Pix2Geomodel: A Next-Generation Reservoir Geomodeling with Property-to-Property Translation. Interesting paper from Al-Fakih et al (2025).
Conclusions
“This study presented Pix2Geomodel, a pioneering conditional GAN framework, to enhance geological modeling of the Groningen gas field’s Rotliegend reservoir. By leveraging Pix2Pix architecture, the framework successfully predicted facies, porosity, permeability, and water saturation from masked inputs and facilitated property-to-property translation, achieving high accuracies (e.g., facies PA 0.88, water saturation PA 0.96) and robust translation performance (e.g., facies-to-Sw PA 0.98). The approach captured the reservoir’s spatial heterogeneity, validated by variogram analysis, and outperformed traditional methods in handling complex subsurface patterns. Despite challenges with porosity and permeability predictions due to microstructural variability, Pix2Geomodel demonstrated significant potential for reservoir characterization under data-scarce conditions. The study’s open-source datasets and code foster reproducibility, aligning with geoscience community goals. Future work will explore 3D modeling (Pix2Geomodel v2.0), multi-modal data integration, and advanced GAN architectures to address limitations and enhance applications in hydrocarbon recovery, geothermal energy, and carbon sequestration.”
