Stochastic (or geostatistical) impedance inversion techniques have great potential in addressing key questions in reservoir characterization. They work at the vertical scale of reservoir models and, therefore, at higher resolution than the seismic data. They produce multiple equiprobable results which provide an assessment of, the uncertainties, and they are ideally suited for integrating non-seismic information in the inversion process. However, two issues have slowed the acceptance of stochastic impedance inversion techniques. First, there is suspicion of‘ unconstrained random noise generators’, which appear to offer extra information for free, and, secondly, managing and extracting value from multiple realizations is difficult. For these reasons, faster deterministic inversion approaches, resulting in a single lower-resolution impedance volume, with less quantified uncertainty, are more commonly used when building reservoir models.
To address the first issue, we have developed ways of integrating 3D constraints from sedimentary modelling with the geostatistical impedance inversion method, since these two approaches bring complementary information on reservoir properties. The resulting high-resolution multiple realizations of impedance are combined with uncertainties from petrophysical regression analysis to produce multiple realizations of reservoir properties (e.g. porosity), and from each an estimate of total pore volume. We illustrate the benefits of this multiple realization work flow applied to data from a shallow marine siliciclastic reservoir. A comparison of the seismic/sedimentologically constrained reservoir models with those constrained by well data only has demonstrated more accuracy and better control on the spatial variability of reservoir properties. In this example, however, adding more constraints results in a broader range of possible reservoir models and a more meaningful uncertainty assessment. We conclude that our models constrained by well data only were derived with unrealistic simulation parameters and an over-optimistic assessment of a priori uncertainty.