Observations of shear-wave splitting in the Northeastern US
Appalachians and in the foredeep of the Urals vary significantly with the
back azimuth and incidence angle of the phase.
To analyze these datasets properly we developed a new technique for
estimating uncertainties of splitting parameters.
Using this technique we find that typical errors of the shear-wave
splitting parameters determined from low-passed broadband data from GSN
station HRV are 
for the fast direction, and
0.1-0.2s for the delay.
Experiments with synthetic seismograms generated in simple multilayered anisotropic structures show that splitting parameters tend to vary significantly with the back-azimuth of the analyzed shear wave. A restricted subset of back azimuths may strongly bias any model derived from observations, especially if the observations are averaged. On the other hand, the azimuthal variation pattern provides important constraints on vertical or lateral variation of anisotropic properties in the Earth.
On the basis of data from well-recorded events with different
back azimuths, splitting parameters appear to
be broadly consistent throughout the Appalachian terranes in the
Northeastern US.
(This consistency weakens for stations
west of the Appalachians.)
A close similarity in back-azimuth dependence of splitting
parameters is found in data from two long-running stations in the
Northeast US - HRV and PAL. Good back-azimuth coverage at
these two stations allows us to separate observations into two
statistically significant populations. Within these populations mean
azimuths are 
and 
, and
delay values vary within each population from near-zero to 
.
The exact values of delays, as well as individual estimates of fast
direction, are affected by the filter parameters chosen when low-passing.
The back-azimuth dependence of splitting parameters for the station ARU
near the Urals is characterized by sharp transitions between different
groups of observations.
Using synthetic seismograms computed in flat-layered media we developed
one-dimensional models of seismic anisotropy distribution under
stations HRV and ARU. The model for HRV contains two layers of
anisotropic material under an isotropic crust, with fast-axis azimuths
of 
and 
for the bottom and the top layers,
respectively. Depending on the choice of symmetry for the elastic
tensors, these axes are tilted (hexagonal symmetry)
or near-horizontal (orthorhombic symmetry). Assuming 30% orthorhombic
olivine and 70% isotropic olivine, a mixture that is about 6%
anisotropic, the vertical dimensions are 60 and 90 km for the top and
bottom layers, respectively.
The model for ARU includes crustal structure that was constrained
using Ps converted phases [Levin and Park, 1997a].
Assuming hexagonal symmetry of the upper mantle anisotropy, the model for ARU
predicts a 
60 km layer with a fast-axis at 
atop a 140 km
layer with a fast-axis plunging 40 
towards 
.
The analysis performed in this paper was made possible by good azimuthal coverage of observations. These are generally obtainable through prolonged observation. Data from short deployments, even in stable continental regions, apparently run the risk of bias from an uneven distribution of seismicity.