FE-AI framework for fast and accurate multiscale simulation of anisotropic ice-sheet dynamics

  • El Fallaki Idrissi, Mohammed (Arts et Metiers Institute of Technology)
  • Tannous, Mikhael (Arts et Metiers Institute of Technology)
  • Castelnau, Olivier (Arts et Metiers Institute of Technology)
  • Mangeney, Anne (Université Paris Cité)
  • Montagnat, Maurine (University Grenoble Alpes)
  • Gagliardini, Olivier (University Grenoble Alpes)
  • Ghnatios, Chady (University of North Florida)
  • Ponte Castañeda, Pedro (University of Pennsylvania)
  • Chinesta, Francisco (Arts et Metiers Institute of Technology)

Please login to view abstract download link

The effective rheology of polar ice exhibits strong viscoplastic anisotropy arising from its complex crystallographic microstructure, which plays a key role in controlling ice-sheet flow dynamics. While advanced mean-field and full-field homogenization models can accurately describe the nonlinear anisotropic flow law of ice and the evolution of crystallographic texture, their high computational cost limits their application in large-scale ice-sheet simulations. To overcome this challenge, we propose an FE-AI framework that integrates machine-learning-based surrogate models within a finite element (FE) ice-flow solver. The AI surrogates are trained to reproduce both the anisotropic constitutive behavior of ice and the evolution of its microstructural texture, effectively replacing expensive multiscale homogenization calculations. Implemented in the Elmer/Ice FE software, the framework is applied to investigate the impact of anisotropy on large-scale ice-sheet dynamics. This hybrid approach enables efficient multiscale simulations of anisotropic ice-sheet flow.