End-to-end Workflow for Stratigraphy-Guided Reservoir Characterization

Accurate prediction of elastic and petrophysical properties is essential for reducing exploration risk and guiding reservoir development. Conventional seismic inversion workflows remain limited by assumptions on wavelets, background models, and geological contrast, often producing uncertain results in stratigraphically complex settings.

We present a stratigraphy-guided deep learning (SGDL) workflow that directly predicts impedance and porosity volumes from seismic and well data, while embedding automatically generated Relative Geological Time (RGT) and horizons. The case study is the Poseidon field in the Browse Basin, offshore northwestern Australia. The main reservoir is the Jurassic Plover Formation, a fluvio-deltaic system characterized by lateral heterogeneity due to synrift faulting, sealed by the Montara Formation.

Applied to Poseidon, the SGDL approach achieved >90% average correlation with measured logs and >80% in blind wells, compared with ∼60% for simultaneous inversion. Automatically generated horizons and RGT halved prediction error from 10% to 5%, resolving thin, high-porosity sands that inversion blurred. Predictions aligned with known gas-charged facies and Bayesian inversion results. Compared with recent AI benchmarks, the workflow delivered property volumes 44× faster while improving accuracy.

This end-to-end workflow demonstrates robust, efficient reservoir characterization with strong potential for CCS and geothermal applications.