Velocity model building via wellbore-supervised learning: Applications with 3D seismic and helicopter-borne transient electromagnetics

Abstract The shallow subsurface, particularly in arid regions, presents substantial challenges for seismic exploration due to pronounced velocity heterogeneities in the vertical and horizontal dimensions. Traditional refraction-based traveltime inversion techniques face significant limitations in resolving shallow velocity reversals, while full-waveform inversion approaches are difficult to apply in land seismic environments. Areas characterized by intensive seismic exploration typically provide velocity calibration from wells in the form of shallow upholes, deep check shots, and sonic logs that can be used to support velocity model building by 3D seismic data. The use of wellbore velocity information for calibrating inversion-based 3D velocity modeling typically relies on interpretative and highly subjective methods. We develop algorithms and software based on statistics and machine learning, which provide a seamless integration of sparse borehole velocity calibrations with 3D surface seismic data sets for velocity model building. The result is the construction of accurate 3D velocity models calibrated by well data. Seismic gathers with multiwave arrivals or other high-resolution geophysical data, such as helicopter-borne transient electromagnetics, are used as input. The effectiveness of the approach is demonstrated through applications to the synthetic SEAM Arid Model and to field data acquired in complex near-surface and overburden conditions.