Uncovering Solar Wind Phenomena with iSAX, HDBSCAN, Human-in-the-loop and PSP Observations

TL;DR

High-cadence Parker Solar Probe data pose challenges for scalable solar wind analysis. The authors present a four-step pipeline combining iSAX symbolic compression, HDBSCAN clustering, and human-in-the-loop validation, anchored by the magnetic deflection angle $\theta_B$ as a unifying diagnostic. The method reproduces known SIRs and CMEs, uncovers new events, and remains robust across time scales from 1 hour to 3 days, yielding expert-validated catalogs. This scalable, interpretable framework advances space weather forecasting and is made reproducible via public code for PSP and future heliophysics missions.

Abstract

The solar wind is a dynamic plasma outflow that shapes heliospheric conditions and drives space weather. Identifying its large-scale phenomena is crucial, yet the increasing volume of high-cadence Parker Solar Probe (PSP) observations poses challenges for scalable, interpretable analysis. We present a pipeline combining symbolic compression, density-based clustering, and human-in-the-loop validation. Applied to 2018-2024 PSP data, it efficiently processes over 150 GB of magnetic and plasma measurements, recovering known structures, detecting uncatalogued CMEs and transient events, and demonstrating robustness across multiple time scales. A key outcome is the systematic use of the magnetic deflection angle ($θ_B$) as a unifying metric across solar wind phenomena. This framework provides a scalable, interpretable, expert-validated approach to solar wind analysis, producing expanded event catalogs and supporting improved space weather forecasting. The code and configuration files used in this study are publicly available to support reproducibility.

Figures (3)

• Figure 1: Representative summary of nine randomly selected clusters from one experiment, out of more than 150 total clusters. For each cluster, the mean time series is shown in blue, with the 5%–95% confidence intervals indicated in red shading. Cluster identifiers are indicated in the top-left corner, and the number of constituent windows (labeled as “segments” in the figure) is shown in the bottom-left corner. Across most clusters, variation around the mean is small, except for clusters 60 and 96, which exhibit slightly larger spread. Notably, the relationship between the mean and confidence intervals is clear, highlighting the characteristic temporal patterns captured within each cluster.
• Figure 2: PSP observations of two SIRs (A and B: Jan 12–13, 2021, and Jan 30–31, 2022) and two CMEs (C and D: Sept 17–18 and Nov 3–4, 2023). For the SIRs, shaded regions indicate slow wind (red), compressed interaction region (yellow), and trailing HSS (blue). The compression region shows increased $|B|$, $N_P$, and $T_P$, while the HSS is faster, lower-density, and exhibits a steadier and more coherent $\theta_B$. The CME panels highlight contrasting structures: the September event features a forward shock (yellow), compressed sheath (green), and magnetic ejecta (purple) with smooth $\theta_B$ rotation, characteristic of a flux rope. The November event has a weak shock, modest compression, a turbulent sheath, and a magnetic ejecta with coherent field rotations, reduced speed, lower density, and suppressed temperature; $\theta_B$ shows moderate but coherent deflection across the ejecta.
• Figure 3: Parker Solar Probe observations of uncatalogued CMEs and magnetic switchbacks. Panels A-B: CMEs on Jan 15–16, 2021 and Sep 24–25, 2020, showing ejecta with enhanced $|B|$, smooth rotations in $B_{r,t,n}$, declining $V_{SW}$, and suppressed $T_P$, followed by trailing compressions with elevated $N_P$ and variable $\theta_B$. Panels C-D: switchbacks on Aug 3, 2021 and Dec 23, 2023, with shaded regions marking transition (yellow) and peak (red) deflections in $\theta_B$. Sharp rotations occur in both cases ($\sim$90° in August, 60°–80° in December), while plasma parameters remain stable, confirming their Alfvénic nature Kasper_2019.