There is one piece of aurora forecasting that works three weeks out, costs nothing, and almost nobody tells travellers about. It is this: if you read about an aurora storm in Iceland, put a note in your calendar 27 days later.
NOAA’s 27-day outlook in early September was the worked example. It had exactly one strong run in it, on 24-26 September — the strongest geomagnetic forecast of the season at the time. Subtract 27 days and you land on the last week of August, which is the storm we wrote about at the time. The same patch of sun was coming back round.
Update, 1 October: that run has now happened, so this page now carries the scorecard. Skip to how it turned out, or read on for how the trick works.
The Trick: The Sun Turns Once a Month
The sun is not solid, so it does not rotate all in one piece — the equator turns faster than the poles. But the part of it that throws solar wind at Earth comes back around roughly every 27 days as seen from here.
Some aurora is caused by a coronal mass ejection: a discrete cloud of plasma thrown out by an eruption. That is a one-off. It happens, it hits, it is gone, and it never comes back.
But a lot of aurora is caused by a coronal hole — a gap in the sun’s magnetic field where the solar wind escapes faster than usual. A coronal hole is not an event. It is a feature, and features sit still while the sun turns underneath them. So it sweeps past Earth, stops pointing at us, and then swings back into view about 27 days later, often still there.
That is the whole mechanism, and it is why the outlook runs to 27 days rather than 30.
The Worked Example: Our Own Storm, Returning
Here is the arithmetic, using observed data rather than the forecast that preceded it. These are NOAA’s estimated planetary A-index values for the storm days in August, against the same product’s forecast for September:
| Late August (observed) | Ap | One rotation later | Ap (forecast) |
|---|---|---|---|
| 28 August | 15 | 24 September | 20 |
| 29 August | 17 | 25 September | 22 |
| 30 August | 15 | 26 September | 22 |
The dates line up almost to the day. That is not a coincidence and it is not clever forecasting — it is a calendar.
What Recurrence Predicts, and What It Does Not
This is where most explanations stop, and it is where the useful part starts. Recurrence predicts the solar wind. It does not predict the sky over Iceland. Three separate things get in the way.
Only half of August’s storm is the kind that repeats. NOAA’s watch comment at the time was blunt about the mix: “HSS on 27 + CMEs on 28-29.” A high-speed stream from a coronal hole on the Thursday, then coronal mass ejections over the weekend. Only the first of those recurs. And there is an awkward detail in the observed record: 27 August, the day the recurring component was due, was flat — planetary Ap 4, with no three-hour Kp above 1.33. The activity did not show up until the CMEs did. The two drivers overlapped, so the credit cannot be cleanly split, but the part of August that is coming back is at best a share of what made August interesting.
Three weeks of lead is a long way out. We have written before about how accurate the aurora forecast actually is at each range, and the honest answer for a 27-day outlook is that it is close to noise beyond about a week. The Kp 5 pencilled in for the 25th and 26th is an estimate made 18 to 21 days ahead. Treat it as a reason to look, not as a promise.
And it says nothing about cloud — which loses more Icelandic aurora nights than solar activity ever does.
The Moon Is Already Waiting
This is the caveat nobody else supplies, and for 24-26 September it is decisive.
The moon is full at 16:48 UTC on 26 September. So the return arrives in the worst possible lunar week. We ran the numbers for Reykjavík:
| Night of | Astronomical dark | Moon-free dark | Moon |
|---|---|---|---|
| 21 Sep | 5.8 h | 4.4 h | 78% |
| 22 Sep | 6.0 h | 2.5 h | 85% |
| 23 Sep | 6.1 h | 0.8 h | 92% |
| 24 Sep | 6.3 h | 0.0 h | 96% |
| 25 Sep | 6.4 h | 0.0 h | 99% |
| 26 Sep | 6.6 h | 0.0 h | 100% |
Zero on all three nights. Not “compromised” — zero. On the night of 25-26 September the moon never drops below 14 degrees above the horizon at any point during the 6.4 hours of astronomical darkness. It is up, it is bright, and it is there for all of it.
Notice the shape of that table, because it is cruel. Iceland has plenty of darkness in late September and plenty of moon-free hours on the 21st and 22nd — and the activity is forecast to arrive precisely as they run out. It is the exact inverse of the mid-September window, where a modest Kp 3 lands on hours of genuinely dark sky.
