Ringdal_etal_MRR2009_Proceedings_02-12.pdf (2.34 MB)
Expanding Coherent Array Processing to Larger Apertures Using Empirical Matched Field Processing
conference contribution
posted on 2021-06-28, 09:38 authored by Frode Ringdal, David B. Harris, Tormod Kværna, Steven GibbonsSteven GibbonsThis is chapter 02-12 of the proceedings of the meeting "2009 Monitoring Research Review: Ground-Based Nuclear Explosion Monitoring Technologies" held 21-23 September, 2009 in Tucson, Arizona. Pages 379-388. The full proceedings document is current available on https://www.osti.gov/biblio/992203.
The abstract of this chapter is as follows:
We have adapted matched field processing—a method developed in underwater acoustics to detect and locate
targets— to classify transient seismic signals arising from mining explosions. Matched field processing, as we apply
it, is an empirical technique, using observations of historic events to calibrate the amplitude and phase structure of
wavefields incident upon an array aperture for particular repeating sources. The objective of this project is to
determine how broadly applicable the method is and to understand the phenomena that control its performance.
We obtained our original results in distinguishing events from ten mines in the Khibiny and Olenegorsk mining
districts of the Kola Peninsula, for which we had exceptional ground truth information. In a cross-validation test,
some 98.2% of 549 explosions were correctly classified by originating mine using just the Pn observations (2.5-12.5
Hz) on the ARCES array at ranges from 350 – 410 kilometers. These results were achieved despite the fact that the mines are as closely spaced as 3 kilometers. Such classification performance is significantly better than predicted by the Rayleigh limit. Scattering phenomena account for the increased resolution, as we make clear in an analysis of the information carrying capacity of Pn under two alternative propagation scenarios: free-space propagation and propagation with realistic (actually measured) spatial covariance structure. The increase in information capacity over a wide band is captured by the matched field calibrations and used to separate explosions from very closely-spaced sources. In part, the improvement occurs because the calibrations enable coherent processing at frequencies above those normally considered coherent.
We are investigating whether similar results can be expected in different regions, with apertures of increasing scale and for diffuse seismicity. We verified similar performance with the closely-spaced Zapolyarni mines, though discovered that it may be necessary to divide event populations from a single mine into identifiable subpopulations. For this purpose, we perform cluster analysis using matched field statistics calculated on pairs of individual events as a distance metric. In our initial work, calibrations were derived from ensembles of events ranging in number to more than 100. We are considering the performance now of matched field calibrations derived with many fewer events (even, as mentioned, individual events). Since these are high-variance estimates, we are testing the use of cross-channel, multitaper, spectral estimation methods to reduce the variance of calibrations and detection statistics derived from single-event observations.
To test the applicability of the technique in a different tectonic region, we have obtained four years of continuous data from 4 Kazakh arrays and are extracting large numbers of event segments. Our initial results using 132 mining
explosions recorded by the Makanchi array are similar to those obtained in the European Arctic. Matched field
processing clearly separates the explosions from three closely-spaced mines located approximately 400 kilometers
from the array, again using waveforms in a band (6-10 Hz) normally considered incoherent for this array. Having
reproduced ARCES-type performance with another small aperture array, we have two additional objectives for
matched field processing. We will attempt to extend matched field processing to larger apertures: a 200 km aperture (the KNET) and, if data permit, to an aperture comprised of several Kazakh arrays. We also will investigate the potential of developing matched field processing to roughly locate and classify natural seismicity, which is more
diffuse than the concentrated sources of mining explosions that we have investigated to date.
targets— to classify transient seismic signals arising from mining explosions. Matched field processing, as we apply
it, is an empirical technique, using observations of historic events to calibrate the amplitude and phase structure of
wavefields incident upon an array aperture for particular repeating sources. The objective of this project is to
determine how broadly applicable the method is and to understand the phenomena that control its performance.
We obtained our original results in distinguishing events from ten mines in the Khibiny and Olenegorsk mining
districts of the Kola Peninsula, for which we had exceptional ground truth information. In a cross-validation test,
some 98.2% of 549 explosions were correctly classified by originating mine using just the Pn observations (2.5-12.5
Hz) on the ARCES array at ranges from 350 – 410 kilometers. These results were achieved despite the fact that the mines are as closely spaced as 3 kilometers. Such classification performance is significantly better than predicted by the Rayleigh limit. Scattering phenomena account for the increased resolution, as we make clear in an analysis of the information carrying capacity of Pn under two alternative propagation scenarios: free-space propagation and propagation with realistic (actually measured) spatial covariance structure. The increase in information capacity over a wide band is captured by the matched field calibrations and used to separate explosions from very closely-spaced sources. In part, the improvement occurs because the calibrations enable coherent processing at frequencies above those normally considered coherent.
We are investigating whether similar results can be expected in different regions, with apertures of increasing scale and for diffuse seismicity. We verified similar performance with the closely-spaced Zapolyarni mines, though discovered that it may be necessary to divide event populations from a single mine into identifiable subpopulations. For this purpose, we perform cluster analysis using matched field statistics calculated on pairs of individual events as a distance metric. In our initial work, calibrations were derived from ensembles of events ranging in number to more than 100. We are considering the performance now of matched field calibrations derived with many fewer events (even, as mentioned, individual events). Since these are high-variance estimates, we are testing the use of cross-channel, multitaper, spectral estimation methods to reduce the variance of calibrations and detection statistics derived from single-event observations.
To test the applicability of the technique in a different tectonic region, we have obtained four years of continuous data from 4 Kazakh arrays and are extracting large numbers of event segments. Our initial results using 132 mining
explosions recorded by the Makanchi array are similar to those obtained in the European Arctic. Matched field
processing clearly separates the explosions from three closely-spaced mines located approximately 400 kilometers
from the array, again using waveforms in a band (6-10 Hz) normally considered incoherent for this array. Having
reproduced ARCES-type performance with another small aperture array, we have two additional objectives for
matched field processing. We will attempt to extend matched field processing to larger apertures: a 200 km aperture (the KNET) and, if data permit, to an aperture comprised of several Kazakh arrays. We also will investigate the potential of developing matched field processing to roughly locate and classify natural seismicity, which is more
diffuse than the concentrated sources of mining explosions that we have investigated to date.
