Mapping High Frackability and High TOC Zones in The Barnett Shale: Supervised Probabilistic Neural Network vs. Unsupervised Multi-Attribute Kohonen SOM

Verma, Sumit (University of Oklahoma) | Roy, Atish (University of Oklahoma) | Perez, Roderick (University of Oklahoma) | Marfurt, Kurt J. (University of Oklahoma)



The Barnett Shale is one of the most important unconventional shale plays in the USA. Identification of the key unconventional gas/tight rock parameters - total organic content (TOC) and frackability (ease with which the rock can be fractured) are key to economic completion processes. Singh (2008) conducted a complete analysis of cores and gamma ray logs over the Barnett Shale covering the study area, defining 14 different parasequences, and found a correlation between the gamma ray log response to TOC and frackability. Rock frackability increases with an increase in the quartz and dolomite content of a rock. In general, these minerals have low gamma-ray readings. In contrast, the organic rich content and clay rich intervals are radioactive; it can be identified through high gamma ray readings. Assuming this observation to be generally applicable, we extend Singh's (2008) core to well log correlation to a well log to 3D seismic correlation through probabilistic neural network (PNN) resulting in a predicted 3D gamma ray volume.

The pre-stack seismic data span approximately 30 mi2.We use 19 wells with gamma ray in the neural network training which are correlated to pre-stack simultaneous seismic inversion volumes and other seismic attributes correlatable to lithology. The generated gamma ray volume showed good correlation with the gamma ray logs, including the 11 blind wells those were not used in neural network training at any stage.

The gamma ray output volume is then correlated to the unsupervised multi-attribute seismic facies analysis. Finally, the predictions are validated through the actual production data (EUR), which confirms our initial hypothesis.