Some Solar Storms Kill, and We Can't Tell Which

Solar energetic particle (SEP) events — bursts of high-energy protons and heavier ions accelerated by solar flares and coronal mass ejections (CMEs) — are among the most dangerous radiation hazards in space. A severe SEP event can deliver lethal radiation doses to unshielded astronauts within hours and cause satellite electronics failures, yet we cannot predict which solar eruptions will produce dangerous particle events. The Sun generates thousands of flares and hundreds of CMEs per solar cycle, but only a small fraction (~1-5%) produce significant SEP events at Earth. Current models cannot reliably distinguish the eruptions that will accelerate particles to dangerous energies from those that will not, because the acceleration mechanisms depend on CME speed, shock geometry, seed particle populations, and interplanetary magnetic field structure — parameters that are poorly measured and interact nonlinearly. Warning time is typically 15-60 minutes after SEP onset is detected at L1 (the ACE/DSCOVR spacecraft location, 1.5 million km sunward of Earth), far too short for astronaut sheltering or satellite safing during EVA or in deep space.

NASA's Artemis program plans crewed lunar surface operations in the late 2020s, and eventual crewed Mars missions face months of exposure to the interplanetary radiation environment. The October 2003 "Halloween storms" delivered doses that would have exceeded astronaut career limits for unshielded personnel; the September 2017 SEP event occurred during Hurricane Irma recovery, complicating response. For satellites, the March 2012 event caused anomalies on multiple spacecraft. SEP events also trigger ground-level enhancements that disrupt high-frequency radio communication and increase radiation doses on polar airline routes. The Heliophysics decadal survey identified "Predict extreme space weather events" as a highest-priority science goal, explicitly linking it to human exploration and national security needs.

Empirical models (PROTONS by NOAA SWPC, SEPMOD, SPARX) use statistical relationships between solar flare X-ray intensity, CME speed, and SEP occurrence probability, but their false positive rates are high (~50-70%) and they miss events where the flare is modest but the CME shock geometry is favorable for particle acceleration. Physics-based models (iPATH, EPREM, ENLIL+SEPMOD) simulate particle acceleration at CME-driven shocks and transport through the heliosphere, but require input parameters (CME mass, speed, angular width, initial magnetic field configuration) that are uncertain by factors of 2-5 from coronagraph observations. The fundamental limitation is observational: we have no direct measurement of the coronal magnetic field where acceleration occurs (only photospheric magnetograms that must be extrapolated upward), no measurement of the seed particle population that gets accelerated, and single-point in-situ measurement of the interplanetary magnetic field (at L1) that cannot capture the 3D structure the particles propagate through. Machine learning approaches trained on historical SEP catalogs show marginal improvement over empirical models because the training set contains only ~100 well-characterized major events — too few for pattern recognition in the high-dimensional parameter space.

Routine measurement of the coronal magnetic field from the photosphere through the corona (~1-10 solar radii), using next-generation coronagraphs, radio interferometry, or spectropolarimetry. A constellation of in-situ monitors distributed around the Sun (not just at L1) to characterize the seed particle population and interplanetary magnetic field structure — concepts like a "Heliophysics Sentinel" or Solar Ring mission. Physics-informed machine learning that combines partial physics knowledge (shock acceleration theory, particle transport equations) with data-driven pattern recognition, constraining the model to respect known physics while learning unmodeled correlations. Improved CME characterization from multiple vantage points (Solar Orbiter, STEREO-A, future missions) that reduce the uncertainty in CME speed and direction.

A student team could build a machine learning SEP prediction model using publicly available data (GOES proton flux, SOHO/LASCO CME catalogs, SDO/HMI magnetograms) and benchmark it against the NOAA SWPC operational forecasts, testing whether additional features (CME source region magnetic complexity, preceding activity) improve prediction skill. Alternatively, a team could develop a simplified particle transport simulation (1D focused transport equation) and explore how uncertainty in CME shock parameters propagates to uncertainty in predicted SEP onset time and intensity at Earth. Relevant disciplines: space physics, machine learning, signal processing, aerospace engineering.