No abstract

A method for identification of pulsations in time series of magnetic field data which are simultaneously present in multiple channels of data at one or more sensor locations is described. Candidate pulsations of interest are first identified in geomagnetic time series by inspection. Time series of these ''training events'' are represented in matrix form and transpose-multiplied to generate timedomain covariance matrices. The ranked eigenvectors of this matrix are stored as a feature of the pulsation. In the second stage of the algorithm, a sliding window (approximately the width of the training event) is moved across the vector-valued time-series comprising the channels on which the training event was observed. At each window position, the data covariance matrix and associated eigenvectors are calculated. We compare the orientation of the dominant eigenvectors of the training data to those from the windowed data and flag windows where the dominant eigenvectors directions are similar. This was successful in automatically identifying pulses which share polarization and appear to be from the same source process. We apply the method to a case study of continuously sampled (50 Hz) data from six observatories, each equipped with threecomponent induction coil magnetometers. We examine a 90-day interval of data associated with a cluster of four observatories located within 50 km of Napa, California, together with two remote reference stations-one 100 km to the north of the cluster and the other 350 km south. When the training data contains signals present in the remote reference observatories, we are reliably able to identify and extract global geomagnetic signals such as solar-generated noise. When training data contains pulsations only observed in the cluster of local observatories, we identify several types of non-plane wave signals having similar polarization.

Assessing the statistical significance of electromagnetic anomalies in the ultralow frequency (ULF) range observed prior to earthquakes is a necessary step toward determining whether these perturbations constitute actual earthquake precursors. A statistical epoch analysis (SEA) was recently performed by Han et al. (2014, https://doi.org/10.1002/2014JA019789) to analyze earthquakes happening between 2001 and 2010 near the geomagnetic observatory of Kakioka, Japan; the authors found a significant number of anomalies 6 to 15 days prior to the earthquake day within 100 km from Kakioka, while no significant pre-earthquake activity was observed for the farther region 100 to 216 km from the observatory. In this current paper, we describe the application of our independent software implementation of their method. Despite using a different outlier rejection scheme, we manage to approximate their results. Upon validation of our program, we conduct multiple sensitivity studies. First, we explore how different outlier rejection schemes impact the results. We then restrict the analysis to only mantle earthquakes, highlighting a marginally significant number of anomalies prior to the earthquake day. Next, we test a higher band-pass filter than the one initially used but find no anomalous pre-earthquake activity in this higher-frequency band. We then use a different catalog to establish the list of qualifying "earthquake days" which also leads the anomalous pre-earthquake episode to vanish, thus raising concerns about the robustness of the results. Finally, we apply the SEA to another time window, ranging from 2013 to 2018: No significant pre-earthquake episode can be observed for this interval. We conclude our study by providing guidelines for upcoming work.

Magnetic field changes as earthquake precursors have been the subject of numerous studies and some controversy. Infrequent large earthquakes and sparse magnetometer coverage along fault zones complicate statistical analysis. We present an analysis of ground‐based magnetic time‐series measurements before 19 earthquakes ≥M4.5 in California drawing from over 330,000 site‐days of measurement spanning a decade. To perform a fair existential test for electromagnetic antecedents we applied a pre‐specified statistical analysis with two key ideas. First, we combine signals from nearby (≤40 km) sites via spectral cross‐power, and then look for large spikes in frequency domain (0.016–25 Hz). The former is only possible with a dense set of sites running over a long period of time. In this statistical case‐control study we used the machine learning concept of rigorously separated train and test sets of earthquakes which were generated via a rule‐based query of the USGS earthquake catalog. Before each declustered earthquake, we constructed one period 24–72 hr before (the “precursor” or “p‐period”) and a series of seven equally‐sized preceding periods (“quiescent” or “q‐periods”). We distilled the data in each period to a frequency‐dependent feature—the 98th percentile of spectral cross power. We trained a model based on Linear Discriminant Analysis and applied the discriminator to the test set revealing a modest effect in the days leading up to an earthquake. While the observed effect size is not directly useful for earthquake prediction (long a scientific goal), it suggests a relationship which should be further investigated for a physical link.

No abstract