CAREER: Hypothesis Testing Mantle Structure with Probabilistic Tomography

Wayne State University, Detroit MI

Investigators

Abstract

Seismic tomography, which uses waves produced by earthquakes to create 3D models of deep Earth structure, is an indispensable tool for understanding the Earth's mantle. Tomographic models are used to address questions about Earth’s tectonic past, geodynamics, and the processes underlying natural hazards, in addition to helping locate earthquakes and enforce test ban treaties. However, due to difficulty in estimating complex model uncertainties, these questions are met largely with qualitative analysis. In order to move towards quantitative hypothesis testing of mantle structure, this CAREER project will apply a probabilistic framework to estimate the full range of Earth models that agree with seismic data and other constraints. These models of seismic properties will be constructed using large earthquake datasets, first for the mantle beneath Alaska, then the entire globe, and novel analyses will be applied to answer geologic and geodynamic questions. A new undergraduate team research course will apply the principles espoused by this project to environmental problems in Detroit, MI. Students at different experience levels will form teams to identify targets, define testable hypotheses, and design geophysical surveys to assess them. Research results will also be incorporated into two planetarium shows, one on the geologic history of Michigan and one on natural hazards, to be presented regularly at the Wayne State Planetarium to K-12 schools, undergraduate classes, and the public. The project aims to create new mantle models with fully quantified uncertainty and forward a framework to compare model probabilities to predictions made by geology and geodynamics. Research activities will be motivated by four major questions: 1) How well do body wave data constrain mantle velocity variations? 2) To what extent do hypotheses posed by plate reconstructions and sinking rates agree with seismic structure? 3) Which mantle convection models comport with the scale and distribution of heterogeneity? 4) Which data sources, seismic phases, and prior information best resolve mantle properties? These questions will be addressed by employing Transdimensional Bayesian Inference (TBI) methods to create global and regional ensembles of mantle P- and S-wave velocity models and transition zone thickness and amplitude. Ensembles will be evaluated using novel analyses, resulting in constraints on global slab properties, estimates of the existence and geometry of plumes, and insights on plate reconstructions. Assessment of the upper mantle will be bolstered by a new TBI method for transition zone topography using machine learning-picked precursor phases. Research activities will produce efficient GPU-accelerated software for 3D TBI tomography, new methods for picking transition zone-reflected phases and inferring phase transition topography, and approaches for incorporating TBI results as prior information. Practices for interrogating, visualizing, and sharing high-dimensional ensembles and posterior distributions will be defined. This project is co-funded by the Directorate for Geosciences to support AI/ML advancement in the geosciences, and by the Division of Earth Sciences program in Education and Human Resources. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.