Northern Lights by Month: September to March

Darkness is arithmetic. The sun's altitude for any minute of any night at any latitude is computable centuries ahead, and it is settled the moment you choose your dates. Cloud is a weather forecast, useful days ahead rather than weeks, and it is the only one of the three that regularly equals zero for a week at a stretch. Geomagnetic activity is barely forecast at all beyond a probability, and the burst that actually lights the sky — a substorm — is watched for rather than predicted. How far ahead each of them is worth anything is the forecast article's subject; this one is about what the calendar can and cannot buy you.

The aurora does not have a season; the sky does. It is overhead as often in June as in December, and for the four and a half months around midsummer Iceland's sun never gets far enough below the horizon for anyone to see it — which is why this page counts hours of darkness rather than nights of activity. All times here are Icelandic local time, which is UTC in every month of the year. The number that decides an aurora night is not sunset and it is not the length of the night; it is how much of that night the sun spends more than 18° below the horizon. In mid-September at Reykjavík the sun is down for eleven hours and the sky is truly dark for 44% of that. In mid-December it is down for close to twenty hours and truly dark for 65% of them. The missing hours are twilight, and at the shoulders of the season twilight is most of the night.

That distinction decides the shape of an evening rather than just its length. In mid-September the dark window at Reykjavík does not open until an hour before midnight, so anything earlier happens in a sky that is still faintly lit and only a strong display will carry. By mid-October it opens in the evening and the night has become usable. By mid-December you can be parked, fed and watching by seven, and still have eight hours of darkness left at midnight. If you have a bedtime, the last column of the table is the one to read.

Akureyri is the instructive comparison, because further north does not simply mean darker. In midwinter it does, by about ten minutes at mid-month. At the shoulders it is the other way round, because a higher latitude keeps the summer sun closer to the horizon for longer. Astronomical darkness returns to Reykjavík on 2 September and to Akureyri on 6 September; it goes from Akureyri on 5 April and from Reykjavík on 9 April. The far north buys you midwinter and charges you the shoulders.

Those dates are worth stating precisely, because the answers in circulation disagree by three weeks — this site's own pages among them — and they disagree because they are quietly answering with different thresholds. The Almanak Háskóla Íslands prints two pairs of columns, birting and myrkur, and dögun and dagsetur, and it is the second pair that governs the aurora. We have not been able to confirm from the almanac itself which solar altitude it sets each pair at, so we state our own figures rather than its: by our computation the sun never reaches 18° below the Reykjavík horizon between 10 April and 1 September. Our own fall and winter guides give a different window, 16 April to 27 August; the two cannot both describe the same threshold, and this is the figure we now stand behind. Late August is a defensible season opening if you mean nautical darkness, which returns to Reykjavík on 15 August. Early September is the answer if you mean the real thing.

The Icelandic Met Office has published monthly mean cloud cover per station since 1949, and the 1991–2020 normals settle the question. Reykjavík's cloudiest aurora-season month is September at 6.01 oktas and its clearest is November at 5.70, with October a hundredth of an okta behind it — which is to say October and November are the same month as far as cloud is concerned. The whole spread is about a third of an okta, four per cent of the sky. Keflavík spans 5.9 to 6.1. Akureyri spans 6.1 to 6.3. Nothing in that table would change a booking. Move between stations instead and the numbers open up properly: Mývatn averages 5.4 oktas in February against Akureyri's 6.2, a hundred kilometres apart.

The standard account of Icelandic climate has the prevailing wind easterly and the winter weather set by Atlantic depressions tracking past to the south, with air lifted over the south coast and the ice caps raining out there and descending drier over the northeast. The precipitation column bears that out exactly. Over the seven aurora-season months Vík í Mýrdal averages 1,654 mm and Grímsstaðir á Fjöllum 223 mm — a factor of seven, on the same set of station files. The cloud column does not follow it at all. Vík, the wettest place in the sample, has a clearer season-mean sky than Grímsstaðir, the driest. Rain shadow moves precipitation; it does not reliably move cloud.

