Abstract
In this paper, we will discuss the probabilistic seismic inversion of Bonga main reservoirs, with the objective of updating the reservoir static models with the net sand prediction from seismic.
A unique solution to the seismic inverse problem does not exist. Uncertainties arise from two sources: noise in the seismic data and ambiguities in the inverse problem itself. Ambiguities are mainly caused by the fact that the seismic data are band limited. Most inversion algorithms, often guided by well and horizon constraints, typically produce a single solution. This solution may represent the most likely subsurface model but it does not give information about other possible solutions.
For the results discussed in this paper, we used a trace-by-trace based inversion that relies upon rock and fluid property relationships that describe acoustic properties (Vp, Vs, density) as a function of reservoir properties (e.g. porosity, net-to-gross etc). A prior model is provided as input. This prior model is the initial reservoir static model from which the rock and fluid properties are obtained. These properties are then perturbed in a statistical manner for a number of iterations, deriving acoustic impedances which are used to generate the corresponding synthetic traces. These synthetic traces are then compared with the actual seismic response, and selected against a matching criterion such as semblance. The probabilistic approach combines the seismic modelling with a statistically correct examination of uncertainties taking into account noise in the seismic data. As no well is used to constrain the inversion, this allows for blind well information to be used as validation points for the inversion results.
This procedure, in addition to providing the "most likely" model, also provides a statistical examination of uncertainties.
