Microseismic monitoring with the quake neural operator

Accurate monitoring of small-scale seismic events is essential for seismological studies. Traditional techniques rely on manual or automated seismic phase picking, which often leads to inaccuracies in microseismic monitoring due to unclear phase onsets. Here we introduce the Quake Neural Operator (QNO), a deep learning algorithm that builds earthquake catalogs directly from continuous data without explicit phase picking. As a multi-task operator, QNO utilizes classification and regression to detect and locate events across arbitrary seismic network geometries. We show that QNO successfully characterizes events where state-of-the-art phase picking fails. Applying QNO to the Geysers geothermal field, we identify nearly an order of magnitude more seismic detections than reported in routine catalogs. These results are validated through comparison with the Phase Neural Operator and visual inspection. QNO holds the potential to reveal undetected seismic activity, enhancing our understanding of subsurface processes critical to both natural phenomena and industry applications.