ARCANE -- Early Detection of Interplanetary Coronal Mass Ejections

Abstract

Interplanetary coronal mass ejections (ICMEs) are major drivers of space weather disturbances, posing risks to both technological infrastructure and human activities. Automatic detection of ICMEs in solar wind in situ data is essential for early warning systems. While several methods have been proposed to identify these structures in time series data, robust real-time detection remains a significant challenge. In this work, we present ARCANE - the first framework explicitly designed for early ICME detection in streaming solar wind data under realistic operational constraints, enabling event identification without requiring observation of the full structure. Our approach evaluates the strengths and limitations of detection models by comparing a machine learning-based method to a threshold-based baseline. The ResUNet++ model, previously validated on science data, significantly outperforms the baseline, particularly in detecting high-impact events, while retaining solid performance on lower-impact cases. Notably, we find that using real-time solar wind (RTSW) data instead of high-resolution science data leads to only minimal performance degradation. Despite the challenges of operational settings, our detection pipeline achieves an F1-Score of 0.37, with an average detection delay of 24.5% of the event's duration while processing only a minimal portion of the event data. As more data becomes available, the performance increases significantly. These results mark a substantial step forward in automated space weather monitoring and lay the groundwork for enhanced real-time forecasting capabilities.

Figures (10)

• Figure 1: This figure shows the real-time solar wind data as available via NOAA SWPC. The top panel shows the total magnetic field strength $|B|$, along with the vector components $B_X$, $B_Y$, and $B_Z$ in Geocentric Solar Magnetic coordinates. The remaining panels show from top to bottom: the bulk velocity $V$, the proton density $N_P$, proton temperature $T_P$, and plasma $\beta$.
• Figure 2: Spacecraft positions of Wind (blue), ACE (yellow), and DSCOVR (red) in Geocentric Solar Ecliptic coordinates from the time DSCOVR became the operational spacecraft (July 2016 onward), along with the position of the L1 point (black).
• Figure 3: Overview of missing data characteristics in the real-time solar wind (RTSW) data set. (a) Histogram of missing data percentage per interplanetary coronal mass ejection event for the RTSW data set. (b) Percentage of missing data depending on the chosen resolution for both the RTSW data set. The dashed vertical gray line indicates the resolution we opted for in this study (10 min).
• Figure 4: Schematic representation of the used model, adapted from rudisser_automatic_2022. A complete description of the used layers and components can be found in rudisser_automatic_2022. Additionally, we show a sample from the data set, containing an example input and output. The input consists of 8 variables. From top to bottom: Magnetic field $B$ plus components $B_X$, $B_Y$, $B_Z$, bulk velocity $V$, proton density $N_P$, proton temperature $T_P$ and plasma $\beta$. The output is a segmented time series that consists of $0$ (white) and $1$ (red), indicating whether a given time step corresponds to an interplanetary coronal mass ejection event.
• Figure 5: Schematic illustration of the postprocessing method used to extract 150 different time series of predictions at varying waiting times $\delta$. Each subplot represents a single prediction time step for a specific $\delta$ value. The red dashed rectangle indicates the sliding window of input data provided to the model. The yellow vertical arrow marks the current time step being classified as either CME or no CME. The horizontal strip below each plot labeled Prediction represents the full prediction time axis: segments before the prediction point correspond to previous predictions (red to indicate positive predictions), while the gray segment under and after the yellow arrow indicates time steps that have not yet been predicted. The neon green double-headed arrow labeled $\delta$ shows the gap between the end of the input window and the current prediction time step, defining the waiting time: how long the model “waits” before making a prediction about a point in time. By shifting the prediction point further from the input window (increasing $\delta$), we can assess how the model´s detection performance evolves with seeing more future data before making a prediction.