Stratigraphy-guided joint inversion of acoustic impedance and resistivity for carbonate reservoir characterization

Abstract

Accurate characterization of complex carbonate reservoirs is essential for reservoir evaluation, fluid identification, and decision-making in hydrocarbon exploration and development. However, elastic properties derived from seismic data are often insufficient for reliable fluid discrimination in heterogeneous carbonate formations. Although resistivity is highly sensitive to fluid content, its availability is generally limited to sparse well locations. To address this challenge, we propose a stratigraphy-guided deep-learning framework based on a CNN-Transformer architecture for the joint inversion of acoustic impedance (AI) and resistivity (RT) from post-stack seismic data. The proposed approach integrates seismic traces and interpreted horizon information within a unified deep-learning architecture. Multi-scale convolutional layers extract local seismic features, whereas Transformer layers capture broader contextual information and long-range dependencies. Stratigraphic encoding derived from interpreted horizons is incorporated to improve geological consistency and lateral continuity of the inversion results. The method was validated using a carbonate reservoir dataset from the Middle East consisting of a 3D post-stack seismic volume, well-log data, and three interpreted horizons. Compared with CNN, Transformer, and CNN-Transformer models, the proposed framework achieved the highest prediction accuracy for both AI and RT. The predicted AI and RT volumes exhibited improved stratigraphic continuity and geological consistency and enabled effective identification of hydrocarbon-bearing sweet spots. These results demonstrate the potential of the proposed framework for multi-property reservoir characterization and sweet-spot prediction in complex carbonate reservoirs.