AI Reservoir Characterization in Geologically Complex Fields

Abstract

Abstract Reservoir characterization in structurally and stratigraphically complex settings is often limited by the turnaround time and assumptions of conventional seismic inversion. Low-frequency model construction, wavelet estimation, phase uncertainty, and smoothing can materially influence the resulting elastic volumes, particularly where reservoir intervals are thin or laterally variable. This study evaluates a stratigraphy-guided artificial intelligence workflow for estimating elastic and petrophysical properties in the Poseidon field of the Browse Basin, offshore northwestern Australia. The workflow combines migrated 3D seismic data, well logs, interpreted horizons, stratigraphic units, and Relative Geologic Time within a temporal convolutional network. Acoustic impedance, porosity, and gamma ray were modeled, with acoustic impedance used for the principal quantitative comparison. Generalization was assessed by withholding wells from training and validation, and the predictions were compared with published simultaneous-inversion and other benchmark results from the same field. Across the evaluated wells, acoustic-impedance correlations exceeded 90% on average, while blind-well correlations were above 80%. The corresponding published simultaneous-inversion benchmark reported average correlations of approximately 60%. Adding stratigraphic context improved the blind Poseidon-1 correlation from approximately 70% to 81% and reduced mean absolute percentage error from approximately 10% to 5%. The predicted volumes delineated low-impedance intervals in the Plover Formation and preserved greater vertical detail than the comparison inversion. The results show that stratigraphic constraints can improve the accuracy and geological coherence of AI and data-driven property estimation. The approach is complementary to physics-based inversion and provides a rapid method for testing reservoir interpretations where conventional assumptions are uncertain.