QuakeFinder Blog

QuakeFinder and Google Research publish study advancing predictive data on earthquakes

California study shows statistically significant electromagnetic signal in days prior to earthquakes

Logo for Project Namazu

Project Namazu was the codename used for the Google / QuakeFinder collaboration

Palo Alto, CA – QuakeFinder, Stellar Solutions’ earthquake forecasting research initiative, and Google Research have published a study in American Geophysical Union’s (AGU) Journal of Geophysical Research: Solid Earth that advances the pursuit of predictive data in alerting the public to impending earthquakes. The team identified a signal of modest but statistically significant size in the days immediately preceding intermediate to large earthquakes in California. The full publication of Case-Control Study on a Decade of Ground-Based Magnetometers in California Reveals Modest Signal 24–72 hr Prior to Earthquakes can be viewed at https://doi.org/10.1029/2022JB024109

“Data science and artificial intelligence are giving scientists new tools to understand the physical world. We are excited that our research combined with Stellar Solution’s unique magnetometer recordings is hinting at previously unknown physical phenomena in earthquake forecasting.”, Jason Miller, Product Management Director for Google Applied Science.

The Google Applied Sciences team led the effort, designed the study, and performed the analysis with exceptional rigor. QuakeFinder provided the data and verified the findings.

“This study provides important evidence that a physical change can be observed in the days before an earthquake. While we still have much work to do in identifying stronger signals, this research supports our vision that earthquake forecasting using the magnetic field may one day be possible,” said Dan Schneider, QuakeFinder’s Director of Research and Development.

QuakeFinder, which is seeking to ultimately save lives by forecasting earthquakes, has collected data since 2005 using over 150 stations around the world. Their high-resolution magnetometer recordings have captured thousands of earthquakes (M3.0 or greater) to support their work in developing algorithms to analyze these signals.

This is the second study that has evaluated the QuakeFinder California dataset and the second one to find a statistically significant correlation between activity in the magnetic field and earthquakes days later. In 2019, QuakeFinder published the first comprehensive evaluation of this dataset in Computers and Geosciences: https://doi.org/10.1016/j.cageo.2019.104317. This used single station data and different algorithms but the same data set and found statistical correlations prior to quakes greater than M4.

The second study (by Google) achieved a Signal to Noise Ratio (SNR) of 0.5 when comparing the magnetic data recorded 24-72 hours before earthquakes (greater than magnitude 4.5) to data recorded earlier. A supplemental analysis added a compensation factor for the average global magnetic activity and this boosted the SNR from 0.5 to 0.96. As an SNR of 1 or greater is typically considered a compelling result in the physical sciences, this analysis points toward the magnetic field being a valuable source of information in the earthquake forecasting process. The study employed a number of strategies to ensure rigor. One was to combine signals from pairs of stations to guard against noise that might pollute the data at a single station, such as automobile traffic, construction and other human activity near the station’s location. Another strategy was to control for variations across different stations, locations and geologies by using a case-control construction in which data from station pairs close to an earthquake (~30km) are compared to other time periods from the same station pair. These strategies help to eliminate false positives.

“As a next step in our research, when correcting for global magnetic activity, we can look to use actual high resolution remote station data provided by QuakeFinder rather than the low resolution global average provided by National Oceanic and Atmospheric Administration, as it may allow the signal to become even more apparent,” Schneider added. He noted that this study was conducted on data collected in California; and its results cannot be assumed for other fault types or geologic conditions.

Stellar Solutions is a Malcolm Baldrige Award-winning company that for 27 years has solved some of the most complex systems engineering problems, from national security to space exploration, across commercial and government sectors. Founder Celeste Ford’s vision, satisfying customer critical needs while realizing employee dream jobs, led to the creation of QuakeFinder in 2001. Stellar Solutions has invested over $30-million in the QuakeFinder program, augmented by partial grants from NASA, the US Department of Defense, PG&E and the Musk Foundation. The initiative has drawn on the expertise of leading aerospace organizations like NASA, geoscientists from Stanford University and University of California at Berkeley, Stellar’s own systems engineers, and partners around the world including from Peru, Taiwan and Greece. Originally inspired by Stanford research after the 1989 Loma Prieta earthquake in California, QuakeFinder founder Tom Bleier devised the idea of placing sensors underground every 20 miles close to the fault trace to capture pre-seismic electromagnetic disturbances.

