Effects of frequency-dependent DAS and geophone data inclusion in elastic FWI

Anton Ziegon, Kristopher A. Innanen

Elastic full-waveform inversion (EFWI) is a promising tool for imaging and monitoring of subsurface operations, as it is able to invert multicomponent data into three independent elastic parameters that allow for a more complete reservoir description. One common issue associated with EFWI is cycle-skipping, a phenomenon which occurs when low-wavenumber information about the model in the form of transmission-like or low-frequency data is lacking. Conventionally, single- or multi-component geophones were used to record the seismic wavefield, however, they are limited on the low-frequency end to 10 Hz due to their natural frequency and their deployment at depth in boreholes poses significant cost and time. Therefore, it is very challenging to overcome cycle-skipping purely with geophone data. With the rise of distributed acoustic sensing (DAS), a new way to sense the wavefield is provided, that allows to sense frequencies below 10 Hz and that can be deployed along boreholes in a cost- and time-effective manner. In this report, a framework is introduced that allows to include DAS and geophone data in the EFWI procedure frequency dependent, meaning that the frequency range can be extended by including additional DAS data below 10 Hz and additional geophone data at the high-frequency end. This approach is tested on two synthetic models that resemble geologic CCS plays. The results indicate, that the frequency-dependent data inclusion creates an imbalance in the inversion that deteriorates model reconstruction and data fit, however, down-weighting of the additional included data and specific frequency band designs are shown to mitigate this issue sufficiently. Further analysis indicates no significant imaging improvements for the two tested models but possible computational benefits due to improved convergence behavior. We suggest further testing on more complex geologic models, as the simple models of this study probably lack the complexity needed to see major benefits of additional low frequency data inclusion.