Northern Lights
▾ In this article

September Is the Worst Month for Northern Lights in Thirteen of the Fourteen Places We Measure

September ranks lowest in the comparative cloud–darkness index at thirteen of fourteen destinations studied. We explain how limited darkness affects that score and why it is not a universal ranking of aurora viewing conditions.

10 min read Updated on
Green and pink aurora reflected in an unfrozen lake with bare autumn trees and no snow on the shore
Open water, bare trees and no snow: the September look. On our figures September is also the weakest month of the aurora season in thirteen of the fourteen destinations we measure. · Photo: Freepik

The northern lights in September come with a problem that stays out of the brochures. Tromsø gets about 1.8 hours of genuine darkness in a September night. Not dusk, not the long blue hour, darkness: the sun more than 18 degrees below the horizon, which is the point at which a faint arc stops competing with the sky it is sitting in. In December the same place gets 8.0 hours, every hour of the window we measure.

September also has the most open sky Tromsø sees all year. Both of those are true at the same time, and only the second one gets sold.

We ran ten seasons of cloud record and the ephemerides together for fourteen aurora destinations. On the combined figure September is the weakest month of the entire season in thirteen of the fourteen. The exception is Yellowknife, and it is worth the detour.

Why September looks like the right month

The advice to travel in September is not stupid and it is not a stitch-up. It rests on two things that are true.

Our cloud rule counts a night when at least two hourly samples between 18:00 and 02:00 inclusive have cloud cover at or below 50%. Samples need not be consecutive; they do not measure two uninterrupted clear hours. After-midnight samples belong to the preceding evening.

The second is the equinox. Geomagnetic activity has a real seasonal shape, with a bump either side of the equinoxes, and September sits inside one of them. On the GFZ Potsdam record since 1932, September's mean Ap is 15.68, which is 17.5% above the annual mean and second only to March. So both halves of the September pitch check out.

What the pitch leaves out is that neither of them is worth anything without hours to spend them in.

The hours the cloud statistic does not count

Pale blue and pink twilight sky over a Nordic fjord and low mountains, with no darkness in the sky

A sky that has not finished getting dark

Astronomical darkness is the sun more than 18 degrees below the horizon. That threshold is not one we chose. Below it the scattered sunlight stops interfering; above it a weak aurora is competing with a lit sky, and usually loses.

At high latitude in September the sun does not get far enough down for long enough. Alta averages 1.7 hours a night, Tromsø 1.8, Levi 2.2. Every Fennoscandian row in our table sits between 1.7 and 2.8 hours. By December every one of those same rows reads 8.0, which is the most our eight hour window can hold.

Up here the twilight is not a rounding error at either end of the night. In September it is most of the night.

Multiply the cloud figure by the darkness figure and you get the number this site uses throughout:

cloud–darkness index = (percentage of qualifying nights ÷ 100) × average astronomical-darkness hours

The cloud–darkness product is a comparison index, not measured or expected hours of aurora viewing. It multiplies a monthly share of qualifying nights by average darkness hours without modelling their joint timing. Adding monthly geomagnetic activity does not turn it into a sighting probability.

Previously published Open-Meteo cloud figures, ten September–March seasons (2015–2025). Original model selection unverified; values not recalculated in this correction.
Destination September: nights with a usable sky September: hours of darkness September: cloud–darkness index That row's own best month
Tromsø34.0%1.80.61October, 2.04
Alta36.0%1.70.61October, 2.20
Bodø37.7%2.60.98November, 2.83
Reykjavík40.7%2.71.10December, 2.67
Akureyri27.3%2.50.68November, 1.45
Rovaniemi41.7%2.81.17March, 2.15
Levi41.3%2.20.91January, 2.13
Saariselkä34.7%2.10.73March, 1.80
Kiruna44.0%2.51.10November, 2.86
Abisko27.3%2.30.63October, 1.26
Fairbanks44.0%3.11.36January, 4.01
Anchorage39.3%3.81.49November, 2.38
Yellowknife55.7%4.02.23February, 3.67
Whitehorse57.7%2.81.62February, 2.46

Read the last two columns against each other. September delivers somewhere between 30% and 66% of what the same destination's best month delivers. Tromsø sits at the bottom of that range: 0.61 in September against 2.04 in October, so a September night there holds about three tenths of an October night. Alta is the same shape, 0.61 against 2.20.

The three rows where September comes off least badly are Whitehorse, Anchorage and Yellowknife, all at roughly two thirds of their own best month. They are also the three most southerly places we measure, which is not a coincidence. Yellowknife and Anchorage get the two longest September nights in the table, 4.0 and 3.8 hours. Whitehorse gets the clearest September sky of anywhere, 57.7% of nights.

