Figure 1. Shear-wave fast direction and delay for SKS waves received at a hypothetical station from a variety of directions and apparent velocities (in km/s). Data are shown as a bar centered on the nominal back azimuth and apparent velocity, with the bar's orientation parallel to the azimuth of the the fast direction, and its length proportional to the delay. Near-zero delays are plotted with open circles. (Left) For an earth model with a single hexagonally-anisotropic layer with horizontal symmetry axis overlying an isotropic halfspace; (Right) for a model in which the symmetry axis plunges 45 degrees. Note that the splitting parameters vary most slowly in the horizontal case.
Figure 2. Synthetic SKS phases for two earth models. (Right) Two 50 km
hexagonally-anisotropic layers overlying an isotropic halfspace.
The symmetry axis is horizontal in both layers, and has an azimuth of
N30E in the top layer and N60E in the bottom layer. (Left) One 100 km
anisotropic layer overlying an isotropic halfspace. The
anisotropic tensor is the arithmetic mean of the tensors in the
two-layer case. Radial (top row) and transverse (second row)
horizontal component synthetic seismograms (bold) are computed by
convolving the impulse response functions (solid) with a pulse that has
a bandwidth similar to typical SKS phases. This SKS pulse ( 
, 20 km/s) (third row) is reconstructed using apparent
splitting parameters computed by the cross-correlation method. Note
that the two layer case has the larger error (bottom row).
Figure 3. Apparent splitting parameters for SKS waves received at a hypothetical station. (Right) For an earth model with two 50 km hexagonally-anisotropic layers overlying an isotropic halfspace. The symmetry axis is horizontal in both layers, an has an azimuth of N30E in the top layer and N60E in the bottom layer. (Left) For an earth model with one 100 km anisotropic layer overlying an isotropic halfspace. The anisotropic tensor is the arithmetic mean of the tensors in the two-layer case. Splitting parameters are computed by the cross-correlation method from synthetic SKS phases. Note that the two layer case has the more complex pattern.
Figure 4. Influence of the filtering applied to the synthetic waveforms on the apparent splitting parameters. Waveforms are simulated in a 2-layer model developed for station HRV (see Table 2). Apparent splitting parameters measured from waveforms with an upper spectral limits of 0.45Hz (bold), 0.35Hz (solid) and 0.15Hz (dotted) are shown, with a lower limit of 0.05Hz common for all. For all cases tested the estimate of the apparent fast direction tends to vary greatly when the delay is the smallest. The lowest bandpass (0.15Hz) gives most unstable results throughout.
Figure 5. Map of NE Appalachian region, with "average" parameters of seismic anisotropy plotted at points where they were constrained by various workers. Arrow azimuths correspond to the fast directions determined for the particular site, and are scaled with estimated delay. The compilation is extracted from the "Anisotropy Resource Page" maintained by Derek Schutt ( http://darkwing.uoregon.edu/~schuttd/aniso_source.html). Individual values are from Silver and Chan, [1991]; Bostock and Cassidy, [1995]; Barruol et al., [1997] and Fouch et al., [1999].
Figure 6. Histograms of single seismogram splitting parameter errors for HRV data. (top)
One 
error in delay time. (bottom) One error in
fast direction azimuth.
Figure 7. Map of NE Appalachian region showing shear-wave splitting data for two earthquakes with different propagation directions (large one-sided arrows) observed in 1995. Splitting azimuth and delay are shown at each station as two-sided arrows aligned with the fast direction and scaled with delay. Symbols are color-coded by event, black for one and white for the other. Note that splitting directions for the two events is quite different, yet is fairly consistent across the region for each event.
Figure 8. Shear-wave splitting data for PAL and HRV. Note that the
pattern is similar for the two stations, but varies rapidly with
back azimuth. Splitting directions for 
phases (two data points from back azimuth 
for HRV,
three data points from back azimuths 
for PAL) closely match those obtained for SKS phases from the same events. Thus contamination of the signal by the D'' anisotropy [Garnero and Lay, 1997] does not seem to occur along these particular paths. S phases from South American earthquakes with hypocenters deeper then 500 km are included to provide coverage from the south.
Figure 9. Shear-wave splitting data for the subset of PAL and HRV data
that fall within 
of each other, superimposed on one another
(black, HRV; grey, PAL). Note overall similarity of pattern.
Figure 10. Histogram of the angle between pairs of fast azimuth measurements for the station HRV. The distribution is bimodal, suggesting that two distinct fast azimuths are present.
Figure 11. SKS waves observed at HRV for 4 different backaziumths, A) SSE, B) WNW, C) NNW and D) NNE. (Left) Observed radial (top row) and transverse (3rd row) horizontal component seismograms. The significant energy observed on the transverse component is an effect of the seismic anisotropy. Corrected radial (2nd row) and transverse (bottom row) component seismograms, where the effect of propagation through the anisotropic medium has been removed. (Right) Particle motion diagrams (top) before and (bottom) after correction. Note that the energy on the transverse component has been greatly reduced, and the particle motion made significantly more linear, indicating that the splitting parameters have been correctly calculated.
Figure 12. Anisotropic tensor orientations for (top) 1 layer and (bottom) two layer models for HRV. The fast, intermediate and slow axes are denoted F, I and S, respectively. Top and Bottom layers are denoted T and B, respectively.
Figure 13. (Left) Shear-wave splitting patterns for best-fitting one-layer (top) and two-layer (bottom) HRV models. (Right) Observed (gray) and predicted (black) shear-wave splitting parameters.
Figure 14. Observed and predicted variation of the apparent fast direction
at HRV. Observations are shown by triangles with error bars. A subset
of "robust" data points (circled) was chosen so that 
. Clearly, "robust" and "poor" data points follow the same
pattern. Crosses show values of fast direction reported for HRV by
Barruol et al. [1997] (as given in electronic supplement table
2). Thick lines show patterns predicted by our orthorhombic (solid)
and hexagonal (dotted) models, thin lines show predictions for
equivalent 2-layer splitting operators [Silver and Savage, 1994].
While all models capture the periodicity of the pattern, the spread of
values and the deviations from the 
pattern are not matched by
the splitting operator predictions.
Figure 15. A schematic representation of the model for seismic anisotropy distribution under HRV.
Figure 16. Regional setting of GSN station ARU (open triangle) and the geometry of inferred anisotropy in the lithosphere. Grey lines - topography contours at 330 and 560 m. Wide grey line: the main Uralian Fault Zone. Open arrow: the symmetry axis of lower-crustal anisotropy. Grey bars: inferred fast axes of anisotropy for the two layers in the mantle.
Figure 17. Shear-wave splitting data for ARU (Arti, Russia). See Figure 1 for plotting conventions.
Figure 18. (Left) Shear-wave splitting patterns for best-fitting one-layer (top) and three-layer (bottom) ARU models. (Right) Observed (gray) and predicted (black) shear-wave splitting parameters.
Figure 19. Symmetry axes of best-fitting hexagonal tensors for ARU. Crust and mantle are labeled
C and M, respectively. Top layer and bottom layers in the mantle are labeled T and B,
respectively. In the bottom layer of anisotropy the symmetry
axis plunges 
to the east, which is equivalent to an upward tilt of to the west.
Figure 20. A schematic representation of the model for seismic anisotropy distribution under ARU (marked by a flag), with the lowermost crust anisotropy (dark arrow marked "slow") from the model of Levin and Park, [1997a].