Detailed petroleum system insights using deep learning: A case study from the Scarborough Gas Field, offshore Australia

Abstract

Geomodelling large, high-resolution, three-dimensional (3D) seismic data can be a time-consuming and tedious process. However, new methods utilizing deep learning are accelerating the mapping and interpretation of geologic features. While most deep learning applications on seismic data have focused on mapping features such as faults and salt, we propose a novel, interactive deep learning methodology that enables the geoscientist to characterize various petroleum system elements by labeling and training networks on associated elements proven by exploration well data. The Scarborough seismic survey within the Carnarvon Basin, offshore Australia, contains several seismic representations of petroleum system elements such as dry gas shows and shallow gas anomalies. Dry gas migrates vertically through overlying strata and along faults. Results from well-trained deep learning networks can not only accurately map various petroleum elements of the basin, but also reveal a higher level of detail. This new level of detail can even help the interpreter identify which faults may be related to migration and leak-off. Compared to traditional petroleum system modeling workflows, these results were obtained in a fraction of the time using interactive deep learning and enables geoscientists to better characterize regional and local trends while also making observations at the petroleum system scale.