Google and Harvard University developed AI models that could predict earthquake aftershocks

This model has been studied from the synthesis of 199 large earthquake events in recent decades, and from the data of 130,000 aftershocks , is thought to be more accurate than the methods being used to predict surplus today.

Researchers from the AI ​​division of Google and Harvard University created an artificial intelligence model (AI) capable of predicting aftershocks up to a year after a major earthquake. . This model has been studied from the synthesis of 199 large earthquake events in recent decades and from data of 130,000 aftershocks, which is thought to be more accurate than the methods being used to predict aftershocks. nowadays.

This earthquake prediction model is thought to be more accurate than the methods being used to predict today's aftershocks.

The aftershocks after the earthquake in the dataset used to develop and train AI took place in a perimeter stretching 50km vertically and 100 km horizontally from each earthquake epicenter.

Data used to train AI models come from notable earthquakes such as the 2004 Sumatra earthquake, the 2011 Japan earthquake, the 1989 Loma Prieta earthquake in the San Francisco Bay Area and the Northride earthquake in 1994 near Los Angeles.

The results were published today in the journal Nature. Research conducted by authors Phoebe DeVries comes from the Department of Planetary and Earth Sciences at Harvard University, along with Google's machine learning researchers Martin Wattenberg and Fernanda Viégas, and division director of recruitment. Goolge Ai Brendan Meade.

No real seismologist is involved in the study, although DeVries and Meade see them as earth scientists.

Lessons learned during AI modeling training will be used to explore a bigger question: What causes earthquakes?

'We can look at what's going on from this network and really understand it, it really shows a number of different physical theories that cause earthquakes, and it leads us in a fun new way. taste , 'Meade said. This artificial intelligence model was developed using historical data of major earthquakes over the years, but later, it will be used to inform future earthquake data.