--> --> Initial Results of Direct Multivariate Modeling of Sandstone Reservoir Quality Using Whole-Rock Elemental Data

International Conference & Exhibition

Datapages, Inc.Print this page

Initial Results of Direct Multivariate Modeling of Sandstone Reservoir Quality Using Whole-Rock Elemental Data


Whole-rock elemental data is now routinely acquired from core in the laboratory (by ICP or XRF), and from cuttings at wellsite (by XRF) while drilling. While primarily used for chemostratigraphic correlation and mineralogy modeling, this data may also contain stand-alone predictive information about sandstone reservoir quality; specifically porosity and permeability. The ability to predict these attributes directly from elemental analysis of cuttings while drilling is of great practical value. The rationale behind elemental prediction of sandstone reservoir quality is based on the premise that mineralogy and flow properties of a given sandstone are each related to the sum total of processes that generated and modified the sand (e.g., weathering, erosion, transport, temporary storage, final deposition, bioturbation, compaction, early diagenesis, burial diagenesis, hydrocarbon charge, etc.). Therefore, a mathematical relationship may exist between elemental composition (tied to mineralogy) and flow properties such as porosity and permeability. An element-based predictive model for porosity and permeability was created to test this premise using data from over 1,450 core plugs, from 12 wells, in two completely unrelated reservoir sequences. Specifically, one sequence was deposited in an estuarine-dominated environment, while the other was deposited in an eolian-dominated environment. While these sequences exhibited similarities in their mineralogical and lithological characteristics, their respective elemental (i.e., compositional) variations span completely separate convex hulls in the 10D compositional space. This attribute of the available dataset mandated the use of “local models” to predict the flow properties of each sequence, as opposed to a single “global model,” which could be used to predict both sequences. Training and testing of the models were accomplished using k-fold cross-validation. The elemental data was partitioned by well to emulate a real-world field-development scenario; where the data from existing cored wells could be used to construct predictive models applicable to the analysis of cuttings from future wells. Based on the initial dataset, the resultant model accuracy was ± 2 vol% for porosity and ± 11 mD for permeability. Additionally, it is believed that the uncertainty related to laboratory reference measurements of low-permeability sandstones is the limiting factor with respect to the accuracy of the current models.