Inversion of surface wave spectra for source parameters of large earthquakes using aspherical earth models

Abstract

Amplitude and phase spectra of Rayleigh and Love waves of periods of 150 to 300 s from eight large earthquakes which occurred between 1990 and 1993 are corrected for wave propagation and then inverted for the source duration, rupture directivity, and moment tensor of the earthquake source. For various propagation anomalies predicted by models of the Earth's aspherical elastic and anelastic structure, the application of corrections for the focusing and take-off-azimuth anomaly, in addition to conventional corrections for the phase anomaly, improves inversions of surface waves. Rayleigh waves provide better constraints than Love waves in the determination of source parameters (the data consist of either only Rayleigh waves or both Love and Rayleigh waves, whenever the full moment tensor is involved). For several earthquakes there is clear discrepancy between source parameters determined from different data sets, particularly from Rayleigh or Love waves. The moment-tensor solution tends to underpredict observed amplitude spectra for data sets with large amplitude anomalies, especially for Love waves. This study shows that (1) the 1990 Sudan earthquake is anomalous for its unusually large non-double-couple component of the moment tensor; (2) the 1990 Iran earthquake has a strike–slip mechanism with a subvertical, NW-striking nodal plane, which is the fault plane inferred from surface ruptures; (3) the 1991 Costa Rica earthquake may have a seismic moment larger than previously determined from long-period surface waves; (4) the 1992 Nicaragua earthquake involves a predominant rupture parallel to the trench and a large centroid-offset up-dip to the trench axis, and the latter process may be responsible for large tsunami generation; (5) the 1993 Hokkaido earthquake has a shallow-dipping and west-plunging nodal plane, which is consistent with the P-wave first motions but differs from the Centroid-Moment-Tensor solutions.