A Machine Learning Approach to Understanding the Physical Properties Associated with Magnetic Flux Ropes in the Solar Wind

Magnetic flux ropes (MFRs) are an important structure in the solar-terrestrial environment characterized by their twisted magnetic field structure. Believed to be the internal magnetic field configuration of interplanetary coronal mass ejections (ICMEs) extending from the sun, they are important for understanding how the Sun and solar wind can impact the Earth. Closely related to the large-scale magnetic flux ropes in ICMEs, or magnetic clouds (MCs), are the small-scale magnetic flux ropes (SMFRs) observed frequently in the solar wind. There is some uncertainty about the origin and propagation of SMFRs. Additionally, existing methods for identifying them from spacecraft measurements have significant limitations due to incomplete physical models. We suggest the use of machine learning in order to gain understanding of how solar wind conditions relate to MCs and SMFRs, because in theory, machine learning models should be able to discover patterns from solar wind data. We carefully define several machine learning problems and apply feature ranking to the solar wind data from multiple spacecraft linked to known SMFRs and MCs. This gives us the relative significance of various solar wind properties with respect to MFRs from different perspectives. Then, we apply recursive feature elimination to determine which features are relevant and compare our results to known properties of SMFRs and MCs in the literature.