The AI model can potentially be used to mitigate the impact of earthquakes. (Image Credit: Nadiia Ganzhyi/Unsplash).
Artificial intelligence developed by scientists from the University of Texas at Austin has correctly predicted 70 per cent of the earthquakes over a seven-month long trial run in China. The AI was trained on statistical bumps in realtime seismic data during the time leading up to previous earthquakes. The AI monitored realtime seismic activity and provided weekly forecasts, and successfully predicted 14 earthquakes within 320 kilometres of the epicentres. The AI was particularly accurate when it came to predicting the strengths of the Earthquakes. There were eight false warnings, and the AI missed predicting one earthquake.
Predicting earthquakes has been a historically challenging endeavour, one that was suspected by some scientists to be impossible. The trial was conducted as part of an internal competition in China. A paper describing the findings from the trial has been published in the Bulletin of the Seismological Society of America. Member of the research team, Sergey Fomel says, “Predicting earthquakes is the holy grail. We’re not yet close to making predictions for anywhere in the world, but what we achieved tells us that what we thought was an impossible problem is solvable in principle.”
When it comes to earthquake prediction 70 per cent accuracy is sufficiently high to help minimise economic and human losses, and can help cities and towns better prepare for earthquakes in the future. The AI model is pretty straightforward, and has been trained on five years of seismic data. The AI then bases its forecasts based on realtime seismic inputs. The researchers are confident that in regions with denser seismic monitoring networks, the predictions and accuracy of the AI model can be significantly improved. These regions include Turkey, Japan and the USA.
Also Read | Can earthquakes be predicted? As forecasts swamp news, we seek answers
The researchers next intend to conduct another trial run in Texas, which boasts of over 300 seismic stations and over six years of continuous data, making it an ideal testbed for verifying the results. The researchers also intend to integrate physics-based models into the technology, which is expected to give a boost to the accuracy of predictions in regions where data availability is poor.
