Anisotropic Shear Velocity Models of the North American Upper Mantle Based on Waveform Inversion and Numerical Wavefield Computations.

Thursday, 18 December 2014
Clouzet Pierre1, Yder Masson1, Barbara A Romanowicz2,3, Scott W French4 and Huaiyu Yuan5, (1)Institut de Physique du Globe de Paris, Paris, France, (2)Institut de Physique du Globe de Paris, Coll├Ęge de France, Paris, France, (3)Univ California Berkeley, Berkeley, CA, United States, (4)Univ. of California, Berkeley, Berkeley, CA, United States, (5)Macquarie University, CCFS, Sydney, NSW, Australia
The Earthscope TA deployment across the continental US now has reached the eastern part of the United States, providing the opportunity for high-resolution 3D seismic velocity imaging of both lithosphere and asthenosphere across the entire north-American continent (NA). Previously (Yuan et al., 2014), we presented a 3D radially anisotropic shear wave model of North America (NA) lithospheric mantle based on full waveform tomography, combining teleseismic and regional distance data sampling the NA. Regional wavefield computations were performed numerically, using a regional Spectral Element code (RegSEM, Cupillard et al., 2012), while teleseismic computations were performed approximately, using non-linear asymptotic coupling theory (NACT, Li and Romanowicz, 1995). For both datasets, the inversion was performed iteratively, using a Gauss-Newton scheme, with kernels computed using either NACT or the surface wave, path average approximation (PAVA), depending on the source-station distance.

Building upon our previous work, we here present a new radially anisotropic lithospheric/asthenospheric model of shear velocity for North America based entirely on regional waveforms from an augmented dataset of ~150 events contained and observed inside the study region, with forward wavefield computations performed using RegSEM down to 40s, starting from our most recent whole mantle 3D radially anisotropic shear velocity model (SEMUCB-wm1, French and Romanowicz, 2014). Several iterations of inversion are performed using a Gauss-Newton scheme. We present and compare two models obtained, on the one hand, using NACT/PAVA kernels as in our previous work, and on the other, using hybrid kernels, where the Hessian is computed using NACT/PAVA, but the gradient is computed numerically from the adjoint wavefield, providing more accurate kernels while preserving the fast convergence properties of the Gauss-Newton inversion scheme. We also present an update to our azimuthally anisotropic shear velocity model under NA (Yuan and Romanowicz, 2010), obtained starting from the final radial anisotropy model and including a large dataset of station-averaged SKS splitting data over NA.