Migration-based near real-time detection and location of microearthquakes with parallel computing

Prompt detection and accurate location of microseismic events are of great importance in seismic monitoring at local scale and become essential steps in monitoring underground activities, such as oil and gas production, geothermal exploitation and underground gas storage, for implementing effective control procedures to limit the induced seismicity hazard. In this study, we describe an automatic and robust earthquake detection and location procedure that exploits high-performance computing and allows the analysis of microseismic events in near real-time using the full waveforms recorded by a local seismic network. The implemented technique, called MigraLoc, is based on the spaceâ time migration of continuous waveform data and consists of the following steps: (1) enhancement of P and S arrivals in noisy signals through a characteristic function, by means of the timeâ frequency analysis of the seismic records; (2) blind event location based on delay-and-sum approach systematically scanning the volume of potential hypocentres; (3) detection notification according to the information content of the hypocentre probability distribution obtained in the previous step. The technique implies that theoretical arrival times are pre-calculated for each station and all potential hypocentres as a solution of the seismic-ray equation in a given 3-D medium. As a test case, we apply MigraLoc to two, low-magnitude, earthquake swarms recorded by the Collalto Seismic Network in the area of the Veneto Alpine foothills (Italy) in 2014 and 2017, respectively. Thanks to MigraLoc, we can increase the number of events reported in the network catalogue by more than 25 per cent. The automatically determined locations prove to be consistent with, and overall more accurate than, those obtained by classical methods using manual time-arrival picks. The proposed method works preferably with dense networks that provide signals with some degree of coherency. It shows the following advantages compared to other classical location methods: it works on the continuous stream of data as well as on selected intervals of waveforms; it detects more microevents owing to the increased signal-to-noise ratio of the stacked signal that feeds the characteristic function; it works with any complex 3-D model with no additional effort; it is completely automatic, once calibrated, and it does not need any manual picking.