Geologically constrained electrofacies classification of fluvial deposits: An example from the Cretaceous Mesaverde Group, Uinta and Piceance Basins

Abstract

Statistical classification methods consisting of the k-nearest neighbor algorithm (k-NN), a probabilistic clustering procedure (PCP), and a novel method that incorporates outcrop-based thickness criteria through the use of well log indicator flags are evaluated for their ability to distinguish fluvial architectural elements of the upper Mesaverde Group of the Piceance and Uinta Basins as distinct electrofacies classes. Data used in training and testing of the classification methods come from paired cores and well logs consisting of 1626 wireline log curve samples each associated with a known architectural element classification as determined from detailed sedimentologic analysis of cores (N = 9). Thickness criteria are derived from outcrop-based architectural element measurements of the upper Mesaverde Group. Through an approach that integrates select classifier results with thickness criteria, an overall accuracy (number of correctly predicted samples/total testing samples) of 83.6% was achieved for a fourclass fluvial architectural element realization. Architectural elements were predicted with user's accuracies (accuracy of an individual class) of 0.891, 0.376, 0.735, and 0.985 for the floodplain, crevasse splay, single-story channel body, and multistory channel body classes, respectively. Without the additional refinement by incorporation of thickness criteria, the k-NN and PCP classifiers produced similar results. In both the k-NN and PCP techniques, the combination of gamma ray and bulk density wireline log curves proved to be the most useful assemblage tested.

Stratigraphic architecture of the Mississippian limestone through integrated electrofacies classification, Hardtner field area, Kansas and Oklahoma

“…KNN is another method of clustering that classifies an unknown data point based on the most frequently occurring class of its nearest previously defined data points (Allen and Pranter, 2016). The number of neighboring points considered (K) is user defined and should be adjusted to find an optimum value for each data set.…”

“…These comparisons are carried out through confusion matrices. A confusion matrix is a tool used to evaluate the effectiveness of various forms of artificial intelligence (Ting, 2011;Allen and Pranter, 2016). By charting predicted classes in columns and measured, or actual classes in rows, confusion matrices are able to quickly identify and quantify instances of successful and unsuccessful classification.…”

Convolutional neural networks as aid in core lithofacies classification

“…Although machine learning has been significantly used in geoscience fields, the application of this technique in core-based lithofacies identification, a key component to better understanding oil and gas reservoirs, is still limited. Machine-learning techniques have been intensely used to aid seismic-facies classification (de Matos et al, 2007(de Matos et al, , 2011Roy et al, 2014;Qi et al, 2016;Zhao et al, 2016Zhao et al, , 2017Qian et al, 2018), electrofacies classification (Allen and Pranter, 2016), lithofacies classification from well logs (Baldwin et al, 1990;Zhang et al, 1999;Bestagini et al, 2017), to predict permeability in tight sands (Zhang et al, 2018), and even for seismicity studies (Kortström et al, 2016;Perol et al, 2018;Sinha et al, 2018;Wu et al, 2018). Cored wells are important because they are the only data that provide the ground truth of subsurface reservoirs including the lithofacies variations.…”

Progress and Challenges in Deep Learning Analysis of Geoscience Images

“…Kachine learning (ML) techniques have been reliably used in geoscience interpretation for almost two decades including seismic-facies classification (de Matos et al, 2011;Qi et al, 2016;Zhao et al, 2017), electrofacies classification (Allen and Pranter, 2016), and analysis of seismicity (Kortström et al, 2016;Sinha et al, 2018). Traditionally, geoscience ML applications rely on a carefully selected set of features or attributes.…”