Vol. 21 No. 1 (2026): July

Cover image by Batyrkhan Gainitdinov on pages 13–28. The image illustrates physics-guided self-supervised upscaling of 3D digital rock models from multi-resolution micro-CT data. Rock fragments acquired at different spatial resolutions are processed by a volumetric neural network that combines contextual representation learning with permeability-aware physical constraints. The trained model converts low-resolution carbonate scans into Darcy-scale multiclass models while preserving the connectivity of high-permeability channels and low-permeability barriers, enabling more reliable flow simulation and permeability estimation with reduced dependence on labelled data.