Lithofacies classification in the Marcellus Shale and surrounding formations by applying expectation maximization to petrophysical and elastic well logs.

摘要

The Marcellus Shale is an organic-rich, marine shale that ranges in thickness from less than 10 ft in southeastern Pennsylvania and western portions of West Virginia to over 350 ft in some areas of northeastern Pennsylvania. In many parts of the Appalachian Basin, the shales of the Mahantango Formation directly overly the Marcellus Shale. These two shales can appear similar in core and hand sample, but contain very different organic and mechanical properties. Therefore, lithofacies classification is an important step for identifying productive zones within the Marcellus Shale and, likewise, is an important process in other unconventional reservoirs. The Expectation Maximization (EM) method was tested in the Marcellus Shale Gas Play and adjacent formations as a viable technique for classifying important lithofacies and locating target reservoir zones, and was found to be successful.;Expectation Maximization (EM) is a pattern-recognition algorithm that uses Gaussian mixture models to classify data, in this case, petrophysical and elastic well logs, into user-defined facies by assuming each well log contains a distribution of Gaussian curves for each facies. Appropriate well logs were selected that were 1) most sensitive to variations in shale lithologies and 2) commonly run in wells across the basin such that all study wells had the same input parameters.;The EM method was first tested along a vertical well section in central Pennsylvania using the gamma ray, density log, neutron porosity log, photoelectric factor, uranium curve from the spectral gamma ray, and s-wave velocity, which was computed directly from the shear sonic log. The EM method was able to classify major rock facies: sandstone, limestone, and shale. A fourth facies, organic-shale, was classified by a uranium log cut-off method. It was later determined that the uranium log had the least impact on the resulting models because it essentially doubled the role of the gamma ray log in the classification process. The shear sonic log was found to be useful for distinguishing sandstone from other lithologies; however shear sonic logs are not as commonly run in the Marcellus Shale and would not be as useful when trying to use the same set of input parameters on a large database of wells across the basin.;This research also focused on refining the EM method to differentiate between shales facies within the reservoir. The chosen wells included the gamma ray log, density log, neutron porosity log, photoelectric factor, resistivity log, and p-wave velocity, which was computed directly from the compressional sonic log. The resistivity log was added because of its ability to distinguish between carbonate and shale lithologies. The study was expanded to twelve wells across Pennsylvania and northern West Virginia, and five facies were identified: gray shale, dark gray shale, black shale, carbonate, and iron-bearing minerals. To verify the geological accuracy of the facies models, the results were compared to core data, well logs, mudlogs, and regional stratigraphy studies when available.;The EM method proved to be a robust facies classification technique in the Marcellus Shale reservoir. It is able to discriminate between reservoir and non-reservoir shale facies based on organic content and brittleness, characteristics that appear mostly homogenous in core. The method can also recognize a standardized, lithofacies-defined top to the Upper Marcellus Member where the overlying Mahantango Formation is gradational and poorly defined on petrophysical logs.