So we counted nights instead of months. Taking NOAA's archive of synoptic observations — 2015 to 2024 for Reykjavík, and 2015 to 2023 for Akureyri, whose station stops filing usable cloud in that archive after 2023 — and keeping only readings made with the sun more than 12° down: at Reykjavík between 13% and 16% of night observations show 2 oktas or less, and between 31% and 39% are solid overcast, in every month of the season. Per night — 1,603 nights with at least three dark observations — 27% gave at least one reading of 2 oktas or less, 45% at least one of 4 or less, and 14% were completely overcast at every single observation. Akureyri is worse on all three counts, and more than twice as likely to write a night off entirely.

The last number in that set is the one that should change how you plan. Over 1,525 consecutive night pairs at Reykjavík, a mostly clear night followed a mostly clear night 35% of the time against a base rate of 18%, and followed a cloudy one only 14% of the time. Cloud arrives in systems and stays for days. Three nights in the same valley are not three independent draws — which is the unstated arithmetic error under a claim we traced to a single page. When we read iceland.org on 11 August 2026 it stated that tour operators average a 40-to-55% success rate per night and compounded that to about a ninety per cent chance over three nights; no operator was named and no data was given for the underlying rate, and the percentages are not on the page today. The fix is not more nights in one place. It is the ability to be somewhere else by dusk.

Edward Sabine noticed in 1856 that magnetic disturbance runs higher near the equinoxes than near the solstices. Russell and McPherron gave it a mechanism in 1973: around the equinoxes the Earth's magnetic axis is tilted so that the interplanetary field arriving from the Sun couples southward into the magnetosphere more effectively, and southward is the direction that opens the door. We measured the size of the effect on GFZ Potsdam's Kp series from 1932 to 2025, counting only the four three-hour bins between 18:00 and 06:00 UT. Mean daily Ap runs about half as high again in March as in December; the table below has the month-by-month figures. Taking both equinoxes against both solstices on a daily basis, the ratio on days that reach Kp 5 at all is 1.77.

NOAA SWPC's own viewing-tips page states the caveat plainly: the equinoxes bring a tendency towards larger storms, but "the number of hours of darkness decreases (increases) rapidly near the spring (fall) equinox so this caveat must be considered." It is printed at the source and dropped by almost everyone who cites it. The peak also does not fall where the theory puts it. Russell and McPherron's model implies maxima around 7 April and 11 October; in the observed record the highest ten-day mean Ap of the entire year falls in 21–31 March, and the second run of high values in 21–30 September, a week or two ahead of the model dates at both ends. If you are choosing between the last week of March and the first week of April, the ninety-four-year record prefers March. That is a real finding and a small one, which is the honest way to hold it. Set against a darkness term that triples across the same seven months, the equinox bonus is the smallest lever on this page.

Cycle 25 reached maximum in October 2024 at a smoothed sunspot number of 160.9 on the SILSO series: above the feeble Cycle 24's 116, below the all-cycle average of 179, and well above the 115 that the 2019 NOAA and NASA panel predicted. By January 2026 the smoothed number was 104.2. SWPC's monthly prediction file puts the coming season near 96 in October 2026 falling to 87 by March 2027 — around 60% of peak, easing towards 55% by spring — and the season after that near 60, which is a third of it.

That is not a reason to hurry, and it is not a reason to wait. Geomagnetic activity does not track sunspot number one for one: the distribution across a cycle is double-peaked, with storms driven by coronal mass ejections clustering near maximum and a second maximum during the declining phase driven by high-speed streams out of coronal holes. What those two kinds of event do to a night, and how much notice each of them gives, is the forecast article's subject. The seasonal consequence is the part that belongs here, and for a country sitting under the oval it is close to neutral: fewer spectacular nights, more frequent moderate multi-day ones. The related claim that the very largest storms arrive two or three years after maximum circulates widely in news and travel writing; we found it asserted by outlets and could not trace it to a paper, so we are not repeating it as fact.