We are not writing these nights off. A bright moon does not switch the aurora off, it raises the floor: faint structure disappears, and a strong display still gets through. If the forecast Kp 5 actually verifies, that is roughly the level at which a display is bright enough to hold its own against a full moon, and you would see something. It is a genuine long shot rather than a dead loss — but it needs the strong version of the forecast to come true, on a night when it has to be strong to count at all. The full mechanics are in our guide to how the moon affects the northern lights.
How It Turned Out
The run arrived. Here is each forecast against the record, as NOAA’s estimated planetary A index and the largest three-hour Kp of the day:
| Date | Forecast, 3 Sep issue | Forecast, 21 Sep issue | Observed |
|---|---|---|---|
| 24 Sep | Ap 20, Kp 4 | Ap 20, Kp 4 | Ap 16, Kp 4.33 |
| 25 Sep | Ap 22, Kp 5 | Ap 18, Kp 5 | Ap 20, Kp 4.00 |
| 26 Sep | Ap 22, Kp 5 | Ap 12, Kp 3 | Ap 13, Kp 4.00 |
The calendar part worked. 23 September came in at Ap 4 and 28 September at Ap 3, so the disturbance landed on the forecast dates, three weeks ahead, with quiet days either side of it. The size was roughly right: the later issue was within four points of the observed Ap on every day. The peak was not. No three-hour period reached Kp 5, so the minor (G1) storm the outlook kept pencilling in never happened.
It was the second time in a fortnight. For 14-17 September the outlook raised the last two days to Kp 5 at only two and three days of lead, and NOAA issued two G1 watches on top. The observed Ap was 11, 8, 9 and 9 against a forecast of 18, 16, 20 and 18, and the highest Kp all week was 4.00. No storm.
So we now have two runs and two different kinds of miss. The first overshot the whole disturbance by about a factor of two. The second got the disturbance right and overshot only its peak. Our reading: a Kp 5 in the 27-day outlook means “probably active”, not “storm”. That fits what we found when we looked at how accurate aurora forecasts are at each range.
And the moon
We called 24-26 September a long shot that needed the strong version of the forecast to come true. It didn’t, and the aurora showed up anyway. On the evening of 26 September, four to five hours after the full moon, app users around Egilsstaðir posted photos of bright bands and rays overhead against a moonlit blue sky, between about 20:15 and 22:05 UTC. Planetary Kp at the time was 2.33 to 2.67.
That corrects us in one direction. A full moon raises the floor, but in the east that night the floor was not high enough to hide what came. Planetary Kp is an average over the whole planet. Iceland sits close under the auroral oval, where an ordinary evening can still put structure overhead. A camera still beats the eye under a full moon, and a strong display is still what you need for faint detail. But “full moon plus a missed storm” did not turn out to be a lost night everywhere.
The Next Tests
Three dates are worth watching next, and between them they show what this trick can and cannot do. Numbers below are from the 27-day outlook issued 28 September 02:21 UTC and the 3-day forecast issued 1 October 00:30 UTC.
2 October: the one the outlook could not see. The 28 September issue has a quiet Kp 2 for 2 October. Later that day a filament erupted on the sun and threw out a coronal mass ejection. On 29 September NOAA issued a G1 watch for 2 October, and the 3-day forecast now peaks at Kp 4.67 between 15:00 and 18:00 UTC. That is before dark in Iceland. After that it eases to 4.00 for 18:00-21:00 and around 3.3 to 3.7 overnight into 3 October. NOAA expects only a glancing blow. This is the limit of the 27-day rule in one example: it follows coronal holes, and a CME is a one-off it cannot know about. The moon is waning but still up all night on 2 and 3 October, with no moon-free dark hours in Reykjavík.
4-6 October: the next rotation. This is one rotation after the early-September activity. Only the coronal-hole part of that can come back, since the CMEs that week were one-offs. The outlook has it at Ap 12, 15 and 12, Kp 4 on each day. NOAA’s latest discussion also models a weak glancing blow from a second CME near Earth on 4 October. The moon is starting to help: Reykjavík gets about 1.6, 3.8 and 5.8 hours of moon-free darkness on those three nights.
