Constraining the Mantle's Rheology Using Methods in Uncertainty Quantification

Citation

Ratnaswamy, Vishagan (2019) Constraining the Mantle's Rheology Using Methods in Uncertainty Quantification. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/F6FW-T648. https://resolver.caltech.edu/CaltechTHESIS:05132019-143045769

Abstract

An accurate estimation of the large-scale forces in the mantle has been difficult to obtain as numerical models either do not use an accurate rheology nor reproduce surface observations. While much work has been done in developing high-fidelity forward models that capture the salient physics of shear-thinning and dynamic weakening, they fail to reproduce observations such as plate motions and topography. In this thesis, we develop an optimization methodology that minimizes the misfit in surface observations such as plate motions and average effective viscosity for certain regions of the mantle. We utilize adjoints to calculate the gradient, while using second-order adjoints to construct the Hessian so as to infer the rheological parameters of the mantle's rheology. Furthermore, we build on this optimization scheme by constructing the Gaussian approximation of the posterior distribution for the inferred rheological parameters using the Hessian and establish the trade-offs between each parameter through their conditional distributions. We further extend this Gaussian approximation to infer extrinsic quantities such as the stresses in the fault zones and the average effective viscosity in the hinge zones to not only quantify the uncertainty, but also to see partitioning of the coupling of each subduction zone.

Thesis Files