Understanding the impact of seismic data errors and tomographic model uncertainties on geodynamic inverse simulations of mantle evolution

• Applicant: Dr. Bernhard Schuberth
• Subject Area: Geophysics
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 456788150

Increasing the resolution of tomographic images has been one of the major efforts in seismology in the last decades, and great progress has been made based on methodological advancements and better data coverage yielding images at unprecedented detail. Nowadays, the focus is shifting towards additionally providing formal means to quantify image resolution and model uncertainty, which are crucial aspects for linking seismic data to geodynamic simulations. Geodynamic inversions that aim at `retrodicting' the past evolution of Earth's mantle require tomographic estimates of its present-day state, and they would benefit greatly from knowledge of the associated error bars. More robust inferences could be drawn from running a suite of simulations instead of a single model realization. The SOLA Backus–Gilbert tomographic scheme has recently gained attention in this respect, as it provides information on both resolution and model uncertainty and can be applied to large inverse problems on the global scale. The generalized inverse operator can explicitly and efficiently be calculated in this method, which makes it possible to project the data uncertainty to the model space. However, so far it is unclear how to make practical use of the tomographic model uncertainty in simulations of mantle convection. The open questions are, for example, how to incorporate the tomographic covariance information in geodynamic inversions, given that only a limited set of retrodictions is computationally feasible, and whether higher tomographic resolution is to be preferred over lower uncertainty.We will answer these questions based on a synthetic closed-loop retrodiction experiment. Starting from a forward mantle circulation model that is considered as known `true' mantle structure, we predict synthetic seismic data that is subsequently projected back to the (tomographic) model space using the generalized inverse operator of a recent SOLA tomography. This will be repeated many times with varying realizations of `noise' (i.e. data errors) added to the seismic data. Finally, the `tomographically filtered' model and the associated covariance will serve as the basis for an ensemble of adjoint geodynamic retrodictions. This theoretical experiment will allow us to quantify —in the light of realistic seismic data noise— differences of initial `true' and retrodicted models and to track the evolution of associated precision and accuracy back in time into the geologic past. The systematic uncertainty quantification, from the seismic recordings to geologically assessable predictions, will provide important understanding on the propagation and evolution of errors in geodynamic retrodictions, which is crucial to draw robust conclusions on the adequateness of model parameters. Our project will thus impact future geodynamic simulations and the way the next generation of data-driven Earth models will be constructed that provide quantitative information to other geoscience disciplines.

• DFG Programme: Research Grants
• International Connection: France
• Co-Investigator: Professor Dr. Hans-Peter Bunge, Ph.D.
• Cooperation Partner: Dr. Christophe Zaroli