Multivariate statistical analyses applied to seismic facies recognition

Abstract

One of the most important goals of seismic stratigraphy is to recognize and analyze seismic facies with regard to the geologic environment. The first problem is to determine which seismic parameters are discriminant for characterizing the facies, then to take into account all those parameters simultaneously. The second problem is to be sure that there is a link between the seismic parameters and the geologic facies we are investigating. This paper presents a methodology for automatic facies recognition based upon two steps. The first, or learning step, begins with the definition of learning seismic traces for each facies we wish to recognize. The choice of learning traces is based upon either well data or a seismic stratigraphic interpretation. A large number of seismic parameters are then computed from the learning traces; multidimensional analyses are carried out in order to validate the choice of learning traces and to select, among all the available parameters, those that discriminate best. At this stage, a modeling step may be carried out to relate the seismic parameters to the geologic features. The second step is a predictive one which allows automatic facies recognition. We compute the previously chosen discriminant parameters on unknown seismic traces and classify the unknown traces with regard to the learning traces. We develop the methodology and successfully apply it to two examples of reservoir facies recognition. Our main conclusion is that seismic traces contain geologic information that can be extracted by multivariate data analyses of a large number of seismic parameters. Automatic facies recognition is reliable and fast; the derived facies map has the great advantage of combining simultaneously several discriminant parameters.

Introductory Chapter: An Introduction to the Seismic and Sequence Stratigraphy and to the Integrated Stratigraphy: Concepts and Meanings

“…The seismic facies analysis deals with both the individuation and the geologic interpretation of the geometry, continuity, amplitude, frequency and velocity of the seismic relectors, more than the outer shape of the sedimentary bodies and the seismic facies associations in a depositional sequence [2,[56][57][58][59][60][61]. In the modern development of this methodology, one aim is represented by the recognition of clusters or groups, representative of signiicant variations in the properties of the rocks, in the lithology and in the content of luids.…”

Uncertainties in Seismic Facies Analysis for Reservoir Characterisation or Monitoring: Causes and Consequences

“…Seismic facies analysis is a generic terminology to describe pattern recognition methods, which map seismic traces described by a set of characteristics or attributes extracted at the reservoir level, to a set of categories, or seismic facies (Dumay and Fournier, 1988). The calibrated relationship between attributes and facies is called a classification function.…”

“…This methodology is based on pattern recognition techniques dedicated to interpreting seismic objects extracted from seismic data (multi-attribute voxels, or pieces of traces at a reservoir level) and described by a number of characteristics, or seismic attributes in a set of classes (Dumay and Fournier, 1988). These classes may be themselves related to geological heterogeneities (sandstones vs. shales) as in Bertrand et al (2002), to petrophysical classes defined by cut-off values (e.g.…”

Recognition and reconstruction of coherent energy with application to deep seismic reflection data

“…It is closely related to the singular value decomposition (SVD), as shown by Freire and Ulrych (1988). In geophysics, it is most often used for wave-noise separation and wave-wave separation (Hemon and Mace, 1978;Freire and Ulrych, 1988;Glangeaud and Mari, 1994) or (multivariate) statistical analyses in reservoir characterization studies (Dumay and Fournier, 1988).…”

“…It then tries to find a projection which minimizes distances within classes, while at the same time maximizing distances between classes, thus requiring only a single projection. For more details, we refer to Richards (1993) and for an application to Dumay and Fournier (1988).…”