Microseismic Detection in Borehole DAS as a Spatial Localization Problem: Cross-System Validation at the 2024 Utah FORGE Stimulation

ISAO KUROSAWA

Distributed acoustic sensing (DAS) turns borehole fibers into dense seismic arrays, but converting terabytes of continuous strain-rate data into reliable real-time detections remains a bottleneck for induced-seismicity monitoring. We argue that the limiting question for DAS microseismic monitoring has been posed incorrectly. Existing detectors ask whether an event is present in a record; on a dense fiber that question is nearly trivial, whereas the question that constrains downstream monitoring — where along the array the energy originates — is not. We therefore reframe detection as spatial localization on the array: the continuous record is decomposed into overlapping channel-time regions of interest (ROIs), each carrying thirteen physical attributes, and the wavefront is recovered as the subset of ROIs that are mutually consistent under propagation. Applied to the 1,496-channel Neubrex DAS array in Utah FORGE well 16B(78)-32 during the April 2024 stimulation of well 16A(78)-32, the detector reproduces 90% of an independent reference catalog derived from a separate fiber and interrogator, while processing 12-s tiles 2.9-3.7 times faster than real time on one GPU. Controlled experiments confirm the reframing: every model we test captures presence almost equally well (area under the curve about 0.98), whereas localization improves sharply once neighboring ROIs constrain one another, with node-level average precision rising from 0.735 for independent scoring to 0.957, and holding at 0.957 on average under leave-one-stage-out evaluation across all eight stimulation stages — a generalization gap of essentially zero. The reformulation, not any network, is what the data reward. A coherence-based triage of detections absent from the reference catalog, calibrated against known events, identifies at least 26 uncatalogued microearthquakes (about 5 per hour of active injection) in stage 4 alone, nearly all sharing a common apparent slowness. The detector provides a validated, real-time front end for DAS-based traffic-light systems and source characterization.