Using PCA and PLS on publicly available data to predict the extractability of hydrocarbons from shales

Prediction of hydrocarbon extraction from shale requires specialized knowledge of shale play characteristics and analysis to assess effective, economical, and sustainable implementation of oil and natural gas production. In this work, we present a statistical approach that can be used as a preliminary investigation into the hydrocarbon resource potential of a shale play based on limited data. Statistical algorithms for Principal Component Analysis (PCA) and Partial Least Squares Regression (PLS) were used to determine if depositional environments and lithographic boundary characteristics of different plays allowed prediction of specific production parameters. This project characterizes Eagle Ford and Utica formations—two high-producing shale plays in the United States—and Banff/Exshaw and Colorado formations—two recently assessed shale plays in Alberta, Canada. Partial Least Squares Regression models were unable to model gas production parameters from predictor variables, highlighting the complexity of gas formations and the need for data on microscale petrophysical characteristics. In contrast, oil production parameters were better predicted, because bulk variables such as mineral composition appeared to correlate with oil location in mineral interfaces. As expected, a PLS model's predictive capabilities increased with specificity of data sets to particular regions of a shale play. This study indicates how PCA and PLS modeling could assist stakeholders to make preliminary decisions regarding hydrocarbon extraction, especially when limited to publicly available petrophysical data.