold that it鈥檚 never going to be possible,鈥� says seismologist at the USGS. Even the more optimistic reckon that this would, at best, be akin to weather forecasts, giving the probability of an earthquake in the coming days and weeks.

But it is worth pursuing, no matter how slim the chances of success. After all, we had a reminder of just how deadly strong quakes can be in early February, when several struck Syria and Turkey, killing more than 54,000 people.

The idea that electromagnetic signals could be produced in the run-up to a quake was put forward decades ago by , a physicist then based at NASA鈥檚 Ames Research Center in California. He suggested that imperfections in the molecular structure of rocks in Earth鈥檚 crust can be disrupted during earthquakes, unleashing electrical currents that can propagate up through the ground and create a charge in the atmosphere. These charges could build up and cause flashes of electricity 鈥� earthquake lights 鈥� and even explain other phenomena associated with seismic activity, like temperature changes and abnormal animal behaviour.

Freund鈥檚 hypothesis has never gained mainstream acceptance, and some question the basic precepts of his model. Still, the broader idea that there could be an electrical connection between the rocks in Earth鈥檚 crust and the atmosphere isn鈥檛 so wild, even if the details aren鈥檛 well understood.

The story of the recent excitement around earthquake electricity starts back in 1985 when an engineer named Tom Bleier read about earthquake lights and had an idea. If earthquakes were creating bursts of electricity at the surface, he reasoned, they were probably also generating electromagnetic fields deep underground at the epicentre. Earth鈥檚 crust will screen out everything but the lowest electromagnetic frequencies. However, an induction magnetometer tuned to those low frequencies might pick up a signal. These devices 鈥� tens of thousands of metres of fine copper wire wrapped round a metal core 鈥� aren鈥檛 hard to build. Surely it was worth a try?

Bleier tried to persuade the USGS to fund research into this idea. But the agency wasn鈥檛 interested and for years it went nowhere. Then, in the late 1990s, Bleier began working as a satellite engineer for a California-based company called Stellar Solutions whose founder, Celeste Ford, was an old friend. He persuaded her to put up philanthropic funding for an earthquake monitoring system and, in 2000, a company called QuakeFinder was born. With it, Bleier began building a network of magnetometers optimised for ultra-low frequencies around California. To get as close to faults as possible, the devices were installed in backyards, farms, hay fields 鈥� anywhere property owners would allow. 鈥淲e鈥檇 knock on their door and say: 鈥楥an we have your permission to do it?鈥�,鈥� says Bleier. 鈥淎nd in the next day or so, we had it in there and working.鈥�

By 2017, QuakeFinder had 125 of the instruments strung along California鈥檚 major faults. It has been gathering data dozens of times per second for over a decade, picking up on even extraordinarily slight electromagnetic fluctuations. 鈥淚t鈥檚 really hard work to collect good data, clean data, to maintain instruments out in the field,鈥� says Simon Klemperer, a geophysicist at Stanford University in California who has independently analysed QuakeFinder鈥檚 data. 鈥淨uakeFinder did this very successfully.鈥�

Earth鈥檚 magnetic field

People can鈥檛 naturally sense Earth鈥檚 electromagnetic field. But if we could, it might sound like an ocean of fluctuating static, never the same from one moment to the next. Everything from solar storms to passing cars alters the frequencies magnetometers pick up. Subtracting that background noise to find the signals of interest underneath is a challenge, says Karl Kappler, QuakeFinder鈥檚 chief scientist.

The company鈥檚 researchers have been wrestling with this for years, but began to make progress around 2019. In a study published that year, they looked at whether the range of electromagnetic frequencies they saw changed in the days before an earthquake. Using a subset of their data that was comprised of nearly 900 quakes of magnitude 4 or greater, the researchers reported between four and 12 days before these tremors. Their analysis showed that the signals had a statistical significance of 3 sigma, meaning there is a 99.7 per cent chance that they aren鈥檛 just a fluke. 鈥淲hat that suggested was that there really was an effect,鈥� says Kappler.

