A statistical assessment of seismic models of the U.S. continental crust using Bayesian inversion of ambient noise surface wave dispersion data

We present a new approach for evaluating existing crustal models using ambient noise datasets and its associated uncertainties. We use a transdimensional hierarchical Bayesian inversion (THBI) approach to invert ambient noise surface wave phase dispersion maps for Love and Rayleigh waves using measurements obtained from Ekstrom (2014). Spatiospectral analysis show that our results are comparable to a linear least-squares inverse approach (except at higher harmonic degrees), but the procedure has additional advantages: (1) it yields an auto-adaptive parameterization that follows earth structure without making restricting assumptions on model resolution (regularization or damping) and data errors (2) it can recover non-Gaussian phase velocity probability distributions while quantifying and separating the sources of uncertainties in the data measurements and modeling procedure and (3) it enables statistical assessments of different crustal models (e.g., CRUST1.0, LITHO1.0 and NACr14) using variable resolution residual and standard deviation maps estimated from the ensemble. These assessments show that in the stable old crust of the Archean, the misfits are statistically negligible requiring no significant update to crustal models from the ambient noise dataset. In other regions of the US, significant updates to regionalization and crustal structure are expected especially in the shallow sedimentary basins and the tectonically active regions, where the differences between model predictions and data are statistically significant.