COMPARISON OF GLOBAL WAVE-FORM INVERSIONS WITH AND WITHOUT CONSIDERING CROSS-BRANCH MODAL COUPLING

In this study we present a new approach to inverting long-period seism ic waveforms. The effect of lateral heterogeneity is partitioned into two. The first part represents the effect of horizontally averaged str ucture along the great circle between the source and receiver, and Is allowed to remain in non-linear form in the formulation. The second pa rt incorporates any further correction due to cross-branch modal coupl ing, which has been neglected in the more conventional path average,ap proximation (PAVA). This term is linearized and then treated asymptoti cally so that the seismogram depends only upon the structure within th e great-circle section determined by the source and receiver (Li and T animoto 1993a). We refer to this new method as the non-linear asymptot ic coupling theory (NACT). The sensitivity kernels predicted by the PA VA and NACT are compared. While the sensitivity kernels are similar fo r surface waves and shallow-turning body waves, they are very differen t for body waves that sample the deep mantle. By examining the inversi on algorithms for the PAVA and NACT, we demonstrate that the computati on time required by the NACT tends to be of the same order of magnitud e as that required by the PAVA, as the number of model parameters incr eases. Based upon a realistically large data set (5041 body-wave seism ograms and 1531 mantle-wave seismograms), formal resolution analyses a re performed using both PAVA and NACT. We find that the NACT is signif icantly more powerful in resolving 3-D structure in the deep mantle. W e compare the models obtained for the same observed data set using the two approaches. As expected, they differ more in the lower mantle tha n in the upper mantle. The difference in their amplitude spectra incre ases with spherical harmonic degree. The model developed using the NAC T predicts the observed surface geoid better than that developed using the PAVA, based upon geodynamic flow modelling.