4. Spectral Decomposition AASPI Overview of AASPI accomplishments and software development Kurt J. Marfurt Attribute-Assisted Seismic Processing and Interpretation

AASPI 2018 Sponsors AASPI Welcome 2012 Current AASPI sponsors. More details on our website mcee.ou.edu/aaspi

He’s not going anywhere He now have more time for research (and documentation!) He will work increasingly with students of fellow faculty Heather Bedle has taken over his teaching duties We will see more students from these two folks!

Postdocs Visiting Scholars Current AASPI postdocs and visiting scholars. More details on our website mcee.ou.edu/aaspi

M.S. in machine learning Current AASPI students. More details on our website mcee.ou.edu/aaspi M.S. in data informatics

M.S. in data informatics Current AASPI students. More details on our website mcee.ou.edu/aaspi

DISC2018 book: This book written by Marfurt for his 2018 DISC tour is available for purchase from the SEG. AASPI users will recognize much of the mathematics (most of which are in the AASPI documentation) and many of the examples from former students and colleagues. If you don’t’ much about attributes, this is a good place to start. Based on contributions from an average of 15 students/year over 20 years working ½ time – so built on 150 man/woman years of effort Mathematical details appear in gray boxes

Recent graduates Xuan Qi (Ph.D.) to Petuum, Inc. Lennon Infante (Ph.D.) to BP Mohsen al-Ali (Ph.D.) to Saudi Aramco Murphy Cassels (M.S.) to Haliburton Lennon Infante (M.S.) to OU PhD program Tobi Olorunsola (M.S.) to U. Arkansas PhD program Kara Rohan (M.S) at Devon Energy 2018 AASPI theses and dissertations. You can download these publications from our website mcee.ou.edu/aaspi

2018 Publications 2018 AASPI publications. You can download these publications from our website mcee.ou.edu/aaspi Plus 11 expanded abstracts and 7 AAPG Explorer geophysical corner notes

New AASPI Software Capabilities New AASPI Algorithms: June 2018 Application Name Application Description Location Software Documentation AASPI References ica3d Independent component analysis. Unlike principal component analysis, which constructs orthogonal components that best represent the data in a mathematical sense, ICA attempts to separate different patterns of signal from each other and from coherent noise such as acquisition footprint using a principal of negentropy. under aaspi_util > Attribute Correlation Tools tab http://mcee.ou.edu/aaspi/documentation/Volumetric_Classification-ica3d.pdf http://mcee.ou.edu/aaspi/upload/AASPI_Theses/2018_AASPI_Theses/David_Lubo-Robles_MS_Thesis_Spring_2018.pdf rfc3d Random forest classification is one of the more promising decision tree supervised learning techniques that uses training data to identify similar patterns in the seismic data volume. under aaspi_util> Attrubute Correlation Tools tab http://mcee.ou.edu/aaspi/documentation/Volumetric_Classification-rfc3d.pdf in progress iterative_fault_enhancemet workflow Using iteratively energy-weighted directional Laplacian of a Gaussian filtering for fault enhancement, which rejects lateral dicontinuity anomalies, smooths faults, and enhances low incoherent anomalies. under aaspi_util > AASPI Workflows tab http://mcee.ou.edu/aaspi/documentation/Image_Processing-fault_connectivity.pdf http://mcee.ou.edu/aaspi/publications/2018/Qi_et_al._2018_Image_processing_for_automatic_fault_extraction.pdf nonparallelism Computation of the variance of the dip vector within an analysis window. This measure is complementary to reflector convergence and has shown some value in machine learning and supervised classification of chaotic features such as mass transport deposits and salt under aaspi_util> Geometric Attributes tab documentation in progress New 2018 AASPI capabilities. Go to our website mcee.ou.edu/aaspi and click on the software documentation and references links. New AASPI Algorithms: December 2018 Application Name Application Description Location Software Documentation AASPI References pnn3d Probalistic neural network supervised learning under aaspi_util > Facies analysis tab  

Geochronostratigraphic horizon mapping Lots of challenges: Algorithmic (for low S/N horizons) Graphical interaction Graphical display Horizon formatting Goal: Release a version by 12/31/2018

