The complicated nature of seismic anisotropy in the Northeastern US has
been noted in a number of previous papers.
Levin et al [1996] reported lateral variations in observed
splitting parameters of individual phases that appeared to correlate
with tectonic units of the region.
Working with averaged splitting parameters, Barruol et al [1997]
suggested that throughout the eastern North America (a region much
broader then this study) there is a good fit between surface geology
and fast direction 
.
Barruol et al [1997] described our study region as "anomalous"
because the averaged fast-axis orientations they obtained were
near-normal to the strike of the local tectonic features.
More recently, Fouch et al. [1999] showed that averaged splitting
parameters in a broad region of the eastern North America may also be
explained by the mantle flow pattern around the keel of the North
American craton.
In their results our study region stands out as an area where the fit
to their single layer flow-controlled splitting model is the poorest.
Further observations will be needed to determine how our two-layer
model joins with the rest of Eastern North America, where other studies
suggest that one layer suffices.
In this paper we examine splitting parameters of individual phases, not station averages. A comparison of our measurements (Figure 5) with those of Fouch et al. [1999] shows a measure of disagreement. We were able to resolve splitting parameters for a number of phases that appear as "nulls" in Fouch et al. [1999], most likely because we used a passband that is richer in high frequencies. Frequency dependence in splitting parameters was shown by [Marson-Pidgeon and Savage, 1997] to influence parameter retrieval. In particular, Levin et al., [1999] show that splitting parameters estimated from synthetic seismograms simulated in the HRV anisotropic model become progressively less stable as the the high-frequency limit of the spectral passband is lowered. Our present results, as well as our previous examination of data from the two long running stations HRV and PAL [Levin et al., 1999]), indicate that azimuthal variations in shear-wave splitting occurs, and must be factored into data interpretation. We find that the vertical stratification of anisotropic parameters is required at all sites, and that it has only minor lateral heterogeneity throughout the region. Our results for the northeastern US indicate that neither surface geology nor the distribution of the shear velocity at depth seem to govern the distribution of anisotropy.
As discussed in the introduction, uniformity of the anisotropic
parameters over a broad region would suggest a dynamic origin of the
fabric in the mantle [Vinnik et al., 1992]. Given the orientation
of the fast direction in the lower layer of the model (N 
E), it
appears natural to associate it with the mantle flow under the North
American continent, that has an absolute plate motion vector of 
[Gripp and Gordon, 1990].
Interpreting the upper layer of the model in terms of dynamic fabric is
problematic, though, as it would require two regions with distinctively
different strain to exist one above the other in the mantle.
Placing the upper layer of our model within the consolidated
lithosphere of the continent raises another problem - the sense of
fabric in this layer needs to be uniform throughout the region composed
of various tectonic units and characterized by significant seismic
velocity variations.
Given the diversity of ages and tectonic histories of the constituent
tectonic units, it is reasonable to assume the that the fabric-forming
event should post-date the assembly of the region in in present form.
Two possible causes for the anisotropy are (1) a continent-continent
collision during the final stages of the Appalachian orogeny and (2)
the rifting that led to the eventual opening of the Atlantic Ocean.
Continental collisions are believed to result in orogen-parallel flow
in the mantle [Vauchez and Nicolas, 1991].
This would lead to the orogen-parallel orientation of fast directions.
In our model the fast direction within the upper layer (N 
E) is
nearly normal to the strike of the tectonic units in the region.
Similarly, studies of shear-wave splitting in the regions of
active rifting [Sandvol et al., 1992, Ben-Ismail and
Barruol, 1997] report fast-axis directions aligned along the axis of
the rift.
This would be approximately north-south in our region, a poor match to
the observed fast-axis direction in the "frozen-fabric" upper layer.
We suggest an alternative mechanism that might form a regional strain fabric in the lithosphere that would align roughly normal to the strike of the orogen. Houseman et al., [1981] propose that, in the process of continent-continent collision, a thickened root of the mantle lithosphere will form, and may delaminate from the overlying crust and uppermost mantle. Recent computer simulations of this process [Schott and Schmeling, 1998] show that a region of significant strain directed towards the center of the orogen forms along the plane of future delamination and detachment. The delamination process involves the entire width of the orogen. If such delamination indeed occurred under this part of the Appalachian Orogen, it would have involved the Grenvillian terranes on its flanks as well as the younger terranes in the center, and would have left in the lower continental lithosphere the strained layer with fabric aligned in general east-west direction. A uniform east-west extensional strain fabric would remain in the ``scarred'' lithosphere left behind after delamination, assuming that the downgoing plume retained a rough 2-D geometry beneath the orogen.