Uncertainty in Near-Surface Refraction Inversion

Abstract

Uncertainty in the tomographic inversion of near-surface seismic refraction data can be separated into aleatory variability, which describes the misfit errors and epistemic uncertainty, which describes the suite of acceptable models. Common implementations of refraction tomography usually focus on reducing aleatory variability and frequently disregard epistemic uncertainty. In this study, the tomograms generated with three models of the seismic velocities in both the weathering and in the sub-weathering, using the generalized reciprocal method (GRM), are consistent with the traveltime data. However, only one tomogram is consistent with the optimum XY value and the attributes derived from the head wave amplitudes and seismic velocities. This study demonstrates that epistemic uncertainty can be explicitly addressed with the GRM, because the most probable tomogram can be selected objectively from a number of acceptable alternatives. The GRM-based tomogram successfully detects, defines and differentiates narrow regions with low seismic velocities which represent shear zones and a massive sulphide ore body. None of these zones is detected with the tomogram generated with the default starting model using smooth vertical velocity gradients. It is concluded that minimizing epistemic uncertainty through the use of the most appropriate starting model is more important than minimizing aleatory variability.