Current efforts to utilize full waveform inversion (FWI) as a tool beyond acoustic imaging applications, for example for reservoir analysis, face inherent limitations on resolution and also on the potential trade-off between elastic model parameters. Adding rock physics constraints does help to mitigate these issues. However, current approaches to add such constraints are based on averaged type rock physics regularization terms. Since the true earth model consists of different facies, averaging over those facies naturally leads to smoothed models. To overcome this, we propose a novel way to utilize facies based constraints in elastic FWI. A so-called confidence map is calculated and updated at each iteration of the inversion using both the inverted models and the prior information. The numerical example shows that the proposed method can reduce the cross-talks and also can improve the resolution of inverted elastic properties.
Introduction
Full waveform inversion (FWI), in principle, aims to utilize all the information in the recorded data to reconstruct the subsurface structure and to estimate the elastic and/or acoustic parameters. To better simulate wave propagation, depending on the efficiency or accuracy requirements, pseudo-acoustic, elastic and viscoelastic equations are used for the forward modeling engine in FWI. With the current availability and improvements in computational power, solving these complex equations is becoming more and more practical. However, more complex equations require more parameters to describe the real earth and inevitably introduce more null space.
Estimating elastic parameters, such as P-wave velocity, S-wave velocity, and density is an ongoing cause of the seismic exploration community. In current practice, due to an inherent crosstalk between parameters, for example, P-wave velocity and density, the density model is not usually updated at all to reduce the nonlinearity (null space) of the inversion which ultimately leads to a better convergence. However, to get a better understanding of the subsurface, multiparameter inversion is necessary. Other common ways to reduce the null space in multiparameter inversion are better parametrization (Operto et al., 2013; Oh and Alkhalifah, 2016) and incorporation of a priori information to constrain the inversion. Utilizing a priori information in the form of preconditioning or regularization has been shown to efficiently reduce the null space (Asnaashari et al., 2013).
In classic AVO inversion, however, a more advanced type of constraints based on facies has proved to be very effective to optimize the seismic inversion (Zabihi Naeini and Exley, 2017). Zabihi Naeini et al. (2016) discussed the main components of FWI as a potential reservoir characterization tool and one of their suggestions was to use facies based rock physics constraints in FWI. In this paper, we, therefore, utilize one such facies based constraint in FWI. We assume that the inverted models adhere to a Gaussian distribution (Tarantola, 2005) and, iteratively, based on the prior information, a so-called facies confidence map is calculated and used as a regularization term in inversion.
