Speaker
Description
Accurate and timely forecasts of coronal mass ejection (CME) arrival times at Earth are are essential for operational space weather services aimed at mitigating impacts on spacecraft, ground-based infrastructure, and radiation-sensitive systems. We present two complementary machine-learning-based tools that address both CME arrival-time prediction and the identification of interplanetary CMEs (ICMEs) in near-Earth in-situ data, forming a coherent Sun-Earth operational forecasting chain.
PhysNetCME is a physics-driven machine-learning model that combines CME initial parameters derived from coronagraph observations with real-time upstream solar-wind measurements to estimate transit times to 1 AU. A drag-based propagation model is embedded within the neural-network training process, ensuring physical consistency while retaining adaptability to data. An ensemble approach provides deterministic forecasts together with uncertainty estimates, supporting risk-informed operational decision-making.
ICMEAlert complements this capability by continuously monitoring L1 solar-wind plasma and magnetic-field data to detect ICME signatures in near-real time. Using unsupervised anomaly detection based on the Isolation Forest algorithm, the system assigns confidence levels to anomalous structures without relying on labeled event catalogs.
Together, these tools enhance forecast lead time, reliability, and uncertainty awareness, contributing to the transition of advanced modelling approaches toward sustained operational space weather services.
| Numerical model | PhysNetCME-ICMEAlert |
|---|
