3D Architecture of Porous and Tight Dolomites Using High-Resolution Machine Learning workflow in a Carbonate Reservoir

Abstract Dolomite rocks play a crucial role in reservoir characterization and 3D modeling in the studied carbonate reservoir. Their rock properties vary greatly due to factors, such as: formation mechanisms, dominant mineralogy, original fabrics and diagenesis. This study uses a Geologically High-Resolution Machine Learning (GHREM) workflow to investigate the 3D distribution of two main dolomite types - porous/permeable and tight - affecting reservoir quality, heterogeneity, and architecture. Stage 1: Multi-seismic attributes such as acoustic impedance, relative impedance, root mean square, and quadrature amplitude volumes were used as the input to generate high-resolution bulk density and acoustic impedance properties. Well-based bulk density and acoustic impedance logs served as calibration targets for this stage. Stage 2: The resulting high-resolution 3D properties were then used as inputs for predicting the final target properties (i.e. the porous and tight dolomite 3D discrete model). The final GHREM-based prediction incorporated 1D dolomite flags as discrete 1D input. A neural network algorithm drove the machine learning process in both stages, achieving better predictive accuracy over alternative approaches. The GHREM workflow utilized an in-house machine learning application which enables an objective characterization of the 3D architectural patterns of the primary dolomite facies between wells (porous and tight dolomites). By integrating multiple seismic attributes, intermediate high-resolution bulk density and acoustic impedance 3D volumes, and 1D dolomite identifier flags, this method provided a robust framework for identifying and quantifying 3D dolomite body geometries. Blind testing validated the efficacy of the employed neural network algorithm, yielding accuracies exceeding 85%. Stratigraphically, distinctive distribution patterns emerged, revealing varying proportions of porous and tight dolomites across different intervals. Notably, tight dolomites predominate within the transgressive systems tract of sequences, particularly in the basal portion of the reservoir, whereas their spatial dominance shifts towards the flanks of structural features. Conversely, porous dolomites primarily occupy the regressive systems tracts of sequences, manifesting as narrow, discontinuous bodies with reduced spatial extent compared to tight dolomites. Ultimately, the resultant 3D porous and tight dolomite models served as critical inputs for constructing the comprehensive Integrated Diagenetic and Depositional Facies (IDDF) model (as outlined in Zeiza et al., 2023), facilitating subsequent generation of corresponding 3D porosity and permeability models. Dolomitization in the regressive systems tract interval, specifically in association with pre-cursor reefal facies (i.e. Cladocoropsis & Stromatoporoid grainstones/rudstones) exhibits a strong correlation with the diagenetic process of dissolution. This diagenetic interaction results in the formation of a touching vug network, which significantly contribute to fluid-flow pathways. Petrophysical analysis utilizing whole-core computed tomography (CT) scanning reveals these vuggy dolostones possess elevated porosity and permeability values.