Machine-learning based discovery of missing physical processes in radiation belt modeling

TL;DR

This paper introduces a physics-informed neural network approach to improve the modeling of Earth's radiation belt electron dynamics, revealing that a drift-diffusion model better describes the phenomena and providing a simpler, more accurate parameterization.

Contribution

It demonstrates how PINNs can infer physical model coefficients from data, leading to improved radiation belt models and insights into electron dynamics.

Findings

Drift-diffusion equation better describes electron dynamics

New parameterization for diffusion and drift coefficients

PINNs effectively infer physical model parameters

Abstract

Real-time prediction of the dynamics of energetic electrons in Earth's radiation belts incorporating incomplete observation data is important to protect valuable artificial satellites and to understand their physical processes. Traditionally, reduced models have employed a diffusion equation based on the quasilinear approximation. Using a Physics-Informed Neural Network (PINN) framework, we train and test a model based on Van Allen Probe data. We present a recipe for gleaning physical insight from solving the ill-posed inverse problem of inferring model coefficients from data using PINNs. With this, it is discovered that the dynamics of "killer electrons" is described more accurately instead by a drift-diffusion equation. A parameterization for the diffusion and drift coefficients, which is both simpler and more accurate than existing models, is presented.