Quantitative comparison of inversion methods for estimating density from seismic data: multi-component data improved seismic prediction

In the Athabasca oil sands, lithology and fluid composition are typically better correlated with density than with other elastic properties, such as P- and S-wave velocity. Therefore, improving the accuracy of density estimates in oil-sands reservoirs has become one of the most important goals in quantitative interpretation, which integrates seismic, well-log and geological data through deterministic or stochastic inversion methods.

In this paper, we applied three methods for density estimation to an oil-sands reservoir in the Athabasca oil sands area; specifically, these are:

1. PP pre-stack AVO inversion;
2. PP-PS joint pre-stack inversion;
3. multi-linear regression and neural networks method.

Direct comparisons of estimated density, as determined using the above three methods, were made to evaluate the predictive ability of each method, and to quantitatively demonstrate the value added by P-S data. Predicted density from each method was cross-corelated with measured density logs from 113 wells; the correlation coefficients are 63% for PP-inversion, 81% for PP-PS joint inversion, approximately 90% for a step-wise regression method. Additionally, the top three of four attributes contributing to density estimates from multi-linear regression were found to be from PP-PS joint inversion. Therefore, we can conclude that the PS data has improved density estimates and thus played an important role in refining density predictions.