Description
Turbulence is a fundamental process observed in astrophysical plasmas. In collisionless environments, turbulence naturally generates thin, intense current sheets where magnetic reconnection can occur. Reconnection is a process in which magnetic field lines break and reconnect, releasing magnetic energy, and is thought to play a crucial role in turbulence dynamics and energy dissipation.
The Earth’s magnetosheath, a highly turbulent region, provides a natural laboratory for studying collisionless plasmas. The NASA's Magnetospheric MultiScale (MMS) mission offers high-resolution, multi-point observations of this region, well-suited for identifying turbulence-driven reconnection. However, identifying reconnection events within observations is challenging and time-consuming due to their localised nature, complex magnetic topologies, and the wide range of scales.
We present an unsupervised machine learning framework to systematically identify reconnection sites in turbulent plasma observations. This approach requires key physical features that highlight reconnection sites as input. Then, the Toeplitz Inverse Covariance-Based Clustering (TICC) algorithm groups time-series data into distinct plasma structures based on their internal correlations. TICC successfully recovers known reconnection events and identifies new candidates. These candidates are then analysed in a local coordinate system aligned with the current sheet, where reconnection signatures become clearer and can be quantitatively characterised. A second unsupervised classification step distinguishes true reconnection events from false positives.
This approach enables the construction of statistically robust catalogues of turbulence-driven reconnection events, supporting systematic studies of energy dissipation in space plasmas.