September has the clearest sky of the year in six of the fourteen places we measure, and it is still the weakest month of the season in thirteen of them. Cloud was never the binding constraint.

— Open-Meteo, ten seasons; original model unverified; values not recalculated.

Yellowknife, the one row that refuses to agree

Yellowknife is the exception, and it gets there for a fairly ordinary reason. It has the best September in the table by a distance: 55.7% of nights with a usable sky and 4.0 hours of darkness, which comes to a cloud–darkness index of 2.23. That is more than Tromsø manages in any month of its entire season.

September is not the low point of Yellowknife's own year, which is what makes it the exception. Its weakest cell falls in late autumn or in March depending on which of our two published tables you read, and those two tables do not currently agree with each other about the size of the November cell. We would rather flag that than quietly pick the version that reads better. What survives either reading is the part that matters here: Yellowknife's September, at 2.23, is not the weakest cell in its row, and it is the only row in the table of which that is true.

Then there is the catch, and it is the same catch this site keeps walking into. Yellowknife carries eleven tours in our catalogue, Whitehorse four, Anchorage three, out of 307 across the network. The places where September works best are the places with the fewest ways to book it. Tours cluster around airports and hotel beds, which is a good reason for an operator and a bad one for a traveller choosing a month.

Green aurora above silhouetted black spruce trees and a low ridge in the Yukon taiga
Cloud rule: at least two hourly samples ≤50% cloud between 18:00 and 02:00 inclusive. They need not be consecutive and do not measure continuous clear hours.

Change the metric and the count changes

Everything above leaves the magnetosphere out, and September is the month with the strongest claim on putting it back in. So we did.

The cloud–darkness product is a comparison index, not measured or expected hours of aurora viewing. It multiplies a monthly share of qualifying nights by average darkness hours without modelling their joint timing. Adding monthly geomagnetic activity does not turn it into a sighting probability.

It moves the answer, and by less than you would think. On the three-axis index September is the weakest month in eleven of the fourteen rows instead of thirteen. The three that escape are Anchorage, whose December falls to 1.42 against September's 1.76; Yellowknife, whose November is 2.05 against September's 2.60; and Whitehorse, whose November is 1.59 against 1.93. All three are North American, which is the earlier story wearing a different coat.

Tromsø, meanwhile, moves from 0.61 to 0.71 in September, and from 2.04 to 2.27 in October. The equinox bonus is real. It does not come close to closing a gap that size.

While we are here, one number that circulates in this trade needs its metric bolted to it. "Activity is roughly double at the equinoxes" is broadly right, and which version of roughly you get depends entirely on what is being counted. On mean monthly Ap across the full series the equinox to solstice ratio is 1.41. On the frequency of three-hour intervals reaching Kp 4 or higher it is 1.71, because the tail of the distribution stretches further than the middle does. Both are correct. An article that publishes one of them without saying which has published a number nobody can check, and that habit is most of the reason this site exists. The figure used above is the mean Ap one, and the Kp page in the guide goes through what the index does and does not measure.

The September moon, and why that number lies about itself

Our moon column counts the share of nights in a month that clear two hurdles at once: at least two hours of astronomical darkness, and a moon that is thin, low or already down.

September is the worst month of the season on that column almost everywhere, and not by a little. Tromsø comes out at 17% and Alta at 13%, the two lowest cells in the whole table. From October through February nearly every row sits between 45% and 61%.

Here is where a careless reading goes wrong, ours included if we are not paying attention. Those September figures are mostly not about the moon at all. They are the darkness hurdle failing. Alta averages 1.7 hours of astronomical darkness in September, under the two hour gate on most nights, so the night is disqualified before the moon is ever consulted. That 13% is a statement about the sun.

The moon itself behaves perfectly normally this month. New moon falls on 11 September 2026, full moon on 26 September. If you are already going, the fortnight around the 11th is the dark half, and those two dates carry no forecasting risk whatsoever. They are arithmetic, and they will still be right in ten years.

Bright green aurora filling the sky above snowy hills near Tromso with a half moon clearly visible below it
A bright moon and a strong aurora in one Tromsø sky. A moon washes out a faint arc and leaves a strong one alone, which is why our column looks at where the moon is, not just how lit it is.
1 2 3 4 5 6 7 8 9 10 11 12
strongest month worth booking weakest, or no season at all

How long each half of this calculation has been available

The seasonal shape of geomagnetic activity has had a published explanation for over fifty years. The cloud record that decides whether you get to see any of it has been freely queryable for about ten. Putting the two on the same line is the only new thing here.

