U-PINO: A U-Shaped Physics-Informed Neural Operator for DAS Microseismic Source Localization With Application to FORGE DAS Data

Microseismic source localization is essential for mapping spatiotemporal stress evolution in stimulated reservoirs. It provides the foundation for near-real-time monitoring of subsurface operations, such as hydraulic stimulation in geothermal systems. Conventional localization approaches either rely on repeated eikonal solves, which are computationally expensive, or on purely data-driven surrogates, which may violate physical constraints near the source and lose robustness in heterogeneous media. In this study, we present a U-Net-based physics-informed neural operator (U-PINO) for microseismic localization from dense distributed acoustic sensing (DAS) measurements. The proposed model combines the long-range modelling capability of the Fourier neural operator (FNO) with a U-Net-style encoder–decoder that recovers fine-scale spatial detail. Training is regularized by an eikonal-consistency loss defined with respect to a known velocity model, thereby enforcing physically admissible travel-time behaviour. We evaluate the method on a complex velocity model from the Utah Frontier Observatory for Research in Geothermal Energy (FORGE). We validate the predicted traveltimes against fast-marching method (FMM) solutions, and benchmark performance against a plain FNO baseline. The results show that U-PINO reproduces FMM traveltimes more faithfully and improves localization accuracy, reducing mean travel-time root-mean-square error by up to 78% relative to FNO. In addition, the proposed U-PINO configuration converges more efficiently, requiring about 50% less training time per epoch than the FNO baseline. Finally, we demonstrate field applicability by localizing a catalogue of microseismic events from the FORGE geothermal site, where the pretrained model yields accurate source locations with only minimal fine-tuning. These results highlight the potential of U-PINO as a robust, physics-guided surrogate for high-precision, near-real-time microseismic monitoring in complex subsurface environments.