You Will Probably See Grey: What the Northern Lights Look Like to the Naked Eye

Human vision runs on two systems that hand over to each other. Above a luminance of about 0.03 candela per square metre the cones are working and you see in colour. Below it you are on rods, which are far more sensitive and cannot encode colour at all. The classic demonstration is exactly the one you will run in a field in Lapland: take a coloured test spot down below that level and the colour drains out of it while the spot itself stays perfectly visible. Between the two lies the mesopic range, where the handover happens gradually and colour is present but weak.

An ordinary aurora sits on the wrong side of that line, or straddles it. The light is there. Your rods pick it up without difficulty, which is why you can see the band at all, trace its shape, and watch it move. What you cannot do is put a colour on it, because the machinery that does colour has already gone off duty.

There is a small cruelty in the arrangement. The aurora's dominant emission is the oxygen green line at 557.7 nanometres, and human sensitivity peaks at around 555, so the sky is glowing at almost exactly the wavelength we are best built to notice. We are still bad at it in the dark. The colours page in the guide goes through where each of those emissions comes from and what altitude it implies.

The aurora is not only an astronomical object. It is also a meteorological one, and the World Meteorological Organization classifies it among the electrometeors in the International Cloud Atlas, alongside lightning and St Elmo's fire. The section on brightness is short, and the part of it that matters to a traveller is one sentence:

The luminance or brightness of polar aurorae is very variable. It may be faint and barely visible, or it may be comparable with that of clouds illuminated by the full Moon, and, occasionally, it may be much greater.

Occasionally. That word is doing a great deal of work, and it appears in a text that carries the legal status of standard practices and procedures under Annex I to the WMO technical regulations. It is not a blogger being pessimistic about your holiday.

The Atlas then gives four steps. Observers know them as the International Brightness Coefficient, and the Australian Bureau of Meteorology's space weather service publishes the same four with numbers attached, in kilorayleighs. The numbers are the interesting part, because they are logarithmic. Every step up is ten times the light of the step below.

Here is where we have to stop short of the number you actually want. Nobody publishes a distribution of these four steps over a season at a given place, and we do not have one either. Making one up would put us in the same position as the operators quoting a 90% success rate with no denominator, which is the practice this site exists to complain about.

What can be said is bounded and checkable. The standard itself treats the bright end as occasional. The scale is logarithmic, so the difference between what you are likely to get and what you were shown is a factor of ten or a hundred in light, not a good night versus a slightly better one. And the physiological threshold does not negotiate: below roughly 0.03 cd/m2 there is no colour available to you at any level of enthusiasm.

Brightness follows activity, loosely. The planetary Kp index is the usual proxy, and the US Space Weather Prediction Center's viewing guidance is unusually blunt about what each band looks like: at Kp 0 to 2 the aurora is far north and "quite dim in intensity", at Kp 3 to 5 it moves equatorward, brightens and starts to move around.

We calculated how often that happens from the GFZ Potsdam Kp and ap series, which runs from 1932 and is open data. Counting three hour intervals that reach Kp 4 or higher, by month of the aurora season:

Read the first column as odds. In September, the best month on this measure, roughly one three hour interval in six reaches Kp 4. In December, which is peak season for booking, it is closer to one in ten. Most of the time the planetary index is sitting at 1, 2 or 3, and the display that goes with it is the faint one.

Two caveats belong immediately next to that, and the second one is the more useful of the two. Kp is a three hour average, while a substorm can flare and fade inside twenty minutes, so a quiet index does not rule out a short and vivid show. And Kp is planetary: it is standardised across thirteen mid latitude observatories, and if you are standing underneath the oval it matters far less than people assume. SWPC says so itself, that a person in the right place can see very nice displays at Kp 3 or 4. Our page on the Kp index covers why chasing that number from a hotel room is mostly a way of ruining an evening.

Colour and activity are the second and third questions. The first is whether there is a hole in the cloud, and that is the variable nobody can influence and almost nobody quantifies. We count a night as usable when cloud cover sat at or below 50% for at least two hours between 18:00 and 02:00, measured across eleven seasons of ERA5 reanalysis. The thresholds are ours and we publish them so you can argue with them.

The result for a short trip is sobering, and it also refuses to behave the way most people's arithmetic assumes. Cloudy nights arrive in blocks. In Tromso, after a clear night the chance of another clear night is 44.3%, and after a cloudy one it is 19.1%. Treating three nights as three independent coin flips overstates your odds by seven points there, and more elsewhere.

Three nights in Tromso is a coin flip on the cloud alone, before anybody asks whether there was an aurora behind it. Reaching 90% there would take eleven nights, which no realistic trip buys. Kiruna, ninety kilometres from Abisko and rarely marketed as an aurora town, gets to 71.2% in the same three nights. The method, and the places where it breaks down, are set out on the when the sky opens page.

The practical problem with a faint display is not that it is disappointing. It is that people miss it entirely, standing underneath it, because it looks like the weather. Four checks settle it in about a minute.

