Artificial intelligence is poised to take over earthquake monitoring. It can help better locate the origin of earthquakes and also predict how devastating they might be.
During an earthquake, different types of seismic waves travel through the earth. The first to arrive at any location are called P-waves, which compress and decompress the Earth鈥檚 crust, causing the ground to move back and forth. The more dangerous are the S-waves that come next, which cause the Earth to move up and down.
Normally, to pick up these waves, seismic networks look for a sharp uptick in the amount of energy in the waves shaking the ground, when compared against the long-term average energy due to background seismic activity. Data from at least three different locations are then used to determine the quake鈥檚 epicentre. Computer algorithms pick the P and S waves, which are then reviewed and refined by expert seismologists.
Advertisement
To improve the process, of Stanford University and his team built and trained an artificial intelligence algorithm using more than 700,000 recordings of earthquakes in Northern California. The algorithm was fed the data and was also told which waves were P waves and S waves amid the seismic din. It learned to recognize the signatures of the two types of waves.
The researchers then tested the algorithm on nearly 80,000 raw recordings, where the P and S waves were not identified beforehand, and compared the results with the standard computer algorithm in use today. The AI did significantly better, particularly for the more important S-waves. It was able to pick nearly 20,000 S-waves that matched the assessment of a human expert, compared to just 1100 correctly picked by the standard computer algorithm.
Precisely determining the speed and direction of S waves is essential for making maps of the earth鈥檚 crust. S waves travel at different speeds through different parts of the earth, depending on the underlying geological structure. By building up a picture of how a given region of earth responds to S waves, it鈥檚 possible to predict how much it鈥檒l move when future earthquakes strike. 鈥淲e would want to do it comprehensively for all of California, and ideally for any earthquake threatened part of the world,鈥� says Beroza.
of the Caltech in Pasadena, California, says that it is 鈥減remature to conclude too much just yet鈥� about the neural network鈥檚 efficacy, but is nonetheless impressed. 鈥淭he results look very promising,鈥� he says.
Reference:
