Classification of Magnetosheath Jets Using Neural Networks and High Resolution OMNI (HRO) Data

Abstract

Read online

Magnetosheath jets are transient, localized dynamic pressure enhancements found downstream of the Earth's bow shock in the magnetosheath region. Using a pre-existing database of magnetosheath jets we train a neural network to distinguish between jets found downstream of a quasi-parallel bow shock (θBn<45o) and jets downstream of a quasi-perpendicular bow shock (θBn>45o). The initial database was compiled using MMS measurements in the magnetosheath (downstream) to identify and classify them as “quasi-parallel” or “quasi-perpendicular,” while the neural network uses only solar wind (upstream) measurements from the OMNIweb database. To evaluate the results, a comparison with three physics-based modeling approaches is done. It is shown that neural networks are systematically outperforming the other methods by achieving a ~93% agreement with the initial dataset, while the rest of the methods achieve around 80%. The better performance of the neural networks likely is due to the fact that they use information from more solar wind quantities than the physics-based models. As a result, even in the absence of certain upstream properties, such as the IMF direction, they are capable of accurately determining the jet class.

Keywords

• magnetosheath jets
• neural networks
• solar wind
• machine learning
• bow shock