Python classes and scripts for the manuscript "Data-Driven Inference of the Mechanics of Slip Along Glacier Beds Using Physics-Informed Neural Networks: Case study on Rutford Ice Stream, Antarctica" by B. Riel, B. Minchew, and T. Bischoff (https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2021MS002621). Included in this repository are scripts for training physics-informed neural networks for 1D and 2D ice flow. For both cases, ice flow follows the shallow shelf/stream approximation (SSA).
All neural network architectures used in the manuscript are implemented in networks.py, and training is implemented in train*.py. Utilities for loading and pre-processing training and validation data (from HDF5) can be found in utilities.py.
numpy
matplotlib
tensorflow
tensorflow_probability
pgan (https://github.com/bryanvriel/pgan)
tqdm
h5py
As of January 2022, the scripts in the examples work for tensorflow=2.6.2, tensorflow_probability=0.14.1, keras=2.6.0, and python=3.9. We have currently run into issues for tensorflow=2.7.0, which we are currently investigating. For any other compatibility issues, please open a new issue.