Processing partially coherent large seismic array for discrimination

Seismic signals in a band of frequencies near 1.0 Hz. can be detected from underground explosions and earthquakes at distances of several thousand kilometers. A stochastic model has been proposed to characterize observed signal variations within a Large Aperture Seismic Array (LASA). The model assumes that the observed signal spectrum at a seismometer is some average spectrum multiplied by a random gain and phase. Within a subarray (7 km. aperture) the mean value of the modulus squared of the random term is approximated by 1.0 +0.18 f2 where f is frequency in Hz. For sensors drawn from the full LASA (200 km. aperture) the value is 1.0 +2.0 f2. Two alternative methods for extracting spectral information above 1.0 Hz for discrimination between event types are compared. Beamforming spectra are obtained from the Fourier transform of the average received signal. An alternative incoherent processing method, spectraforming, is to calculate the average spectrum from individual seismometers. Both can be corrected for bias. It is demonstrated that although beamforming will give more noise rejection than spectraforming, that the latter can be superior in terms of output signal to noise ratio when input signal variations between sensors are large. Spectraforming may be of significant value for obtaining spectral information in the 1.0 - 3.0 Hz band for events with Richter magnitudes in the range 4.0 - 4.5. This magnitude range is of considerable current interest for the purpose of nuclear test detection and discrimination.