Joint Facies/Elastic Inference in Waveform Inversion

Seismic AVO inversion for elastic parameters jointly with litho-fluid categories from migrated seismic data is now an established technique. Compared to conventional techniques based on adding smooth background models to impedance inversions, it has several advantages, including the ability to constrain elastic parameters to welllog-data distributions, and ability to directly predict fluids. These methods rely on the migrated amplitudes being inverted being faithfully scaled to reflectivity, and are vulnerable to the presence of non--primary seismic energy which is not modelled by conventional Born-style imaging, such as mode conversions or multiples. In any deconvolutional style inversion such wave energy adds to the noise rather than the signal. We show that it is possible to perform joint elastic/facies inversion on raw shot records, using a full-wave modelling operation in the likelihood of a hierarchical Bayesian inversion, with optimisation performed using the expectation-maximisation algorithm. Such full wave techniques can in principle model all wave modes, and should theoretically have higher S/N ratios than their AVO equivalents. Illustrative examples in high--contrast lithologies using joint acoustic/facies full wave inversion show that cleaner inversions are produced in this regime compared to convolutional methods based on traditional imaging.