Apparent anisotropy in inhomogeneous isotropic media

Surface waves propagating through a laterally inhomogeneous medium undergo wavefield complications such as multiple scattering, wavefront healing, and backward scattering. Unless accounted for accurately, these effects will introduce a systematic isotropic bias in estimates of azimuthal anisotropy. We demonstrate with synthetic experiments that backward scattering near an observing station will introduce an apparent 360° periodicity into the azimuthal distribution of anisotropy near strong lateral variations in seismic wave speeds that increases with period. Because it violates reciprocity, this apparent 1ψ anisotropy, where ψ is the azimuthal angle, is nonphysical and is, therefore, a useful indicator of isotropic bias. Isotropic bias of the 2ψ (180° periodicity) component of azimuthal anisotropy, in contrast, is caused mainly by wavefront healing, which results from the broad forward scattering part of the surface wave sensitivity kernel. To test these predictions, we apply geometrical ray theoretic (eikonal) tomography to teleseismic Rayleigh wave measurements across the Transportable Array component of USArray to measure the directional dependence of phase velocities between 30 and 80 second period. Eikonal tomography accounts for multiple scattering (ray bending) but not finite frequency effects such as wavefront healing or backward scattering. At long periods (>50 sec), consistent with the predictions from the synthetic experiments, a significant 1ψ component of azimuthal anisotropy is observed near strong isotropic structural contrasts with fast directions that point in the direction of increasing phase speeds. The observed 2ψ component of azimuthal anisotropy is more weakly correlated with synthetic predictions of isotropic bias, probably because of the imprint of intrinsic structural anisotropy. The observation of a 1ψ component of azimuthal anisotropy is a clear indicator of isotropic bias in the inversion caused by unmodeled backward scattering and can dominate and mask the 2ψ signal. Observers are encouraged to estimate and report 1ψ anisotropy in their inversions for azimuthal anisotropy, to model finite frequency effects using methods that are tailored to the method of measurement, and to estimate 1ψ and 2ψ anisotropy simultaneously.