How it works
Two tools, one model: live odds for tonight and historical odds for your trip. Here is every step and every number, so you can judge it yourself.
What counts as a "good night"
At least one hour when all three are true at once:
- Dark: the sun is more than 12° below the horizon (nautical darkness).
- Mostly clear: no more than 3/8 of the sky covered by cloud.
- Active enough: the 3-hour Kp index is at least what that place needs.
The Kp each place needs
Auroras follow magnetic latitude, not geographic latitude, and the real magnetic field is lumpy: a simple dipole formula puts Reykjavík near 69° magnetic latitude, the real field at 64.0°. That is about two steps of Kp. We trace the Earth's magnetic field line from 110 km above each place with the IGRF model and use its apex latitude.
Then: Kp needed = (69.3° − magnetic latitude) ÷ 2.35. We fitted those two numbers so the model reproduces the Finnish Meteorological Institute's published aurora statistics, using 10 years of FMI ground observations of cloud and the real Kp record:
| FMI station | Magnetic lat. | Kp needed | Clear dark nights | Auroras: model | Auroras: FMI |
|---|---|---|---|---|---|
| Kilpisjärvi | 66.3° | 1.3 | 1176 | 74.7% | 3 nights in 4 |
| Utsjoki Kevo | 66.9° | 1.0 | 1282 | 76.9% | 3 nights in 4 |
| Sodankylä | 64.5° | 2.1 | 1167 | 52.2% | every second night |
| Helsinki Kaisaniemi | 56.9° | 5.3 | 1177 | 3.4% | about once a month |
Clouds: from satellites to what you see
For the 11-year history we use NASA POWER's hourly cloud amount, which exists for every place on Earth. It is an average over a cell about 50 km wide, so "50% cloud" can mean a clear sky over you and cloud next door. We matched it month by month to the hourly "total cloud cover" reported by five FMI weather stations: the NASA value below which the same share of dark hours counts as mostly clear.
| Month | Sep | Oct | Nov | Dec | Jan | Feb | Mar |
|---|---|---|---|---|---|---|---|
| NASA cloud below | 58% | 65% | 63% | 60% | 67% | 67% | 70% |
Hour by hour, NASA and the ground stations then agree on "mostly clear or not" 74–85% of the time.
Does it hold up?
Same nights, same Kp, two cloud sources: the stations' own observations (ground truth) and our NASA-based version. 4-night odds by month:
| Sep | Oct | Nov | Dec | Jan | Feb | Mar | |
|---|---|---|---|---|---|---|---|
| Kittilä (Levi): ground truth | 59% | 57% | 59% | 60% | 72% | 78% | 83% |
| Kittilä (Levi): our model | 53% | 51% | 52% | 52% | 70% | 74% | 77% |
| Kilpisjärvi: ground truth | 69% | 83% | 80% | 83% | 89% | 90% | 86% |
| Kilpisjärvi: our model | 67% | 79% | 83% | 80% | 83% | 88% | 79% |
| Sodankylä: ground truth | 60% | 53% | 54% | 60% | 66% | 70% | 75% |
| Sodankylä: our model | 51% | 44% | 49% | 56% | 66% | 69% | 69% |
Our numbers come out 4 points below the ground truth on average (from 9 below to 3 above), so treat them as slightly cautious. As an outside check, the tour operator Aurora Nights reports about 83% for a 4-night trip to Abisko from its own records; our model gives 85% for January and 84% for February.
Trip odds
For every place and every day of the year we look at all stays that started within ±7 days of that date in 2015–2025 and count the share with at least one good night. 2015–2025 covers one full solar cycle, from the quiet minimum around 2019 to the peak around 2024–2025.
Tonight's forecast
Every 10 minutes we fetch NOAA SWPC's 3-day Kp forecast and the live 1-minute Kp, and MET Norway's hourly cloud forecast for each place in turn (each place about once an hour). For each hour of tonight we multiply: the chance that the forecast Kp reaches the Kp needed (forecasts are uncertain, so this is a smooth curve, not a cut-off), the clear-sky fraction, how dark it is, and a small penalty for a bright moon high in the sky. The night's chance is the best hour.
Limits
- Kp is planetary and 3-hourly. Local substorms can light up a "Kp 2" night far more than expected, and a "Kp 5" forecast can fizzle.
- The cloud calibration uses Finnish stations. Coastal Norway, Iceland and North America may differ somewhat.
- Light pollution, mountains blocking the northern horizon and your own patience are not in the model.
- Near the magnetic pole (northern Greenland, Svalbard) the aurora often sits south of you, so "Kp needed" is less meaningful there.
Data and licences
- Kp history: GFZ German Research Centre for Geosciences, Potsdam (CC BY 4.0).
- Kp forecast and live Kp: NOAA Space Weather Prediction Center (public domain).
- Cloud forecast: MET Norway Locationforecast (CC BY 4.0).
- Cloud history: NASA POWER (MERRA-2), hourly CLOUD_AMT.
- Ground truth: Finnish Meteorological Institute open data and aurora statistics (CC BY 4.0).
- Magnetic field: IGRF (International Geomagnetic Reference Field), via ppigrf.
- Map: Natural Earth (public domain).