Exciting news! We're transitioning to the Statewide California Earthquake Center. Our new website is under construction, but we'll continue using this website for SCEC business in the meantime. We're also archiving the Southern Center site to preserve its rich history. A new and improved platform is coming soon!

Earthquake Forecasting Using Machine Learning Algorithms

Alexei C. Shatz, Julie Pastorino, Jared T. Santa Maria, Yuner Lu, Yonatan A. Gugsa, Daniel E. Molina, William H. Savran, & Thomas H. Jordan

Published August 1, 2019, SCEC Contribution #9313, 2019 SCEC Annual Meeting Poster #307

The San Andreas Fault System (SAFS), which comprises of the San Andreas, Hayward, Elsinore, and San Jacinto faults, has been experiencing an absence of Mw 7.0 or greater earthquakes for a period of 100 years or more. Due to the absence of large earthquakes, we expect an increased probability of an earthquake of Mw7.0 or greater on the SAFS. The SAFS poses a significant source of seismic hazard in California, resulting in increased risk to millions of residents and billions of dollars of potential damage. We focus on forecasting time-dependent earthquake probabilities for 30-year intervals, given the recurrence intervals of the fault system in the previous 100-year period. Our primary goal is to forecast the likelihood of Mw 7.0 or greater events on the SAFS given no observed earthquakes of Mw 7.0 or greater in the previous 100 years. We apply three different machine learning algorithms to forecast time-dependent earthquake probabilities — logistic regression, neural network, and a random forest classifier. These models determine the marginal probability of rupture on a specified fault section given information about the state of the SAFS. We compare the different approaches using a log-likelihood ratio relative to a time-independent forecast. We find the logistical regression method shows increased log-likelihood scores with respect to the baseline forecast, while the random forest and the neural network show the opposite result. This research will give insight into the effectiveness of applying machine learning techniques to forecast time-dependent earthquake probabilities.

Shatz, A. C., Pastorino, J., Santa Maria, J. T., Lu, Y., Gugsa, Y. A., Molina, D. E., Savran, W. H., & Jordan, T. H. (2019, 08). Earthquake Forecasting Using Machine Learning Algorithms. Poster Presentation at 2019 SCEC Annual Meeting.

Related Projects & Working Groups
Communication, Education, and Outreach (CEO)