Automated Identification of Foreshock Transients

Foreshock transients (FTs) are short-lived mesoscale structures near Earth's bow shock, typically generated by interactions between solar wind discontinuities and either the bow shock or foreshock backstreaming ions. They are characterized by a hot, low-density core, with reduced magnetic field strength and plasma velocity, and bounded by compressed edges.

 

In this study, we develop a machine learning pipeline to identify FTs using Cluster 1 spacecraft data from 2003–2009. We start with a catalog of 83 FT events and 300 solar wind/foreshock intervals, each has a time duration of 6 minutes and including magnetic field, plasma parameters, and 31 channels of backstreaming ion energy spectrogram as features. Seven 1D Convolutional Neural Networks (1D CNNs) are trained using a leave-one-year-out cross-validation approach. After that, each model is validated on solar wind/foreshock (SWF) regions corresponding to the held-out year. The model detects about 280 new FTs between 2003–2009 with precision of around 0.3. These detections, along with false positives, are then added to the training set to improve performance. When applied to 2010 SWF data, the updated model identifies 24 true positives with a precision of 0.5, compared to a precision of 0.2 when the additional training data is not included.

 

This study demonstrates the feasibility of an automated approach for FT detection. The updated model can be applied to data from other years or different Cluster spacecrafts. The resulting comprehensive FT catalog will support future studies on the properties of FTs, while the downstream false positives can serve as a calibration of the SWF catalog.