Advancing Statistical Rock Physics Capabilities for Lithofacies Prediction

When applying statistical rock physics an improved classification result can be obtained when different data types (attributes) can be evaluated and applied. Different types can bring additional but complementary information that can help to reduce the uncertainty. Standard statistical rock physics workflows usually only take seismic properties (Vp, Vs and density) into consideration. These attributes are successful in lithology classification but are relatively insensitive to pore-fluid classification. Electrical resistivity often serves as a good hydrocarbon saturation indicator, due to its sensitivity to pore-fluid. We therefore extend the classic statistical rock physics workflow to incorporate the electrical resistivity, thus further reduce the uncertainties during the classification and prediction process. Classification results after often displayed using a Bayers confusion matrix. This matrix shows only classification but no information is given about data overlap. We propose an enhanced confusion matrix, based on conditional probability, which show both facies classification results and the separation and overlap of the results. The workflow developed is illustrated using log data where one dataset is used to classify data from another location.