How our forecasts work

A forecast you can't audit is marketing. This page documents exactly how every verdict on this site is computed, from raw feed to final answer.

Data sources (all live, all public)

Solar wind, measured before it arrives: NOAA SWPC real-time feeds from spacecraft (currently IMAP, previously DSCOVR and ACE) parked at the L1 Lagrange point: a spot 930,000 miles sunward of Earth where solar and terrestrial gravity balance, so a satellite can hover there and sample the solar wind on its way past. We read the magnetic field (Bz, Bt) and plasma (speed, density) every minute, and compute Earth-arrival lead time as 1.5 million km ÷ wind speed, typically 30–70 minutes.
Kp index: SWPC's 1-minute estimated planetary K, plus the 3-day Kp forecast in 3-hour bins.
OVATION Prime: SWPC's 30-minute aurora probability grid; we read the cell over each location.
Hemispheric power: total gigawatts driving the northern aurora, from the SWPC nowcast.
Clouds & humidity: Open-Meteo hourly forecasts (low/mid/high cloud layers), refreshed every 30 minutes.
Sun, moon & darkness, computed locally with the astronomy-engine ephemeris library: twilight boundaries, moon phase, illumination, rise/set.

Magnetic latitude and your thresholds

Aurora visibility depends on magneticlatitude, computed here with a centered-dipole model using the IGRF-14 geomagnetic pole (80.7°N, 72.7°W, 2025 epoch), accurate to about a degree across the US. From the standard mapping between Kp and the auroral oval's equatorward boundary, we derive three thresholds per location:

Eye (horizon): the oval's boundary reaches within ~3° poleward of you. Aurora visible low on the northern horizon to dark-adapted eyes. (Geometrically, aurora at ~300 km altitude is visible from ~9° away, but atmospheric extinction and horizon glow make ~3° the practical limit for typical displays, a calibration that matches the standard Kp-city tables.)
Camera: within ~6°. A long exposure integrates sub-visual glow your eyes can't. This is the threshold most sites don't model, and most often the honest answer at mid-latitudes.
Overhead: the boundary reaches your magnetic latitude itself.

The nowcast boost

Kp is a 3-hour, after-the-fact average, and it lags what the sky is doing. When the upstream monitor shows strongly southward Bz (≤ −10 nT) with fast wind (≥ 500 km/s), conditions over the next 30–90 minutes will run hotter than the posted Kp. Our effective-Kp adds a physics-based boost of up to ~1.75 Kp in those conditions, and flags northward Bz (which suppresses aurora even at elevated Kp) as a downgrade. This is why our verdict can say "go outside in 40 minutes" while a Kp-only site still says quiet.

The verdict pipeline

Daylight gate: no verdict until the sun is 12° below the horizon; we tell you when that happens.
Cloud gate: above 85% cloud cover the sky is closed, and we say so (plus whether a drive would beat the overcast).
Geomagnetic grading: effective Kp vs your three thresholds produces the dual eye/camera verdict.
Moon adjustment: bright moonlight suppresses faint displays; verdicts and the composite score account for it.
Best window: we scan tonight's dark hours for the stretch with the clearest sky and highest forecast Kp.

Tonight's visibility odds

Each location page shows a per-night probability of seeing aurora, decomposed so you can audit it:

Activity odds: the highest forecast Kp across tonight's dark hours (with the near-term solar-wind boost applied to the next ~2 hours), passed through a logistic curve centered on your location's threshold. The curve's 0.7-Kp width reflects typical Kp forecast error, so a forecast sitting exactly at your threshold reads 50%, one full Kp above reads ~81%.
Sky odds: average sky openness across the clearest half of tonight's dark hours, because observers chase the clearest stretch rather than standing under the nightly average.
Moon factor: naked-eye odds are trimmed 12.5% under a half-lit moon and 25% under a bright one; camera odds are not, since long exposures punch through moonlight better than eyes do.

Eye odds = activity × sky × moon. Camera odds = activity (camera threshold) × sky. Activity is capped at 98% because the sun does not sign contracts. No dark hours in the next 24 h (high-latitude summer) means no odds are shown.

Live sighting reports

Every location page collects one-tap reports (I see it / looked, nothing) and shows the last two hours of them from within ~275 miles. This is the one thing no model can give you: whether aurora is on display right now, seen by human eyes near you. Read it as a second layer on top of the forecast. A cluster of "I see it" reports means go outside immediately, even if the hourly odds looked modest; a quiet board on a high-odds night usually means the substorm hasn't fired yet.

Reports are anonymous and each one is stored with a server-side snapshot of the conditions at that moment, so the picture can't be skewed by someone reporting from a different sky than the one they're standing under.

Where reports add the most

Sighting reports are most valuable exactly where models are weakest. Substorm onset is effectively unpredictable minute to minute, and satellite cloud estimates miss local gaps and low fog. A neighbor 40 miles away confirming a clear view north tells you something the forecast structurally cannot, which is why the reports sit at the top of every location page rather than buried at the bottom.

Known limits

Substorms can fire with only minutes of warning; no service can predict exact onset times.
OVATION under-forecasts extreme events and over-smooths sharp oval boundaries.
Cloud forecasts degrade past ~48 hours; treat day-3 windows as guidance, not promises.
Local light pollution varies block to block; our thresholds assume a reasonably dark viewing spot.

Questions or corrections? See why we built this.