Abstract
Common earth modeling practice propagates rock properties into the model from depositional lithofacies obtained from log and core analyses. The primary objection to this method is that the rocks, particularly carbonates, do not truly reflect the depositional framework. Diagenetic and other post-depositional processes have extensively modified the rocks. These processes tend to homogenize depositional fabrics, thus obliterating specific relationships needed to accurately predict and propagate properties.
Geologically reliable lithofacies for conventional and unconventional reservoir modeling can be obtained with unsupervised multivariate classification procedures applied to suites of log curves. Data analytics applied to log data can provide lithologies and flow units across the reservoir. Important rock properties that are critical inputs for a static earth model are included as part of the lithofacies definition.
Because lithofacies obtained from logs are partially based on the analysis of crossplots calibrated to rocks when possible, log-based facies definitions are subject to the experience-based bias of the project interpreter. Crossplots used for facies analysis often show complex patterns that suggest the multivariate nature of relationships between individual logs. Incorrect facies specification can lead to an inaccurate depiction of the lithologic geometry, which is a key shortcoming of this approach. The paper describes data analytics as applied to log data, developed as an aid to understanding lithologies and flow units across a stratigraphic interval.
A workflow was created that enables natural groups inherent in the data to be obtained. This approach does not require a priori rock type information. Multivariate procedures are applied to a selection of curves that comprise a log suite over an interval of interest to obtain log response groups based on log curve variation. These response groups can be used to propagate critical porosity and permeability data into the model.
The workflow described in this paper was successfully applied in several mature areas that include unconventional plays in the Delaware and Midland basins of west Texas and the Burgos basin of Mexico. New well locations and horizontal drilling targets selected with the recommended methods have performed significantly better than wells drilled without using this approach.
The new approach to rock typing presented in this paper is compatible with modern earth modeling methods and can improve drilling success by highlighting areas with more favorable rock properties. The sequence in which individual methods are applied is important in this workflow. Assumptions regarding distributional properties of the individual data elements are not required. Experience shows that this workflow provides improved understanding of lithological variation over the volume of interest, increasing the probability of better well performance.
