Comparing the Odisi Integrated Stochastic Seismic Inversion Method to an Equivalent Bayesian Inversion Method

BP has developed a one-dimensional stochastic inversion method (ODiSI) for jointly estimating reservoir properties and lithology with associated uncertainties. ODiSI generates a large number (∼10000) of pseudo-wells consistent with the input prior information (well data and seismic horizons) at each trace location. It then generates synthetic (colour-inverted angle-stack) seismic data for each pseudo-well and calculates the misfit between these and the observed seismic data at that trace location. A smaller number (∼100) of the pseudo-wells with the lowest misfit are then used to calculate the estimates of the reservoir properties and associated uncertainties. An alternative approach is to pose the problem in a Bayesian framework in which case the selection of pseudo-wells used to calculate the estimates is determined by the absolute value of the data misfit (with respect to some estimate of the noise on the data) and the prior probability of the proposed pseudo-wells. We compare the Bayesian and ODiSI inversion approaches in terms of the output estimates and uncertainties obtained in each case.