Global Upper Mantle Azimuthal Anisotropy From Probabilistic Tomography

Friday, 19 December 2014: 8:15 AM
Caroline Beghein, University of California Los Angeles, Earth, Space, and Planetary Sciences, Los Angeles, CA, United States and Kaiqing Yuan, University of California Los Angeles, Los Angeles, CA, United States
The new model of Yuan and Beghein (2013), hereafter YBaniSV13, is the first global model to constrain 3-D azimuthal anisotropy in the deep upper mantle. It is compatible with previous models in the uppermost 200km of the mantle, but also displays 1% anisotropy above, inside, and below the Mantle Transition Zone (MTZ). Another interesting characteristic of this model is the change in fast seismic direction detected, on average, at ~250km depth and at the MTZ boundaries. These results have important consequences for our understanding of mantle deformation and convection patterns in the mantle. It is therefore important to assess the robustness if these features. We already tested that the model does not strongly depend on the reference 1-D mantle model, on the presence of discontinuities in this reference model, or on the crustal model and Moho depth used to calculate the laterally varying partial derivatives.

In this work, we apply a model space approach, the Neighborhood Algorithm (NA) of Sambridge (1999), to determine quantitative model uncertainties and parameter trade-offs. First, the NA generates an ensemble of models with a sampling density that increases toward the best fitting regions of the model space, and then performs a Bayesian appraisal of the models obtained that allows us to determine the likelihood of azimuthal anisotropy in different region of Earth’s interior. Such approaches have the advantage of sampling the model null-space, and therefore provide more reliable model uncertainties than traditional inverse techniques. We use YBaniSV13 as initial model, and search the model space around it, allowing for large enough deviations to test the robustness of the anisotropy amplitude. We compare results from a model space search based on the chi-square misfit and from a model space search based on the variance reduction, which is another useful measure of data fit that is independent of data uncertainties. Preliminary results for the chi-square driven search show that most likely model that is close to YBaniSV13, but uncertainties are rather large even in the shallow mantle where azimuthal anisotropy is usually considered to be well-resolved.