Correlated Errors in Seismic Anisotropy Data and Implications for Current Absolute Plate Motion

A longstanding problem in global tectonics is whether 'absolute' plate motions, the motions of the plates relative to an external reference frame, usually taken to be the lower mantle, can be usefully estimated. For several decades, the most widely used method for estimating absolute plate motions has been from the trends (and in some cases the propagation rates) of hotspot tracks. An alternative method for estimating absolute plate motions arises from the orientation of seismic anisotropy inferred from shear-wave splitting (mainly from SKS), which in many places may indicate the direction of the motion of the lithosphere relative to the sub-asthenospheric mantle. To estimate absolute plate motion, Kreemer (2009) compiled a data set of 474 shear-wave splitting data, which we refer to as the SKS data set. Prior studies assume that errors in the azimuths inferred from shear-wave splitting are uncorrelated, which results in minuscule confidence limits. We test this assumption and find instead that the residuals to azimuths within any one plate are strongly correlated. Here we take these correlations into account by adopting a two-tier analysis, similar to the typical method used for analyzing paleomagnetic data. First we find the pole of rotation and confidence limits for each plate individually. Typically this results in a confidence region that is narrow in one well-constrained direction and elongate in the other. Second we perform a global inversion in which each plate is represented not by multiple individual estimates of the orientation of seismic anisotropy but by a single best-fitting pole and confidence limits. Plates are included in the inversion only if their data meet certain minimum criteria, the most important of which is that the anisotropy data differ significantly from a uniform random distribution in azimuth. Thus, several plates with sparse data are omitted. We consider two different approaches to weighting the plates. In the first, we give the short axis of the confidence ellipse equal weight for each plate. In the second approach we weight the short axis of the confidence ellipse for each plate proportionately to its root-mean-square (RMS) velocity. We also consider three different sets of plates to invert: (1) all plates meeting our minimum criteria for the anisotropy data, (2) the oceanic plate subset of the first set, and (3) the continental subset of the first. The different weighting schemes give results with overlapping confidence limits as do the results from different subsets of data. We refer to our preferred set of angular velocities as SKS5W-MORVEL, determined from a subset of the SKS data set from five oceanic plates weighted proportionally to the RMS velocity of each plate and constrained to consistency with the MORVEL global set of relative plate angular velocities. The new set of angular velocities indicates a 0.20×0.12 °/Ma [95% c.l.] right-handed net-rotation of the entire lithosphere about 55.9°S, 58.0°E. We conclude that realistic uncertainties in absolute plate motion from seismic anisotropy are ≈≤ ×0.15 °/Ma (95% c.l.).