Search for a command to run...
Thin-bed reservoirs remain one of the most persistent challenges in seismic interpretation and reservoir characterization. Resolut ion theory formalized by Widess (1973) indicates that beds thinner than one quarter of the dominant seismic wavelength (λ/4) c annot be reliably resolved, while beds thinner than λ/8 are can be regarded as seismically invisible. In practice, the resolution limitation leads to vertical smearing, thickness overestimation, and ambiguity in lithology and fluid interpretation of thin b eds. However, theory and practice in spectral analysis, thin-bed reflectivity inversion, and deep learning have shown that seismic data retain information below the classical resolution limit λ/8, provided that noise is controlled and interpretation frameworks are adapted accordingly. In this manuscript, we apply deep learning-based seismic fluid detection for beds thinner than what is conventionally considered resolvable. Whereas earlier deep learning work (Heir et al, 2025) demonstrated reliable prediction of approximately λ/8, the present study shows consistent detection and fluid characterization of beds of λ/16 thickness. The novelty lies not only in resolving thin beds, but in doing so in a field-scale, 3D context following an AI-based reservoir characterization workflow that integrates seismic data and stratigraphic context, and rock-physics-based interpretation (Heir et al., 2024, Heir and Aghayev, 2025). We apply the method to the Julimar gas-condensate field in the Carnarvon Basin, Australia, characterized by thin, laterally variable fluvial-deltaic sands. Results are benchmarked against conventional simultaneous multi-stack sparse-spike inversion. We demonstrate that the AI workflow predicts the fluids of sands as thin as λ/16: 4.6 m thick gas sands at more than 3000 meters de pth in a blind well test. This is well below conventional seismic resolution limit, while maintaining geological consistency and blind-well validation accuracy above 90%.