Distributed acoustic sensing: modelling, full waveform inversion, and its use in seismic monitoring

Matthew Eaid

Distributed acoustic sensing (DAS) is a rapidly evolving seismic acquisition technology. Employing rugged and small optical fibers, DAS others an opportunity for access to acquisition geometries often not accessible to more conventional geophone sensors. This is a highly attractive property of DAS fibers, that others access to the often unrecorded transmission wavefield modes that are crucial to support land full waveform inversion (FWI). The inherently different sampling that DAS fibers provide leads to a requirement for modeling strategies that differ from those used to simulate point sensor data. To maximize the potential of DAS which only samples tangential strain, and is therefore a single component sensor, fibers are often shaped to improve their wavefield sampling. In this thesis I propose a robust method for simulating DAS data from arbitrarily shaped fibers, that couples a geometric model of the fiber to an elastic wavefield propagator to provide the strain sensed along the tangent of a fiber. The simulation methods are used to generate a large synthetic dataset that supports a machine learning study for extracting source mechanism information from DAS data, and to support FWI.

The shape of the DAS fiber, both on large-scales such as when it tracks a deviated well, or on small scales when it is wrapped in some characteristic shape affects the sensitivity of the fiber to different wavefield components, which has an important influence on the recorded data. The data recorded by a DAS fiber is a function of the fiber shape, and it is therefore expected that fiber shape will have an important influence on parameter estimates in FWI. To address this question, I first develop a method for incorporating data from an arbitrarily shaped fiber in FWI. Using a 2D isotropic-elastic FWI over synthetic toy models, I then examine the role of fiber gauge length, fiber shape, and their interplay in FWI. It is determined that fiber shape has an important influence on parameter resolution, but that the optimal fiber shape is acquisition geometry and model dependent. The work presented in this thesis lays out a sandbox for appraising fiber geometry prior to field deployment, and allows for the optimization of fiber shape to support FWI. It is also shown that short gauge length fibers (where short is described in relation to the fiber geometry) can push DAS fibers towards multi-component point sensors. FWI results obtained with short gauge length fibers and orthogonal-point-sensing geophones agree favorably.

To provide confidence to the synthetically derived conclusions, these insights must also transfer to field data. The methods for the inclusion of DAS data in FWI are used to invert field data from a straight DAS fiber both in isolation and in various combinations with collocated accelerometer data from a field research station focusing on the monitoring of injected carbon dioxide. The field FWI is also 2D and assumes isotropic-elastic wavefield propagation, and incorporates a single-parameter parameterization that leverages prior information from well-log data. Models obtained from inversions of each dataset are correlated in their overall structure, but each provides differing models of the subsurface. Incorporation of both datasets into a single objective function is observed to stabilize the inversion, and leads to a more robust estimate of the elastic subsurface parameter distribution. The models obtained in this study can be used as baseline models of the research station to support further time lapse analysis.