Abstract

Successful hydrocarbon production in the Eagle Ford relies on technological advances such as directional geosteering, horizontal drilling, and hydraulic fracturing as well as the identification and delineation of organic-rich intervals and fracture zones. Total organic carbon (TOC) is an excellent indicator for the organic richness and the hydrocarbon potential of shale in unconventional reservoirs; however, common methods for TOC estimation have many underlying assumptions and rely on empirical formulas. It is useful to develop a robust data-driven approach that could be used to reliably identify TOC–rich zones in unconventional plays such as the Eagle Ford. Using gamma ray, deep resistivity, and sonic wireline well log data from La Salle County, South Texas, we generated a layer unit database and label the layer units using core measured TOC values. We applied a data-driven binary support vector machine (SVM) machine learning approach to identify TOC–rich zones within the Eagle Ford. A support vector machine is a classification tool capable of mapping data to a higher dimension and finding an optimal hyperplane that classifies the data. We evaluated the performance of the SVM classifier and obtain F1 scores (measures of combined precision and recall) of 97.4% and 99.4%, respectively, for the high and low TOC classes. The methodology successfully identifies TOC–rich zones that match with geological observations, the ΔlogR method and independently obtained core TOC measurements. We demonstrate the successful cross-well application of the methodology within the Eagle Ford play area. We also demonstrated the successful cross-basin application of the methodology using data from the Barnett Shale Formation, Fort Worth Basin, North Texas, and the Duvernay Shale Formation of the Western Canadian Sedimentary Basin, Canada.