Datapages, Inc.Print this page

Determining Lithofacies and Quantitative Mineralogy of Sedimentary Rocks from Bulk Elemental Geochemistry: A Neural Network Approach

Smith, Christopher N.*1; Assous, Said 2
(1) Geoscience Development, Weatherford International, Houston, TX.
(2) Geoscience Development, Weatherfod International, East Leake, United Kingdom.

Elemental analysis of sedimentary rocks has been used to correlate stratigraphic units, determine sediment source provenance, evaluate the tectonic setting of a basin, and for sedimentological classification (recently summarized by Pe-Piper et al., 2008). The use of certain elements (e.g. Ti, K, Zr, Hf) to derive information on source and tectonic setting has been shown to be regionally specific and far from universal (Ryan and Williams, 2007; Pe-Piper et al., 2008) and previous attempts to calculate quantitative mineralogy has been inaccurate with errors up to ±75% when compared to modal compositions. Rather than relying on sedimentary norms and linear models, we propose a neural network-based approach to process major, minor and trace element data quickly and accurately (±10%) to determine lithofacies and quantify mineralogy of a sedimentary sample. Two different algorithms are presented for this purpose and a performance comparison is discussed.

Based on the performance of the models, accurate quantitative mineralogy from elemental analysis is best preformed by a neural network approach. This is due to the complexity of mineral chemistry and the non-uniqueness of the distribution of major, minor and trace elements within the minerals of the sample. The neural network models are also able to predict lithofacies, determined from previous sedimentological studies, in shoreline clastic environments.

The strength of the predictive capabilities of the neural networks is that detailed facies models and stratigraphy can be correlated to offset wells through the use of bulk elemental analysis. Element-derived mineralogy and stratigraphy can be used as inputs for reservoir modeling and optimization strategies. Our models have been shown to have a high accuracy predicting lithofacies and mineralogy of sedimentary rock standard reference materials, cuttings and cores, allowing for predictive modeling of reservoir properties.

Pe-Piper, G., Triantafyllidis, S., and Piper, D.J., 2008, Geochemical identification of clastic sediment provenance from known sources of similar geology: The Cretaceous Scotia Basin, Canada: Journal of Sedimentological Research, v. 78, p. 595-607, doi:10.2110/jsr.2008.067.

Ryan, K.M., and Williams, D.M., 2007, Testing the reliability of discrimination diagrams for determining the tectonic depositional environment of ancient sedimentary basins: Chemical Geology, v. 242, p. 103-125, doi: 10.1016/j.chemgeo.2007.03.013.


AAPG Search and Discovery Article #90142 © 2012 AAPG Annual Convention and Exhibition, April 22-25, 2012, Long Beach, California