Interpreting the Subsurface at High Lithological Resolution by Integrating Information from Well-Log Data and Rock-core Images

Spectral facies-interpretation and -classification methods have been proposed to improve the lithological resolution in subsurface characterization. In the spectral facies interpretations, the intensity values of the RGB spectrum and the local entropy from rock-core images are used, and the results are compared to conventional electrofacies interpretations. During the classification, a practically applicable model that identifies the lithofacies with high lithological resolution is constructed by using a deep neural network (DNN) model, with the interpreted spectral facies and well-log data from the corresponding depths used as response and explanatory variables for the training, respectively. Core images and six types of well-log data from the Satyr 5 well in Western Australia are applied for the actual implementation. Through comparative interpretations, three spectral facies are identified as separable lithofacies (i.e., shale-dominant, shale-sand mixture, and clean sand-dominant facies) and two electrofacies (i.e., shale-dominant and sand-dominant facies) are identified by a conventional method. In the classification based on the spectral facies, the trained DNN model showed high prediction accuracy for all the lithofacies. Based on these observations, more precise lithological interpretation and classification can be conducted with the developed methods. The developed methods have a potential to improve subsurface characterization when high lithological resolution is essential.