Application of amplitudes in shallow seismic refraction inversion

Abstract

I generate three starting models for the inversion of a set shallow seismic refraction data using wavepath eikonal traveltime tomography. Two models are generated with the generalized reciprocal method (GRM), which uses a multi-layer model with discrete velocity changes while the other is generated with a one dimensional (1D) inversion algorithm similar to the tau-p method, which uses a model with vertical velocity gradients. There is little correlation between the results in either the absolute depths, the relative depths or the refractor velocities. I attribute the differences in the absolute depths to the occurrence of a velocity reversal in the overburden, the existence of which is indicated by the concave-upwards shape of the traveltime graphs and by the extremely rapid decrease in amplitudes. I attribute the differences in the relative depths and seismic velocities to the poor starting model generated by the 1D tau-p inversion algorithm, which is unable to resolve the fundamental ambiguity between seismic velocity and structure on refracting interfaces. The GRM-derived refractor velocities are compatible with the head coefficients determined with amplitude products. The existence of velocity gradients is suggested by the convergence of the traveltime graphs with increasing shot-to-detector distance. I attribute this convergence in part to the large variations in signal-to-noise ratios, which affect the accuracy of the traveltime data. The shot amplitudes decrease rather than increase with distance, which supports the occurrence of distinct interfaces with velocity gradients over relatively small depths. Therefore, while velocity gradients are likely to occur, the results generated with 1D tau-p tomography are probably not representative of the subsurface conditions. The considerable differences between the results generated with the GRM and those obtained with 1D tau-p tomography are compelling demonstrations of the non-uniqueness of model-based inversion and of the importance of the selection of the initial model. Accordingly, all methods of model-based inversion of shallow refraction data could usefully include starting models generated with the GRM or the refraction convolution section.