Split shear wave arrivals are analyzed in seismograms from local earthquakes in southern Hawaii recorded at five temporary arrays and one permanent network station. We identify split shear wave arrivals by their orthogonally polarized pulses, linear particle motions, and similar waveforms and estimate the delay time for the slow shear wave arrival (S2 using a waveform cross-correlation method. Consistent leading shear wave polarizations were measured at the majority of our stations. Comparison of observed and predicted shear wave polarizations confirms that the former are due to anisotropy rather than earthquake source mechanism. Agreement between fast shear wave (S1) polarizations and independently estimated directions of the maximum horizontal compressive stress (&sgr;H) for the Ainapo and Punalauu Gulch arrays leads us to conclude that the predominant source of the observed anisotropy for these two areas is stress-aligned cracks consistent with the extensive dilatancy anisotropy (EDA) hypothesis. Two distinct S1 polarization directions were observed over distances less than 1 km for the Bird Park and South Flank arrays. S1 polarizations parallel to the NE striking Kaoiki Pali fault system for the Bird Park array combined with a nonhorizontal maximum principal stress (&sgr;1) for the South Flank region suggest stress-induced cracks aligned by nearby faulting as a source for the observed anisotropy.

Large station-to-station variations in S1 polarization and the relationship between delay time and event depth for arrays in the Kaoiki and South Flank regions provide evidence for anisotropy that is predominantly shallow rather than pervasive. Average delay times for the five arrays vary from about 100 to 230 ms, with standard deviations of the order of 30 ms. Estimated anisotropic velocity variations and crack densities exceeding 10% indicate that the upper crust of southern Hawaii is highly fractured. A search for possible temporal changes in delay time associated with the 1983 Kaoiki main shock (ML=6.6), at a station near the epicenter, finds no evidence for change. ¿ American Geophysical Union 1995.