The convoy was more than 30 kilometres from the Kunlun fault in Tibet when the Jeeps suddenly lurched. They had hit a series of parallel cracks, remnants of a magnitude 7.8 earthquake that struck the year before. "It was like driving on steps," recalls Yann Klinger, a geologist from the Paris Institute of Geophysics in France. The cracks were clear signs that the ground had been squeezed like a sponge then released, violently wrenching it apart. Yet they were much too far from the fault line to be explained by the quake. Mystified, the team took some measurements and moved on.
It transpired that Mr Klinger and his team had stumbled upon the aftermath of a supershear earthquake - one that slipped at such blistering speeds that the rip in the Earth overtook its own seismic waves. This created the earthquake equivalent of a sonic boom, capable of striking anything in its path like a hammer blow. While some seismologists had suspected such a quake could happen, physical evidence of their power had been lacking.
Until supershear quakes came on to the scene in the late 1990s, earthquakes were thought to come with an inbuilt speed limit, some 3.5 kilometres per second. For many years only one observation contradicted this conventional wisdom: in 1984, Ralph Archuleta at the University of California, Santa Barbara, reported that the Imperial Valley earthquake that struck California in 1979 briefly ruptured faster than 3.5 kilometres per second.
These observations languished in obscurity for nearly two decades until a wager between an engineer and a geologist meant that they were finally tested in a laboratory. Ares Rosakis at the California Institute of Technology in Pasadena had been investigating how explosions affect materials that have been glued together, and had seen supershear ruptures occur along the glued interface. So why not in the Earth itself? His sceptical colleague Hiroo Kanamori, in the geology department at Caltech, disagreed. The bet was set - an expensive bottle of wine was at stake.
To simulate an earthquake, the scientists took two slabs of a polymer that transmits light when under pressure and pressed them together, the joint representing a geological fault. They shone a light through the fault zone and then triggered a tiny electrical pulse to produce a rupture along the fault line. The patterns made by the light allowed them to see the seismic waves produced as the rupture moved through the fault. Sure enough, the quake produced seismic waves - first compressional waves, followed by the shear waves. And as Mr Kanamori had predicted, the rupture itself trailed well behind its seismic waves.
With Mr Rosakis on the verge of losing the bet, they put the slabs under slightly higher pressure by squeezing the fault tighter. Then, when they triggered a rupture, something odd happened: a fresh "daughter" crack suddenly appeared ahead of the main "mother" rupture, travelling much faster, leapfrogging the "forbidden" speed. Not only that, it continued to produce new shear waves, which added to the first batch to produce a new, more powerful shock wave that trailed behind the rupture in the shape of a boat's wake.
These lab experiments began to show that earthquakes could, in theory, go supershear. But it was the Earth itself that provided the real-world evidence. In 1999, the most seismically active continental fault of the 20th century - the North Anatolian fault in Turkey - slipped to cause the magnitude 7.6 Izmit earthquake. It all added up to a quake that went supershear, says Michel Bouchon at the University of Grenoble in France, who led one of two teams that independently showed that Izmit reached velocities of up to five kilometres per second.
There is now evidence that at least three major quakes around the world since Izmit have gone supershear, including Kunlun, where Klinger's team had found the then-mysterious cracks. Thankfully, there have only been a handful of such quakes recently and most have been in remote areas. This will not always be the case, of course. Some geologists suspect that the devastating San Francisco earthquake of 1906 may have been a supershear.
Understanding earthquakes after the event is only half the battle, however. What everyone wants to know is where the next one might hit. Now the seismologists David Robinson and Shamita Das and their colleagues at the University of Oxford think they have come up with an answer. They compared known supershear quakes for similarities and used these to try and anticipate where in the world the next one is most likely to strike.
All of the supershear ruptures seen so far have been on long, straight sections of faults. This might be because a rupture cannot accelerate to supershear speeds on a convoluted fault path. They looked for unbroken faults on land that do not deviate by five degrees or more over a distance of 100 kilometres. That narrowed it down to 26 sections on 11 different fault systems around the world, including parts of the San Andreas Fault in California. He called them "superhighways".
Worryingly, when they added the population distribution within a 50-kilometre radius of these faults, they found a network of superhighway faults primed to rumble near major cities. Seven of the 26 superhighways lie within reach of heavily populated areas, each potentially affecting more than two million people. One runs straight through the middle of San Francisco, while the cities of Yangon and Mandalay in Myanmar sit at either end of the longest superhighway.
If Mr Robinson's maps are correct, it could mean that regions previously thought to be outside of the worst effects of an earthquake, and maybe even beyond its reach altogether, could be caught unawares by a supershear quake. Unfortunately, most city planners and civil engineers are unlikely to take heed of the warnings of seismologists based on laboratory experiments. What is needed now is more data on actual quakes that go supershear. As geologists wait for the next big one to strike, however, they are hoping that they will be proved right in an uninhabited desert and certainly nowhere near a big city.