Region identification in spacecraft data using supervised machine learning

The classification of near-Earth plasma regions, i.e., distinguishing the region in which a spacecraft is located at any given time, is beneficial for both understanding the dynamics of the interaction between the Earth’s magnetosphere and the solar wind, and for modeling the characteristic boundaries separating these regions. We use measurements from the THEMIS B spacecraft between 2008 and 2010 (340 days in total) with a time resolution of one minute. The data include solar wind velocity and density, magnetic field magnitude, and standard deviation of magnetic field magnitude calculated over one-minute intervals. These data are used for manual labeling of four distinct plasma regions: solar wind, foreshock, magnetosheath, and magnetosphere. Ion energy flux data are used to classify the foreshock, if necessary. An automated classification of the respective regions based on measured plasma and magnetic field parameters is then achieved using either neural network or random forest classifiers. The performance of these classifiers is evaluated and compared. Generally, very high accuracy is achieved, but distinguishing between solar wind and foreshock remains an issue.