Automatic Interpretation in Structurally Complex Seismic Volumes

Abstract

Seismic interpretation is typically a tedious and subjective process that requires interpretive experience and knowledge of the geology in the focus area. In this study, we present a fully automatic three-dimensional method to first classify seismic sequences, and then track seismic horizons using machine learning, image processing and speech recognition algorithms. The presented method does not require manual input data, such as labels or seed points, and therefore reduce the need for specialist geological knowledge. We target structurally complex areas where seismic horizons are offset by large faults, and we test our method on rotated Mesozoic fault blocks in the southwest Barents Sea. Our horizon tracker correlate horizons across different fault blocks and is insensitive to amplitude changes along seismic horizons. We constrain the horizon tracking within defined seismic sequences to avoid tracking across sequence boundaries such as erosion surfaces. For a fully automatic seismic sequence classification, we use a texture descriptor combined with unsupervised machine learning. First, small sub-cubes with similar local binary texture are clustered, then the clustered data is subjected to filter operations that respect sequence stratigraphic principles. After the filtering, each connected cluster is considered a seismic sequence. For validation, we compare this unsupervised approach to results from manual classification and supervised classification. We track seismic horizons with a sliding dynamic time warping grid. Dynamic time warping (DTW) is a pattern matching operation that finds the optimal alignment of two traces by nonlinearly stretch and shrink one trace until it is “warped” into the other. This operation produces a warp path that describes the relationship of all the reflective events in the two matched traces. We exploit this warp path to track specific events, and thus interpret seismic horizons automatically. The sliding dynamic time warping grid iteratively revisit traces in the seismic volume, which provides a measure of accuracy while tracking. The accuracy measurement is used to filter the results and to quantitatively evaluate the interpreted seismic horizons. Our horizon tracker allows us to track multiple horizons simultaneously, and it is less dependent on the actual seismic amplitudes compared to existing autotracking tools, as it relies on the pattern of the seismic traces instead of specific peak or trough values.

AAPG Datapages/Search and Discovery Article #90350 © 2019 AAPG Annual Convention and Exhibition, San Antonio, Texas, May 19-22, 2019