Every year, thousands die from earthquakes either directly, or indirectly from the fires, landslides and tsunamis that result from it. An earthquake occurs when a fault slips, releasing a burst of energy in waves that cause the ground to shake. Earthquakes are recorded by instruments called seismographs and are measured on the Richter scale — the higher the number on the scale, the more catastrophic the quake.
Because of the destructive nature of earthquakes, it makes sense that preparations should be done to keep earthquake casualties and damage to a minimum. Of course, this requires knowing when and where it will occur. The problem is, in spite of continuous research and technological advancements, the best that geologists can do in terms of earthquake prediction is still just make general guesses based on tectonic plate movement, fault zone locations and earthquake history. And by general guess, we mean that on any specific fault scientists just know that another earthquake will occur sometime in the future but unfortunately they do not know when.
It’s obvious that the world needs more accurate earthquake forecasts than mere reliance on earthquake history patterns or changes in animal behavior. And Bertrand Rouet-Leduc of Los Alamos National Laboratory in New Mexico along with some of his colleagues are aiming to address that.
What the research team did was to make use of an artificial quake system that mimics some real-life earthquake behaviors. The group focused on the sounds produced as an earthquake approaches (equivalent to the sound emitted by a fault as it moves and slips). They recorded those sounds, fed the data into a machine-learning algorithm to see if it could recognize some kind of pattern, and asked it to make a prediction on whether an earthquake was likely to occur or not. Amazingly, the machine was able to make accurate predictions.
As MIT Technology Review reported, the group was quoted as saying: “We show that by listening to the acoustic signal emitted by a laboratory fault, machine learning can predict the time remaining before it fails with great accuracy.” Their theory is that there might be earthquake precursors that are too small to be noticed, which is why geologists may have dismissed those, while the machine regarded them as warning signs.
As promising as the result of their experiment was, the group is quick to point out that it does not automatically mean that the same technique can be used to predict real earthquakes. Still, it is enough to go on direction-wise and they are now planning to use the same kind of analysis on actual quakes that are most similar to the laboratory quakes. Specifically, they are looking to analyze the Parkfield segment of California’s San Andreas Fault which constantly experiences earthquakes and is probably emitting chatter or noise similar to the ones emitted in the lab experiment.
While the next stage of their experiment may go either way, we hope Rouet-Leduc’s team gets positive results so that one day their method may lead worldwide to accurate and timely earthquake predictions.