Petrophysically significant rock types are critical factors in reservoir characterization especially in complex carbonate rocks such as make up the Tengiz Field in the Caspian region. However, the current 3D reservoir model for Tengiz is based on simple ?gross? facies defined by the main depositional areas of the field. New rock types were defined using a combination of rock fabric, porosity type, mineralogy and MICP measurements and serve to bring the detailed petrophysics in to the static model. This paper describes the integrated approach that has been developed to predict rock types from logs calibrated to cores, in order to populate the reservoir model.
The rock type prediction method applied in this study includes the integration of: Discriminant Analysis, the Indexed Self Organizing Map method and FMI image analysis. Discriminant Analysis provides the optimal log combination for prediction of given groups (rock types. The SOM method combines unsupervised spatial clustering with calibration by indexing self-organizing maps of log data with core facies and/or rock type definitions, and has the advantage that it captures non-linearity. Indexation is the crucial step here; without this ?calibration-by-indexation? step it is hard to interpret the log data clusters (?electrofacies? in terms of rock types. Indexation necessitates the close integration of core and log data. MICP data is integrated into the rock type definition. FMI image texture analysis by SOM gives a ?texture log?, which is integrated together with conventional logs.
Ten cored wells from Tengiz Platform and Flank areas were used in the study. After Discriminant Analysis, rock types were predicted from an Indexed Self Organizing Map with 80%+ accuracy. Validation of log responses by iterative comparison using core data and the ?Log Spectrum Analyser? led to further improvement in prediction. Blind tests were performed on wells and confirm the robustness of the prediction model.
The output predictions are accompanied by the probability of prediction of each group at each depth. Rock types are derived that provide distributions of the porosity, permeability and saturation properties in each unit. The likelihood of encountering a particular rock type is also quantified.
The next generation Tengiz reservoir model will incorporate log-derived rock-types as described in this study.