Close

Aberrant noise detected 32 minutes before local quake

A project log for Device for Seismic Noise Analysis

Could a digital device to analyze the statistics of the magnitude and 3-D origins of seismic noise predict some local earthquakes?

michael-doodyMichael Doody 06/25/2017 at 21:052 Comments

This morning at 10:00:57 UTC (6:00:57 Eastern), a magnitude 2.6 tremor originating near Lenoir City, TN was recorded on all regional seismometers in the USGS networks.

Here is the USGS summary page of that event.

This event was 21 km in depth and was approximately 20 miles from our seismic noise device. Here is the signal as recorded on the Knoxville strong motion machine - a few miles from here. Note the absence of any signal before or after the event.

Here is the event as recorded on our machine - the data shown is from 4:00 to 7:00 AM. with data points every 1000 milliseconds. The x-axis is number of seconds from 4 AM and the Y axis is vector magnitude units (in machine voltage units, not the magnitude units used to refer to earthquake strength). This data is from the <15 Hz device. Unfortunately, the low frequency machine was not recording data during this event due to a bad clock module.

32 minutes before the main event, small increases in noise magnitude, associated with directional anomalies were noted. After-tremors were recorded on our device which did not show up on the Knoxville machine, but which did show up on some of the more sensitive regional machines.

Here is a zoomed in view of a scatter plot of the vector magnitude data from the aberrant noise precursor before the main event. Once above 33 units, the data is outside the normal bell curve of the noise.

Here is a line plot of the 5 second running average of the logarithm of the combined (magnitude and location) probability parameter. There is strong evidence of a prolonged statistical anomaly beginning about 32 minutes before the main event (red arrow), although of course it cannot be certain that the anomaly is actually related to the following event.

Here is the same data without the running averaging.

Notice that the lowest point in the precursor is around 10^-7. In other words, there is about a one in ten million chance that the magnitude and direction of that time point's data is part of the usual seismic noise. This is not even taking into account the tremendous improbability of having a cluster of "unlikely" measurements closely related to one another in a short time period. One of the time points for the actual quake reaches a peak at about a one in 10 billion chance.

These numbers are all the more impressive if one takes into account the meaning of the logarithmic data. If 50-50 odds was a line one inch long as on this graph, one in 10 million odds would be a line 158 miles long and one in 10 billion odds would be a line 158,000 miles long. That's almost a light second!

Seismic events are happening across the globe constantly. These far away events are what make up the majority of seismic noise. It's clear that some kind of cluster analysis will be helpful to deal with random and isolated (one point only) deviations from normal statistics and will allow a machine to identify local precursor events with extreme clarity. Far-away (weak intensity and brief apparent duration) seismic events would be experienced almost equally by the individual members of a local network of machines, while the machines will give very different responses to nearby precursor events.

A central computer responding to data from a local network will then need to quantify the heterogeneity of the probability data coming in from its individual machines in response to weak intensity signals. If the local devices are seeing a weak (low amplitude) but real ( a cluster of very low probability time points) event in widely different ways, it then can assume that a local seismic event may be happening.

Then decide whether to issue a warning.

Discussions

Lmdtexasusa wrote 06/28/2017 at 01:10 point

This is amazing.

Was this the first time you experienced a local tremor & were able to pick up the pre-tremor noise?

Also, how much does the whole set-up you have there cost, approximately? This would be a good school project for high school science teachers to undertake with their students in earthquake prone areas around the world. 

  Are you sure? yes | no

Michael Doody wrote 07/06/2017 at 16:05 point

No, not the first time. East Tennessee has a fair number of low intensity tremors.  It is difficult to say if the noise anomalies were real, but there have been several times during development when it seemed that a precursor event might have been detected.  It's hard to draw any conclusions when the device itself has not been the same from day to day.

I think that the set-up could be produced for less than $50 or so in large quantities. It's a little to complicated to set up for school  use.

  Are you sure? yes | no