Evaluating models of the US Continental Crust using Ambient Noise Datasets: A Transdimensional Approach

Abstract:

Seismological information on crustal structure, when combined with other geophysical, petrological and geochemical constraints, provides our best insight into the structure and composition of Earth’s continental crust. Traditionally, published models of the continental crust (e.g. Vp profiles) constrained by seismic reflection, refraction experiments and (or) receiver functions are used to infer the thickness and lithology of the upper, middle, and lower crust. In order to be most useful, however, inferences of crustal structure and composition made using seismological models as data need to be presented alongside their uncertainties. Quantifying this uncertainty is the challenge.

In this work, we use the transdimensional hierarchical Bayesian inverse (THBI) approach to construct phase velocity maps for surface waves (5s - 40s), while quantifying uncertainties in the phase velocities (model variance and co-variance), as well as data noise (which can affect the model results), all without having to make a-priori choices of model parameterization or regularization.

We show that phase velocity maps constructed with THBI are comparable to those from linear, least-squares inversion at long wavelengths, though differences are present across length-scales. Average phase dispersion curves extracted from our maps are statistically significantly different from predictions from a selection of global and regional crustal models (e.g., CRUST1.0 and NACR14), particularly in the western USA and across major sedimentary basins. These model-data inconsistencies suggest that tectonic regionalization and averaging can introduce significant errors into the crustal models.

We also preview a second stage analysis where we use the phase-dispersion curves (and their uncertainties) as data to constrain 1-D velocity profiles (with radial anisotropy) across distinct geological provinces in the continental US. We demonstrate, again using the THBI approach, the consequences of more accurate treatment of uncertainty on inferences on the 1-D velocity models.