About Stellar Solutions, Inc.: Stellar Solutions, Inc. is a global aerospace and systems engineering services provider to commercial, defense, intelligence, civil and international sectors. Stellar Solutions provides systems engineering capabilities and strategic support for global communications, remote sensing, national defense, and space exploration. These solutions to customer critical needs connect technology and customers across multiple domains. With physical operations throughout the U.S. and around the globe, Stellar Solutions is a recognized leader in delivering end-to-end critical expertise and problem-solving skills. https://www.stellarsolutions.com/

QuakeFinder: Against All Odds

A recent story about QuakeFinder in the LA Times conveyed the many obstacles that Stellar Solutions has faced over the past two decades in our quest to definitively prove electromagnetic precursors to earthquakes. And in spite of these numerous and significant scientific, technical, political and financial hurdles, we have made tremendous headway toward this noble, long-term goal that could benefit countless lives.

It is important to remember that science is a journey. The article compares our experience to the search for extraterrestrial intelligence and the hunt for the meaning of earthquake lights, and I’ll add that this endeavor is very much like many of the grandest challenges in human history – achieving heavier than air powered flight, the race to space, mapping the human genome, and effective treatments for devastating diseases from polio and cholera to smallpox, HIV and cancer. It’s the nature of innovation: we have achieved a great deal in many areas but in others we still have a ways to go.

While the financial obligation and constraints inherent in this important work have not been overstated, I must elaborate on QuakeFinder’s recent and compelling technical achievements as well as a path forward with partners that can only serve to continue and leverage the significant progress made.

Yes, Stellar Solutions spent big to see if there are electromagnetic precursors to large earthquakes. And yes, they do exist! Our research attempted to prove that electromagnetic signals (in this case magnetic pulses) exist in the days prior to earthquakes larger than M4.0.

The research efforts involved developing algorithms that could find these small signals from 70 Terabytes of data collected from our national network of sensors from 2005 to 2018. The results published late last year in a peer-reviewed journal (Computers and Geosciences) found that for earthquakes larger than M4.0 and within about 40 km of a magnetometer instrument, a measurable increase in magnetic fluctuations occurred in the window 4-12 days prior to the earthquakes. Rigorous statistical methods were used and, in one case, QuakeFinder achieved a 2.86 sigma confidence threshold on these results—or saying it another way—the odds of getting our results if there really is no correlation were only 1 in 475 (0.2%). This suggests that the increased electromagnetic activity does precede earthquakes and is not just experimental happenstance or unrelated to the earthquake process. We are not aware of any other similar research study of this size and scope, nor one that has been formally published or achieved this level of certainty.

Now the challenge is to refine the algorithms to discriminate the unusual activity from the large amount of background noise (e.g. BART trains, lightning, solar storms, and other man-made magnetic noise). To do this, we need funding and partners.

While our research did not specifically focus on earthquake lights, we developed a hypothesis that if there are deep underground electrical activity (e.g. large currents released prior to earthquakes), perhaps very sensitive induction magnetometers might be able to detect these current surges. Today, after 20 years of building a network of very sensitive induction magnetometers, which are spaced about every 20 miles along the faults, QuakeFinder finally had enough data and earthquake events to test the hypothesis.

And QuakeFinder is no longer alone in this quest. Researchers in Japan (Han and Hattori) did a similar analysis over a decade in a relatively quiet area, removed from electric trains and during the night when the trains were not operating. They too found statistical evidence of both magnetic pulses and longer disturbances in the 2 weeks prior to earthquakes.

