Harnessing expressive capacity of Machine Learning modeling to represent complex coupling of Earth's auroral space weather regimes

TL;DR

This paper develops advanced deep learning models to improve global predictions of Earth's auroral space weather, especially extreme events, by leveraging innovative loss functions, multi-task learning, and spatio-temporal modeling techniques.

Contribution

It introduces novel ML approaches, including a multi-task model and loss functions, to enhance space weather prediction accuracy and address longstanding challenges in modeling extreme auroral events.

Findings

ML models outperform traditional methods in predicting extreme auroral events

Loss function engineering significantly improves model accuracy

Spatio-temporal modeling captures complex space weather dynamics

Abstract

We develop multiple Deep Learning (DL) models that advance the state-of-the-art predictions of the global auroral particle precipitation. We use observations from low Earth orbiting spacecraft of the electron energy flux to develop a model that improves global nowcasts (predictions at the time of observation) of the accelerated particles. Multiple Machine Learning (ML) modeling approaches are compared, including a novel multi-task model, models with tail- and distribution-based loss functions, and a spatio-temporally sparse 2D-convolutional model. We detail the data preparation process as well as the model development that will be illustrative for many similar time series global regression problems in space weather and across domains. Our ML improvements are three-fold: 1) loss function engineering; 2) multi-task learning; and 3) transforming the task from time series prediction to spatio-temporal prediction. Notably, the ML models improve prediction of the extreme events, historically obstinate to accurate specification and indicate that increased expressive capacity provided by ML innovation can address grand challenges in the science of space weather.