Abstract
In this article we develop a Physics Informed Neural Network (PINN) approachto simulate ice sheet dynamics governed by the Shallow Ice Approximation. Thisproblem takes the form of a time-dependent parabolic obstacle problem. Priorwork has used this approach to address the stationary obstacle problem and herewe extend it to the time dependent problem. Through comprehensive 1D and 2Dsimulations, we validate the model's effectiveness in capturing complexfree-boundary conditions. By merging traditional mathematical modeling withcutting-edge deep learning methods, this approach provides a scalable androbust solution for predicting temporal variations in ice thickness. Toillustrate this approach in a real world setting, we simulate the dynamics ofthe Devon Ice Cap, incorporating aerogeophysical data from 2000 and 2018.