Where the numbers come from
Every forecast on this site comes from public data, NOAA's own space weather feeds, four open weather models, published astronomical formulas, and one thing that is ours: what people standing under this sky actually saw. Nothing we can't point to. Here is exactly what feeds the number on your screen, and how often it updates. For how those inputs are combined, see how this forecast works.
NOAA SWPC, Kp index
The planetary Kp index, the standard measure of geomagnetic activity, straight from NOAA's Space Weather Prediction Center: the 1-minute observed estimate for this moment, smoothed over the last quarter hour, and their 3-day forecast for the hours ahead. The page always says which of the two a number rests on. One detail most forecast sites skip: NOAA publishes Kp two ways, a floored integer that rounds 1.67 down to a flat 1, and the true fractional value. We use the fractional one.
NOAA SWPC, 3-day geomagnetic forecast
NOAA's daily storm probabilities and the sentence that explains them. Each day gets a chance that the planet reaches a G1, G2 or stronger storm at some point in 24 hours, plus a rationale in NOAA's own words: a coronal hole stream, a CME glancing blow. The outlook shows both, labelled for what they are. A 40% chance of a G1 storm is a whole-day, whole-planet figure, and it is the number that circulates as "40% chance of aurora". It is not your odds of seeing anything from where you stand; that is what the percentage on the page is for.
SOLAR-1, real-time solar wind
The magnetic field carried by the solar wind, measured about 1.5 million km from Earth at the L1 point by NOAA's SOLAR-1 spacecraft, the operational monitor since DSCOVR was retired in June 2026. Bz is its north-south tilt: when it turns south, Earth's field opens up and lets the wind in, and that is the earliest real-time sign a display is coming, often 15 to 60 minutes ahead. We average it over ten minutes, because one minute of solar wind is noise, and the sustained-south alert only fires once it has held.
Open-Meteo, cloud cover from four weather models
An aurora behind cloud isn't a sighting. For each location's exact coordinates we pull hourly cloud in three layers, low, mid and high, from four independent weather models, seven days out, and average them. Cloud is counted by what it blocks: thin high cloud only dims the aurora, a low overcast hides it, so the figure on the page is the aurora-effective one and can be lower than a weather app's total. When the four models disagree badly about the hour that matters, the page says so.
NOAA OVATION, the auroral oval
The measured extent of the auroral oval right now, from NOAA's OVATION model, drawn on the aurora map as a glowing field. It is a picture of where the aurora is, not an input to tonight's probability: we tested it against our own observations and it separated good nights from poor ones less well than the live activity and solar wind readings do.
IGRF-14, magnetic latitude
The twenty tracked destinations use verified magnetic latitudes. Any other point you search on the map gets an estimate from the current International Geomagnetic Reference Field, using the eccentric-dipole form and pulled toward the nearest verified locations so a search close to one of them agrees with it. Estimated values are labelled as estimates.
Ten years of history, for the month-by-month curves
Each location's best months and its typical clear-sky odds come from ten years of hourly cloud from the Open-Meteo archive, counted only while the sun is deep enough below the horizon for aurora, and from the GFZ Potsdam Kp record over the same seasons, which says how often activity reaches that location's threshold. These curves are climate, not tonight's forecast, and the pages keep the two apart.
Sun position, twilight and darkness
Aurora needs real darkness, and dark is not a switch. Sunrise, sunset and the sun's depth below the horizon are calculated directly in your browser from your coordinates and the date, using the standard solar-position equations. From civil twilight, six degrees down, to full night at eighteen, the forecast lets the number through gradually as the sky deepens. Civil twilight is an honest zero. This mostly matters for a few weeks each spring and late summer, when high-latitude locations lose and regain true night, and in midsummer the page says the sky never gets dark rather than printing a small number.
Moon phase and altitude
Moon phase, illumination, rise, set and altitude are calculated in your browser using Jean Meeus's astronomical algorithms, the standard reference method. A bright moon washes out a faint aurora, and how much depends on how high it sits as well as how full it is: the forecast weights moonlight by altitude using the extinction term of the Krisciunas and Schaefer sky-brightness model, so a full moon on the ridge costs far less than one overhead, and a moon below the horizon costs nothing.
Field observations, our own
Under every forecast there is a question, can you see it right now, with three answers: yes, nothing yet, and camera only. Each answer is stored with everything the forecast said at that moment, anonymously, and many come from guides who are out every night the tours run. The reports are compared with the forecast level by level and the numbers on the site are adjusted to match how often each level actually came true. When that adjustment is in effect the page says so under the headline. This is the one input nobody else has, and it grows every night.
If you want the deeper mechanics, Bz explained and the Kp index go further. And if you want to know who's actually running this, see about.