Illuminating the Subsurface by using DAS, Machine-Learning and Crosswell Seismic Tomography

Summary Crosswell seismic tomography is an established technique for imaging two-dimensional subsurface velocity fields between boreholes, with P-wave tomography being widely used due to mature equipment and efficient processing workflows. In contrast, S-wave crosswell tomography remains limited, largely because of operational complexity and data acquisition challenges. Distributed Acoustic Sensing (DAS) provides an alternative to conventional downhole sensors by enabling dense spatial sampling and high repeatability, but its lower signal-to-noise ratio makes reliable first-arrival picking difficult. As a result, DAS-based crosswell tomography has seen limited application. This study presents two novel applications of DAS for crosswell tomography using datasets acquired at the ECCSEL Svelvik CO2 Field Lab in Norway. First, supervised machine learning is employed to identify first arrivals in DAS data, achieving accuracy comparable to conventional sensors. Convolutional neural network–derived pick probabilities are used to weight travel-time picks, improving tomographic inversion results. Second, the study reports the first SV-wave crosswell survey using DAS, demonstrating significant time savings for S-wave tomography. By combining DAS with a borehole-coupled SV-wave source, the approach enhances resolution and spatial coverage while simplifying field operations by removing the need for water-filled boreholes and complex sensor coupling.