AuroraHow accurate are aurora forecasts?
A good aurora forecast for the same evening is useful, but never certain. We tested ours against automated all-sky cameras in Tromsø, Skibotn, Abisko and Kiruna over three winters (2023–2026). On nights we rated Excellent, the cameras recorded clearly visible aurora 79% of the time. On nights rated Low, 15%.
Most aurora forecasts are not checked like this. They show the Kp index, a number for the whole planet, rather than the chance of seeing something where you stand.
How did we measure forecast accuracy?
- Ground truth: the UEC Tromsø AI dataset, which classifies every image from all-sky cameras in Tromsø, Skibotn, Abisko and Kiruna (Nanjo et al. 2022). A night counts when clearly visible aurora appears in at least one 15-minute slot. Clouds count as they fell, as they would for you.
- Forecast: the rating Aurora showed in the afternoon before each night, from a model that never saw that night's data.
- Moonlight: under a bright moon, the camera classifier files some aurora as bright sky. We checked moonlit frames by eye and count those detections as about half (see the full-moon post).
| Rating shown in the afternoon | Nights with clearly visible aurora |
|---|---|
| Excellent | 79% |
| Good | 63% |
| Possible | 42% |
| Low | 15% |
The rates move from winter to winter: a quiet winter reads too high, an active one too low. Pooled over three winters, the ratings are honest. Sources and definitions are on our methodology page.
Are Kp forecasts accurate?
The Kp index tells you how disturbed the Earth's magnetic field is, on a scale from 0 to 9, for the whole planet in 3-hour blocks. It is useful for knowing how far south aurora might reach. It does not tell you about clouds, darkness or the moon where you are.
In our tests, for the same number of aurora nights caught, our forecast gave 61% fewer false-alarm nights than a forecast built on Kp plus cloud cover, and 80% fewer than Kp alone. On its own, Kp could not tell good nights from bad in northern Scandinavia at all.
How far ahead can you trust an aurora forecast?
- Tonight: the most reliable. By the afternoon, the solar wind measured upstream of Earth and the evening's cloud forecast are both fairly well known.
- Tomorrow and the night after: still useful. In our tests, nights rated Excellent one and two days ahead had clearly visible aurora 76% and 72% of the time.
- Three days or more: we found no usable skill, so Aurora doesn't rate those nights. Clouds and solar wind can't be forecast that far ahead for a specific night.
- The exact time: can't be forecast hours ahead. Aurora comes in bursts, so plan to be out in the darkest, clearest hours and watch.
Why camera checks matter
A camera doesn't care what the forecast said. Checking against one is the only way to know whether "Excellent" means anything. It is also why our numbers come with limits:
- they are tested in northern Norway and Sweden only. Elsewhere the app shows lower confidence, and we make no accuracy claim;
- 2023–26 were winters near solar maximum, and activity declines from about 2027;
- the cameras see from fixed spots, and you may stand under a different patch of sky.
How should you use an aurora forecast?
- Use a local forecast that includes clouds and darkness, not Kp alone.
- Plan your trip on history (best time to see the northern lights), and your evening on the forecast.
- On a Good or Excellent night, go out in the darkest hours and give it time.
A forecast, not a guarantee. A good forecast is a reason to look up.
