Illustration: bryan christie design

Kobe, though, drove home the point that “we don’t know where earthquakes will occur in the future,” says Fujinawa. He had spent much of his career doing earthquake­prediction research, looking at changes in Earth’s electro­magnetic fields and other phenomena as possible quake precursors. But he acknowledges that decades of research failed to produce a method of reliably predicting a quake days or even just hours in advance. “I still think the idea is worthwhile,” he adds, but he says it would take billions of dollars to build the kind of research infrastructure needed to confirm or refute current prediction theories.

And so a different idea began to take hold: creating an automated system that could quickly detect the first rumblings of a budding earthquake and then issue a real-time warning.

The concept is simple: an earthquake generates several types of waves, which travel at different speeds. In particular, non­destructive P waves (P for primary) propagate out from a ground rupture at about 6 or 7 kilometers per second, while the much stronger S (secondary) waves that cause most of the damage travel only at about 3 or 4 km/s. The farther you are from the epi­center, then, the greater the difference between the P and S waves’ ­arrivals, and the longer it takes the strong shaking to reach you. So the seismic station closest to the epicenter picks up the P waves first and relays a warning to a central data center, which then quickly estimates the likely epicenter and magnitude.

In a matter of seconds, the data center issues an alert to locations nationwide, ideally before the S waves arrive there and often even before the P waves are felt.

Although it would improve the accuracy of the estimates to wait for two or more stations to report in, that’s a luxury the system can’t afford. “There’s a tradeoff between rapidness and accuracy,” says Osamu Kamigaichi, senior coordinator for international earthquake and tsunami information at the Japan Meteorological Agency (JMA), which oversees the distribution of the early­warning alerts. “Basically, the quicker, the better.” The initial alert is followed by several follow-up alerts—at 10 seconds, 30 seconds, 50 seconds, and so on—and any corrections can be made then.

As it is, the advance warning amounts to mere seconds or, at most, tens of seconds. After an earthquake struck off the coast of northeastern Japan on 16 August 2005, the closest seis­mometer, in the port city of Ishinomaki, took 4.5 seconds to issue an alert; 10 seconds later, the S wave arrived there. But in the city of Kawasaki, about 170 kilometers from the epicenter, it arrived 22 seconds after the first alert.

Twenty-two seconds may not sound like a lot, but for anything controlled by a microprocessor, it’s usually more than enough time to react, Kamigaichi notes. Elevators, for instance, tend to jam between floors when strong shaking sends their cables swinging. After a ­magnitude 6.0 earthquake in Chiba prefecture in 2005, 78 elevators got stuck. But with, say, a 6-second lead time, a moving elevator could stop at the ­closest floor and open its doors. Since April, in fact, all 227 000 elevators in Japan must now be equipped with control systems that accept JMA’s signals.

The alerts could also trigger automated responses at power and chemical plants, dams, computer and communications networks, trains, and hospitals. And that’s not all. Fujinawa’s REIC now has more than 120 ­member companies, all of which see a lucrative market in selling early-warning products and services. These include home systems that could automatically shut off gas valves and unlock doors, portable earthquake beepers that would flash a light or sound a chime when an alert is received, and data-­distribution services that take the JMA alerts and relay them to ­customers. Last October, Nikkei Weekly reported that some analysts predict a 300‑­billion-yen ($2.5 billion) market for such applications.

According to Caltech’s Heaton, early-warning systems “hold the greatest promise in the largest earthquakes.” In those massive upheavals, he explains, the fault rupture occurs over a large area and can take some time to develop—maybe as long as 5 minutes. “So while it’s still rupturing, you can predict how large the earthquake is likely to become, given its current size.”

Of course, to ensure that you can detect the first rumblings of any earthquake, you need a dense network of seismometers blanketing the entire country. Japan’s arrangement of about 1000 stations also includes high-speed communication links connecting each station to a central data-processing center in Tokyo, plus software to do the seismic calculations and distribute the alerts. The seismic stations cost several billion dollars; the five-year effort to create the early-warning network and its attendant applications began in 2002 and is expected to eventually cost about $160 million.

Japan didn’t start entirely from scratch. The country already had a number of stations, used by government and academic scientists to study seismic phenomena and to prepare public bulletins immediately following an earthquake. (If you’ve been in Japan during a sizable earthquake, you may have noticed that TV stations there broadcast announcements within a few minutes of the actual event.) Separately, Japan Railways had developed its own automated early-warning sensors along its bullet-train routes; they offered strong proof that the concept of early-warning alerts could work. [For more on that system, see “A Brief History of Earthquake Warnings,” on the IEEE Spectrum Web site.]

In the current network, JMA runs about 200 of the stations; another 800 or so stations are maintained by the National Research Institute for Earth Science and Disaster Prevention (NIED). A few dozen others belong to universities. The stations come in various shapes and sizes. JMA’s are located on the surface; inside each concrete container sit a coffee-can-size strong-motion seismometer (which detects the strong shaking most likely to damage structures) and a similarly proportioned ­seismic-­intensity meter (designed to measure lower levels of shaking). NIED’s stations are all buried in boreholes—some as shallow as 100 meters, some as deep as 3 km—to help isolate the instruments from artificial noise.

The communications links also differ. JMA uses leased lines, which can transmit a signal from the station to the JMA data center in about 0.2 seconds. NIED’s signals travel through the regular Internet—which takes an average of 1.5 seconds but is much less expensive.