Accelerated Subsurface Characterization Powered by Science-Aware Machine Learning

Abstract Accurate subsurface characterization requires precise interpretation of seismic events and estimation of elastic and petrophysical properties. Faults and horizons define structural and stratigraphic frameworks and together, they are vital for reservoir characterization and geoscience studies. The elastic properties estimated from seismic data are further used to define the reservoir porosity and lithology. The goal of this paper is to demonstrate how science-aware machine learning (ML) algorithms can accelerate subsurface characterization, for increased efficiency. A streamlined workflow, incorporating machine learning, was adopted to accelerate the subsurface characterization. This workflow was applied to the Groningen onshore gas field, within the southern North Sea basin and to a dataset from offshore Australia. This ensured workflow validation across various environments and with diverse data challenges. This machine learning based workflow enabled rapid and automated analysis of seismic and well data, reducing the time spent on mundane interpretation tasks, by identifying patterns and relationships that are difficult to detect manually. Incorporating science awareness into these machine learning algorithms also enhanced the quality and reliability of the results, compared to relying solely on machine learning approaches. Using this streamlined workflow allowed for more time to analyze the results and focus on generating insights.