Real-Time 24-hour Ahead Forecasting of Solar Energetic Particle Events with SEPNET-PRISM-R

Abstract

Solar energetic particle (SEP) events, operationally defined as proton flux exceeding 10 pfu in the >10 MeV channel, threaten spacecraft systems, astronaut safety, and high latitude aviation, yet continuous real-time forecasting with adequate lead time remains difficult. We present SEPNET-PRISM-R, a near-real-time system that issues hourly, 24 hour ahead probabilistic SEP forecasts. The model integrates multimodal solar data - including active region patches, flare features, coronal mass ejection (CME) characteristics, and historical proton flux - to concurrently predict SEP occurrences, peak proton flux, soft X-ray flux, and CME velocity, with the primary objective being SEP forecasting. Building on existing literature, SEPNET-PRISM-R enhances operational performance by incorporating a postprocessing module that enables cross-validated decision making and interpretable quantification of feature importance. The training dataset spans from February 1986 to September 2025, with subsequent forecasts generated sequentially through the present via a real-time space weather prediction platform that provides predictive probabilities, feature attributions, and historical analogs, enabling more reliable and informed operational decision-making. The proposed framework substantially suppresses false alarm rates while maintaining robust event detection capabilities, as evidenced by case studies and aggregate statistics.