A REVIEW ON AI-ASSISTET SEISMIC INVERSION FOR QUANTITATIVE RESERVOIR CHARACTERIZATION

Abstract

Artificial intelligence (AI) is increasingly transforming seismic inversion by enabling faster and more flexible estimation of subsurface elastic and reservoir properties. This review examines the development of AI-assisted seismic inversion for quantitative reservoir characterization, covering machine learning, deep learning, transformer-based, and physics-informed approaches. Applications to acoustic impedance, P- and S-wave properties, lithology, facies, porosity, and fluid characterization are discussed together with comparisons to conventional inversion methods. Current limitations, including limited labeled data, poor generalization, synthetic-to-field discrepancies, physical inconsistency, interpretability, and uncertainty, are critically evaluated. Emerging approaches integrating seismic physics, rock physics, multimodal data, and explainable AI show strong potential for improving reliability. Overall, the future of quantitative reservoir characterization is expected to rely increasingly on hybrid, physics-informed, and uncertainty-aware AI frameworks rather than purely data-driven inversion.