Probabilistic Geodynamic Modeling as a New Step Towards Interdisciplinary Investigation of the Upper Mantle

A number of geological and geophysical techniques have been used to investigate different aspects of the Earth's upper mantle. Combining observations from various fields can therefore provide a more comprehensive understanding of the mantle structure, composition, and dynamics. However, uncertainties associated with all experimental and observational data can introduce nonuniqueness to their interpretation. Any interdisciplinary research must account for the uncertainties in their data and model assumptions to develop a more unbiased and robust understanding of the upper mantle. Unfortunately, such data analyses have often been overlooked by previous studies, making it difficult to gauge the reliability and scope of their work. To address this oversight, we propose a novel probabilistic approach to geodynamic modeling based on Bayesian inference that presents a means to combine experimental rock mechanics, theoretical mantle dynamics, and observational geophysics into a self-consistent and statistically-sound framework (Jain and Korenaga, 2020). As an example, we simulate steady-state plate-driven mantle flow beneath the Pacific mid-ocean ridge and, based on the results, compute the radial anisotropy in the suboceanic upper mantle. Our simulations use flow laws for diffusion and dislocation creep in olivine to approximate upper mantle viscosity. The versatility of the probabilistic modeling scheme allows us to explore the influence of various model assumptions, including those on upper-mantle water content, background shearing of the lithosphere-asthenosphere boundary, and the uncertainties associated with experimentally-derived flow laws for olivine rheology, on our predictions of radial anisotropy. Comparison of our predictions with seismically-derived data on radial anisotropy yields new geophysically-relevant constraints on all the aforementioned variables. The uncertainties associated with our results highlight avenues for further research. Our new probabilistic approach to modeling can easily be adapted to any interdisciplinary research in the geosciences.