Multimineral Characterization of Shales for Reactive Transport Modeling Based on Micro-XRF Interpretations

Abstract:

Accuracy of modeling reactive transport in the subsurface is rooted in our understanding of the mineral abundances, their spatial heterogeneity, and the extent to which minerals are accessible to introduced fluids. In this work, we demonstrate application of a new machine learning approach to characterize sedimentary rock samples in regard to their mineral distributions, accessible mineral surface areas, abundances, grain size distributions, and solid-solution presence. The approach uses raster scanned µXRF data with areas of tens of millimeters obtained at 2 µm resolution. Rock samples include shale from the Eagle Ford Formation, and a shale mudrock from the Upper Wolfcamp formation. The shale sample is important because of its distributed arsenic-bearing sulfide phases embedded in soluble carbonate phases. This phase is mapped in 2D and statistically characterized in 3D by applying principles of stereology. The mudrock is unique because it is a precipitation-filled fracture, consisting of solid-solutions and a partially open fracture, exposing multiple empty pixels. Empty pixels serve as an important subset of data for fracture or pore space mapping. All data were obtained from the GSECARS beamline 13-ID-E at Advanced Photon Sources (APS). The datasets used for training the algorithm come from coupled µXRF-µXRD analyses of the Eagle Ford shale and a mineral mixture, formulated from six minerals of similar and overlapping chemistries. We highlight that the multimineral identification per pixel is a significant advancement to existing single layer mineral mapping tools, and the ease of application to new µXRF data, without the need for retraining on a mineral it has already been trained on, is another added benefit of this approach.