Characterising Small-Scale Structures in the Turbulent Magnetosheath Using Unsupervised Machine Learning

In collisionless plasmas, turbulence generates intermittent small-scale structures such as intense, thin current sheets, within which magnetic reconnection can occur. These structures, and reconnection in particular, are thought to play a key role in turbulence dynamics, energy dissipation, and particle energisation. The Earth’s magnetosheath, a highly turbulent region downstream of the bow shock, provides a natural laboratory for studying these nonlinear plasma processes. The Magnetospheric MultiScale (MMS) mission offers high-resolution, multi-point observations that are ideally suited to resolving small-scale structures in this environment. However, identifying and characterising such structures in spacecraft observations remains challenging due to their localised nature, complex magnetic topology, and the wide range scales involved.

We propose an unsupervised machine learning approach to systematically identify and characterise these structures, with specific emphasis on magnetic reconnection sites within turbulent plasma observations. Our method uses the Toeplitz Inverse-Covariance Clustering (TICC) algorithm, which models each cluster as a time-invariant correlation network, enabling the detection of complex patterns in turbulence. We evaluate TICC’s ability to identify reconnection events against existing datasets and interpret its clusters using the network-based feature scores. Finally, we assess the turbulence properties associated with the identified structures and the prevalence of magnetic reconnection across multiple intervals. This study aims to provide key insight into how the role of turbulent plasmas may vary across different turbulent environments.