Inverse Modelling of Stratigraphy: A Tool to Predict Facies Distribution and Assess Uncertainties for Reservoir Characterisation

Charvin, Karl¹, Kerry L. Gallagher¹, Gary J. Hampson², Jo Ann Hegre³, Richard Labourdette⁴ (1) Imperial College London, London, United Kingdom (2) T.H. Huxley School, Imperial College, London, United Kingdom (3) Total E&P UK plc, Aberdeen, United Kingdom (4) Total S.A, Pau, France

The prediction of sub-seismic facies architecture and associated rock-property distributions from well data is challenging in all reservoirs. A common approach to this problem is to predict reservoir architecture from well data via stratigraphic interpretations that infer environmental parameters (e.g. relative sea-level, sediment supply). In this context, reservoir geoscientists often have to deal with questions of uniqueness, uncertainty, non-linear interaction between processes, and model sensitivity. Moreover, the increasing demand for precision in reservoir management requires reservoir geoscientists to assess the accuracy of, and assess uncertainties on, reservoir architecture models, which pushes classical geostatistical tools to their limits.

In this contribution, we address these issues using a recently developed numerical method. The new method combines a “process-response” forward model of shallow-marine stratigraphy (BARSIM) simulating wave-dominated sedimentary processes, with non-linear stochastic inverse techniques (Markov chain Monte Carlo) that describe how information on a parameterised physical system can be derived from observed data and prior information.

This new methodology has been validated on synthetic case studies and has been applied to outcrop data (Aberdeen Member, Blackhawk Formation, Book Cliffs, Utah). This method provides the full posterior probability distribution on all the environmental parameters in the forward model, which allows quantitative assessment of uncertainties in stratigraphic architecture. At reservoir scale, sampling of the parameter posterior probability distributions for multiple realisations of modelled reservoir architecture allows the quantification of uncertainties in facies distributions. The results have direct applications to static and dynamic reservoir modelling.