Petro-elastic facies classification in the Marcellus Shale by applying expectation maximization to measured well logs

Summary A new methodology for a log-based facies classification using a statistical algorithm, Expectation Maximization, is proposed to classify lithologic-facies utilizing commonly available wireline logs. This method was tested in the Marcellus Shale, an unconventional reservoir located in the Appalachian Basin of eastern North America. The method relies on Gaussian mixture models with the assumption that each facies has a unique Gaussian distribution of petroelastic properties. The technique was checked against mud logs, well log interpretations, and regional stratigraphy for accuracy and produced promising results.