Why can't scientists predict earthquakes, and what can they do instead?

Scientists cannot predict earthquakes. A true prediction means saying when, where and how big a future earthquake will be, within limits, and nobody has ever done that reliably. In the 1970s many scientists thought it was just a matter of time. By the 1990s, continuing failures made many question whether it is possible at all.

The best-known test happened at Parkfield, on the San Andreas Fault in California. Moderate earthquakes had struck there in 1857, 1881, 1901, 1922, 1934 and 1966, which looked like a rhythm. In 1985 two scientists predicted a 95 percent chance of the next one by 1993 at the latest. The USGS and California installed one of the most sophisticated networks of instruments to catch it. Then 1993 came and went. The quake arrived on 28 September 2004, ten years late, with no obvious warning signs. Nature does not keep a calendar.

Science is also entangled with human risk. After the 2009 L'Aquila earthquake in Italy, where 308 people died, six scientists and an ex-government official were convicted of manslaughter in October 2012 for downplaying the likelihood of a major earthquake. One account stresses that the charge was not so much failing to predict the quake as giving undue assurance beforehand. In November 2014 an appeals court quashed the scientists' convictions; only the official's stood, with a reduced sentence. The case shows how hard it is to talk about risk when you cannot predict.

What we can do is warn in seconds. When an earthquake starts, the fast but weaker P waves arrive before the slower, damaging S waves. A network of seismometers detects the P waves and sends an alert faster than the shaking travels. That can give a few seconds to move somewhere safer.