• Stars. You can see stars through an aurora. You cannot see stars through cloud. This is the fastest test there is and it almost never fails.
• Drift. Cloud moves with the wind, in one direction, at the speed the rest of the sky is moving. An aurora ignores the wind. A quiet arc can hang in the same place for an hour, then brighten and change shape without going anywhere.
• The lower edge. A quiet auroral arc has a defined bottom edge and a top that fades out. Cloud rarely does that, and never that cleanly.
• The phone. Three to five seconds on a tripod or braced against a fence post. Green means aurora, grey means cloud, and the test takes less time than arguing about it.

Everything above is an argument about expectations, not an argument against the trip. Four things follow from it, and all four change what you actually see.

The first is time. Dark adaptation is neither instant nor linear. Cone sensitivity improves over the first few minutes, then at around five to eight minutes the rod branch takes over and the threshold falls again, bottoming out near 10 to the minus 5 candela per square metre after about forty minutes in the dark. That last stretch is the difference between noticing a faint arc and standing under one without seeing it. A phone screen at full brightness undoes a good part of it in a second, which is the honest reason a guide asks people to put them away.

The second is where you point your eyes. Rod density is lowest at the very centre of the retina, so the faintest things register better slightly off to one side. Astronomers call it averted vision, and it works on aurora.

Third, the phone is an instrument as well as a souvenir. A five second exposure will tell you whether the grey band overhead is worth waiting under before it does anything, which is worth knowing when it is 20 below and the bus leaves at one.

The fourth is about what you compare the night against. A quiet arc on a clear night, watched for half an hour with adapted eyes, is a better hour than a bright display seen through thin cloud. Neither of them looks like the listing photograph, and only one of those facts is anybody's fault.

Those three photographs were all made the same way, by holding a shutter open for several seconds. Two of them show a sky that a person standing there would have described in shades of grey. That is not an accusation against the photographer. It is the reason this article exists.

September is a hard month at these latitudes and we have gone through why in some detail: the sky is at its clearest and there is almost no darkness to spend it in, 1.8 hours a night on average in Tromso against 8.0 in December. Two dates in the next few weeks are exact rather than statistical, which makes them the only part of the plan that carries no forecasting risk at all. New moon falls on 11 September 2026 and full moon on 26 September, so the dark half of the month is the fortnight starting now. The autumn equinox is at 00:09 UTC on 23 September, and equinox months carry more geomagnetic activity than solstice months by a factor of 1.41 on mean monthly ap, or 1.71 measured on the frequency of intervals reaching Kp 4. Those are two different metrics and we have quoted both, because an article that publishes one without saying which has published a number nobody can check.

• We have no measured distribution of brightness steps over a season, and neither, as far as we can find, does anyone else. "Probably grey" rests on the standard's own wording, on the ten times spacing between steps, and on where the colour threshold sits. It is not a frequency we counted.
• The 0.03 cd/m2 figure is a textbook boundary between cone and rod vision, not a property of the aurora. Individual eyes vary, age matters, and the mesopic handover is gradual rather than a switch.
• Cloud figures come from a reanalysis on a grid of roughly 28 km. Good for climate, useless as a forecast, and blind to a valley microclimate.
• The thresholds in the cloud calculation are ours: 50% cover, two hours, an 18:00 to 02:00 window. Different choices give different numbers.
• A usable sky is not an aurora seen. Every probability in the table above is a ceiling.
• All of it is measured at a fixed point, which is unfair to a tour willing to drive 200 km to get out from under the cloud. That remains the strongest argument for taking one.

• Brightness scale and descriptions: World Meteorological Organization, International Cloud Atlas, polar aurora brightness, section 3.2.4.3.1, at cloudatlas.wmo.int. The shaded text there is Annex I to the Technical Regulations, WMO No. 49.
• Kilorayleigh values for the four steps: Australian Bureau of Meteorology, Space Weather Services, Characterising the Aurora, at sws.bom.gov.au.
• Cone and rod thresholds, dark adaptation: Michael Kalloniatis and Charles Luu, Light and Dark Adaptation, in Webvision, Moran Eye Center, University of Utah, at webvision.med.utah.edu. The 0.03 cd/m2 duplicity boundary, the five to eight minute rod branch and the forty minute asymptote are taken from that page.
• Kp bands and what they look like: NOAA Space Weather Prediction Center, Tips on Viewing the Aurora, at swpc.noaa.gov.
• Geomagnetic frequencies: computed by us from the GFZ Potsdam Kp and ap series, 1932 to 2026, open data.
• Cloud and darkness: ERA5 reanalysis, produced by ECMWF for the Copernicus Climate Change Service, queried through the Open-Meteo historical archive, eleven seasons from 1 September 2015 to 30 April 2025. Moon phases and the equinox from standard ephemeris algorithms.
• Every link above was requested and returned a page on 9 September 2026.