Speaker
Description
Understanding how solar wind parameters control ionospheric structure remains a fundamental challenge in space weather prediction. This study applies unsupervised machine learning techniques to analyse EISCAT Svalbard radar observations (1998-2023) during winter months, revealing unexpected hierarchies in solar wind-ionosphere coupling.
K-means clustering with UMAP dimensionality reduction were employed to identify distinct ionospheric regimes across four regions: auroral oval, cusp aurora, polar cap, and high electric fields. Statistical significance of solar wind parameter effects was quantified using Kruskal-Wallis tests with effect size calculations (η²). Seven solar wind parameters were analysed: temperature (T), velocity (V), density (N), pressure (P), plasma beta, Mach number, and Akasofu epsilon.
Key findings suggest a potential shift in understanding solar wind control mechanisms. Solar wind temperature appears as the dominant driver across all ionospheric regions, with remarkably large effect sizes (η² = 0.41-0.97). This represents 41-97% of the variance in ionospheric clustering patterns, substantially exceeding traditional parameters. The hierarchy generally follows: T > V > P/N > Akasofu epsilon.
Regional variations indicate distinct statistical sensitivities to solar wind parameters. High electric field regions show extreme temperature sensitivity (η² = 0.88-0.93). The polar cap exhibits the strongest sensitivity to solar wind density variations (N: η² = 0.10), possibly indicating direct statistical control by basic plasma parameters. The auroral oval shows balanced sensitivity to multiple solar wind parameters (T, V, P all significant), while the cusp aurora demonstrates intermediate parameter dependencies between the polar cap and oval regions.
Surprisingly, the Akasofu epsilon parameter ranks 4th-6th in importance (η² = 0.02-0.11), significantly below basic plasma parameters. This might suggest that local polar ionospheric structuring responds more directly to solar wind plasma properties than to integrated energy coupling functions designed for global auroral activity.
The dominance of temperature possibly indicates that solar wind structures (CMEs, corotating interaction regions) might be the primary drivers of polar ionospheric variability. High-temperature periods correlate with compressed plasma carrying enhanced magnetic field strength, driving stronger magnetosphere-ionosphere coupling than steady-state solar wind conditions.
Implications for space weather encompass multiple areas. Temperature-based forecasting may provide superior polar ionospheric predictions compared to traditional approaches. Conventional velocity and magnetic field coupling functions may underestimate structured solar wind impacts on polar regions. Regional ionospheric responses require differentiated modelling approaches that account for local coupling physics rather than global parameterizations. Single-point polar observations capture different physics than global auroral indices, suggesting complementary rather than redundant measurement strategies.
This analysis demonstrates that machine learning approaches might reveal hidden parameter hierarchies in complex space physics datasets. The unexpected dominance of solar wind temperature challenges conventional wisdom about solar wind-ionosphere coupling and suggests new avenues for improving space weather models in polar regions.
Future work will extend this analysis to different seasons and magnetic local times, and try to understand how temperature-based forecasting models compare against traditional approaches for operational space weather prediction.
