Well Log-Guided Synthesis of Subsurface Images from Sparse Petrography Data Using CGANs

Summary This work presents a novel approach using Conditional Generative Adversarial Networks (CGANs) to address data scarcity challenges in subsurface fluid flow studies. Our method generates representative thin section images across the full depth of subsurface formations using limited existing images and corresponding porosity profiles. The CGAN model demonstrates robust performance in generating geologically consistent thin section images across a wide range of porosity values (0.004–0.745), achieving 80% accuracy within a 10% margin of target porosity values. The successful integration of well log data with the trained generator enables continuous visualization of pore-scale features along the wellbore, effectively bridging the gaps between discrete core sampling points. This capability significantly enhances our understanding of spatial variations in rock properties and provides valuable insights for reservoir characterization, particularly beneficial for implementing energy transition technologies like carbon capture and storage and underground hydrogen storage.