Deep Learning‐Based 3D Microseismic Event Direct Location Using Simultaneous Surface and Borehole Data

Abstract Microseismic monitoring is crucial for characterizing and assessing fracture systems developed during subsurface operations like geothermal reservoir development, sequestration, and hydraulic fracturing. In such long‐term projects, surface arrays provide a large aperture for accurate epicenter location but struggle with poor event depth estimation and low signal‐to‐noise ratio; while borehole arrays deliver better signal quality and enhanced depth estimation, but have severely limited aperture and azimuthal ambiguity for epicenter location. To bridge the capacities of surface and borehole arrays, we propose a deep learning model to directly locate microseismic events in 3D using seismic recordings from both surface and borehole sensors, based on the elastic medium assumption. Specifically, the DEtection TRansformer (DETR) network is utilized as the base model to identify and locate events simultaneously. The proposed method, based on distinct feature extraction blocks for the surface and borehole data, has potent compatibility for embracing diverse data sets. It is also able to effectively model the complicated dynamics from the moveout patterns of microseismic energy to responsible source locations, and the concerted interactions of these moveout signatures across various sensor layouts. The statistical analysis on results from synthetic passive seismic data demonstrates the efficacy of the method. An application on the field microseismic events from the Utah FORGE site further validates its accuracy and reliability, in which a domain adaptation technique is used to bridge the gap between training on synthetic data and application on field data for better generalization.