Data and figures for reconstruction and multi-scale transfer learning of complex fracture networks

The cvPINN-TL-DFN framework introduces a transfer learning approach that bridges lab experiments and field-scale models. It leverages a multi-fidelity loss to adapt features from lab-trained cvPINNs to subsurface discrete fracture networks (DFNs) at the kilometer scale, ensuring consistency in toughness regimes and the preservation of critical fracture connectivity for geothermal applications.