A METHOD FOR EXTRACTING 3D MODAL MINERALOGY AND TEXTURAL DATA FROM PLUTONIC ROCKS

Quantitative analyses of the mineralogy and textures of igneous rocks can provide valuable insights into both the generation and evolution of magmas as well as timescales of magmatic processes. Typically, such analyses are conducted either using crystal separates, which eradicate context, or two-dimensional (2D) sections, which require the application of stereological corrections when assessing three-dimensional (3D) textures. Many of these issues can be minimized by leveraging in situ, non-destructive 3D analytical techniques, such as x-ray tomography. However, it often is non trivial to use such techniques to produce useful 3D data from plutonic rocks.This challenge is due to limitations on sample size, long imaging and analysis times, and difficulties in image processing as a result of low density contrast of constituent minerals (e.g., quartz and feldspars).

Here, we present a novel workflow for producing 3D reconstructions of igneous rocks. Our methodology uses serial grinding (with typical step sizes ranging from 10-30 microns) and optical imaging of large (i.e., centimeter-scale) slabs, in conjunction with a neural-network-based image processing pipeline, to estimate modal mineralogy, bulk rock composition, and crystal size distributions of granitic samples. This workflow is semi-automatic and leverages a custom convolutional neural network that can be applied to other 3D datasets. We demonstrate our methodology on a granite sample from the Golden Horn Batholith (Washington, US). In addition to presenting quantitative outputs, we share synthetic experiments that both demonstrate errors associated with 2D estimates and consider the smallest recoverable 3D grain size. Furthermore, we describe challenges associated with 3D analyses. We suggest that our methodology can provide useful information when non-destructive, density and phase-based techniques cannot be used.