Forecasting solar energetic particles using multi-source data from solar flares, CMEs, and radio bursts with machine...

Area of Science:

• Space Physics
• Heliophysics
• Astrophysics

Background:

• Solar energetic particle (SEP) events pose risks to space technology and astronauts.
• Predicting SEP events is difficult due to the imbalanced nature of the data.
• Existing prediction methods require consistent and reliable approaches.

Purpose of the Study:

• To develop and evaluate a consistent machine learning (ML) framework for predicting SEP events.
• To compare the performance of different ML algorithms in SEP prediction.
• To identify key solar activity indicators crucial for accurate SEP event forecasting.

Main Methods:

• Applied machine learning algorithms: Random Forests (RF), Decision Trees (dtree), and Support Vector Machines (SVM) with linear and nonlinear kernels.
• Utilized diverse datasets including solar flares, coronal mass ejections (CMEs), and radio bursts (sweep and fixed frequency).
• Evaluated model performance using standard metrics: Probability of Detection (POD), False Alarm Rate (FAR), True Skill Statistic (TSS), and Heidke Skill Score (HSS).

Main Results:

• The Random Forests (RF) model demonstrated consistent superior performance across all tested datasets compared to other algorithms.
• RF achieved high skill scores (POD, TSS, HSS) and low false alarm rates (FAR) for both sweep and fixed-frequency datasets.
• Key features identified for SEP prediction include CME linear speed, angular width, flare intensity, soft X-ray (SXR) flux, and radio burst characteristics.

Conclusions:

• Machine learning, particularly RF, offers a robust and effective approach to the imbalanced problem of SEP event prediction.
• CME characteristics, flare intensity, and radio burst properties are critical predictors for SEP events.
• The developed method provides a consistent framework for improving space weather forecasting and mitigating SEP event risks.