The quest for accurate earthquake forecasting is an extraordinarily difficult and unprecedented task–far larger than what a single, small aerospace company can afford. It is true that Stellar Solutions is “hitting the pause button” and reducing staff to a minimum level, after having spent approximately $30M over 20 years building the network of sensors and developing these initial algorithms. But, all of this is far from our last contribution to this cause or the end of the QuakeFinder story.

We are hopeful that the initial published QuakeFinder research results, as well as corroborating results in Japan (and soon possibly, China) will attract other funding from either private or government sources, to continue and advance the effort. Better results may occur with the application of Artificial Intelligence techniques. We expect at least one such data mining company to publish their initial results using the QuakeFinder magnetometer data sometime this spring. And our work and methods have application to other areas of societal benefit as shown by our recent selection as a final winner in a federal competition for the World Magnetic Model. The never-ending struggle and adventure of science can be fraught with risk and uncertainty. However, its discoveries, breakthroughs and possibilities have the potential to change everything that we know and care about for the better. We look forward to this continuing journey!

QuakeFinder Publishes Peer-reviewed Electro-Magnetic Research

Groundbreaking research covering major faults in California between 2005-2018 examines potential clues in Earth’s magnetic field

QuakeFinder has just published landmark research on earthquake forecasting in Computers & Geosciences, the peer-reviewed academic journal. The study, which examines electromagnetic earthquake precursors covering the major faults in California from 2005 to 2018, advances QuakeFinder’s research into predictive indications in Earth’s magnetic field that may be identified several days prior to an earthquake.

See the Press Release!

The paper, “An algorithmic framework for investigating the temporal relationship of magnetic field pulses and earthquakes applied to California,” has completed peer review and received final acceptance for immediate online publication and will appear in print in the December issue (Volume 133) of the peer-reviewed journal Computers & Geosciences. The paper can be found at https://doi.org/10.1016/j.cageo.2019.104317.

To our knowledge, this is the largest ever study of electromagnetic earthquake precursors to be published and is quite an exciting development for our team that’s been hard at work collecting and analyzing this data for over fifteen years. The data provide hints that earthquake precursors exist in the magnetic field to a statistically significant degree, and this paper represents a major step forward in the challenge faced by QuakeFinder to conclusively demonstrate the existence of these precursor signals and then work towards isolation that will allow individual earthquakes to be forecast.

The treated data set exceeds prior published research by over an order of magnitude, and suggests, with 98.6% (2.2 sigma) confidence, that magnetic field exhibits precursory behavior in the period of four to 12 days prior to earthquakes. While this approach is not yet accurate enough to forecast individual quakes, our research indicates that results will improve with enhanced signal processing.

In comparison, Earthquake Early Warning (EEW) systems based on seismic observations currently provide, at best, seconds of warning before earthquakes. QuakeFinder’s work is poised to transform definitive earthquake forecasting from a science fiction dream into an achievable goal and the team is continuing its work to perform statistically valid research beyond individual quakes or small sample sizes.

QuakeFinder’s Daniel Schneider will present these results at the American Geophysical Union (AGU) conference this December in San Francisco on Friday, December 13, 2019 (Abstract NH52B-05 – QuakeFinder’s Algorithm Results for Forecasting Earthquakes).

Computers & Geosciences publishes high impact, original research at the interface between Computer Sciences and Geosciences. Its articles apply modern computational and informatics-based computer science paradigms to address problems in the geosciences.

Computers and Geosciences Cover Image

Two Recent Earthquakes Shake California

QuakeFinder is reviewing data from its stations close to two recent earthquakes in California. A magnitude 4.5 quake struck at 10:33pm PDT on Monday, Oct. 14 near our Concord station and the second, a magnitude 4.8 was felt only 14 hours later at 12:42pm on Tuesday, Oct. 15 near our Pinnacles station. Both stations were successfully recording magnetic field data during the earthquakes and in the days leading up to them.

QuakeFinder is studying the magnetic field, searching for reliable pre-seismic signals with the ultimate goal of building an earthquake forecasting system. The data from these quakes will help us better understand the behavior of the magnetic field and further isolate earth-sourced magnetic signatures.