Detection of small-magnitude events using unsupervised machine learning for waveform-based source imaging through grouped time reversals

Abstract Efficient arrival picking is one of the crucial steps for seismic data processing both in active and passive seismology. We employ the unsupervised Fuzzy C-means clustering to improve the picking of the arrivals of small magnitude earthquake events, and then source location is computed by grouped time-reversal imaging approach by first splitting the receivers into groups and then backward propagating the wavefield. Once the arrivals are auto picked, P-wave signals are identified and extracted from the entire waveforms. Finally, a high-resolution source location image is obtained by multidimensional cross-correlation of the wavefields. The performance of the approach has been tested on a suite of synthetic data and field data sets obtained from the North-West Himalayan region of Jammu and Kashmir. The estimated uncertainties reveal the improvement of the locations over the conventional travel time inversion method. We demonstrate that these two independent approaches can effectively be used to analyze seismological data sets in a complex medium, and even works with sparse seismic networks for locating the small-magnitude earthquakes. The present approach can be used to locate the seismic sources utilizing the single-component waveform data.