Application of seismic attributes and machine learning clustering techniques to the characterization of faults in a post-salt reservoir, Jubarte Field (Campos Basin)

Perico, Edimar

Application of seismic attributes and machine learning clustering techniques to the characterization of faults in a post-salt reservoir, Jubarte Field (Campos Basin)

dc.contributor.advisor	Bedle, Heather
dc.contributor.author	Perico, Edimar
dc.contributor.committeeMember	Carpenter, Brett
dc.contributor.committeeMember	Pranter, Matthew
dc.date.accessioned	2021-05-12T19:22:02Z
dc.date.available	2021-05-12T19:22:02Z
dc.date.issued	2021-05-14
dc.date.manuscript	2021-05-11
dc.description.abstract	Seismic reflection datasets represent an important source of information capable of revealing structural features in the subsurface, such as fault zones. Faults can be interpreted using the original amplitude volumes, especially when blocks of rocks have moved dozens or hundreds of meters relative to each other. One of the seismic characteristics that contribute to fault recognition is the presence of significant acoustic impedance contrasts in the faulted sedimentary layers. However, the recognition of subtle discontinuities represents a more challenging task when performed in intervals with a low signal-to-noise ratio and weak contrasts of acoustic properties. In these cases, seismic attribute analysis can enhance structural mapping. Seismic attributes are different measurements of the original seismic components, such as amplitude and phase. These attributes are routinely applied to highlight geological features of interest, not only for structural studies but also for stratigraphic analysis. This research aims to compare instantaneous and geometric attributes (with different spectral components and azimuth volumes) to define what improvements can be obtained for fault characterization. The southeastern part of the post-salt section in the Jubarte Field (Campos Basin) represents the area selected to evaluate the impact of different parameters. In that region, the implementation of a 4D/4C system called Jubarte PRM resulted in many seismic volumes associated with various surveys. The main input used within this research is the P-wave full-stack volume, but complementary investigations included different azimuth data sets. Four wells helped calibrate log curves and seismic cubes and generate synthetic seismograms to characterize the Maastrichtian reservoir response. Data conditioning represents a significant step to prepare the datasets to compute seismic attributes. Spectral balancing and structure-oriented filtering (SOF) are algorithms applied in this initial phase. Attenuation of random seismic noise, enhanced visualization of stratigraphic reflectors, and more visible faults (higher contrasts with relation to the background) are examples of improvements obtained with the data conditioning. The computation of most attributes requires specific parameters. For instance, curvature components are estimated using different algorithms (structural vs. amplitude) with short or long operators. The results show that different attributes can highlight distinct aspects of the seismic traces. For example, the cosine of instantaneous phase indicates discontinuities, and it is less affected by different amplitude contrasts. This attribute provides a uniform visualization of the discontinuities, keeping visible some additional features, such as stratigraphic elements. The energy-ratio attribute delineated discontinuities with sharper anomalies compared to other similarity attributes. Broadband volumes provide superior results for the delineation of non-vertical faults. The discontinuities' geometry is better defined using the entire amplitude spectrum than volumes composed by specific frequency ranges. Furthermore, comparisons using vertical sections and time slices demonstrate the benefits of having a full-stack azimuth data set (e.g., less seismic noise and more uniform fault segments). Individual azimuth sectors reveal small discontinuities, especially when oriented perpendicular to the acquisition direction. In short, I noticed after various tests that a multi-attribute approach was adequate for a more reliable and complete description of faults and fracture zones. Unsupervised machine learning techniques complemented the analysis due to the larger number of volumes available and its sub-products (seismic attributes). I used principal components analysis (PCA) to identify the volumes with higher variability. Self-organizing maps (SOM) were included within the workflow to see if this clustering technique can delineate fault surfaces. Although the method has some limitations related to non-vertical discontinuities and noise interference, it can improve and optimize the visualization of structural discontinuities as observed in the reservoir interval. Lastly, it is necessary to check the results in all the steps because various methods can increase non-desired features such as seismic noise. Evaluation of the seismic features needs to be considered in-context interpretation to define the geological meaning for the structures described in seismic cubes. Overall, I noticed that an integrated investigation using data conditioning methods, different seismic attributes, and analysis of multiples volumes using unsupervised clustering methods improved the subsurface structural investigation of the Jubarte Field.	en_US
dc.identifier.uri	https://hdl.handle.net/11244/329540
dc.language	en	en_US
dc.subject	Seismic attributes	en_US
dc.subject	Faults	en_US
dc.subject	Unsupervised machine learning	en_US
dc.thesis.degree	Master of Science	en_US
dc.title	Application of seismic attributes and machine learning clustering techniques to the characterization of faults in a post-salt reservoir, Jubarte Field (Campos Basin)	en_US
ou.group	Mewbourne College of Earth and Energy::School of Geosciences	en_US
shareok.orcid	0000-0002-1848-6075	en_US