A Data Driven Model for Micro-Porosity Prediction of Complex Carbonate Reservoirs Using Rock-Stable Isotopes, XRF & MICP Data

Porosity classification and quantification, including micro-porosity characterization, are key elements in reservoir description. Among all subsurface geoscience data, core data are often considered the reference commonly used to calibrate other measurements, despite data scarcity compared to other information such as drill cuttings. In this study, we introduce a data-driven methodology based on Rock Stable Isotopes (RSI), X-Ray Fluorescence (XRF) and Mercury-Injection Capillary Pressure (MICP) core measurements to predict micro-porosity in carbonate reservoirs. The study was conducted using a dataset consisting of 273 core plug samples across multiple carbonate formations. For each core sample, about 10 grams were sub-sampled and homogenized to perform RSI and XRF measurements. Along with the micro-porosity measurements obtained from MICP data of their sister samples, the dataset was utilized to build a neural network AI model to derive MICP based micro-porosity in carbonate reservoirs. Once verified and tested, the model can then be run on drill cuttings for real time applications, such as geo-steering. The created model was able to predict micro-porosity in the studied samples at an R-squared value of 0.91 using data points from a single formation, while understandably higher uncertainty has been introduced when multiple formations were included, illustrating that the model depends on rock depositional environments, diagenesis, thus rock types. The prediction model confirms the robustness of the established relationship between RSI, XRF and micro-porosity data. A detailed examination of the results shows that micro-porosity generally decreases from Limestone to Dolostone in this data set, and it appears that XRF correlates to micro-porosity better than RSI. This core plug-based workflow can help deliver critical reservoir characterization properties using drill cuttings from mud logging data, for micro-porosity is typically non-productive. Once a correlation is established, the model can help to study possible reasons for reservoir quality production challenges.