From Subsurface Prediction Models to MLOps: A New Era for Geoscience Data

The abstract discusses integrating machine learning (ML) and MLOps workflows in geoscience for subsurface prediction models. Traditional methods often rely on limited datasets and static models, creating challenges in updating predictions when new data becomes available. The proposed MLOps approach ensures continuous integration and delivery of ML models, automating data ingestion, model training, validation, and updates. Using seismic data and attributes, the workflow achieves accurate predictions, as demonstrated with North Sea data for porosity modeling. This dynamic methodology enhances decision-making, resource management, and geoscience collaboration by continuously refining models with new data and interpretations.