Improved seismic discrimination using pattern recognition

The problem of discriminating between earthquakes and underground nuclear explosions is formulated as a problem in pattern recognition. As such it may be separated into two stages, feature extraction and classification. The short-period (SP) features consist of m_b and autoregressive parameters characterising the preceding noise, signal and coda. The long-period (LP) features consist of LP power spectral estimates taken within various group velocity windows. Contrary to common usage we have extracted features from horizontal Rayleigh waves and Love waves as well as vertical Rayleigh waves. The classification is performed by approximating the statistical distribution of earthquake and explosion feature vectors by multivariate normal distributions.The method has been tested on a data base containing 52 explosions and 73 earthquakes from Eurasia recorded at NORSAR between 1971 and 1975. Several of these events are difficult on the m_b : M_s diagram m_b(PDE) and M_s (NORSAR) have been used]. The data set was divided into a learning and an independent data set. All of the events both from the learning data set and the independent data set were correctly classified using the new procedures. Furthermore, the increase in separation as compared to the m_b : M_s discriminant is significant.