Initial steps in computing “well-log attributes” Well log utilities las import las export Kuwahara estimate of upward fining slopes Plot well log API/ft No wells, Poor resolution area Kuwahara filtering We have begun building well log import/export utilities as well as the computation of well log attributes. Import and export capabilities will be there by 12/31/2018. Saurabh Sinha has prototyped Kuwahara and Expectation Maximization algorithms to quantify upward-fining and upward-coarsening sequences. Please see his poster and give him some feedback! (Sinha and Pires de Lima, 2018)

Enhanced AASPI Software Capabilities Enhanced AASPI Algorithms: June 2018 Application Name Application Description Software Documentation AASPI References curvature3d Vector aberrancy scaling now in world coordinates http://mcee.ou.edu/aaspi/documentation/Volumetric_Attributes-curvature3d.pdf http://mcee.ou.edu/aaspi/publications/2018/Qi_and_Marfurt_2018-Volumentric_aberrancy_to_map_subtle_faults_and_flexures.pdf ; http://mcee.ou.edu/aaspi/documentation/Data_Conversion-Converting_poststack_data_from_AASPI_to_ZGY_formats.pdf dip3d Signficantly improved the gradient structure tensor estimate of dip computation. For data with moderate dips, the results between this technique and the previous semblance search technique are comparable. The GST based algorithm is signficantly faster for data exhibiting steep dip. In addition to a measure of confidence and the moderately useful chaos estimate, the GST algorithm also outputs a measure of planarity, call gst_similarity. As of January 2018, the gst algorithm is set to be the default algorithm. Because steep dips (associated with noise, such as internal to salt domes and mass transport complexes) may occur in the data, one should alway run filter_dip_components after dip computation to remove non-physical artifacts in the data; otherwise, these artifacts will be amplified by subsequent curvature and aberrancy calculations. http://mcee.ou.edu/aaspi/documentation/Volumetric_Attributes-dip3d.pdf all algorithms We now support the Petrel ZGY-format when running under the Windows environment. To use this format internally, you must set an appropriate parameter under the aaspi_default_parameters tab and place a checkmark in front of "Output binary data as ZGY format". Because of scaling problems, at present, we only support 32-bit ZGY format. Attributes with well-known ranges (e.g. coherence, frequency, phase, azimuth, dip, will be saved as 8-bit volumes in the future) http://mcee.ou.edu/aaspi/documentation/Data_Conversion-Converting_postack_data_from_ZGY_to_AASPI_formats.pdf;   spec_cwt A significant rewrite of the documentation. Implementation of Matos' (2011) inverse continuous wavelet deconvolution bandwidth extension algorithm. Successful application requires (1) internally interpolating the data to a finer sample increment, sampling the spectrum by 5 or more frequencies per octave, and (3) selecting appropriate bandwidths for the initial decomposition and subsequent bandwidth extension. Good defaults are provided by the GUI. http://mcee.ou.edu/aaspi/documentation/Spectral_Attributes-spec_cwt.pdf   filter_single_attribute Modified to allow subtraction of a filtered volume from the original data. This allows us to "enhance" coherence anomalies subparallel to reflector dip using an alpha-trim mean or LUM filter and subtract it from the original data, leaving more steeply dipping faults in the resulting image. http://mcee.ou.edu/aaspi/documentation/Image_Processing-filter_single_attribute.pdf fault_enhancement Coherence or other edge attributes can now be weighted by the energy, envelope, or other amplitude measure, thereby more heavily weighting discontinuities that occur at the peak energy part of the wavelet. Such weighting further diminishes the stairstep artifacts seen on vertical slices through coherence. The 2nd moment tensor used in defining the orientation of a hypothesized fault is now computed about the center of mass within the analysis window, rather than about the center of the analysis window itself http://mcee.ou.edu/aaspi/documentation/Image_Processing-fault_enhancement.pdf kuwahara3d Now (optionally) allows a time-variant adaptive window size that increases with decreasing average spectral content of the survey computed in time or depth. http://mcee.ou.edu/aaspi/documentation/Image_Processing-kuwahara3d.pdf most poststack algorithms Now print most errors such that they appear in a pop-up window as well as in the *.out file cigar_probe Now allows multiple input volumes, simplifying the workflow http://mcee.ou.edu/aaspi/documentation/Attribute_Correlation-cigar_probe.pdf Enhanced 2018 AASPI capabilities. Go to our website mcee.ou.edu/aaspi and click on the software documentation and references links.

Spectral decomposition limited to zone between two picked surfaces Release spec_cwt by 12/15/2018

Today’s program