Abstract/Summary

Understanding space-weather phenomena is a growing requisite given our day-today reliance upon space-based infrastructure. This entails identifying the causal factors of space-weather phenomena, quantifying the magnitude of response of space-weather events, and jointly using this information for forecasting. Machine learning (ML), as a set of mathematical and statistical tools, has been successfully used across many fields of research, demonstrating vast potential to improve our understanding of space-weather phenomena. We apply unsupervised ML (dimension-reduction and clustering) to derive robust solar wind classifications – providing further insight into space-weather driving. Our unsupervised techniques are applied to a theoretically-motivated set of ex�tant composition variables - which are non-evolving with solar wind propagation. We demonstrate that solar-wind-speed-based classifications lose latent information regarding solar source regions. Our dimension-reduction suggests a more informative latent-space to represent streamer-belt-origin solar wind. Subsequently, we investigate the outer boundary of the outer radiation belt (OBORB). Modelling of the energetic-electrons in the outer radiation belt is crucial to the effective operation of many Earth-orbiting satellites, and the outer boundary conditions for such models are critical to accurate simulation. We ap�plied simple ML models to a dataset of electron distribution functions, testing a range of potential boundary locations – yielding an empirical identification of the quiet-time boundary location. Next, we employed Bayesian neural networks to construct parameterised, probabilistic models providing synthetic nowcasts of the electron fluxes at the boundary. These models bridge the gap between the empirically identified OBORB location and the information required by modellers to construct the outer boundary conditions. This work showcases how a broad spectrum of ML techniques can be applied to a variety of space-weather related problems. We present novel scientific results with significant implications for future studies into the solar wind and radiation belts, and ultimately, space-weather forecasting.