Webinar by Paul Cupillard

Look at the video here.

Abstract

Within the last decade, full waveform inversion (FWI) turned to be tractable at the exploration scale, resulting in robust images of the distribution of seismic wave velocities within the 3D subsurface. Because of the increase of computational power, higher and higher frequencies can be taken into account, leading to more and more precise images. Nevertheless, seismic recordings are intrinsically band-limited, so the resolution of the images is limited too and significant uncertainties remain when interpreting them in terms of geological structures. In this webinar, we will outline a method to reduce such uncertainties. The method relies on a Bayesian inverse approach in which structural parameters, such as horizon and fault surface geometry, form the model space, and where the dataset is the FWI image itself. In this context, the appropriate forward modelling tool for mapping the model space into the data space is the non-periodic homogenization operator. This operator enables to compute the effective elastic medium of any structural model filled with elastic properties. The obtained smooth medium is what the waves “see", i.e. the result of a FWI in case of sources and receivers all around the medium. Weighting homogenization results by the actual sources and receivers configuration therefore leads to synthetic FWI images to be compared to the data. To sum up, the ingredients of our inverse approach are:

Priors on the structural parameters to be estimated along with a way of sampling them,

1. The homogenization operator for computing synthetic FWI images from structural models,
2. An appropriate function to estimate the local (i.e. spatialized) likelihood between the synthetic images and the real FWI result,
3. The Bayes' theorem to get the posterior distribution of the structural parameters.

We will discuss these four aspects with a particular focus on point 2, showing how the homogenization works along with an application to the SEG-EAGE overthrust model. Relying on this model, we show preliminary results of our inverse approach. This latter emphasises the benefit of considering geomodelling techniques in geophysics to better understand the subsurface.

Look at the video here.