Hundreds of earthquakes occur every day. Fortunately, most are so small that we would never know about them without the help of sensitive seismometers. In daily earthquakes only three to six feet of the fault plane slips; humans cannot feel the shaking. In magnitude 5.0 earthquakes a mile or two of the fault plane ruptures; humans can easily feel movement, but modern buildings can withstand it. At magnitude 8.0 the rupture propagates for hundreds of miles across the fault plane, and the tear can extend up to the surface. It will rip a building in two.
By monitoring the buildup of strain between earthquakes, seismologists know that many areas of the crust are close to failure. But the detailed structure of the faults deep below the surface also plays an important role in both the nucleation and propagation of earthquake ruptures—a structure that cannot be sampled directly. For this reason, most seismologists do not believe it is possible to create a forecasting system capable of predicting a large earthquake hours or days before it strikes. For the foreseeable future, the best anyone will be able to do is to quickly detect a large earthquake and sound the alarm.
A few unique characteristics of earthquakes aid in this task. What we perceive as one extended jolt actually comes in stages. Energy from a break in the crust travels through the earth in two forms: P-waves and S-waves. Both types leave the fault surface at the same time, but there the similarities end. P-waves, like sound waves, are compression waves. They travel relatively quickly, but they do not carry much power. During an earthquake, you feel the P-waves as a sudden, vertical thump. S-waves are more like ocean waves, slow movers that contain most of the energy and bring the strongest shaking. The ground motion is horizontal and vertical, and they can bat entire buildings around like they were dinghies in the surf.
In addition, not all waves look alike; they take on different shapes depending on the size of the slip patch. The P-wave radiation for small slip patches has relatively low amplitude and high frequency—a small but sharp pulse. Bigger earthquakes rupture larger areas of a fault and have more slip, so the P-wave is larger in amplitude and lower in frequency. It is akin to the difference between the squeak of a small bird and the roar of a grizzly bear.
A single seismometer could estimate the magnitude of the earthquake based on just this information. Any P-wave with high amplitude and low frequency would trigger a warning. This single-station approach is the fastest way to give warnings near the epicenter. Yet the character of earthquake ruptures varies—not all magnitude 5.0 earthquakes look the same—and the specific sediments underneath the seismometer modify the P-wave. This variability increases the risk of both false alarms—warnings when there is no earthquake—and missed alarms when a damaging earthquake is under way.
To reduce the likelihood of both false and missed alarms, we can combine data recorded by several seismometers located a few miles apart. In this setup the sediments beneath each instrument would be different, so we can obtain an average estimate of the magnitude. This approach requires seismic networks that transmit instrument data to a central site and then integrate them. Yet it takes a few seconds to transmit and analyze the data, and in every passing second the damaging S-wave travels another two to three miles.
The best approach is thus to combine the single-station and network-based approaches, which provides the potential for both rapid warnings in the region near the epicenter and tens of seconds of warning to locations farther away.
Any system has to make a trade-off between the accuracy and the warning time available. As the seismic network collects more data on an earthquake, the predictions will improve, but the time until shaking will decrease. Some users may tolerate more false and missed alarms to have more warning time. For example, schools may prefer to get the warning sooner so children can take cover. A few false alarms a year provide the regular drills necessary so that everyone knows what to do. Nuclear power stations, in contrast, require only a second to shut down the reactor—but doing so comes at great cost. Operators there will want to wait until extreme shaking is certain.