Forecasting Radiation Belt Electron Flux Using Machine-learned Parameters for Radial Transport Equation
Abstract
We solve the inverse problem of learning the optimal drift and diffusion coefficients for the radial transport equation of energetic electrons in the Earths radiation belt, using a Physics-Informed Neural Network (PINN). The learned parameters are then parameterized as simple functions of L-shell and Phase Space Density, and this new parameterization is used in a forward model for forecasting unseen values of the electron flux. We show that the parameterization discovered by PINN is more robust, accurate, and efficient than earlier empirical parameterizations used in the literature. This work constitutes a prime of example of how to enhance our space weather prediction capabilities, by combining prior physics knowledge with data-driven discovery of unknown coefficients.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMNG45B0578C