“Behind-the-outcrop” outcrop: A creative machine learning application for ground penetrating radar data visualization as an artificial photorealistic rock volume

Outcrop analogs are the high-resolution equivalent of carbonate reservoirs often utilized to decipher sub-seismic lateral and vertical depositional facies heterogeneity. Ground penetrating radar (GPR) is one of the most popular geophysical methods to extend the dimensionality of the depositional bodies identified on the 2D outcrop face. Correlating radargram signals to outcrop depositional facies requires exclusive and subject-specific expertise unfamiliar to most field geologists. This study presents a creative application of forward modeling and the conditional generative adversarial network (CGAN) to construct a photorealistic 3D “behind-the-outcrop” model from radargram and a drone-based digital outcrop model (DOM). This study tested the method to “see” the stromatoporoid/coral buildups in the interior of an outcrop cliff in central Saudi Arabia. The digital “behind-the-outcrop” model provides a straightforward medium for geologists to visualize and interpret rock formation instead of interpreting radargram signals.