SEPNET

SEPNET is a multi-task deep learning model that predicts the probability of a solar energetic particle (SEP) event exceeding 10 pfu (≥10 MeV protons) at Earth in the next 24 hours. It uses as input (1) SHARP parameters from SDO/HMI magnetograms (hmi.sharp_720s_nrt) and (2) flare information from the SWPC flare list (LMSAL/HEK). The model uses flare and SHARP features from the preceding 24 hours; it is not triggered by individual CME or flare events. Outputs include a probability value, symmetric uncertainty (from the spread of estimated probabilities over the input window), and an all-clear flag.

• Forecasts depend on the availability and timeliness of SHARP and flare catalog data. • Model is trained on historical data; performance may vary during unusual solar activity. • Probability refers to the chance of exceeding 10 pfu; no explicit peak intensity or time-of-peak is given

• Magnetogram: SDO/HMI SHARP near-real-time product (hmi.sharp_720s_nrt); last data time as available at forecast issue. • Flare catalog: SWPC flare list from LMSAL/HEK (e.g. GOES/XRS); all flares in the 24 hours preceding the issue time are used. Optional NOAA active region numbers when available.

• Probability of SEP > 10 pfu (> 10 MeV protons) at Earth in the next 24 hours (mean of model-estimated probability over the preceding 24 h). • Symmetric uncertainty (standard deviation of estimated probability over that window). All-clear flag (based on probability vs. a configurable threshold). Prediction window: issue time to issue time + 24 hours.

Initial release. Model outputs probability of SEP >10 pfu in the next 24 hours using SHARP and flare catalog inputs.