Elijah Pomerantz ’22, Physics

Elijah Pomerantz ’22

Major

Physics

Project Title

Ice Hardness Inversion in 1D With Physics-Informed Neural Networks

Presentation Link

View Elijah's Presentation

I worked with a physics-informed neural network (PINN) to develop an inversion model for determining a temperature-dependent parameter that characterizes the hardness of an ice shelf. PINNs can embed the knowledge of any physical laws that govern a given data set in the learning process. We worked on developing the tools to solve idealized problems as we move toward applying the PINN to real-world data for ice in Antarctica. The goal of this work was to identify places in ice shelves where we expect ice to be softer, which will help determine where ice shelves may be susceptible to breaking. I found that our PINN model can make good predictions about the ice-hardness parameter for our idealized problems, even when subject to noisy, sparse or oscillatory data. I learned how to apply novel machine learning techniques to solve challenging physics problems. It was fascinating to see the ways that computer science, physics and climate science can intersect to help us better understand how our polar regions will react to a warming climate.