Integrating Machine Learning and Geostatistics for Enhanced Unconventional Reservoir Characterization: A PCA-EFA Approach

The mineralogical and geochemical complexity and spatial heterogeneity of organic-rich shale formations and other unconventional reservoirs suggest that integrated multivariate statistical methods and spatial modeling can improve drill site selection and production forecasting. Traditional methods, such as bivariate analysis (e.g., total organic carbon (TOC) vs. porosity) or simple cut-off models (e.g., fixed thresholds for TOC or porosity), often struggle to capture spatial variability, particularly in data-limited environments. This study uses principal component analysis (PCA) for descriptive visualization in reduced space and employs exploratory factor analysis (EFA) to extract latent compositional patterns from the combined geochemical and mineralogical dataset for the Utica and Point Pleasant formations. Given the high dimensionality and multicollinearity typical of such data, a data reduction method is essential to simplify interpretation, reduce noise, and highlight the most informative underlying geochemical structures that govern the compositional variability. The extracted factors from EFA are subsequently integrated into variogram modeling and conditional simulation to quantify spatial uncertainty and improve the reliability of reservoir quality predictions. These variables reveal spatial patterns that help identify drilling targets with higher potential for effective hydraulic fracturing and enhanced fluid storage.