Unsupervised Seismic Facies Classifications of the Lower Strawn Fm. Bend Arch-Fort Worth Basin, TX

Abstract

The Bend Arch- Fort Worth Basin is a tectonically active area that exhibits a complex sediment distribution along north-central Texas. While Pennsylvanian plays have been one of the main exploratory targets, the depositional history and 3D seismic expression of the Desmoinesian time Lower Strawn Formation described by Gun (1979) and Pranter (1989) as a turbidite / submarine fan complex is only partially documented. In this paper, we subject a post-stack 3D seismic volume conditioned to structural oriented filtering and time-frequency domain based spectral balancing from which we extract a set of attributes and P-impedance inversion. Appropriate attributes are then used for unsupervised facies classification with the aim of geometrically differentiate the sand bodies and delineate the architectural elements present along the target unit. After data conditioning, we compute a suite of candidate attributes, including P-impedance inversion. The next “exploratory data analysis” step involved comparing candidate attributes to areas where well control showed the desired sand to be present or absent. Our analysis showed that several of the eight texture attributes that measure the lateral homogeneity, entropy, and other properties of the reflector exhibited value. To simplify the classification, we reduced these eight measures into one represented by the first principal component. The resulting five attributes used in classification were P-impedance, precondition seismic amplitude, peak spectral frequency (from the time-frequency analysis), the first principal component of the eight texture attributes, and the relative stratal location. We evaluated three different classification techniques with the goal of compare a diverse set of results. K-means clustering analysis, Self-organizing Maps (SOM) and Principal Component Analysis (PCA). The results show that some classification techniques can highlight architectural elements stronger than other. Although K-means was able to highlight possible channels, point bars, scroll bars and crevasse splays features, SOM and PCA were able to discriminate more subtle facies changes along a specific area.

AAPG Datapages/Search and Discovery Article #90291 ©2017 AAPG Annual Convention and Exhibition, Houston, Texas, April 2-5, 2017