Solar Energetic Particle Prediction in the Inner Heliosphere Using Deep Learning and PSP/IS⊙IS Data

Solar events, such as coronal mass ejections and solar flares, accelerate large numbers of energetic charged particles, producing solar energetic particle (SEP) events that can harm astronauts, damage satellites, and potentially damage infrastructure on Earth. Previous machine learning (ML) SEP prediction models aimed to predict geoeffective events using data from Earth-orbiting or Sun-Earth Lagrange 1 (L1) point missions. However, predicting SEP events throughout the heliosphere is essential for supporting missions to Mars and other deep-space destinations. This study uses EUV images from the Atmospheric Imaging Assembly (AIA) onboard the Solar Dynamics Observatory (SDO) to predict particle intensity values observed by the Integrated Science Investigation of the Sun (IS (Formula presented.) IS) onboard the Parker Solar Probe (PSP), over various points in the ecliptic plane between Sun and Earth, using deep learning. The ML models extract features from SDO/AIA (Formula presented.) images through a series of convolutional layers, combine them with PSP's trajectory, and process the fused representation using transformer and dense layers. The output is a regression estimate of the energy-weighted average particle intensity (Formula presented.), which is subsequently thresholded to yield a binary classification indicating whether (Formula presented.). The best-performing model achieves a probability of detection of 0.76, true skill score (TSS) of 0.47, F1 score of 0.47, false alarm rate (FAR) of 0.65 and precision of 0.35. These results are comparable to ML models that predict geoeffective events, opening avenues toward ML-aided predictions in the inner-heliosphere.