Probabilistic facies analysis in high-dimensional data using scatter-matrix-based separability measure

Abstract

The crossplotting of elastic properties such as impedance, Vp/Vs ratio, density (#x03C1;), Lambda-Rho (#x03BB;#x03C1;) and Mu-Rho (#x03BC;#x03C1;) has been used to characterize the fluid and lithology in seismic area. In general, 2D or 3D crossplot of the elastic properties was used because of a limitation of data visualization, and the best combination of the properties for the crossplot was selected subjectively or empirically, which can be biased depending on the interpreter’s view. In this abstract, we propose a new workflow to overcome the limitations of the subjective selection of the axis parameters on the crossplots. We first apply non-linear transformation of the elastic properties from well logs, and select the best separable combination of the transformed properties. As the transformed elastic properties used for the combinations are high-dimensional (larger than 4D), it is not trivial to select the most separable properties visualizing them on the crossplots. Instead of the crossplotting, we use scatter-matrix-based-measure to quantify the separability of the combinations, and select the best combination with the largest separability. Finally, using multivariate probability density function (PDF) of the best combination from well logs and inverted elastic property volumes from seismic, we probabilistically analyze the facies on the seismic volume based on Bayesian inference. A field data has been used to demonstrate the effectiveness of our workflow.