September's problem is not its odds, it is its budget. Its per-hour geomagnetic odds are among the best of the year, and that is precisely what makes it a trap: the ranking pages quote the rate and never the hours. Under five hours of true darkness that does not open until an hour before midnight is not much of a budget to spend good odds on, and the thirty-year normals make it Reykjavík's cloudiest month of the season as well. What September genuinely offers is the easiest driving of the season, open highland routes and mild nights to stand around in, none of which is an aurora argument.

There is a second arithmetic running underneath the first, and for once it points the same way. Rental rates in Iceland track the tourist season and not the sky, so the months with the most dark hours are also the cheapest months to put a van under them. It is the one advantage of the deep-winter months nobody argues about, and it softens the trade December asks for: the worst driving light of the year, which the winter guide budgets in detail.

Two caveats travel with all of it. The first is that our two cloud datasets rank the months differently: the twenty-four-hour normals make September the cloudiest of the seven and November the clearest, while our night-only count makes February the cloudiest and December the clearest. The disagreement is itself the finding: the monthly differences are inside the noise in both datasets. The second is that an index of expected hours is not a probability of seeing anything. Three nights in November can be three overcast nights, and three nights in September can contain one perfect one.

We computed the moon's position for every five minutes of astronomical darkness at Reykjavík across the 2026–27 season. It is above the horizon for between 41% and 53% of dark hours depending on the month — 53% in October, 41% in March — and there is no month in which it is meaningfully absent. That is not luck, it is geometry: a full moon is by definition opposite the sun, so it is up for the whole of the dark window in every month of the year, everywhere. This season's new moons fall on 11 September, 10 October, 9 November, 9 December, 7 January, 6 February and 8 March; published times differ by several hours between sources, so treat the dates as good to a day and no better.

What does change through the season is how high it rides, and that matters more than the phase calendar. At 64°N a December full moon culminates 53° up and lights the whole landscape; a March full moon reaches 24° and sits low in the south; a September one reaches 32°. So the moon costs most in exactly the months whose darkness budget is largest, and least in the months that have least to lose. It is a much smaller effect than cloud, which writes off between half and three-quarters of the season's dark hours, where the moon is above the horizon for around half of them and only dims what it lights. If your dates are flexible to the week, take the new-moon fortnight; if they are not, do not reshuffle a trip around it, and remember that a low quarter moon is useful the moment you want a lit foreground.

Take each month's hours of astronomical darkness, multiply by the share of night observations showing no more than half the sky covered, multiply again by the share of night-time Kp bins reaching 2, and index the result to the best month. The working is below, because the multiplication is the argument and it is the step nobody else takes. Three groups fall out: a winter trio at the top, an autumn-and-late-winter trio a third lower, and September on its own at the bottom of the season it is most often recommended for. At Akureyri the leading trio reorders and November takes it outright.

Our position. If the aurora is genuinely the only reason for the trip and you will accept four to six hours of driving light, book November: it gives up about a tenth of December's opportunity and returns half as much daylight again, and it sits ahead of the only fortnight in the off-season where getting a van becomes work. The caveat is that the deep-winter months are also when several companies pull part of their range and when heat that works with the engine off stops being a preference — the van menu narrows exactly where the darkness budget peaks. If the trip has to earn its keep in daylight as well, book March, and when you are told September and March are the two best months, take the March half and leave the September half. Our own aurora road-trip guide and our autumn guide currently give the shoulder-month answer; on this arithmetic we think it is wrong, and both will be corrected. December and January are what most people actually book, and they are not wrong about the sky; they are simply paying for it in daylight and in road risk.

One thing outranks all of it. A month is a darkness budget and nothing more. Nothing on this page forecasts a night, and the difference between a good trip and a blank one is almost always which side of the country you were on at nine in the evening.