- Copyright © 2002 Society of Exploration Geophysicists
In recent years, we have witnessed a data explosion driven by advances in hardware and software capabilities. Whereas only a few years ago it was quite normal to interpret seismic data on one volume only, it is now common to work on several volumes simultaneously. In addition, modern seismic interpretation software packages allow us to compute a plethora of different attributes. All these volumes and attributes offer a different view of the data, often revealing interesting features but also leading to confusing and sometimes conflicting information.
This paper describes two concepts, each meant to combine all these different types of information into one single so-called “meta-attribute.” Firstly, supervised neural networks are used to combine different attributes into one new meta-attribute. Secondly, mathematical and logical combinations of different attributes result in a user-defined new meta-attribute. Both concepts are described on the basis of a fault interpretation example.
Back in the days of 2D seismic, interpretation of fault systems was difficult. The lateral continuity of faults needed to be inferred with a significant amount of interpretation. With the introduction of 3D seismic data, the continuity problem was solved but only partly so. In 3D data, the “raw” seismic data themselves rarely provide the optimal view for fault interpretation. Positioning of faults and decisions on fault continuity or breakup remain difficult. With increasing understanding and computer power, extracting attributes for the special purposes of enhancing fault mapping has been applied in the industry. However, attributes needed to be selected and parameters set and a basically simple procedure became an expert's job with multiple, often confusing results. In the following sections a few popular attributes for fault mapping are described.
Single attributes that enhance faults in seismic data are mainly based on some form of detection of discontinuity in seismic events. Semblance is one …