To overcome the potential contamination of the direct S waves by source‐side anisotropy in shear‐wave‐splitting analysis, we describe a new approach that we call the reference station technique. The technique utilizes direct shear waves recorded at a station pair and depends on maximizing the correlation between the seismic traces at reference and target stations after correcting the reference station for known receiver‐side anisotropy and the target stations for arbitrary splitting parameters probed via a grid search. The algorithm also provides a delay time between both stations caused, for example, by isotropic heterogeneities. Synthetic tests demonstrate the stability of the estimated parameters, even where variability in near‐surface properties (thickness and velocity of sediment layer) exists. We applied the reference station technique to data from seismic experiments at the northern margin of Tibet. Average splitting parameters obtained from the analysis of direct S‐wave results are consistent with those obtained from previous SKS splitting measurements. Where differences exist, shear‐wave fast polarization estimates resolved from direct S indicate a higher degree of internal consistency for closely spaced stations than those derived from SKS. This is probably due to the much larger number of direct S waves available for splitting measurements compared to SKS for the same observational period, resulting in higher quality measurements. We also demonstrate the ability of the technique to provide improved splitting measurements for temporary stations by following a bootstrap approach in which only a few stations with well‐constrained SKS splitting parameters are used as seeds to determine the splitting parameters of a large array in an iterative manner. In addition, the S measurements sample the anisotropic layer with different angles of incidence and back azimuths, thus potentially providing additional constraints on more complicated anisotropic structures, and the interstation delay times could be used for tomographic studies to reduce the bias from anisotropic structure.
Online Material: Multisplit software package (C++) with instructions.