The origins

  1. 1932

    The Kp record starts

    The GFZ Potsdam series of Kp and ap values begins. Every monthly geomagnetic figure in this article rests on it, and it is open data with no key required.

  2. 1973

    Russell and McPherron explain the equinoxes

    The geometric account of why the interplanetary magnetic field couples more efficiently with Earth's near an equinox. It is the reason September has a case to argue at all.

The growth

  1. 2015

    Our cloud window opens

    Previously published Open-Meteo cloud figures, ten September–March seasons (2015–2025). Original model selection unverified; values not recalculated in this correction.

Today

  1. 2026

    The two get multiplied together

    Cloud and darkness on the same line for fourteen destinations. The result contradicts the standard September advice in thirteen of them.

What September is actually good for

None of this makes September a wasted trip, and if you are reading it in September with the booking already made, these are the parts worth acting on.

The sky is open, and a clear sky is the one variable you cannot influence at all. Six of our fourteen rows have their clearest month of the season right now. What you are short of is hours, and hours are something you can partly buy back by staying up. In September the dark window is narrow and it sits close to local midnight, which is not where a lot of evening tours sit.

The ground is also not frozen yet, and that is why September aurora photographs look unlike January ones. Lakes are still liquid and still reflect. There is colour on the tundra. It is far warmer than the same place in midwinter, which decides how long anybody actually stands still outdoors, and that is a variable none of our tables contain.

Last, September is not one month. The autumn equinox falls just after midnight UTC on 23 September 2026, and the darkness is growing quickly across these weeks: Tromsø's monthly average goes from 1.8 hours in September to 6.4 in October. The end of the month is a different proposition from the first week of it, and a monthly average is precisely the thing that flattens that out. If your dates can move, move them later.

What these numbers do not say

  • This cloud-screening result is not a measured aurora-sighting rate or a strict upper bound on sightings. Displays can be visible through partial cloud or outside the sampled window, and a tour may move. The figures do not measure those outcomes.
  • The original weather-model selection is unverified, so no single grid resolution can be assigned to these cloud figures. They are not local station observations and cannot establish whether Abisko has a small clear-sky microclimate.
  • Our cloud rule counts a night when at least two hourly samples between 18:00 and 02:00 inclusive have cloud cover at or below 50%. Samples need not be consecutive; they do not measure two uninterrupted clear hours. After-midnight samples belong to the preceding evening.
  • The cloud–darkness product is a comparison index, not measured or expected hours of aurora viewing. It multiplies a monthly share of qualifying nights by average darkness hours without modelling their joint timing. Adding monthly geomagnetic activity does not turn it into a sighting probability.
  • All of it is measured at a fixed point, which is unfair to a tour that will drive 200 km to get out from under the cloud. That is also the strongest argument anybody has for taking a tour instead of standing outside the hotel.
  • A monthly average cannot tell you that 26 September beats 5 September, and in September that gap is unusually wide.

Questions we get about September

Is September a bad month to see the northern lights?

It is the weakest month of the aurora season in thirteen of the fourteen places we measure, on cloud multiplied by darkness. That is not the same as saying you will see nothing. It means an average September night in those places holds between a third and two thirds of what an average night in that destination's best month holds, and the shortfall is almost entirely darkness rather than cloud.

Why does everyone recommend September then?

Because the cloud statistic is good, and honestly so. In six of our fourteen destinations September is the clearest month of the whole season. The equinox also raises geomagnetic activity, and that is real too. What the advice skips is that at these latitudes there are only 1.7 to 4.0 hours of true darkness in a September night to spend either advantage in.

How much darkness does an Arctic September night actually have?

Averaged over the month, inside the 18:00 to 02:00 window we measure: 1.7 hours at Alta, 1.8 at Tromsø, 2.2 at Levi, 2.8 at Rovaniemi and Whitehorse, 3.1 at Fairbanks, 3.8 at Anchorage, 4.0 at Yellowknife. In December those same places sit between 6.2 and 8.0 hours, and the Nordic ones are all at 8.0, the ceiling of our window. Darkness here means the sun more than 18 degrees below the horizon, which is the standard definition rather than a threshold of ours.

Is there anywhere September still works?

Yellowknife, and to a lesser degree Whitehorse and Anchorage. Yellowknife gets 55.7% of September nights with a usable sky and 4.0 hours of darkness, which comes to a cloud–darkness index of 2.23, more than Tromsø reaches in any month of the season. It is the only one of the fourteen rows where September is not the weakest cell in its own year.

Does the equinox make September worth it?

