Statistically Significant Results Applying Machine Learning to Identify Earthquake Precursor Periods in QuakeFinder’s Magnetometer Dataset

Abstract:

A collaboration of researchers present the results of applying machine learning to QuakeFinder's (QF) magnetometer dataset towards the detection of anomalous electromagnetic activity occurring ahead of earthquakes. The presentation of these encouraging results includes a discussion of statistical significance, training a classifier and handling of a dataset where the positive population is many times smaller than the negative population. The work also outlines challenges encountered and strategies adopted to separate the data into training and test datasets, accounting for various noise sources, and the tuning parameters involved in configuring the classifier.

QuakeFinder has acquired over 80 TB of data from nearly 15 years of observation of Earth’s magnetic field with outstanding spatial and temporal resolution. This humanitarian R&D project seeks to identify anomalous electromagnetic activity occurring ahead of earthquakes. QF's observatory network consists of 150 stations along the major faults in California, Greece, Taiwan, Chile and Peru. Each station is equipped with 3 feedback induction magnetometers, 2 ion sensors, a geophone, and temperature and humidity sensors. The data are continuously recorded at 50 samples per second with GPS antennas supplying reference timestamps and transmitted daily to the QF data center. QF stations have been in proximity to thousands of seismic events and the dataset affords a rare opportunity to study the question of seismically correlated electromagnetic activity using statistically valid approaches.