In both Appalachians and Uralides we find evidence for at least two distinct layers of seismic anisotropy in the mantle. The exact depth and thickness of layers is subject to assumptions about the percent of aligned minerals within the volume. Anisotropic intensities in our models follow measurements done on hand-samples of peridotite from ophiolites [Christensen, 1984], and are likely to represent the upper bound in percent of alignment. If the alignment of minerals is weaker, the required thicknesses of the anisotropic layers will increase. The tilt of inferred anisotropy-inducing fabric with respect to the horizontal is a function of the symmetry system chosen for the anisotropy. For the Appalachian stations, both orthorhombic anisotropy, with nearly horizontal fast axes, and hexagonal anisotropy, with tilted fast symmetry axes, satisfy the data. We suspect that this tradeoff will be common in studies of this type, and difficult to resolve without a clear indication of the best symmetry choice from mineral texture studies. Significantly, in both modeling exercises the horizontal azimuth of the fast axis does not depend on the symmetry system and the method of computing synthetic seismograms. We thus believe that robust elements of our models are the minimum number of anisotropic layers, their vertical sequence, the orientation of fast direction and the cumulative anisotropy within each of them.
The overall regional consistency of observations in the Northeastern Appalachian region (Figure 7) argues against a "local" character for shallow-mantle structures revealed by shear-wave splitting. Rather, the structure modeled using the HRV dataset appears to be common to the area of Appalachian terranes. On the other hand, some regional variation is present (e.g., Levin et al, [1996]). Whether it is caused by the "true" lateral variation in anisotropic features of the subsurface, or simply reflects changes in geometry of observation, is an open question.
In case of the Uralian foredeep we have a "spot" measurement, and
arguing for its regional extent is harder. Major geologic structures
of the Ural Orogen are very consistent along strike, making it almost
2-dimensional.
An average splitting direction of 
(sub-parallel to the strike of the Urals) reported by [Vinnik et al., 1992] at the station SVE near Yekaterinburg (Figure 16) falls on the opposite side of the main
Uralian Fault Zone. This major suture divides accreted terranes from the East European Platform [Zonenshain et al., 1984]. The splitting at SVE most likely reflects a different structure in the crust and the uppermost mantle.
Given the uncertainties discussed above, the interpretation
is necessarily tentative.
If we consider the anisotropy to be "frozen in" or "fossil," confined
within the continental keel, in both Appalachian and Urals we may infer
at least two distinct past tectonic episodes.
Abbott [1992] describes a conceptual model for accumulating
material with different deformation fabrics within the body of a
continent, via underplating the continent with oceanic-plate material
during successive episodes of subduction.
A similar model, only with oceanic "slabs" stacked in the lateral
direction, has been advocated for Western Europe [e.g. Babuska et
al., 1993, Plomerova et al, 1996].
The "frozen fabric" explanation appears plausible for the upper
layers in both modeled regions. In Urals the lowermost crust has the same
orientation ( 
, from Levin and Park [1997a]) as the
upper layer of our mantle model.
Interestingly, the sense of anisotropy
reverses from the crust (slow axis) to the mantle (fast axis). We
believe that crustal anisotropy in the crust is
imposed by fine layering of materials with contrasting properties,
while in the mantle it is imposed by
preferred alignment of peridotite minerals. The strike of fabric in
lower crust and uppermost mantle under ARU is oblique to the strike of
the orogen, and may reflect a tectonic episode predating its
formation.
Analysis of crustal P-SH conversions [Levin and Park, 1997b]
under HRV did not reveal strong crustal anisotropy of the kind seen
under ARU. The upper layer of the mantle anisotropy has fabric
orientation that is roughly normal to the geomorphological features and
main tectonic boundaries in the region. Numerous subduction episodes,
both east-verging and west-verging, took place during the formation of
the Appalachians, and the fabric we reconstruct is likely to be a
remnant of one of them.
An alternative mechanism for anisotropy in the mantle - active flow in
the asthenosphere [e.g., Vinnik et al, 1992] - appears more
suitable for the lower layers in our models.
For the Appalachians the strike
of fabric in the lower layer of our model aligns closely with the
absolute plate motion vector of 
[Gripp and
Gordon, 1990]. It also aligns with the hypothetical edge of the
North American continental keel.
The keel edge would direct the orientation of asthenospheric flow if
one assumes that mantle moves "west" relative to the North American craton, and
around its keel [Fouch et al., 1999].
The Urals are located in the middle of the Eurasian
continent which is near-stationary with respect to the hotspot
reference frame.
The direction of possible motion for Eurasia is approximately
east-west, if one ignores the error bars on the rate of motion in
Gripp and Gordon, [1990].
This direction, though poorly constrained, aligns with the fast axis of
the lower layer in the anisotropic structure we infer under ARU.
Our results clearly contradict the notion that fast-axis strike for shear-wave splitting in a region of compressional deformation should align with the strike of the orogen [Vauchez and Nicolas, 1991; Nicolas, 1993; Vauchez and Barruol, 1996]. In two regions of Paleozoic deformation we find patterns of seismic anisotropy consistent with vertically heterogeneous anisotropic structure. It is possible that in such complex regions a subset of measurements will indeed line up with the strike of an orogen. Our experiments with multilayered anisotropic structures show that direct mapping of splitting parameters (or their averages) onto tectonic features can be misleading.