| Abstract: Sub-Alfvénic intervals in the solar wind, most often found within the cores of magnetic clouds during interplanetary coronal mass ejection (ICME) events, can significantly influence solar wind-magnetosphere coupling by modifying energy transfer processes and wave activity. These intervals are uncommon and are typically recognized only after they occur, which limits the ability to investigate their onset dynamics or anticipate related geomagnetic disturbances.
In this study, a physics-guided machine learning framework is developed to forecast transitions into sub-Alfvénic flow using upstream solar wind parameters from 1-minute OMNI data. Sub-Alfvénic conditions are identified by the Alfvén Mach number < 1, with event onsets defined at transitions from super-Alfvénic to sub-Alfvénic regimes. For each event, pre-onset phases extending up to 45 minutes before the transition are used to construct labelled samples, allowing the problem to be formulated as probabilistic regime forecasting rather than deterministic onset detection.
Features are derived from sliding 60-minute history windows and include trends and variability in magnetic field strength, solar wind speed, proton density, plasma beta, Alfvén speed, and the deviation of Alfvén Mach Number from unity. Baseline machine learning models, including logistic regression and random forest classifiers, are trained using time-blocked cross-validation across multiple magnetic cloud (MC) events and evaluated using precision recall metrics, lead-time versus recall performance, and comparisons with simple physical-threshold predictors.
Preliminary results suggest that the physics-guided approach improves the early detection of emerging sub-Alfvénic intervals compared to threshold-based baselines, particularly for gradual transitions characterised by strengthening magnetic fields, reduced plasma density, and decreasing plasma beta. The framework demonstrates a meaningful forecasting capability with lead times of 30 to 60 minutes, offering a promising pathway toward automated detection tools and improved characterisation of solar wind-magnetosphere coupling during magnetic cloud passages.
Keywords: Alfvén Mach Number, ICME, Magnetic Cloud, Machine Learning
|
