Novel approaches for elastic wavefield imaging and inversion

Incorporating elasticity into wavefield extrapolation methods includes pitfalls that have no parallel in acoustic scenarios. First, coupled effects among the multiple elastic wave modes (which become three in the presence of anisotropy) often result in crosstalk contamination of the inverted reflectivity and earth model parameters. Second, conventional acquisition does not capture the full wavefield information required for proper extrapolation of recorded seismic signals, resulting in artificial non-physical wave modes. These are the two main problems facing accurate elastic wavefield imaging and inversion today. Fortunately, new inversion methods address the first problem, while the second is being addressed with new acquisition technologies. Multiparameter inversion is subject to leakage (i.e., crosstalk) among inverted parameters (Operto et al., 2013; Pan and Innanen, 2016). Radiation pattern analysis is important to understand the ambiguity among physical properties with respect to the reflection angle and model parameterizations. This analysis also shows that the amplitude responses overlap for different earth model contrasts, thus expressing the difficulty to isolate sensitivity kernels for different parameters. We address this problem by employing the energy norm, and we will present results from an elastic reflection waveform inversion (simultaneous reflectivity and earth model inversion) that obviates the distinction among the various wave modes in the reflectivity model inversion. In this case, the reflectivity is represented by a scalar image formed by the energy norm with contribution of all wave modes, and the low wavenumbers of the various earth model parameters are updated using Born modeling from the scalar elastic image. This method avoids the combined effect of crosstalk and high-wavenumber (cycle-skipping) updates into the inversion. The reflectivity model is also free of low-wavenumber artifacts, that would otherwise contaminate the reflectivity model and consequently harm Born modeling. Non-physical events refer to the numerical extrapolation of wave modes that have not been originally recorded (Yan and Sava, 2008; Ravasi and Curtis, 2013). A theoretically accurate solution to this problem relies on the acquisition of the stress wavefield components, which are becoming feasible with ocean-bottom node acquisition (obtaining the trace of the stress tensor i.e., the pressure), and with distributed acoustic sensing (DAS). Although the latter is labeled as acoustic, Lim and Sava (2018) show the possibility of reconstructing all components of the strain tensor with the fine-sampled measurements from the fiber optic cable. We will show how multicomponent DAS nullifies the occurrence of non-physical modes, leading to improvement in images and inversion gradients.