Seismic Inversion in a Geological Bayesian Framework

Reservoir characterisation is invariably a data integration exercise; no one type of data provides enough information on its own. Therefore it must also be a multidisciplinary process which, ideally, should be inclusive and transparent allowing each discipline to be able to understand and control their contribution and adjust it as new data and new insights become available. Effective integration requires estimates of the uncertainties of each data type in order to control the relative weighting of their respective contributions. Bayesian methods provide a framework for this and several seismic inversion strategies utilising Bayesian methods have been proposed. They typically set up the Bayesian scheme in a geophysical framework such that the integration occurs near the beginning of the process. An alternative proposed here is a Bayesian scheme framed in a geological domain with priors and likelihoods defined directly in terms of reservoir parameters. This requires that reservoir properties and associated uncertainties are estimated, as far as possible, independently from each data type and are then combined towards the end of the process. The talk will outline the process and provide some illustrations.