Towards a Fully Automated Pipeline for Short-Term Forecasting of In Situ Coronal Mass Ejection Magnetic Field Structure

TL;DR

The study demonstrates a fully automated pipeline that links real-time CME detection and remote-sensing information with short-term in situ forecasting of CME magnetic structure at L1. By integrating ELEvo for arrival-time predictions, ARCANE for autonomous MO detection, and 3DCORE for iterative flux-rope reconstruction via ABC-SMC, the pipeline achieves early, data-driven forecasts for a subset of events. Across 3870 DONKI entries (2013–2025), 61 cases with ARCANE detections and ICMECAT ground truth yield forecasts based on early MO data with typical errors around $5$ hours in timing and $\sim10$ nT in field strength, underscoring both the potential and the limitations of simple flux-rope representations in real-time contexts. The results emphasize that forecast quality is constrained by catalog uncertainties, event complexity, and model fidelity, and point to future work on more flexible CME morphologies, upstream monitors, adaptive fitting strategies, and direct coupling to geomagnetic impact metrics.

Abstract

We present an automated pipeline for operational short-term forecasting of coronal mass ejection (CME) magnetic field structure at L1, coupling arrival time prediction, in situ detection, and iterative flux rope reconstruction, following near-real-time remote-sensing CME identification. The system is triggered by new entries in the CCMC DONKI database and first applies the drag-based ELEvo model to determine whether an Earth impact is expected and estimate arrival time. This estimate defines a temporal window constraining the search for CME signatures in real-time L1 in situ solar wind data, where the magnetic obstacle (MO) is automatically detected using the deep learning model ARCANE. Upon MO onset, iterative reconstructions with the semi-empirical flux rope model 3DCORE are performed, using a Monte Carlo fitting scheme, producing continuously updated forecasts of the remaining magnetic field profile. We evaluate the pipeline using 3870 archived DONKI entries and archived NOAA real-time in situ data from 2013-2025, assessing forecast performance at different stages of MO observation. For 61 events with an associated ground-truth counterpart in the ICMECAT catalog, forecasts based on initial MO data already achieve performance comparable to full-event reconstructions. Typical errors are ~5 hours in timing of magnetic field extrema and ~10 nT in field strength metrics, with limited systematic improvement as more of the event is observed. Results show substantial event variability and systematic underestimation of extrema, indicating deviations from ideal flux rope assumptions. These findings demonstrate the potential of fully autonomous real-time forecasting while highlighting limitations imposed by event complexity and model representational capacity.

Figures (12)

• Figure 1: Parameter distributions used in the ELEvo simulations. (a) Histogram of $\gamma$ values based on vrsnak_2013_propagationinterplanetarycoronal, where each bar corresponds to the indicated $\gamma$ interval (e.g. 0.0-0.2, 0.2-0.4, etc.) and the bar height denotes the percentage of events in that interval. (b) Histogram of background solar wind speed ($V_\mathrm{sw}$ ) values from the same source, with the bars representing the labeled speed intervals. (c) Skew-normal fit to the $\gamma$ distribution shown in (a), from which simulation values were sampled. (d) Skew-normal fit to the $V_\mathrm{sw}$ distribution in b), used for sampling $V_\mathrm{sw}$ values in the simulations.
• Figure 2: Example visualization of an ELEvo simulation on April 23, 2023 15:00 UT. The yellow dot in the center represents the Sun, and the green dot indicates Earth's position. Each blue semi-ellipse corresponds to a CME, with the solid blue line showing the ensemble mean and the shaded blue area representing $\pm2$ standard deviations across the ensemble members.
• Figure 3: Distribution of arrival time prediction errors from ELEvo simulations, compared against observed IPS arrival times listed in the DONKI catalog.
• Figure 4: Overview of detection performance as a function of the waiting time $\delta$. (a) F1 Score and mean relative delay (expressed as a percentage of the event duration) as a function of $\delta$ for the sheath region. (b) F1 Score and mean relative delay (in percent of duration) as a function of $\delta$ for the magnetic obstacle (MO).
• Figure 5: Example 3DCORE reconstruction of the event launched on April 21, 2023, observed by Wind. (a) In situ magnetic field measurements (solid lines) together with the corresponding 3DCORE fitting results. The shaded region indicates the $2 \sigma$ spread of the ensemble, and the dashed colored lines represent the ensemble mean. The vertical black dashed lines mark the start and end times of the MO, which define the reconstruction interval. The vertical gray dashed lines denote the fitting points used to evaluate the quality of the fit throughout the iterative optimization. Magnetic field components are shown in geocentric solar magnetic (GSM) coordinate system. (b) Top down view on the solar equatorial plane showing the 3D mean 3DCORE solution at the time of the observation at Wind (2023 April 23 15:06 UT), 10 hours before the MO arrived at Wind. The orange solid lines within the 3D shape represent three specific field lines of the CME.