Abstract
Summary The problem of discriminating between earthquakes and underground nuclear explosions is formulated as an exercise in pattern recognition approach analysis. an advantage of our procedure is flexibility, by combining both adaptive noise suppression and event classification incorporating feature selection criteria. The analysis has been applied to a learning set of 44 nuclear explosions (eight test sites) and 35 earthquakes in Eurasia recorded at the NORESS array (Fig. 1). the signal features considered were the normalized power in eight spectral bands in the 0.2-5.0 Hz range of the P wave (6 s) and the P coda (30 s). Physically, it means that we exploit potential differences in the shape of earthquake and explosion spectra, respectively. Other features included are peak P and P-coda amplitude frequencies and relative P/P-coda power. These 19 features were extracted either from conventional array beam traces or the optimum group filtered traces (OGF- removal of coherent low-frequency noise). Using the feature selection algorithm, based on estimates of the expected probability of misclassification (EPMC), only two to four features were needed for optimum discrimination performance. the dominant features were coda excitation and P- and P-coda power at lower signal frequencies. Furthermore, feature parameters extracted from the OGF traces had a slightly better performance in comparison to those extracted from beam traces. Finally, there were no misclassifications for OGF-derived features when the explosion population was limited to East Kazakh events, while including events from the other test sites lead to a decrease in discrimination power.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have