Crustal and mantle structure of the North America Continent using SEM based Full Waveform Seismic Tomography and Trans-Dimensional Inversion.

Wednesday, 17 December 2014: 9:00 AM
Marco Calo1, Thomas Bodin1, Barbara A Romanowicz1,2, Clouzet Pierre3, Carene S Larmat4 and Monica Maceira4, (1)Berkeley Seismological Laboratory, UC Berkeley, Berkeley, CA, United States, (2)Institut de Physique du Globe de Paris, Coll├Ęge de France, Paris, France, (3)University Denis Diderot Paris VII, Paris Cedex 13, France, (4)Los Alamos National Laboratory, Los Alamos, NM, United States
The construction of a 3D upper mantle continental scale tomographic model using long period waveform data involves the choice of a reference model, which is generally smooth in the vertical direction, including only the Moho and transition zone discontinuities. The resulting 3D model therefore cannot explicitly account for strong lateral variations in features such as the mid-lithospheric boundary or the lithosphere-asthenosphere boundary, when it is sharp, so that trade-offs between smooth volumetric heterogeneity and the topography of such boundaries cannot be resolved.

We present the first results of a time domain waveform tomography of the north American (NA) continent obtained using the spectral element method (SEM) for wavefield computations and a 3D starting model that contains upper mantle discontinuities determined in a prior step from receiver function analysis. The model is constructed using a procedure that consists of 5 steps:

1) We apply k-means cluster analysis to a recent global tomographic model to regionalize NA into three main provinces (Oceanic, transitional, continental).

2) We calculate 1D Vs and radial anisotropy (Xi) profiles at 26 stations deployed in the NA continent using a trans-dimensional Markov-chain algorithm. Here, we jointly invert three datasets; (i) Rayleigh and Love phase velocity dispersion, (ii) Rayleigh and Love group velocity dispersion, (iii) Averaged seismograms for calculating receiver functions using the cross-convolution method (Bodin et al., 2014).

3) A 1D model is obtained for Vs and Xi in each of the 3 provinces by averaging the models obtained in 2) for that province and these layered 1D models are connected laterally using smoothing operators based on inter-station distance to generate a layered 3D starting model.

4) We calculate smooth equivalent models of Vs and Xi using the homogenization procedure of Capdeville et al. (2013)

5) We perform a SEM based waveform tomographic inversion of our regional distance database obtaining smooth 3D velocity perturbations which are then added to the layered 3D starting model.

After several iterations, the model obtained includes lateral variations in the depth of significant boundaries, constrained by high frequency data, as well as lateral volumetric velocity variations, constrained by the longer period waveforms.