Steven John Gibbons , Frode Ringdal, Tormod Kværna

Correlation detectors are becoming a standard method for identifying seismic signals from repeating sources. These highly sensitive, source-specific detectors frequently facilitate a reduction in the detection threshold by around an order of magnitude. Detections are typically declared when the value of the correlation coefficient (CC), or a related statistic, exceeds significantly some measure of the variability of values over a longer time window. The performance of correlation detectors is often compromised by the presence of short duration, high-amplitude signals, which influence excessively the value of the CC. We suggest replacing the original seismograms with waveforms in which the value of each sample is replaced by the ratio of that value to a centred moving average of absolute values of the original waveform. These ratio-to-moving-average (RMA) seismograms are relatively featureless over long time intervals, but resemble greatly the original waveforms over short time windows and hence still capture the characteristic seismic fingerprint of a given source. We demonstrate a correlation detection calculation, which fails due to the presence of a high-amplitude signal interfering with part of the correlation window, but which succeeds when RMA seismograms are used due to the diminished influence of the interfering signal. We also demonstrate an example from an aftershock sequence where the CC traces are heavily modulated due to the high dynamic range of the original waveforms. This makes the setting of detection thresholds difficult and results in multiple peaks, which do not correspond to events in the vicinity of the master event. Repeating the calculation using RMA seismograms results in CC traces with a more well-defined detection threshold and most of the spurious detections are lost. The ability to set lower detection thresholds without increasing greatly the number of false alarms facilitates the robust detection of lower magnitude events.