Appraising the reliability of converted wavefield imaging: application to USArray imaging of the 410-km discontinuity

S U M M A R Y We develop a generic method to appraise the reliability of wavefield imaging methods and use it to validate some novel observations on the 410-km discontinuity. The core concept of the error appraisal method is to produce a simulated data set that replicates the geometry of the real data. Here we implemented two simulation methods: (1) flat layer primary P to S conversions, and (2) a point source scattering model for P to S conversion data based on the Born approximation and ray theory propagators. We show how the approach can be extended for any simulation algorithm. We apply this new approach to appraise recent results using a 3-D, three-component P to S conversion imaging method applied to data collected by the USArray. Multiple metrics show that the amplitude of P to S converted energy scattered from the 410-km discontinuity varies by 18 dB with a systematically lower amplitude in an irregular band running from Idaho through northern Arizona. In addition, we observe strong lateral changes in the ratio of amplitudes recovered on the radial versus the transverse component. We compute point resolution functions and a checkerboard test to demonstrate we can reliably recover relative amplitudes with a lateral scale of the order of 200 km and a vertical scale of approximately 10 km. Irregular coverage locally distorts the amplitudes recovered in the checkerboard, but a 156 km scale checkerboard pattern is recovered. Flat layer simulations show we can recover relative amplitudes to within a range of 1 dB and the reconstructed transverse to radial amplitude is everywhere less than 0.1. A model with north–south oriented ridges with a 3◦ wavelength and 12.5 km amplitude shows of the order of ±6 dB amplitude variations and small, but clear correlation of the transverse/radial amplitude ratio topography in the model. Finally, we model the 410-km discontinuity as a rough surface characterized by variations in amplitude and depth derived from the USArray data. The rough surface model recovers the scale of the observed amplitude variations, but does not explain the observed large variations in transverse component amplitudes imaged by the USArray data. The results indicate the 410-km discontinuity is definitely not a single interface separating isotropicmedia. We argue that it is likely better viewed as a rough surface with a structural fabric that creates anisotropic behaviour in some places.