Data-driven identification of stratigraphic units in 3D seismic data using hierarchical density-based clustering

Abstract

Seismic sequences are stratigraphic units of relatively conformable seismic reflections. These units are intervals of similar sedimentation conditions, governed by sediment supply and relative sea level, and they are key elements in understanding the evolution of sedimentary basins. Conventional seismic sequence analyses typically rely on human interpretation; consequently, they are time-consuming. We have developed a new data-driven method to identify first-order stratigraphic units based on the assumption that the seismic units honor a layer-cake earth model, with layers that can be discriminated by the differences in seismic reflection properties, such as amplitude, continuity, and density. To identify stratigraphic units in a seismic volume, we compute feature vectors that describe the distribution of amplitudes, texture, and two-way traveltime for small seismic subvolumes. Here, the seismic texture is described with a novel texture descriptor that quantifies a simplified 3D local binary pattern around each pixel in the seismic volume. The feature vectors are preprocessed and clustered using a hierarchical density-based cluster algorithm in which each cluster is assumed to represent one stratigraphic unit. Field examples from the Barents Sea and the North Sea demonstrate that the proposed data-driven method can identify major 3D stratigraphic units without the need for manual interpretation, labeling, or prior geologic knowledge.