Waveform inversion of surface wave data: test of a new tool for systematic investigation of upper mantle structures

Abstract

SUMMARY In most tomographic inversion of surface wave data, the long-period seismograms are first interpreted in terms of dispersion and/or attenuation curves before performing an inversion in terms of laterally and depth-varying properties. An alternative to this approach is to perform a direct waveform inversion or, as in Cara & Levěque, (1987), to use another set of secondary observables built up from the seismograms. In this paper, we systematically test with actual Rayleigh wave records this latter technique by considering laterally homogeneous models. We use for this purpose a set of recent long-period digital records from the Geoscope station Dumont d'Urville, Antarctica, for surficial events in the south Indian Ocean and the southeast Pacific Ocean, and for intermediate-depth events in the Vanuatu and Kermadec trenches. In addition to the obvious advantages of being able to deal with situations where overtones are present in the seismogram, it is found that the waveform inversion procedure allows us to reach a better depth resolution than in classical inversion of dispersion curves, even when only the fundamental mode is present in the seismogram. Quite good resolution is obtained to depth as large as 300km for S velocity when using surficial events located at a few thousand kilometres from the station, while classical surface wave studies do not allow us to resolve S velocity at depth larger than 150 km for these events. When intermediate-depth events are used at distances of about 5000 km, the presence of overtones in the signal allows us to get resolution to depth as large as 600 km for S velocity. Poorer resolution is obtained in both situations for the attenuation factor. S velocity appears furthermore to be more robust than attenuation at depth where good resolution is achieved. Due to the great sensitivity of surface wave amplitude to departure from the assumption of lateral homogeneity, more sophisticated direct modelling would be required to get more confidence in the inverted attenuation models.