U-PINO DAS Localization
Physics-guided DAS microseismic localization
About
U-PINO combines Fourier spectral blocks, a U-Net-style encoder-decoder, and an eikonal-consistency loss to predict a full travel-time field from DAS arrival-time observations (Al-Qadasi et al. 2026). The source is localized at the predicted travel-time minimum.
Why it matters
Fast localization supports monitoring of stimulation and stress evolution in geothermal reservoirs. Purely data-driven surrogates may lose physical consistency, while repeated numerical eikonal solutions can be too expensive for near-real-time use.
Key result and audience
The accepted paper reports up to a 78% reduction in travel-time RMSE relative to a plain FNO baseline and roughly 50% less training time per epoch for the evaluated configuration. The repository supports Utah FORGE experiments, field-event examples, and FNO comparisons for DAS and induced-seismicity researchers.
Public links
The paper is accepted/in press in Geophysical Prospecting; a DOI was not present in the public record reviewed on 17 August 2026. This public repository is a fork of the authors’ upstream implementation.