Emboldened, QuakeFinder turned over its data to researchers from Google, who trained a machine-learning algorithm to sort through it and identify relevant signals. Turning the number-crunching over to this computer let them comb the data with far greater sensitivity and optimise the algorithm specifically for the problem at hand. Crucially, they only chose signals picked up by two or more magnetometers and they split the data set in two, using one half to train the algorithm and then testing it on the second half, which the algorithm hadn鈥檛 seen before.

In this study, the researchers again saw intriguing evidence that . The results, published in 2022, also achieved around a 3-sigma confidence level. Kappler says the second paper felt like a breakthrough for the company. It put the firm in the position of being 鈥渨ell past the threshold of evidence鈥�, he says.

Klemperer sounds a note of caution about QuakeFinder鈥檚 results. His own didn鈥檛 turn up the same precursor signals. That could just be down to differences in data processing methods. But he also points out that the company is looking retrospectively at earthquake data for any signal that looks suspicious, rather than coming up with a hypothesis and testing it. This is a common criticism of earthquake precursor research. 鈥淚f you鈥檙e starting with the time of an earthquake, and looking back, that鈥檚 just not the right way to do science,鈥� says Hough.

That is because it leaves scientists vulnerable to bias, she says, picking out signals that fit a hypothesis and ignoring those that don鈥檛. The gold standard of proof, of course, would be to use these signals to successfully predict an earthquake in advance 鈥� something QuakeFinder hasn鈥檛 yet managed.

Dan Schneider, QuakeFinder鈥檚 director of research and development, takes this point. The company鈥檚 work thus far is more about proving that earthquake precursors exist and that prediction is theoretically possible, than about forecasting any individual tremors, he says. 鈥淭his doesn鈥檛 find any particular needles in any particular haystacks,鈥� says Schneider. 鈥淏ut it does point in the direction that there are needles in these haystacks to be found.鈥�

The QuakeFinder team also argues that similar results from Japanese researchers are further evidence for precursors. Using data recorded between 2001 and 2010 by six magnetometers arrayed near Tokyo, this study found a significant before large earthquakes compared with afterwards. 鈥淵ou can鈥檛 discount three independent studies,鈥� says Bleier. 鈥淭here鈥檚 something there.鈥�

Where does all this leave us? Some scientists argue that, despite years of dismissing the possibility, intriguing evidence that earthquakes can be predicted keeps popping up. Geophysicist Angelo De Santis at the National Institute of Geophysics and Volcanology in Rome, who studies pre-earthquake signals, says that there are probably many different kinds of precursors. 鈥淚t is not only a [single] precursory anomaly that we are looking for, but it is a pattern,鈥� he says. 鈥淲e are able to see a sort of chain, a sequence of different kinds of anomalies.鈥�

De Santis and others have published research identifying what they say is , beginning with changes to atmospheric temperature and humidity, followed by increases in infrared radiation from Earth鈥檚 surface and elevated levels of methane and carbon monoxide, then, finally, anomalies in our planet鈥檚 ionosphere, the highest slice of the atmosphere. Together, these kinds of changes might represent a more trustworthy indicator than any one signal alone.

Still, the science of earthquake precursors remains on wobbly ground. Stellar Solutions, QuakeFinder鈥檚 primary source of funding, paused its financial support in 2021. QuakeFinder鈥檚 employees are now doing research in their spare time as they work other jobs, while the magnetometers in their network are beginning to go silent one by one as their batteries die.

In general, scientists seem torn over the value of this kind of research. Even Hough, who is sceptical that we will ever find reliable precursors, can鈥檛 help but sometimes ponder the possibility of success.

Three decades ago, she went to Joshua Tree in California to investigate the aftermath of a large earthquake. She heard from local ranchers that their horses had spent the night before the quake 鈥渟creaming鈥�. Is it possible the animals somehow knew what was coming? Hough鈥檚 scientific brain urges her to throw out this kind of fanciful idea. 鈥淭he screaming horses鈥� it鈥檚 not a meaningful scientific observation,鈥� she says. 鈥淏ut at the same time, you wonder.鈥