Prediction of Geoeffective CMEs Using SOHO Images and Deep Learning

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.

• The paper presents a novel ensemble deep learning model, GeoCME, to predict geoeffective CMEs using SOHO images.
• It leverages transfer learning with ResNet152 and InceptionResNetV2, achieving high MCC, TSS, and an 8.7% false positive rate.
• The study underscores GeoCME's operational potential for advancing space weather forecasting and protecting infrastructure.

Prediction of Geoeffective CMEs Using SOHO Images and Deep Learning

Introduction

The paper examines the prediction of geoeffective coronal mass ejections (CMEs) using solar images and a deep learning framework known as GeoCME. This model addresses the challenge of determining whether a CME, upon arrival at Earth, will cause a geomagnetic storm characterized by a Dst index less than -50 nT. The GeoCME framework is built on observations from the Solar and Heliospheric Observatory (SOHO) instruments, including LASCO C2, EIT, and MDI, utilizing ensemble and transfer learning methodologies to enhance prediction capabilities.

Data Collection

The study focuses on halo and partial halo CMEs from Solar Cycle 23. Key data sources include 136 CME events selected from the SOHO/LASCO CME catalog and supported by Dst index values from the RC list maintained by Richardson and Cane. This data selection is crucial for understanding the distribution and intensity of geomagnetic storms resulting from CME events.

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.

Methodology

Transfer Learning

The paper leverages transfer learning due to the limited dataset size, adapting pre-trained models such as ResNet152 and InceptionResNetV2 for geoeffective CME prediction. These models, originally trained on ImageNet, are repurposed to identify relevant solar features in CME images. Key elements include residual blocks in ResNet which facilitate gradient flow, enhancing network depth capabilities.

Figure 3: Illustration of a residual block (left) and an InceptionResNet module (right), demonstrating improved training efficiency and feature capturing at multiple scales.

Ensemble Model

GeoCME integrates ResNet152 and InceptionResNetV2 to extract features from SOHO images. The ensemble approach improves overall model performance by combining output through concatenation layers, convolutional blocks, and ensemble layers. This architecture design allows for precise feature extraction and decision making based on multi-source data from LASCO C2, EIT, and MDI.

Figure 4: Illustration of the GeoCME architecture showcasing the ensemble of pipelines dedicated to each SOHO instrument, culminating in an aggregated prediction output.

Results

GeoCME is tested with deterministic and probabilistic prediction models. For deterministic predictions, GeoCME achieved high MCC and TSS metrics, reinforcing the model's efficacy. The model demonstrated an acceptance that approximately 82% of CME events are geoeffective, with an 8.7% false positive rate, suggesting a sensitive model that does not miss geomagnetic storms.

Figure 5: The confusion matrix obtained by GeoCME used as a deterministic prediction model on the test set.

In probabilistic forecasting, metrics such as the Brier score and Brier skill score validate GeoCME, proving the model's relative advantage in capturing complex solar-terrestrial interactions.

Discussion and Conclusion

The paper concludes that the GeoCME framework successfully bridges machine learning techniques and solar imagery to predict geoeffective CMEs, offering both deterministic and probabilistic forecasting based on sophisticated model architectures. It highlights the importance of using ensemble learning and transfer learning to optimize predictions.

GeoCME's potential for operational usage is noted as significant due to its ability to use direct observations for prediction, avoiding complicated parameter calculations. Its real-world application could improve space weather forecasting, thereby safeguarding technological infrastructure against geomagnetic storm impacts. Future enhancements may expand dataset sizes or integrate additional data sources to further refine prediction accuracy and model robustness.