Fault characterization in a postsalt reservoir interval, Juabarte Field (Campos Basin) using seismic attributes and machine learning

Seismic attributes are routinely applied for interpretation tasks. Changes in amplitude and phase components reveal faults, and provide insights into hydrocarbon reservoir management. We investigate how different seismic attributes improve the recognition of faults. Data conditioning and unsupervised machine learning methods complement the analysis. The area covered by the 4D/4C Jubarte Permanent Reservoir Monitoring (PRM) system in the northern part of Campos Basin was used to test the impact of different algorithms and parameters. Changes in seismic anomalies associated with post-salt reservoirs reveal the presence of faults and fractures. However, seismic noise and geological units with weak acoustic impedance contrasts required the application of additional methods. Spectral balancing and structure-oriented filtering (SOF) increased the lateral continuity of some stratigraphic reflectors and attenuated random noise, which improved fault surface visibility in many areas. Seismic attributes, both geometric and instantaneous, uncover additional features of fault surfaces. Comparisons between different azimuth-restricted volumes reveal that faults can be delineated when the acquisition direction is positioned perpendicular to structure. Attributes computed using the full-stack volume show less noise content and more rectilinear fault segments. Most-positive and most-negative curvature components indicate more details of major features, and have the advantage of indicating possible up-thrown and down-thrown sides of a deformational zone. The large number of seismic cubes and attributes motivated the use of principal component analysis (PCA) and self-organizing maps (SOM), which complements the identification of faults segments with clusters composed of specific neurons aligned within structural discontinuities. The improvements obtained demonstrated the importance of having a workflow that combines different methods. For the Jubarte Field, a multi-attribute approach demonstrates advantages for delineating the lateral extension of faults and a more precise discontinuity location. Finally, the impact that seismic noise and stratigraphic features may have in the characterization of discontinuities associated with faults was noted.