A neural network approach for elucidating fluid leakage along hard-linked normal faults

Abstract

Increasing displacement and strain accumulation in normal faults can result in the formation of hard-linked structures that are preferred loci for fluid leakage. Three-dimensional seismic data from offshore New Zealand reveals Miocene geological units that were structurally deformed to form several hard-linked fault zones. Fluids are observed to migrate through these breached zones into younger strata. Here, we use an automated approach by designing two different meta-attributes, the Thinned Fault (TFC) and Fluid (FC) Cubes, to capture the detailed geometry of hard-linked fault zones, and of fluid flowing through these same structures. The two meta-attributes are prepared through an amalgamation of different seismic attributes, which are trained based on the interpreter's skills and experience following a supervised scheme of neural learning. The meta-attributes enhance sub-surface geological features and reveal their structural geometries. Faults in the study area are therefore observed to strike to the NE with different geometries, e.g. forming curved shapes (F1 and F2), sigmoid shapes (F3), and Y shapes (F4). Relay ramps between these faults are intensively breached, allowing for important fluid migration through hard-linked structures. We demonstrate that our interpretation approach does not only honour the interpreter's knowledge about key geological processes, but also adds value in revealing the 3D architecture of hard-linked normal faults.