Regional records of-well documented mine blasts in the area of Caucasian Mineral Water (Northern Caucasus, Russia) are analyzed to test the efficiency of the spectrogram method and Pg/Lg spectral ratios as discriminants between small earthquakes (magnitude <4.5) and comparable magnitude chemical explosions. Being a very useful tool for frequency content studies, the spectrogram method sometimes failed to recognize ripple-fired or spatially separated mine blasts. Due to short delays commonly used (20–40 ms) and complicated crust structure in the region, prominent spectral scalloping of the spectrogram is not observed regularly and is not coherent at different stations. Mean vertical-component Pg/Lg spectral ratio in the band 8 to 20 Hz serves quite well as a discriminant but its value, in general, significantly depends on the particular mine and epicenter location. We have found that no single method can yet assure reliable discrimination between seismic signals from earthquakes and explosions in this region. A self-training algorithm, based on the comparison of `demeaned' envelopes, which grouped events by means of hierarchical cluster analysis, P/S discriminant, was implemented to classify a data set of 18 months of routine observations. A total of 366 out of 374 records were successfully classified and separated into 19 groups of events. This algorithm is especially useful when data analyzed are from sparse networks or from a region of complicated crust structure where averaging over the network can lead to systematic error.
Read full abstract