CALIBRATING BACKSCATTERED ELECTRON IMAGE GRAYSCALE VALUES WITH MINERAL COMPOSITIONS TO GENERATE PHASE COMPOSITION MAPS AT THE THIN SECTION SCALE

Crystallization history of magmatic bodies (plutonic-volcanic systems) is commonly inferred from microscopic textures and rock and mineral compositions. Backscattered electron (BSE) images can be generated much more rapidly and inexpensively than x-ray composition maps and their grayscale values hold significant composition and spatial information. We use energy-dispersive mineral analyses to relate grayscale values directly to phase identification and composition through regression analysis. The regression model is then applied to grayscale values in BSE images to generate images thresholded by phase composition (not elemental composition as in x-ray maps). Similar previous work (Ginebre et al., 2002, Contrib. Min. Petrol., 143, 436-448) imaged a single phase (oscillatory-zoned plagioclase) with high resolution at the micrometer-scale. We generate centimeter-scale phase composition maps of several minerals in gabbro samples, an increase of 6 orders of magnitude in area imaged. Our regression analysis works well for plagioclase solid solution, but pyroxenes are more complicated. Augite, Pigeonite, and Orthopyroxene with similar Mg/Fe ratios have overlapping gray scale values because the Mg/Fe ratio exerts much more control over mass and grayscale values than Calcium. To solve this, we include a dummy variable to distinguish high-Ca from low-Ca pyroxenes along with the composition data to build the regression model. The proportion of Ca used as a cutoff to assign the dummy variable is determined by minimizing the sum of the squares of the residuals in the regression analysis; in our example a Ca proportion of 0.25 is the cutoff that maximizes the separation of high- and low-Ca pyroxene. To apply the model for thresholding pyroxenes in a BSE image, an energy-dispersive x-ray map for Ca is used as an independent way of identifying pyroxene types. We present images from our regression model showing the spatial distribution of specific minerals and specific compositional ranges of minerals with solid solution. Applications are discussed in a separate presentation by Srogi et al. We are extending our work to include modeling three co-existing pyroxenes, solving the grayscale overlap between Plagioclase and K-feldspar, and modeling other minerals such as quartz, Fe-Ti oxides, and olivine.