Characterizing a Mississippian tripolitic chert reservoir using 3D unsupervised and supervised multiattribute seismic facies analysis: An example from Osage County, Oklahoma

Abstract

Seismic interpretation is based on the identification of reflector configuration and continuity, with coherent reflectors having a distinct amplitude, frequency, and phase. Skilled interpreters may classify reflector configurations as parallel, converging, truncated, or hummocky, and use their expertise to identify stratigraphic packages and unconformities. In principal, a given pattern can be explicitly defined as a combination of waveform and reflector configuration properties, although such “clustering” is often done subconsciously. Computer-assisted classification of seismic attribute volumes builds on the same concepts. Seismic attributes not only quantify characteristics of the seismic reflection events, but also measure aspects of reflector configurations. The Mississippi Lime resource play of northern Oklahoma and southern Kansas provides a particularly challenging problem. Instead of defining the facies stratigraphically, we need to define them either diagenetically (tight limestone, stratified limestone and nonporous chert, and highly porous tripolitic chert) or structurally (fractured versus unfractured chert and limestone). Using a 3D seismic survey acquired in Osage County Oklahoma, we use Kohonen self-organizing maps to classify different diagenetically altered facies of the Mississippi Lime play. The 256 prototype vectors (potential clusters) reduce to only three or four distinct “natural” clusters. We use ground truth of seismic facies seen on horizontal image logs to fix three average attribute data vectors near the well locations, resulting in three “known” facies, and do a minimum Euclidean distance supervised classification. The predicted clusters correlate well to the poststack impedance inversion result.

Peculiar geological features in the Nugget Sandstone of the Moxa Arch

“…While looking at different seismic attributes used to be good enough in earlier days, with the recent developments in computational geophysics, multiple attributes can be layered or co-rendered over each other to be able to visualize or identify a seismic facie that would have otherwise gone unnoticed by even the best of geology interpreters (Roy et al, 2013;Zhang et al, 2015;Zhao et al, 2016). We begin our paper by discussing the geology of the Moxa Arch, followed by the description of the available data.…”

Characterizing a turbidite system in Canterbury Basin, New Zealand, using seismic attributes and distance-preserving self-organizing maps

“…Feeding multiple attributes into a classification algorithm enables interpreters to analyze different aspects of seismic response (energy, frequency, phase, geometry, texture, etc.) simultaneously, generating a map of facies or correlating seismic responses to engineering/production data (Roy et al, 2013;Zhang et al, 2015).…”

“…The Kohonen self-organizing map (SOM) (Kohonen, 1982) is one of the most widely accessible techniques in commercial software packages that produces reasonably high quality seismic facies maps. The SOM preserves the topological connections among clusters, which is a preferred attribute for visualization when using similar colors for similar facies (Strecker and Uden, 2002;Roy et al, 2013). However, traditional Kohonen SOM does not preserve the distance in input space, which may result in an over-/ undershrinking of the clusters in the SOM latent space.…”

Lithofacies classification in Barnett Shale using proximal support vector machines

“…Self-organizing maps (SOM) is perhaps the most popular classifier of seismic data. Commonly called "waveform classification", the waveform in SOM is not limited to seismic amplitude samples, but can be a suite of impedance samples, or a vector of appropriately scaled seismic attributes (Roy et al, 2013). In its simplest implementation, SOM is unsupervised, preventing interpreter control.…”

“…In its simplest implementation, SOM is unsupervised, preventing interpreter control. Roy et al (2013) introduced supervision by the addition of a maximum likelihood classifier. In contrast, SVM and PSVM are supervised classifiers which assign classes with geological meanings before classification.…”