Bridging Ice Crystal Evolution and Ice Flow Through a Multiscale Model

Predicting future changes in ice sheet dynamics requires a robust understanding of how polycrystalline ice deforms under different mechanical and thermal conditions. Current ice-dynamics models often rely on empirical flow laws, but a growing body of laboratory and field evidence reveals that ice deformation is strongly influenced by evolving microstructural properties, such as crystallographic fabric and grain size. In this study, we present a two-way coupling framework that bridges the microscale and macroscale dynamics of ice deformation. Our approach combines a full-field viscoplastic model of ice microstructure with a Stokes flow solver for glacier dynamics, linked through a Fourier Neural Operator (FNO) surrogate model. The FNO captures complex, temperature- and stress-dependent microstructural evolution with high fidelity and low computational cost. We validate the microscale model against two benchmark laboratory experiments: one emphasizing static recrystallization under high-temperature conditions, and the other on dynamic recrystallization under high shear strain rates. We then apply the coupled framework to a one-dimensional vertical ice column spanning a range of flow and thermal regimes, demonstrating that microstructural feedback substantially modifies the predicted strain rate and stress profiles in ways not captured in empirical flow laws. We then apply the model to five ice-core sites across both Greenland and Antarctica, and compare simulated grain size to measured profiles. These comparisons illustrate how our coupled multiscale model enables the identification of microstructural features that preserve evidence of past flow.