Celebration of Scholars
P38 - Using Machine Learning to Model Terrestrial Gamma-Ray Flash Photon Escape From Earth's Atmosphere
Name:
Lucas Peterson
Major: Mathematics & Physics
Hometown: Franklin
Faculty Sponsor: Julie Dahlstrom
Other Sponsors: Brant Carlson
Type of research: SURE
Funding: WSGC
Abstract
Terrestrial Gamma Ray Flashes (TGFs) are millisecond-long emissions of gamma rays from thunderclouds typically observed by satellites. Photons observed by satellites are subject to scattering and absorption as they escape the atmosphere, so satellite data is often interpreted with the aid of simulation. Such simulations are often slow, especially for sources deep in the atmosphere, due to the high likelihood of absorption. To speed up these simulations, we train a Conditional Generative Adversarial Network (cGAN) to reproduce distributions of photons at satellite altitude resulting from sources with energies from 0.1-10 MeV and incidence angles of 0-90 degrees. Our method can closely reproduce the 1D and 2D histograms of TGF photons. While it comes at a high upfront computational cost of collecting the data needed to train it, the resulting cGAN can generate photons at least 1000x faster than GEANT4 for sources at 12.5km altitude. Overall, our results support the application of machine learning techniques for TGF simulation tasks.Submit date: March 17, 2025, 8:45 a.m.