A physics constrained deep learning framework characterizing Antarctic ice shelf fractures

We provide a MATLAB MAT-file dataset that serves as input for physics-informed neural network (PINN) simulations of the Amery Ice Shelf, East Antarctica. The dataset, "pinn_input_dataset_Amery_Ice_Shelf_IPR_constrained_v1.0.mat", contains structured velocity and ice thickness fields, as well as discrete constraint points derived from airborne ice-penetrating radar (IPR) measurements. These inputs are designed to guide the PINN in reconstructing ice shelf dynamics, where direct observational coverage is limited. The MAT-file can be read directly in MATLAB. Contents of the data can be found in "pinn_input_dataset_Amery_Ice_Shelf_IPR_constrained_v1.0_contents.txt". The Jupyter notebook "run_pinn_fracture_simulation.ipynb" contains the main code for running the PINN model.