Seismic Facies Classification using High-Resolution Processing and Transfer Learning with Res-Unet

Seismic facies analysis plays a significant role in not only initial basin exploration and prospect evaluation, but also reservoir characterization and ultimately field development. However, large-scale spatial distribution of effective reservoirs and non-reservoirs can not meet the requirements of reservoir quality classification in field development period. In this study, a workflow for seismic facies classification in productive reservoirs has been developed and applied to X block in Ecuador. First, a high-resolution processing workflow is conducted to improve the reliability of thin layer delineation. Then, high-precision facies labels from well locations are extended to the entire 3D seismic area for grasping the reservoir heterogeneity. After that, a transfer learning model is created by training Res-Unet on original dataset with real seismic data. Finally, the proposed method is applied to X block in Ecuador during the development period. The application of this methodology provides the early step for building a reliable and predictive seismic faices model for new well proposal. The use of the propsoed approach, in which high-resolution seismic processing and transfer learning are considered together, could optimize and support seismic faices model building with positive impact on productive target pinpointing and consequently on asset production and overall value.