JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 98, NO. Bll, PAGES 19,913-19,932,NOVEMBER 10, 1993 Seismic Propagation Across theEastPacificRise' Finite Difference Experiments andImplications for Seismic Tomography WILLIAM $. D. WILCOCK, 1MARTIN E. DOUGHERTY, 2SEAN C. SOLOMON, 3 G. M. PURDY, 4 AND DOUGLASR. TOOMEY 5 We usea full-waveform acousto-elastic finite difference technique to investigate seismic propagation across theEastPacificRise at 9ø30'N for a two-dimensional velocity model based on thatproposed by Vera andothers (1990). The primary feature of the model is an upper crustal low-velocity region, corresponding to the axial magma charnber, which includes a small magma body located 1.6km beneath theseafloor at therise axis. The high velocity gradients in thisregion result in a complex pattern of propagation whichincludes considerable scattering of energy above and below the magma chamber. A qualitative comparison of finite difference seismograms withdata collected by receivers located 9 km and 20 km off axis during a tomography experiment at 9ø30'Nshows generally goodagreement. For paths that cross the rise axis,the first arrival in the finite difference solutions diffracts above the magma chamber. This phase has a very low amplitude and at larger offsets falls below the ambient noiselevelsobserved during thetomography experiment.In such cases, thefirst arrival with significant energy is a diffraction from below the magma chamber. A high-amplitude Moho- turning (PrnP) phase which results from the large velocity change across the Moho beneath the rise axisis apparent in both the finite difference solutions and the observations. Ray-theoretical calculations of the paths of the diffracted arrivals are veryunstable, and for the diffractions above the magma chamber no solution canbe found with a single-precision algorithm. Synthetic delay-time inversions using an approximate ray-tracing algorithm demonstrate theimportance of ensuring that picked arrival times areassigned to paths that pass to the correct side of the magma body. Synthetic inversions of spectral estimates of t* show that Q-1 models are compromised not only if the ray paths are inco•ectbut alsoif t* estimates include significant contributions from more thanonephase. Deterministic scattering from the magma chamber may contribute noticeably to spectral estimates of t*, buttheresults of thefinitedifference experiments implythathigh levels of attenuation observed for phases passing below themagma chamber are predominantly the result of intrinsic attenuation. INTRODUCTION Seismic experiments have played a central role in constraining the crustal structure of the East Pacific Rise (EPR). At many locationsthe axial crust is characterized by a strong seismic reflector at 1-2 km depth [Herron et al., 1978;Detrick et al., 1987, 1991]. Because the amplitude behavior of the reflected phase indicates a decrease in velocity withdepth [Haleet al., 1982; Vera et al., 1990], the reflector is interpreted as the roof a small lenticular magma body. P andS wavevelocities within molten basalt are about 3 km/s [Murase a•id McBirney, 1973; Manghnani et al., 1986] and 0 km/s, respectively, much lower than valuesof over 6 and 3 km/s found in the overlying crustal material [Vera et al., 1990]. Expanding spread profiles(ESPs) [Harding et al., 1989;Veraet al., 1990]andtomographic data [Burnett et al., 1989;Toorney et al., 1990;Caress et al., 1992] also require a substantial increase in seismic velocities immediately below and to the sides of the axial low velocity zone. Thus,in addition to thehighvertical velocitygradients in the upper crust and Moho transition zone that are characteristic of all oceanic crust, the velocity structure of the EPR is strongly 1School of Oceanography, University of Washington, Seattle. 2Department of Geosciences, Boise State University, Boise, Idaho. 3Department of Terrestrial Magnetism, Carnegie Institution of Washington, Washington, D.C. 4Department of Geology and Geophysics, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts. 5Department of Geological Sciences, University of Oregon, Eugene. Copyright 1993by theAmerican Geophysical Union. Paper number 93JB01820. 0148-0227/93/93 JB-01820505.00 two-dimensional and includes very high velocity gradients surrounding the axial magma body. Such characteristics present considerable challenges to theinterpretation of seismic data. To date, the strongestconstraintson the crustal velocity structure of the EPR have beenderived from the analysis of ESPs oriented parallel to theriseaxis [Hardinget al., 1989; Vera et al., 1990]. Under the assumptions that along-axis variations in the velocity structure are negligible and that the recordedseismic waveshave propagated in the verticalplane of the profile, such data can be forward modeled using the reflectivity technique [Fuchsand Miiller, 1971]. A one-dimensional layeredvelocity model is constructed which matches both the travel time and amplitude characteristics of the recorded phases. The two- dimensional cross-axis structure may be inferred by interpolating between ESPs obtained at different distances from the rise axis. In contrast, the derivation of two-dimensional models from refraction lines oriented perpendicular to the rise axis has been limited to the modeling of travel time data [e.g., McClain et al., 1985]. Since seismicamplitudes can be strongly sensitiveto velocity gradients, the exclusion of amplitude data significantly degrades the resolution of theresulting models. More recently,tomographic data sets which include a wide variety of source-receiver configurations have beeninverted for models of P wave velocity [Burnettet al., 1989; Toorney et al., 1990; Caress et al., 1992] and attenuation [Wilcock, 1992; Wilcock et al., 1992]. Becauseof the difficulty of calculating stable ray-theoretical paths between many source-receiver pairs the velocity inversions either have been limited to a single iteration with respect to a one-dimensional starting model [Burnettet al., 1989; Caresset al., 1992] or have employedan approximate ray-tracing algorithm to undertake multiple iterations [Toomey et al., 1990]. While recent developments in approximate ray-tracing techniques [e.g., Vidale, 1988; Moser, 1991] should yield improved tomographic models [Toorney et al., 19,913