Prediction of Geoeffective CMEs Using SOHO Images and Deep Learning

TL;DR

This work tackles the challenge of predicting whether a CME arriving at Earth will trigger a geomagnetic storm by introducing GeoCME, a deep-learning framework that ingests SOHO images from LASCO C2, EIT 195 Å, and MDI magnetograms. The method uses transfer learning with two pre-trained backbones (ResNet152 and InceptionResNetV2) in an ensemble architecture to extract and fuse features across instruments, producing a probabilistic geoeffectiveness forecast that can be thresholded for a deterministic decision. Key results show strong performance: $MCC=0.807$ and $TSS=0.714$ for deterministic predictions, and $BS=0.094$ with $BSS=0.493$ for probabilistic forecasts, with the best results achieved when all three data sources are used together. The study demonstrates the feasibility of image-based, operationally relevant geoeffectiveness prediction, and highlights the value of ensemble and transfer-learning approaches in extracting hidden signals from solar imagery to better understand CME–solar-terrestrial coupling.

Abstract

The application of machine learning to the study of coronal mass ejections (CMEs) and their impacts on Earth has seen significant growth recently. Understanding and forecasting CME geoeffectiveness is crucial for protecting infrastructure in space and ensuring the resilience of technological systems on Earth. Here we present GeoCME, a deep-learning framework designed to predict, deterministically or probabilistically, whether a CME event that arrives at Earth will cause a geomagnetic storm. A geomagnetic storm is defined as a disturbance of the Earth's magnetosphere during which the minimum Dst index value is less than -50 nT. GeoCME is trained on observations from the instruments including LASCO C2, EIT and MDI on board the Solar and Heliospheric Observatory (SOHO), focusing on a dataset that includes 136 halo/partial halo CMEs in Solar Cycle 23. Using ensemble and transfer learning techniques, GeoCME is capable of extracting features hidden in the SOHO observations and making predictions based on the learned features. Our experimental results demonstrate the good performance of GeoCME, achieving a Matthew's correlation coefficient of 0.807 and a true skill statistics score of 0.714 when the tool is used as a deterministic prediction model. When the tool is used as a probabilistic forecasting model, it achieves a Brier score of 0.094 and a Brier skill score of 0.493. These results are promising, showing that the proposed GeoCME can help enhance our understanding of CME-triggered solar-terrestrial interactions.

Figures (13)

• Figure 1: Chart showing the total counts of halo/partial halo CMEs among all CMEs during Solar Cycle 23 (1996-2008) according to the SOHO/LASCO CME catalog.
• Figure 2: Distribution of the Dst index values caused by the 136 halo/partial halo CME events in our dataset.
• Figure 3: Breakdown analysis of the geoeffective and non-geoeffective CME events in our dataset where a solid circle represents a geoeffective CME event and a cross mark represents a non-geoeffective CME event. These events were distributed over 10 years, from 1997 to 2006.
• Figure 4: SOHO observations on the CME event that occurred at 08:06:00 UT on 17 September 2002. Shown from left to right are a LASCO C2 image, an EIT 195 Å image, and a full-disk MDI magnetogram.
• Figure 5: Illustration of a residual block (left) and an InceptionResNet module (right). The residual block consists of two $3\times3$ convolutional layers followed by a residual connection that allows gradients to flow directly through the network, improving training efficiency. The InceptionResNet module includes parallel convolutional paths with $1\times1$ and $3\times3$ filters, which, combined with a residual connection, capture image features at multiple scales to maintain efficient gradient flow.