A simple method to discriminate non-seismic pulses and long-period noise in broadband seismic records and its application to centroid moment tensor inversion of earthquakes in Colombia

Abstract

Non-seismic pulses and long-period noise are artifacts produced by broadband seismometers and degrade centroid moment tensor (CMT) solutions of earthquakes estimated from waveform inversion. In addition to the source amplitude ratio proposed by a previous study, we found that kurtosis is useful for discriminating artifacts from seismic signals. For 157 tectonic earthquakes with moment magnitudes of 4.6–7.5 in and around Colombia in 2014–2022, we manually selected waveforms from the Colombian broadband seismic network and estimated CMT solutions by waveform inversion in a long-period (50−200 s) band. We found that 98% of the selected waveforms had kurtosis values and signal-to-noise ratios larger than 2.5 and 3, respectively, and 92% had source amplitude ratios smaller than 11. We used these values as thresholds and automatically selected waveforms using the three parameters for the 157 earthquakes. Their CMT solutions showed a clear improvement compared to those estimated without any selection. They were better than those estimated with the previous method using only the source amplitude and the signal-to-noise ratios. The proposed parameters were implemented to obtain real-time automatic solutions that were similar to corresponding manual CMT solutions, and 87% of them were estimated within 20 min. Our simple method based on the statistical parameters applied to automatic CMT inversion contributes to improved earthquake monitoring and tsunami warnings in Colombina.