Uncertainty Estimation in the Quantitative Interpretation of Inverted Reservoir Property Volumes

Date

2016-12

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Volumetric uncertainty in reservoir property volume estimation using 3D seismic data, well logs, and P-wave inversion outputs can be calculated using synthetic modeling, multi-attribute linear regression, and collocated cokriging. This can be accomplished using a multi-attribute linear regression to create the initial reservoir property volume, and then using this volume as a covariate to a simulated collocated cokriging approach from which an uncertainty in the volume estimate is computed. In a synthetic test example, the initial reservoir property volume estimated from the synthetic dataset exhibited a 0.92 correlation coefficient to the known reservoir properties. Due to the high correlation between the hard data and soft data the collocated cokriging output was almost identical to the multi-attribute non-linear regression. Except around the vicinity of the well were it had an overall smoother output. An uncertainty volume generated from the standard deviation of thirty realizations of the collocated cokriging process run using SGS (Sequential Gaussian Simulation) effectively predicted the lower error regimes in the vicinity of the well locations, however the areas with high error away from the well locations were not captured by this process. When applying the framework validated above to a real dataset in the Mississippi Limestone near Morrison Oklahoma the output of the non-linear regression based method had a correlation coefficient of 0.81 to measured well logs. The collocated cokriging process created a higher vertical resolution output than the non-linear regression output because the vertical sampling is closer to that of the well, approximately 74,000,000 cells in the grid compared to 14,000,000 in the synthetic. The overall approach shows the potential to calculate volumetric uncertainty in reservoir property volume estimation using 3D seismic data, well logs, and P-wave inversion outputs, which can be computed on a regular basis using multi-attribute linear regression and collocated cokriging.

Description

Keywords

Multivariate Statistics, Geostatistics

Citation