Fault interpretation and feature learning on full azimuth stacks

Abstract

Determining the location, size, and orientation of features within a subterranean formation can be determined by using more than one set of azimuthal data collected along at least two different angle ranges of seismic detection. The azimuthal data collected along one azimuthal range can be stacked and combined. A feature probability map can be generated for each azimuthal data collection using a machine learning system. Feature probability maps generated using azimuthal data collected along different azimuthal angle ranges can be used to optimize a machine learning estimator to generate ensemble azimuthal datasets. More than one estimator can be used thereby generating more than one ensemble azimuthal dataset. These results can be combined using a weighting algorithm applied using a machine learning model resulting in a combined feature probability map that can reduce the uncertainty of the characteristics of the feature of the subterranean formation.