Automated classification of local earthquake data in the BUG small array

SUMMARY The Bochum University Germany (BUG) stations monitor the mining-induced seismicity in the Ruhr basin of NW Germany. Four of the stations form an array of small aperture. Automated classification of local earthquakes into different source regions and recognition of noise bursts can be described by a two-step approach; the recognition of sonogram patterns at each single station and the subsequent, rule-based coincidence evaluation. The detection on single-traces performs by adaptation of sonogram patterns recognized as in image processing. The algorithm tunes a set of reference events to a wide variety of noise conditions. The adaptive algorithm consists of two steps: (a) each reference pattern is scaled to the actual amplitude of signal energy, (b) weak phases falling below the updated detection threshold are excluded. This procedure is repeated for each pattern and every time increment. The best pattern fit is taken as the final identification. The rule-based coincidence evaluation is introduced by concept and completely described by its 14 rules and all implicit assumptions. Given the additional information of single-trace classifications, the system can avoid any false alarms that were previously be caused by the casual coincidence of local noise bursts. Both stages of the automated classification scheme were tested on routine observatory data of a one month period. For sonogram recognition, the knowledge base consisted of 12 seismograms. The majority of earthquakes were recognized at all sites, the identification of quarry blasts was excellent and most site-specific noise bursts were rejected. The coincidence evaluation improved the network-wide classification of earthquakes to 85 per cent which is above any single-station optimum.