PATHFINDING BONACCORDITE ABUNDANCE: USING MACHINELEARNING TO MAKE MULTISCALE PREDICTIONS

Situated in the lowest serpentinized layers of the Barberton Greenstone Belt (South Africa), the ~3.5 Ga Bon Accord nickel oxide deposit comprises a globally unique combination of Ni-rich minerals. In particular, the abundance of the Ni-Fe3+-spinel trevorite has contributed to the interpretation of this mined out deposit as either the result of a Fe-Ni meteorite impact, or as a fragment of the Archean Earth’s core. However, recent petrographic and geochemical studies have pointed out the association of Bon Accord with metasomatised ocean floor komatiite lavas and suggested instead the deposit may be the remains of an ancient hydrothermal vent. One mineralogical line of evidence supporting this model is the only known natural appearance of the mineral bonaccordite, a boron-bearing Ni-oxide. Anthropogenic appearances of this mineral are known to occur as precipitates on the fuel rods of nuclear reactors, demonstrating the role of high temperature fluid flow in the growth of this mineral.

We present the results of a work program designed to constrain the abundance of bonaccordite in a sample of the Bon Accord material by using a multiscale-multimodal approach guided by a physics-informed machine learning model. Trevorite and bonaccordite should theoretically be distinguishable using electron back-scatter diffraction, due to their low symmetry relationships. However, this requires knowing a priori the specific sample locations of the bonaccordite . We present a multiscale approach where we leverage physics-based priors to predict locations of bonaccordite grains from energy dispersive spectroscopy despite standard x-ray spectrometers being insensitive to boron. We feed these predictions forward to other microanalysis techniques where we can test them. The framework set out here is universally applicable to multiscale correlative challenges where nano-scale features need to be isolated from a complex background. For Bon Accord in particular, the results of this work have the potential to shed new light on fluid-rock (hydrothermal) processes in the Paleoarchean.