Exploring the Possibility of Forecasting Strong Earthquakes with Machine Learning

Machine Learning can be used to predict the long-term strength of earthquakes in North America, South America and Japan, as well as Northern India, Southern China, and Southern China.

Our Machine Learning models indicate that there are periods with earthquakes of magnitude >=7, and periods without such earthquakes in the seismic zones analyzed. Our Machine Learning models also predict a new phase of seismic activity for earthquakes >=7 between the years 2040+-5 and 2057+-5, 2024 +-1 and 2026+-1, 2026+-2 and 2030 +-2, 2024 +-2 and 2029+-2, 2024 +-1 and 2028+-2, 2022 +-1 and 2028+-2 for five seismic zones: United States, Mexico South America, Japan and Southern China-Northern India. Our algorithms can also be used to make probabilistic predictions in any seismic zone.

Our algorithm to analyze strong earthquakes can be applied in smaller or specific seismic areas where historical moderate earthquakes of magnitudes between 5-7 occur. This is the case for the Parkfield section on the San Andreas fault, California, United States. Our analysis explains why Bakun and Lindh’s (1985) proposal of a moderate earthquake in 1988 +- 5, as suggested by Bakun, was never realized and why the long-awaited Parkfield characteristic earthquake happened in 2004. Our Bayesian Machine Learning model, which adopts a 35-year periodicity, predicts possible seismic events between 2019 and 2030, with a higher probability around 2025+-2. Parkfield is a great place to test, develop, and demonstrate earthquake forecasts. In a few short years, we will be able to show that our algorithm is able to forecast both strong and moderate earthquakes.