E-EEWS: an Earthquake Early Warning algorithm based on Ensemble Machine Learning

Jean-Paul Ampuero

We introduce the Ensemble Earthquake Early Warning System (E-EEWS) to detect, locate and estimate the magnitude of an earthquake using 3 seconds of P-wave recorded by a single station. The system is based on an Ensemble Machine Learning algorithm that uses attributes from temporal, spectral and cepstral domains extracted from ground acceleration time series. The training set comprises datasets from Peru, Chile and Japan, and the global STEAD dataset. E-EEWS consists of three stages: detection, P-phase picking and source characterization (magnitude, epicentral distance and back-azimuth estimation). It achieves an overall success rate in the discrimination between earthquakes and noise of 99.9 $%$. For P-phase picking and source characterization, the estimates are virtually unbiased. By updating estimates every second, the approach gives time-dependent magnitude predictions that follow the earthquake source time function.

Dr. Ampuero received the Ph. D. degree in geophysics at Univ. Paris VII Denis Diderot and Institut de Physique du Globe de Paris, France, in 2002. After postdocs in Princeton Univ., ETH Zurich and Univ. of Tokyo, he became a Professor at the California Institute of Technology. Since 2018, he is a Senior Researcher at IRD and holds an Excellence Chair at UCA. His research at Geoazur focuses on theory, modeling and observations to understand earthquake physics. He is a Fellow of the American Geophysical Union.