The ground heaves, an earthquake is born. Underground sensors along fault lines detect rumblings humans can't and relay signals to a central computer. Precious seconds before anything is felt, wailing sirens blare that a big one is on its way.

That sliver of time could be used to warn people to flee from windows and take cover. Companies such as gas and electric utilities could take actions to protect their systems. And speeding trains could have enough time to brake to a halt.

Such alert systems already exist in parts of Japan, Mexico, Taiwan and Turkey where the main users are businesses such as railway companies, power plants and manufacturers.

But that's not the case in the United States — except for a handful of schools, firehouses and airports that use commercially available, battery-powered seismic gadgets that warn a limited region.

This summer, the U.S. Geological Survey is cautiously taking another look at early warning, beginning with a three-year test to gauge how well three experimental systems around California would work in the real world.

Scientists will gather data without broadcasting alerts to residents or businesses. While a working system is still years and tens of millions of dollars away, many see the pilot project as a first step toward catching up with the rest of the world.

Many existing global networks sprang from killer quakes, and some scientists fear it would take a catastrophe to jolt the U.S. into action.

"If the capital of the United States were Los Angeles, we would have had an early warning system a long time ago," said Tom Heaton, professor of earthquake engineering at the California Institute of Technology and developer of one of the test systems.

It's not the first time the idea has surfaced in the United States. A decade ago, while Southern California was upgrading its seismic monitoring stations following the deadly 1994 Northridge quake, some scientists had high hopes the U.S. would develop its own system. But liability issues and fears of false alarms hampered progress. Finally, federal funding ran out.

Early warning systems can't predict quakes. Instead, a sprawling network of sensors near the epicenter estimates a temblor's size once the ground ruptures and sounds an alarm before shaking starts by exploiting the lag time it takes for different seismic waves to travel to the surface.

This is possible because shock waves coursing through the Earth move slower than electronic signals transmitted via a computer data network.

Weak tremors called primary, or P waves, spread from the epicenter and travel faster than the destructive shear, or S waves, which cause severe shaking. A warning is issued when the P wave signal passes a certain intensity, though there's debate about how much to read into the signals.

The amount of forewarning depends on the distance from the epicenter, so early warning won't work for areas directly above the ruptured fault, where the P and S waves are nearly simultaneous and shaking is the most intense. According to some estimates, communities radiating miles from the deadly San Andreas fault — which cuts through most of California and caused San Francisco's disastrous 1906 quake — could receive up to 60 seconds of warning.

Supporters say that gives communities a fighting chance to reduce injuries and damage. Skeptics contend seconds are hardly enough time to make a real difference and say there's no evidence that existing systems have saved significant lives.

"When people think of early warning, they think they'll get time to do something useful before shaking starts," said David Wald of the USGS National Earthquake Information Center in Colorado. "But in reality, you don't get a lot of time."

Japan pioneered seismic early warning in the 1980s with a system that automatically halted its high-speed trains during major quakes. Mexico, Taiwan and Turkey followed, though their alert systems are much less complex.

Since 2004, Japan has been experimenting with an advanced system that warns a select group of 250 government agencies, businesses and schools. The government plans to expand the service this summer and eventually make it broadly available after educating the public about its limitations, said Makoto Saito of the Japan Meteorological Agency.

Early warning isn't foolproof: About 10 percent of the 245 warnings that the newest Japanese system sent between February 2004 and last August were false alarms.

Mexico's $2 million alert system, installed after the 1985 quake flattened much of Mexico City, was suspended twice during its inaugural year after a computer glitch failed to detect a major quake and a false alarm caused a subway stampede. Mexican officials tweaked the system so that an alert is sent only if there is confirmation from two seismic stations instead of just one.

The USGS project will test three experimental early warning systems — two developed by Caltech and one by the University of California, Berkeley. Initial funding will start at $250,000.

Scientists have tested the three computer algorithms using historical earthquake data to see how well they can accurately estimate a quake's magnitude, or how much energy is released. For example, the Berkeley model can correctly guess a quake's size from as little as four seconds of data.

The trick will be to see how well the models fare in real time.

The three models will be incorporated into existing seismic networks around Los Angeles and the San Francisco Bay area, which monitor the hundreds of quakes that hit the state each year including those too small to be felt. The systems currently gather information about a quake's strength, location and intensity minutes after it occurs.

It'll be up to the experimental systems to sort incoming data and estimate a quake's size within seconds. Scientists will then compare the computer's guess to the actual magnitude and monitor for false alarms.

Even if the models work, the U.S. must first upgrade its seismic monitoring networks and resolve the nagging cost issues of building a national system. Congress has only funded a fraction of the $170 million needed to upgrade and install digital sensors in seismically active zones.

Still, early warning proponents are encouraged.

"Early warning isn't a panacea," said Richard Allen, a Berkeley seismologist who developed one of the models. "But I do think it can reduce the impacts of earthquakes."