An automatic non-volcanic tremor (NVT) detection algorithm is developed. NVT occurs in the 1-10 Hz frequency range, but high ambient noise dominates above 2 Hz at many of the MesoAmerican Seismic Experiment (MASE) stations, so the seismograms are bandpass filtered 1-2 Hz. The site amplification effect is determined by comparing the coda envelopes of regional earthquakes at all stations and then removed. A median filter is applied to the absolute amplitude of the bandpass filtered data to remove the spikes of local earthquakes. Day-long data files of each station are detrended and the median value is removed to suppress the effect of storms and local noise. An average of all data files per day is then obtained to amplify the coherent energy from NVT and diminish residual noise. An amplitude cutoff for NVT is empirically determined and applied to all daily data within the study period to generate the NVT catalog. The automatically determined catalog is compared with a catalog created by visual analysis of daily spectrograms. A detailed comparison is done of the month with the greatest difference between the two catalogs, May 2006. It is found that the automatic detection algorithm had fewer false picks and undetermined NVT than the initial visual NVT detection. Thus, the automatic NVT detection algorithm developed has fewer errors and can be effectively used for further analysis of the tremor activity.
Husker, A. Peyrat, S. Shapiro, N. Kostoglodov, V.