A novel workflow integrating machine learning-guided characterization of carbonate reservoirs using matching pursuit as an approach for post-stack inversion

Carbonate reservoir's complex pore system and high heterogeneity have led to the need for new specialized studies to improve reservoir characterization. However, this goal is often restricted by a lack of data. In this study, a novel procedure applied to a data-constrained petroleum field in the Gulf of Mexico is presented. By implementing a matching learning-guided method, the procedure, characterizes two pay zones in carbonate reservoirs. This was achieved by using matching pursuit to condition and execute a spectral inversion on a stacked seismic volume. From this, density and effective porosity volumes were estimated using interval velocities (obtained through migration) and acoustic impedance (derived from spectral inversion). These attributes were then processed using machine learning, combined with a geostatistical analysis employing Bayesian inferences, to generate geological models that honour the geology while capturing associated uncertainties. The results allowed for the discretization of a carbonate facies with high porosity and low clay content, indicative of potential hydrocarbon-bearing zones, thus highlighting opportunities for exploration.