Zone Identification and Oil Saturation Prediction in a Waterflooded Field: Residual Oil Zone, East Seminole Field, Texas, USA, Permian Basin

Roueché, Jacqueline N. (Lynxnet LLC, U.S. Geological Survey) | Karacan, C. Özgen (U.S. Geological Survey)



Recently, the miscible CO2-EOR tertiary process used in the main pay zone (MP) of suitable reservoirs has broadened to include exploitation of the underlying residual oil zone (ROZ) where a significant amount of oil may remain. The objective of this study is to identify the ROZ and to assess the remaining oil in a brownfield ROZ by using core data and conventional well logs with probabilistic and predictive methods.

Core and log data from three wells located in the East Seminole Field in Gaines County, Texas, were used to identify the MP and ROZ in the San Andres Limestone, and to predict oil saturations. The core measurements were used to calculate probabilistic in-situ oil saturations within the MP and the ROZ as a function of depth. Well logs, in combination with core data and calculated saturations, on the other hand, were used to develop two expert systems using artificial neural networks (ANN); one to identify the ROZ and MP, and the other to predict oil saturation. These systems were also supported by a classification and regression tree (CART) analysis to delineate the rules that lead to classifications of zones.

Results showed that expert systems developed and calibrated by combining core and well log data can identify MP and ROZ with a success score of more than 90%. Saturations within these zones can be predicted with a correlation coefficient of around 0.6 for testing and 0.8 for training data. The analyses showed that neutron porosity and density well log readings are the most influential ones to identify zones in this field and to predict oil saturations in the MP and ROZ. To explain the relationships of input data with the results, a rule-based system was also applied, which revealed the underlying petrophysical differences between MP and ROZ.

This new predictive approach using machine learning techniques, could potentially address the challenges that previous studies have come up against in defining the ROZ within the formation and quantifying remaining oil saturations. The method can potentially be applied to additional fields and help reliably identify the ROZ and estimate saturations for future resource evaluations.