Machine Learning to Better Predict Earthquakes

Published Oct-06-19

Breakthrough:
Machine learning and neural networks to improve earthquake hazard assessments.

Company:
The Zoo Team, Austria

The Story:

Machine Learning to Better Predict Earthquakes Earthquake prediction deals with specifying the time, location and magnitude of future earthquakes. This is an immature branch of science and while it is possible to identify areas where earthquakes are more likely to take place it is currently not possible to predict exactly when or where the natural disasters are going to happen.

“Neither the USGS (the U.S. Geological Survey) nor any other scientists have ever predicted a major earthquake. We do not know how, and we do not expect to know how any time in the foreseeable future,” reads a statement on the website of the USGS.

There are many factors that prevent us from predicting earthquakes, chief among them is the lack of unequivocal precursory signals and insufficient knowledge about how they occur and what happens during the start of an earthquake.

Open Innovation Approaches

However, that's not to be too downcast because there are numerous endeavors worldwide attempting to boost our predictive power. They include the use of open innovation and machine learning.

Current scientific studies related to earthquake forecasting focus on where the event will take place, and how large it will be. In 2019 the Los Alamos National Laboratory, the Department of Energy and university partners hosted a contest on the Kaggle platform that challenged participants to predict the when. Specifically, exactly when a laboratory-simulated quake would strike.

More than 4,500 international teams took part and they had to predict the timing of earthquakes from historic seismic data from a well-known experimental set-up used to study earthquake physics.

“Crowdsourcing for new approaches in earthquake forecasting helps us leverage a wide range of expertise in addressing one of the most important problems in Earth science, because of the devastating consequences of large quakes,” said Bertrand Rouet-Leduc, a Los Alamos researcher who prepared the data for the competition.

Winners

Submissions were evaluated using the mean absolute error between the predicted time remaining before a laboratory earthquake and the actual time remaining. There were three winners who came very close to perfect time forecasting and who shared the $50,000 prize money.

The top team, called The Zoo devised a solution that involved a decision-tree approach applying four statistical features. “The final prediction model is based on a combination of state-of-the-art machine learning models in the areas of neural networks and gradient boosted decision trees,” commented Austrian computer scientist and team leader Philipp Singer.

Better Predictions to Save Lives

Bertrand Rouet-Leduc was left in no doubt about the potential of the winning submissions.

“The winning teams’ results could have the potential to improve earthquake hazard assessments that could save lives and billions of dollars in infrastructure.”

Share on      
Next Story »

What Can we Solve for You?
Crack_in_Bridge_Street_from_the_2010_Canterbury_earthquake.jpg