21-24 October: September’s run, round again. This is the feature we scored above, one more rotation on, and the outlook has a Kp 5 on 22 October at more than three weeks of lead. Given the record above, read that as “probably active”. The moon repeats September’s pattern, only a little earlier: full moon is 26 October, and moon-free darkness in Reykjavík drops from 3.2 hours on the night of the 21st to 1.6 on the 22nd and zero on the 23rd and 24th. A solar rotation (about 27 days) is slightly shorter than a lunar month (about 29.5), so a recurring stream comes round roughly two days earlier in the moon’s cycle each time. That is why this run keeps landing near the full moon.
Result, 6 October: 4-5 October was the first under-call. The 28 September issue had Ap 12 and Kp 4 for 4 October, and Ap 15 and Kp 4 for the 5th. NOAA’s estimates came in at Ap 36 and Kp 5.67 on 4 October, and Ap 24 and Kp 5.33 on 5 October. Three-hour periods reached Kp 5 or higher repeatedly, and NOAA issued its first Kp 5 alerts of the autumn, a minor (G1) storm. The disturbance returned on the forecast dates, as in September, but this time the outlook was too low rather than too high, by roughly a factor of three on the 4th. So the scorecard now reads: two over-calls and one under-call. Taken together with the 2 October CME, which the outlook could not see at all, the lesson is the same as before. A 27-day figure tells you which dates to watch, not how strong the night will be. We cannot say what the sky looked like everywhere: the west of Iceland was under heavy cloud on the 4th, which is why the Tonight page shows cloud as well as activity.
Result, 13 October: another CME the outlook could not see. The 28 September issue had 10 October as quiet. That afternoon a coronal mass ejection reached Earth, with a sudden impulse at 15:16 UTC, and drove the planetary Kp to 7.00 for 15:00-18:00 and 6.67 for 18:00-21:00 UTC, a strong (G3) storm and the biggest of the autumn so far, at Ap 46. NOAA’s own watch for that day was a G1. Like the forecast for 2 October, the peak came before dark in Iceland: Reykjavík was not fully dark until 21:05, by which time Kp had eased to 5.33. It is the second CME in a fortnight that the 27-day product had no way to see, which is exactly the limit described above.
The coronal hole behind mid-September came back on schedule. The 28 September and 5 October issues had Kp 3 for 11-14 October. By 12 October NOAA had raised 13 October to Kp 5 in both the 27-day outlook and the 3-day forecast, on the onset of a coronal-hole high-speed stream. Its first six hours came in at Kp 3.33, short of the forecast so far. We will score it when the day is complete.
The 21-24 October run has been cut. The outlook issued on 5 October had Kp 5 on 21 and 22 October. The one issued on 12 October has Kp 3, 4, 4 and 3 for 21 to 24 October, at Ap 15, 20, 15 and 10. That is the same stretch of the sun as 4 October’s storm, one rotation later, and the moon is still the problem, as described above. We will add the result here when it is in.
Next on the board: 31 October and 1 November. The 12 October issue puts its only Kp 5 days there, at Ap 30 and 25, one rotation after the 4-5 October storm. The moon is waning by then and rises at about 20:20 and 22:40 UTC, so Reykjavík gets only 0.7 and 2.9 hours of moon-free darkness at the start of those nights.
What to Actually Do With This
The 27-day rule is worth knowing because it works at a lead time nothing else reaches, and because it costs you one line in a calendar. Use it the way a forecaster would:
- When a storm happens, note the date. Add 27 days. That is your candidate.
- Check the moon for that date before you get excited. This is the step everyone skips, and it is the one that decided 24-26 September before any space weather was involved.
- Wait for the 3-day forecast. About three days out, the date enters the deterministic products, and that is when a rotation estimate turns into a real forecast. Our guide to reading an aurora forecast covers which numbers matter at which range.
The honest advice is unchanged: book the darkness, not the storm. Darkness is astronomical fact and will not be revised. For the month ahead, see our October 2026 guide — the moon clears out from 8 to 17 October.
And on the night itself, the only number that matters is the one where you are standing. Our live aurora forecast scores 137 monitored locations across 14 regions every five minutes on solar wind, OVATION output and observed cloud, and the Aurora Iceland app will tell you which of them is clear and active right now.