It helps and it does not rescue the month. Folding a geomagnetic factor into the calculation moves September from weakest month in thirteen rows to weakest in eleven, and the three rows it rescues are all North American. Tromsø's September goes from 0.61 to 0.71 while its October goes from 2.04 to 2.27, so the ranking between them does not budge.

When in September should I go, if I am going?

As late in the month as the dates allow, and if possible around the new moon of 11 September 2026. Darkness increases sharply across September, to the point where Tromsø's October average is more than three times its September average, so the last week of the month is meaningfully better than the first. The moon dates are exact and involve no forecasting; the cloud is the part nobody can plan around.

Where do these numbers come from, and can I check them?

Historical cloud figures: previously published calculations from the Open-Meteo archive, September 2015–April 2025 (ten September–March seasons). The original model selection is unverified; these values have not been recalculated. See the corrected method and source limits.

Aurora tours in Yellowknife

The one destination in our data where September is not the weakest month of its own year, and one of the thinnest catalogues we carry. Both of those facts are on the record above.

4-Day Yellowknife Aurora Tour with 3 Hotel Nights and Mandarin or English Guide

4-Day Yellowknife Aurora Tour with 3 Hotel Nights and Mandarin or English Guide

5.0 (4) 5.0 stars out of 5, 4 reviews

Three evening aurora sessions from about 10 pm, three hotel nights, 24-hour airport transfers and a city tour. Mandarin or English guiding; meals not included.

€ 778.91
Aboriginal Owned Aurora Hunting Tours from Yellowknife

Aboriginal Owned Aurora Hunting Tours from Yellowknife

4.8 (412) 4.8 stars out of 5, 412 reviews

A wholly Aboriginal owned company whose guides grew up here and hunt the aurora for a living. Free photographs, tripods lent, hot chocolate and maple cookies.

€ 116.38
Intimate Aurora Cabin Night with Bannock and Fish Chowder
Selling fast

Intimate Aurora Cabin Night with Bannock and Fish Chowder

4.9 (299) 4.9 stars out of 5, 299 reviews

A small warm cabin, hot bannock and local fish chowder at midnight, and northern clothing to pose in. Five guests minimum, or book the whole cabin outright.

€ 92.81
Northern Lights Bus Tour on the Ingraham Trail from Yellowknife
Free cancellation

Northern Lights Bus Tour on the Ingraham Trail from Yellowknife

4.8 (439) 4.8 stars out of 5, 439 reviews

The trail runs past a chain of lakes, and stopping at several means the same aurora over three sets of water and rock. Free photographs and pickup included.

€ 61.77
Skydome Aurora Viewing near Yellowknife with Hotel Pickup
Selling fast

Skydome Aurora Viewing near Yellowknife with Hotel Pickup

4.9 (16) 4.9 stars out of 5, 16 reviews

Three heated domes a short drive from town, one large and two mini, with guests free to move between them and the outdoor space all night. No car needed.

€ 81.27
Yellowknife Aurora by Bus: Lake Hopping with Hotel Pickup

Yellowknife Aurora by Bus: Lake Hopping with Hotel Pickup

4.9 (151) 4.9 stars out of 5, 151 reviews

In autumn the lakes mirror the sky; once the ice locks, the bus parks out in the middle of one. Four guests minimum, or you are moved to another tour.

€ 89.07

Ratings collected by Viator and Tripadvisor

Sources

  • Historical cloud figures: previously published calculations from the Open-Meteo archive, September 2015–April 2025 (ten September–March seasons). The original model selection is unverified; these values have not been recalculated. See the corrected method and source limits.
  • Darkness, moon phases and the equinox: calculated from the standard ephemeris algorithms in Jean Meeus, Astronomical Algorithms, chapters 25 and 49, with no external dataset. Astronomical darkness is the sun more than 18 degrees below the horizon. Lunar interference combines illuminated fraction with the moon's altitude, sampled every twenty minutes, and a night counts as undisturbed below a mean of 25%.
  • Geomagnetic activity: the Kp and ap series published as open data by GFZ Potsdam, covering 1932 to 2026. Monthly mean Ap, the share of three-hour intervals reaching Kp 4 or above, and the equinox to solstice ratios quoted here are all computed from that series.
  • Tour counts: this site's own catalogue, read on 2 September 2026.
  • The method behind the cloud figures, including where the thresholds bite hardest, is set out on the when the sky opens page of the guide, and the month by month decision on best time to go.

Data verified on . Previously published Open-Meteo cloud figures, ten September–March seasons (2015–2025). Original model selection unverified; values not recalculated in this correction. About This Site

The month matters more than the operator

How much darkness each destination gets per night, which nights the moon leaves alone, and every new moon of this season.

Keep reading