LOS ANGELES (AP) _ Soviet and U.S. scientists say they have moved closer to predicting strong earthquakes by developing a mathematical formula that showed how certain seismic patterns preceded 16 of 20 big jolts.

''This is a step toward improving the prediction of large earthquakes in time and space,'' said geophysicist Leon Knopoff, a geophysics professor at the University of California, Los Angeles, and a co-author of the study being published Thursday in the British journal Nature.

The new method hasn't yet been used to predict earthquakes and won't allow scientists to predict that a quake will occur in a precise time and place.

But it will allow them to make more general predictions saying that a strong earthquake is likely within three years somewhere in a broad region, such as Southern California or the northern Appalachians, Knopoff said Wednesday.

That would be an improvement over scientists' present ability, which is mostly limited to making long-range forecasts.

For example, the U.S. Geological Survey predicted in July that within 30 years, a major or great quake measuring 7 to 8 or more on the Richter scale is 60 percent likely along the San Andreas Fault in Southern California and 50 percent likely along the Hayward or San Andreas faults in the San Francisco Bay area.

Knopoff conducted the study with Vladimir Keilis-Borok and I.M. Rotwain, both of the Soviet Academy of Sciences' Institute of the Physics of the Earth, and California Institute of Technology geophysicist-geologist Clarence Allen, the former chairman of the National Earthquake Prediction Evaluation Council.

If the new method is proven valid, it will help government officials and private citizens because ''even a vague description of the area that could be at risk and a refinement in the time scale is valuable,'' said Mike Guerin, spokesman for California's Office of Emergency Services.

Knowing that a big quake is likely within three years in a certain region ''sets a deadline in which they need to accomplish certain preparations and plans,'' he added.

But such a vague prediction would make it ''very difficult to do anything more than is already being done in California in terms of preparedness,'' countered Randall Updike, executive secretary of the quake prediction council and deputy chief of the Geological Survey's Office of Earthquakes, Volcanoes and Engineering.

The council reviewed the new method during a meeting in June. Members decided that while it was a step forward and should be studied further, it would be premature to use it immediately to predict three-year periods in which large quakes were likely, said James Davis, a council member and chairman of the California Earthquake Prediction Evaluation Council.

The Soviet-UCLA-Caltech team devised a complicated mathematical formula to identify ''times of increased probability'' - roughly three-year periods in which certain patterns of seismic activity are likely to culminate in a strong earthquake.

They developed the formula by studying seismic patterns that preceded 14 earthquakes of 6.4 magnitude or more that rocked California and Nevada between 1938 and 1984.

The scientists then tested the formula by applying it to 20 quakes in five other regions of the world since 1963. They found the formula, used in hindsight, successfully ''predicted'' 16 of those quakes. In other words, the formula found that 16 of those quakes were preceded by ''times of increased probability'' for large temblors.

The patterns incorporated into the formula included above- or below-normal numbers of smaller jolts, the rate at which such quakes increased or decreased, the size of those smaller quakes, the number of aftershocks following such quakes and whether such quakes in one area were followed immediately by quakes in a distant area.