The details of how the current is generated remain something of a mystery. One theory is that the deformation of the rock destabilizes its atoms, freeing a flood of electrons from their atomic bonds, and creating positively charged electron deficiencies, or holes.

One of us, Freund, working at NASA Ames Research Center in Mountain View, Calif., demonstrated through laboratory rock-crushing experiments that the sundering of oxygen-to-oxygen bonds in the minerals of a fracturing rock could produce holes. These holes manage to propagate through rock up toward the surface, while the electrons flow down into Earth's hot mantle. The movement of these charges, measured at 300 meters per second in the lab, causes changes in the rock's magnetic field that propagate to the surface.

Another theory is that the fracture of rock allows ionized groundwater thousands of meters below the surface to move into the cracks. The flow of this ionized water lowers the resistance of the rock, creating an efficient pathway for an electric current. However, some researchers doubt that water can migrate quickly enough into the rock to create large enough currents; for this theory to be correct, the water would have to move hundreds of meters per second.

Whatever the cause, the currents generated alter the magnetic field surrounding the earthquake zone. Because the frequencies of these magnetic field changes are so low—with wavelengths of about 30 000 kilometers—they can easily penetrate kilometers of solid rock and be detected at the surface. Signals at frequencies above a few hertz, by contrast, would rapidly be attenuated by the ground and lost.

We can detect such electromagnetic effects in a number of ways [see illustration, "Signs of Quakes to Come"]. Earthquake forecasters can use ground-based sensors to monitor changes in the low-frequency magnetic field. They can also use these instruments to measure changes in the conductivity of air at the earth's surface as charge congregates on rock outcroppings and ionizes the air.

Using satellites, forecasters can monitor noise levels at extremely low frequency (ELF)—below 300 Hz. They can also observe the infrared light that some researchers suspect is emitted when the positive holes migrate to the surface and then recombine with electrons.

Scientists around the world are looking at all of these phenomena and their potential to predict earthquakes accurately and reliably. One group is at QuakeFinder, a Palo Alto, Calif.­based company cofounded by one of us, Bleier, in 2000. QuakeFinder researchers have begun directly monitoring magnetic field changes through a network of ground-based stations, 60 so far, in California [see photo, "Earthquake Investigator"]. In 2003, the company joined forces with Stanford and Lockheed Martin Corp.'s Sunnyvale, Calif., center to launch an experimental satellite designed to remotely monitor magnetic changes. A larger, more sensitive satellite is in the design stages. QuakeFinder hopes to develop an operational earthquake warning system within the next decade.

The 1989 Loma Prieta earthquake near San Francisco sent out strong signals of magnetic disturbances fully two weeks before the 7.1-magnitude quake occurred. The idea that such signals existed was still a new one then, certainly not well enough accepted to justify a decision to issue a public warning.

We happen to have excellent data from that quake. Stanford professor Anthony C. Fraser-Smith had buried a device called a single-axis search-coil magnetometer to monitor the natural background ULF magnetic-field strength at about 7 km from what turned out to be the center of that quake. He selected this spot simply because it was in a quiet area, away from the rumblings of the Bay Area Rapid Transit trains and other man-made ULF noise. He monitored a range of frequencies from 0.01 to 10 Hz, essentially, the ULF band and the lower part of the ELF band.

On 3 October, two weeks before the quake, Fraser-Smith's sensors registered a huge jump in the ULF magnetic field at the 0.01-Hz frequency—about 20 times that of normal background noise at that frequency. Three hours before the quake, the 0.01-Hz signal jumped to 60 times normal. Elevated ULF signals continued for several months after the quake, a period rife with aftershocks, and then they disappeared.

The Loma Prieta quake was a stunning confirmation of the value of ULF signals in predicting earthquakes. This validation of the theory prompted Bleier to establish a network of earthquake sensors in the Bay Area, an effort that grew into QuakeFinder.

Other researchers around the world who monitored changes in the magnetic field at ULF frequencies had noticed similar, but not as extreme, changes prior to other events. These observations occurred shortly before a 6.9-magnitude quake in Spitak, Armenia, in December 1988 and before a devastating 8.0-magnitude earthquake in Guam in August 1993.

Author Bleier recorded spikes of activity, four to five times normal size, in the 0.2- to 0.9-Hz range for 9 hours before a 6.0-magnitude earthquake in Parkfield, Calif., on 28 September 2003. Solar storms sometimes cause ripples in the magnetic field at those frequencies, but there had been no appreciable solar activity for six days prior to the quake.

In Taiwan, sensors that continuously monitor Earth's normal magnetic field registered unusually large disturbances in a normally quiet signal pattern shortly before the 21 September 1999 Chi-Chi, Taiwan, earthquake, which measured 7.7. Using data from two sensors, one close to the epicenter, and one many kilometers away, researchers were able to screen out the background noise by subtracting one signal from the other, leaving only the magnetic field noise created by the imminent earthquake. Two teams, one in Taiwan and one in the United States, calculated that the currents required to generate those magnetic-field disturbances were between 1 million and 100 million amperes.

Besides detecting magnetic-field disturbances, ground-based sensors can record changes in the conductivity of the air over the quake zone caused by current welling up from the ground. These sensors can vary in form, but those we use are made from two 15-centimeter by 15-cm steel plates locked into position about 1 cm apart. A 50-volt dc battery charges one plate; the other is grounded. A resistor and voltmeter between the battery and the first plate senses any flow of current.

Normally, the air gap between the plates acts as an insulator, and no current flows. If, however, there are charged particles in the air, a current begins to flow, creating a voltage drop across the resistor that registers with the voltmeter. The currents created in this way are not large—on the order of millivolts—but are detectable.

Last year QuakeFinder installed 25 ELF detectors with such air- conductivity sensors in California's Mojave Desert to determine if increased air conductivity actually precedes earthquakes and contributes to the formation of the so-called earthquake lights [see photo "Mysterious Lights"]. But to date, no large earthquakes have struck near these sensors, so no data are available yet.