Fault classification from 3D imaging of vertical DAS profile

Abstract

Faults play an important role in geothermal fluid transport and can also present a contrast in acoustic impedance such that seismic methods can approximately locate their presence in the subsurface. Brady Natural Lab is a geothermal reservoir that has numerous faults that allow for both recharge and deep-toshallow heat exchange via subsurface fluids. In March 2016 at Brady, a continuous active seismic survey collected 191 3- mode vibe points, while a vertical DAS cable was in place 150 to 280 meters below surface. Imaging of both synthetic and field data was performed to analyze if certain fault dips and strikes would be detectable given the shot geometry. Coherent structures exist that are consistent with 3 faults: the fault farthest from the DAS well at distance of 750m to the northeast, and two Western-dipping faults below the well that contains the DAS. Lastly, convolution neural networks were used to obtain an agnostic, quantitative measure of the reliability of detecting faults from images derived from DAS. A transfer learning approach utilized layers of convolutional neural networks trained on the ImageNet repository. Presentation Date: Thursday, October 18, 2018 Start Time: 8:30:00 AM Location: 212A (Anaheim Convention Center) Presentation Type: Oral