Machine learning classification of Austin Chalk chemofacies from high-resolution x-ray fluorescence core characterization

The Upper Cretaceous Austin Chalk (AC) Group is an unconven-tional reservoir that extends across Texas and Louisiana. It is composed of interbedded layers of marly chalks to calcareous-siliciclastic mudrocks that vary in the degree of lamination, bioturbation, mineral abundance, and organic matter richness. Inte-grating lithologic observations with geochemistry is critical for interpreting depositional environments and modeling reservoir properties. Central to this integration is the ability to characterize the geochemistry of core samples at a resolution that captures thin-layered heterogeneity common to mudrock systems. Here, we developed a training data set using a semisupervised chemofa-cies clustering approach that is explored with a deep neural net-work model to predict chemofacies across multiple cores of the AC Group. Eight chemofacies are identified that capture differences in inorganic geochemistry, mineral abundance, rock fabric, and or-ganic matter richness; three classify differences in the marly chalks, four classify differences in the calcareous-siliciclastic mudrocks, and one is transitional between marly chalk and calcareous-siliciclastic mudrocks. Two distinct siliciclastic-carbonate mixing trends are identified that differ in modal abundances of tectosilicates and total clay. Two chemofacies are distinguished based on differences in Mo and V trace element enrichment, suggesting differences in bottom-water redox chemistry. Collectively, this approach pro-vides a means to integrate geochemical measurements and litholog-ical observations to interpret the depositional environments of mudrock systems and is an important step toward upscaling core data to characterize reservoir quality.