Updated on October 5, 2026
The data
Our sky statistics come from NASA POWER, which serves hourly values for a given latitude and longitude. We use two parameters for the years 2001 to 2024: cloud amount, which comes from the CERES SYN1deg satellite product, and air temperature at 2 metres, which comes from the MERRA-2 reanalysis. We request one point per place, at the town's coordinates, in UTC.
Definitions
- Dark hour: an hour in which the sun is more than 10 degrees below the horizon at the middle of the hour. We compute the sun's altitude with a standard low-precision formula.
- Night: the period from local solar noon to the next local solar noon. A night belongs to the month in which it starts.
- Usable night: a night with at least 2 consecutive dark hours in which cloud amount was 50% or less.
- Stay figures: for each month we take every possible stay of N nights that starts in that month, across all 24 years, and count the share with at least one usable night. Because these are real sequences of days, runs of bad weather are included. We do not multiply single-night odds together, which would assume each night is independent of the last and would overstate the result.
- Dark hours per night: counted for the 15th of the month in 5-minute steps.
- Temperature: the mean over dark hours only.
A night with fewer than 2 dark hours is not counted. A month is left blank when it has fewer than 2 dark hours per night at mid-month, or when 90% or fewer of its nights can be counted. That is why April is empty for eight of the twelve places, and filled only for Reykjavík, Fairbanks, Yellowknife and Whitehorse. We run the same calculation for Skibotn, an inland valley about 1 h 45 by road from Tromsø, to test what a chase inland can change.
The limits, stated bluntly
- It is an area average. Cloud amount is the mean over a cell of 1 by 1 degree, about 110 km from north to south. Local effects are smoothed away. If a valley sits in a rain shadow, as is claimed for Abisko, this dataset cannot show it properly.
- Satellites find cloud harder to detect over snow in the dark, and that is every place here in winter.
- It is sky only. Nothing here measures whether the aurora was active. A usable night is a night when you could have seen aurora if there was any.
- The 50% threshold is a judgement call. We chose it because, on a cell average, it matches a stricter test at a single point.
Cross-check
We ran the same test on ERA5 reanalysis for 2010 to 2024, with a 40% threshold at a point. This check covers only the first six places on the site: Tromsø, Reykjavík, Rovaniemi, Abisko, Fairbanks and Yellowknife. The six places added in October 2026 have not been checked this way. It gives the same broad order: Fairbanks and Yellowknife far ahead of the Nordic places. It puts Abisko lower and Tromsø a little higher than our figures. Read small differences between the Nordic places with that in mind.
Geomagnetic latitude and Kp
The geomagnetic latitudes and the Kp values at which aurora is typically overhead are approximate. We found no official per-city table. Corrected geomagnetic latitudes are published for magnetometer stations, not for towns, so we now take them from station tables: the IMAGE network table hosted by the Finnish Meteorological Institute (calculated for 2001) and the observatory pages of the International Service of Geomagnetic Indices (calculated for 2020.5).
Kiruna, Tromsø and Abisko have a station in or beside the town, and Reykjavík has an observatory nearby. For Levi, Saariselkä, Akureyri, Rovaniemi and Luleå we estimated the value from the nearest stations. Whitehorse rests on a station about 30 km north of the town, and we saw its value only in a search result. For Fairbanks and Yellowknife we still use a third-party site, auroraforecast.me, which does not name its model or year. The years differ, so read every latitude as give or take half a degree.
The Kp values are our own derivation, from a published table of where the edge of the auroral oval sits at each Kp level. Where a place sits near a boundary between two levels we took the more cautious one. They are not official figures. See the Kp index, explained.
Tour data
Prices, scores, review counts, durations, group sizes, cancellation windows and no-aurora wording come from the English Civitatis catalogue as it stood on 5 October 2026, when it held 34 listings for our twelve places. Prices are "from" prices per person and they change. Our reading of each policy is explained in what happens if the aurora does not show.
The Tonight tool
The Tonight page combines two live feeds: the planetary Kp forecast from NOAA's Space Weather Prediction Center and the hourly cloud forecast from MET Norway's Locationforecast service. Our server fetches both and keeps the result for about 30 minutes.
For each place we take the next unbroken run of dark hours, using the same 10-degree rule, and apply these rules:
- Sky: we find the two consecutive hours with the lowest average cloud. Up to 30% is clear, up to 60% is partly cloudy, above that is overcast.
- Activity: the highest forecast Kp during the night is compared with the Kp that place typically needs. At or above it is enough, one or more above is strong, one below is low, and anything less is quiet.
- Verdict: overcast means clouds, whatever the Kp. A quiet forecast means quiet. A clear sky with enough activity is a go. Everything else is a maybe.
The tool is a quick filter. Check the official pages linked from each card before you go out.
Reporting an error
Write to us through the contact page with the page, the figure and a source we can check. We correct confirmed errors and change the date on the page.