A Novel Reduction-Based Scheme for In-Situ Solar Wind Origin Classification using Machine Learning
Abstract
Determining how in-situ solar wind differs will aid in understanding the physical processes at the coronal features from which they originate. Here, a solar wind origin classification scheme is presented, employing statistical methods alongside a novel machine learning approach where dimensionality reduction algorithms (Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE)) are used in conjunction with one another, and clusters are determined using Density Based Clustering and Noise (DBSCAN). The input data is in the form of permutations of parameters from missions such as Advanced Composition Explorer (ACE), Ulysses, WIND and STEREO. Such data includes in-situ plasma parameters (speed, density, temperature, entropy, etc.), as well as elemental composition, and derived quantities such as cross helicity and residual energy. As a result, a multi-category wind classification scheme is constructed, and it is shown that in-situ solar wind can be characterized into distinct, physically meaningful groups using this reduction combination and clustering technique. With the cluster identities, a comparison is made to the source regions via a backmapping technique: ballistic mapping traces the samples to (a range of) source-surface point(s), and a Potential Field Source Surface (PFSS) model is used to locate the data to the photosphere. Uncertainties with this approach are addressed. With the resolved zones for the footpoints, remote sensing data is used to analyze the thermal conditions and electron density of the source location (via full disk images from Solar Dynamics Observatory (SDO) Atmospheric Imaging Assembly (AIA) and Extreme ultraviolet Imaging Telescope (EIT)). The outcome of this test for the agreement of each cluster to common properties of the source regions is then presented. Commonalities are found between the source region and the in-situ samples, and are then discussed. Sources and controls for instances of subjectivity are shown, as well as implications moving forward. The conclusion of this is that the customized approach can be further improved upon and adapted to availability of new data, alongside the scenarios to which the categorization scheme can be applied to.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMNG42A..08C