Tensor Solar Flare Classifier
Binary 24-hour forecast of M-class solar flares from the unified state tensor. LOOCV AUC 0.995.
What it does
Scores the probability of a >=C-class X-ray flare in the next 24 hours given the current tensor state. Trained on real NOAA SWPC GOES flux outcomes, not synthetic forcings.
Physics basis
Inputs include SWPC solar-wind parameters (IMF Bz, proton density, Kp), ACE L1 spacecraft data, cosmic-ray muon modulation, Schumann-resonance response to precipitating particles, ionospheric TEC, and SDO active-region tracking where available.
When it fires
Every 15 minutes. High-confidence predictions (prob >=0.40) are logged to the ledger as tensor_solar_flare_classifier.
What the customer receives
- Probability estimate every 15 min for >=M-class flare in next 24 h
- Optional >=C-class variant for high-cadence warning (lower threshold)
- Integrated Kp + aurora extent forecast
- SHA-256 signed ledger entries
Operational numbers (live)
Engagement paths
Per-query, subscription, territorial-exclusive, and royalty-on-find structures are all available. Specific commercial terms are scoped after a technical-fit conversation.
Honest caveats
- Label temporal correlation is high because a single active solar region produces many flares in sequence; effective independent sample count is lower than n=70.
- Base rate shifts with solar cycle phase -- an AUC measured during solar maximum does not transfer one-for-one to solar minimum.
- 24-hour lead time is shorter than the longest-lead-time flare classifiers (up to 72h) but the tensor ingests wider multi-physics context so skill is comparable at shorter horizons.
Michael Jessop — michael@deepmapai.com · partner portal