An international team of researchers recently examined the potential for new a model for predicting earthquake aftershocks using a subset of machine learning known as deep learning. While government agencies and earthquake researchers currently use a longstanding model known as ETAS (Epidemic Type Aftershock Sequence) to predict earthquake aftershocks, it has been found to grapple with the large earthquake datasets available today.
Inspecting earthquake damage after an earthquake in Puerto Rico in 2020. (Credit: United States Geological Survey)
This new model, known as RECAST (Recurrent Earthquake forecast), demonstrates the potential for using deep learning to better predict earthquakes and has been recently published in detail in Geophysical Research Letters.
“The ETAS model approach was designed for the observations that we had in the 80s and 90s when we were trying to build reliable forecasts based on very few observations,” said Dr. Kelian Dascher-Cousineau, who is a recent PhD graduate at UC Santa Cruz, and lead author of the study. “It’s a very different landscape today.”
For the study, the researchers compared the new deep learning RECAST model to the traditional ETAS model on earthquake catalogs from Southern California. They discovered that while RECAST performed moderately better than ETAS at predicting aftershocks, they also found the time computational energy required to produce those predictions was enormously better than ETAS for bigger earthquake catalogs, which is attributed to RECAST’s deep learning capabilities where it can learn how it goes, meaning RECAST is more flexible than ETAS.
“We’ve started to have million-earthquake catalogs, and the old model simply couldn’t handle that amount of data,” said Dr. Emily Brodsky, who is a professor of earth and planetary sciences at UC Santa Cruz and a co-author on the study.
It is because of this flexibility from the deep learning capabilities that RECAST could be used to learn from data spanning multiple areas around the world and provide better predictions about poorly examined earthquake areas. For now, the team aspires to continue conversations regarding the use of deep learning for earthquakes.
How will RECAST better predict earthquake aftershocks in the coming years and decades? Only time will tell, and this is why we science!
As always, keep doing science & keep looking up!
Sources: EurekAlert!, IBM, Geophysical Research Letters