Traveltime-based microseismic event location using artificial neural network

Location of earthquakes is a primary task in seismology and microseismic monitoring, essential for almost any further analysis. Earthquake hypocenters can be determined by the inversion arrival times seismic waves observed at stations, which non-linear inverse problem. Growing amounts data real-time processing requirements imply use robust machine learning applications characterization seismicity. Convolutional neural networks have been proposed hypocenter determination assuming training on previously processed event catalogs. We propose an alternative approach, does not require pre-existing observations, except velocity model. This particularly important monitoring when labeled events are available due to lack seismicity before commenced (e.g., induced seismicity). The algorithm based feed-forward network trained synthetic times. Once trained, deployed fast location using P-wave (or S-wave) benchmark method against conventional technique show that new approach provides same or better accuracy. study sensitivity dataset, noise detected events, size network. Finally, we apply real it able deal with missing efficient way help fine tuning early stopping. achieved re-training each individual set picked arrivals. To reduce time used weights tune them. allows us obtain locations near real-time.