JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 98, NO. B6, PAGES 9793-9823, JUNE 10, 1993 High-Resolution Global Upper Mantle Structure and Plate Tectonics YU-SHEN ZtIANG Institute of Tectonics, University of California, SantaCruz TOSHIRO TANIMOTO Department of Geological Sciences, University of California, SantaBarbara Seismological Laboratory, California Institute of Technology, Pasadena A global high-resolution S wavevelocity model RG5.5 is obtained for theupper 500 km of Earth's mantle using a 5 ø x5ø equal-area block parameterization. Thedata set consists of some 18,000 seismograms associ- ated with 971 events with magnitudes largerthan 5.5. Fundamental modes (Love and Rayleigh waves)are usedwith periods from 75 to 250 s. The horizontal resolution length is around 1000 km, and the vertical resolution varies with depth from 60 to 250 kin. Model RG5.5 hasmanyfeatures consistent with previous three-dimensional global and local seismic studies, but many new features are found. The S waves under mid-ocean ridges have broadslow velocityand have very slow velocity in the upper 100 km below the sur- face. The minimum velocity is at depths near 50 km or shallower. The lateralextent of the slowvelocityre- gion across ridges increases with spreading rate. The S wave velocities underridges are strongly correlated with spreading rates at shallow depth, but the correlation decreases with depth andalmost disappears at 100 kin. The slow velocities shiftoff the current spreading positions below 100 km depth under the Mid-Atlantic Ridge and may record past positions of the ridgeand/or be related to hotspots nearthe ridge. Some major hotspots are associated with slow-velocity anomalies with magnitudes of about1-2% slower thanthe global average and with lateraldimension largerthan 1000 km at depths between 100 and 200 kin. Differences in the upwelling structure between ridges andhotspots are indicated. The S wavevelocity structures may sug- gest an active mechanism for the EastAfrican Rift Valley and a plateextension mechanism for the Baikal Rift Valley. 1. INTRODUCTION Seismological study over most of this centuryhas led to a refined picture of Earth underthe assumption of spherical sym- metry, suchas the existence of the crust,the uppermantle, the lower mantle, the fluid outer core, the solid inner core, and also the detailed profilesof seismic wave velocities and density as a function of the radius (or depth) [e.g., Jeffi'eys and Bullen, 1940; Anderson and Toks•iz, 1963; Gilbert and Dziewonski, 1975; Dziewonski and Anderson, 1981; Kennett and Engdahl, 1991]. Progress in extensive collection of digital data and growing computer power in the last decade has led to three- dimensional imagesof Earth's interior from the upper mantle [e.g., Masters et al., 1982; Nakanishi and Anderson, 1982, 1983, 1984;Nataf et al., 1984, 1986; Woodhouse and Dziewon- ski, 1984; Tanirnoto and Anderson, 1985; Tanirnoto, 1986a,b; Montagner and Tanimoto, 1990, 1991; Zhang and Tanimoto, 1991] to the lower mantle and the core [e.g., Clayton and Co- mer, 1983; Dziewonski, 1984; Creager and Jordan, 1986; Morelli and Dziewonski, 1987; Giardini et al., 1988; Ritzwoller et al., 1988; Gudmundsson, 1989; Tanimoto, 1990; Li, 1990; Su and Dziewo.,,.ski,1991; R.L. Woorward et al., "Constraints on the large-scale structure of Earth'smantle,submitted to Journal of Geophysical Research, 1992]. These seismic tomographic inferences have alreadybenefited other disciplines of the Earth sciences. Some getphysicists, for example, have employed these results to model mantleconvection, plate motion, undula- tionsof the core-mantle boundary, and the geoid[Hager et al., Copyright 1993 by theAmerican Geophysical Union. Paper number 93JB00148. 0148-0227/93/93JB-00148505.00 1985; Hager and Clayton, 1988; Lay et al., 1990; Fortearm Peltier, 1991]. One of the shortcomings of previous tomography studies, however, is their restriction to low-order spherical harmonic ex- pansions, which resolve only large wavelength anomalies (larger than 4000 kin). Previous uppermantlestudies, such as Woodhouse and Dziewonski [19841, and Nataf et al. [1986] truncated expansion at spherical harmonic degrees (/) 8 and6, aespectively. The earlydata set did notjustify expansions to de- grees higher than6 [Natafet al., 1986; Nakanishi and Ander- son, 1984]. There have beenprogressive attempts to increase the cutoff, for example,to l = 10 by Tanimoto and Anderson [1985] and, recently,to l = 12 by Wong [1989] and Su and Dziewonski[1991]. This involveslarger and larger data sets. But the horizontal resolution is still only continentalscale (about 3000 kin) at degree 12. In order to correlate seismologi- cal results with tectonics,extending the analysisto higher spherical order and resolving small scale features areneeded. The aimsof this study are to obtain a global, high-resolution, uppermantle S wave velocitystructure, and to improve our understanding of Earth's dynamics. The development of global networks of digital seismographs in thelast decade has resulted in theaccumulation and processing of unprecedented amount of highquality data, and has improved spatial andazimuthal cov- erage. These successes have made our present study possible. We have collected about 18,000 seismograms associated with 971 events with magnitudes larger than5.5, a vastincrease in the data set relativeto the first generation of three-dimensional models only a decade ago. A two-step inversion for velocity structure has been used in this study. The first stepis to measure the global Love and Rayleigh wave phase velocity variations for periods ranging 9793