The uncertainty in layered models from wide-angle seismic data

Abstract

The analytical method of estimating the uncertainty in layered models is addressed to models obtained using a layer-stripping modeling strategy or forward modeling. It is based on a simple principle of small error propagation. There are two variants of the method: a simplified one that includes refraction and vertical reflections and one that also includes wide-angle reflections. Both give a quantitative estimation for the existing models. To allow for a simple analytical estimation, refracted waves are described using a head-wave approximation in constant velocity layers; wide angle reflection paths are also simplified. In the case of trial and error forward modeling, this method can help determine how well the used parameterization is reflected in the data and avoid over-fitting the structures. This is especially important because the forward modeling is very subjective and there is no method to assess the parameterization without generating alternative models. For inversion problems using the layer-stripping method, the analysis allows for a correct propagation of errors and will help to evaluate the effect of including a priori information with known uncertainty. As a result, the layer-stripping modeling strategy is worse than simultaneous inversion for layered models because it gives larger uncertainties.

Monte Carlo approach to assess the uncertainty of wide-angle layered models: Application to the Santos Basin, Brazil

“…An alternative approach, that associates multi-parameter uncertainty estimate with a Markov chain Monte Carlo inversion scheme, has been proposed by Mosegaard & Tarantola (1995), and applied by Clark et al (2013) with the implementation of Hauser et al (2011). Uncertainties associated with different modelling approaches were also studied by Majdański (2013).…”