Abstract. Lateral heterogeneity in the earth'~-upper mantle is investigated by inverting dispersion curves of long-period surface waves (100-330 s). Models for seven different tectonic regions are derived by inversion of regionalized great circle phase velocity measurements from our previous studies. We also obtain a representation of upper mantle heterogeneities with no a priori regionalization from the inversion of the degree 6 spherical harmonic expansion of phase and group velocities. The data are from the observation of aoout 200 paths for Love waves and 250 paths for Rayleigh waves. For both the regionalized and the spherical harmonic inversions, corrections are applied to take into account lateral variations in crustal thickness and other shallow parameters. These corrections are found to be important, especially at low spheric<;J.l harmonic order. the "trench region" and fast velocities down to 250 km under shields. Below 200 km under the oceans, both S velocity and S anisotropy support a model of small-scale convection in which cold blobs detach from the bottom of the lithosphere when its age is large enough. The spherical harmonic models c],early demonstrate (a posteriori) the relation between surface tectonics and S velocity heterogeneities in the first 250 km: all shields are fast; most ridges are slow; below 300 km, a belt of fast mantle follows the Pacific subduction zones. However, at greater depths, large-scale heterogeneities that seem to bear no relationship to surface tectonics are observed. The most prominent feature at 450 km is a fastvelocity region under the South Atlantic Ocean. Smaller-scale heterogeneities that are not related to surface tectonics are also mapped at shallower depths: an anomalously slow region centered in the south central Pacific is possibly linked to intense hot spot activity; a very fast region southeast of South America may be related to subduction of old Pacific plate. Between 200 and 400 km, a belt of SV>SH anisotropy follows part of the ridge and subduction systems, indicating vertical mantle flow in these regions. The spherical harmonic results open new horizons for the understanding of convection in the mantle. Perspectives for the improvement of the models pre sen ted are discussed.'Now at

Abstract. Long-period surface waves are used to map lateral heterogeneities of velocity and anisotropy in the upper mantle. The dispersion curves are expanded in spherical harmonics up to degree 6 and inverted to find the depth structure. The data are corrected for the effect of surface layers and both Love and Rayleigh waves are used. Shear wave velocity and shear polarization anisotropy can be resolved down to a depth of about 450 km. The shear wave velocity distribution to 200 km depth correlates with surface tectonics, except in a few anomalous regions. Below that depth the correlation vanishes. Cold subducted material shows up weakly at 350 km as fast S-wave anomalies. In the transition region a large scale pattern appears with fast mantle in the South-Atlantic. S-anisotropy at 200 km can resolve uprising or downwelling currents under some ridges and subduction zones. The Pacific shows a NW-SE fabric.

No abstract

Many inversion studies of first P-arrival times have been made for the threedimensional upper mantle structure beneath the Japanese Islands by using the linear damped least-squares method (AKI and LEE, 1976). One of the most recent contributions relying on this method is that by HASEMI et al. (1984) on the structure beneath the Tohoku region. In the linear inversion it is assumed that the ray paths in the inverted model are the same as those in the initial model. The ray tracing experiments attempted in the early 1970 's (JACOB, 1970 showed that the ray paths are strongly guided by the high velocity slabs descending beneath the island arcs.The aim of this study is to examine the performance of a nonlinear inversion of travel times of first-arrivals for the island arc structure. We apply the Dijkstra method (DIJKSTRA, 1959) of network theory in the forward calculation of ray paths and travel times instead of the shooting or bending method (JULIAN and GUBBINS, 1977). We subdivide a two-dimensional space of 100 km x 100 km into a number of blocks of 10 km x 10 km as shown in Fig. 1. We assume an average layered velocity profile of the area under consideration a priori. We make the velocity profile by taking the value of the JMA standard P-wave structure (HAMADA, 1984) for each 10 km starting from a depth of 5 km.The algorithm proposed by DIJKSTRA (1959) is considered to be the fastest of the algorithms which find the shortest paths connecting a node and the other all nodes of a network (IRI and KOBAYASHI, 1976). For the algorithm, one should consult DIJKSTRA (1959) or IRI andKOBAYASHI (1976). We seek the shortest time paths connecting a node ns and the other all nodes ni(j_??_s) of the network. When the node ns is assigned to an event, and the nodes n;(j 0 s) include station positions, we are able to obtain the shortest time paths from the event to all the stations. When the node ns indicates a station and all the event locations are included in nj(j # s), the algorithm gives us the shortest time paths from all the events to the station. In this study we shall put the nodes on the boundaries of the blocks. Figure 1 shows a laterally heterogeneous model that corresponds to a vertical section of the uppermost mantle perpendicular to the strike of island arcs. The 195