Quiet hints preceded Kobe earthquake.

Months before last January's devastating earthquake, the ground beneath Kobe, Japan, started showing signs of an impending crisis. Subsurface water displayed chemical changes that intensified in the days before the disaster, two teams of Japanese scientists report in the July 7 Science.

The chemical hints came to light only after the Jan. 17 quake, so they could not help scientists predict the catastrophe, which killed more than 5,000 people. The findings are important, however, because they represent some of the best-documented cases of precursory phenomena--signals produced before a quake, says Urumu Tsunogai from the University of Tokyo.

Tsunogai and Hiroshi Wakita studied the concentration of dissolved chloride and sulfate ions in groundwater. The researchers had not collected water samples before the quake, so they took the innovative step of buying bottled water pumped from an aquifer near the quake's epicenter. Kobe's high-quality mineral waters are collected for drinking and for brewing sake, a rice wine. Because the bottles are stamped with a date, Tsunogai and Wakita could track chemical changes that occurred before and after the tremor.

From mid-1993 to mid-1994, concentrations of chloride and sulfate ions remained constant. Five months before the quake, the concentrations started climbing; they peaked in late February, after the jolt.

George Igarashi of Hiroshima University and his colleagues documented prequake changes in the concentration of dissolved radon gas measured at a monitoring site 30 kilometers from the tremor's epicenter. Radon readings began rising slowly in October 1994 but jumped dramatically 9 days before the quake. Readings then dropped abruptly.

Both sets of researchers suggest that the chemical signals stem from geologic changes leading up to the quake. According to Igarashi, microcracks may have developed in the rock months before the main shock, allowing increased amounts of radon to dissolve in underground fluids. To explain the drop 9 days before the quake, he proposes that accumulating strain in the crust started to seal cracks immediately before the tremor.

According to Tsunogai, the new finds demonstrate that the crust began to prepare itself long before the Kobe earthquake. "This time scale may be very important for understanding the mechanism of earthquakes," he says.

Chi-Yu King of the U.S. Geological Survey in Menlo Park, Calif., calls the new findings "a small but significant step in a long journey toward achieving earthquake prediction."

Researchers in Asia have observed several cases of groundwater precursors in the past. But U.S. geoscientists have remained skeptical of such reports because they could not rule out other causes of the precursors.

King cautions that scientists must observe many well-documented cases of chemical precursors before they can try using them for predictions. Even then, a prediction program would have to include various prequake clues, because groundwater changes might not precede all earthquakes.