How to Read the Aurora Forecast

Darkness is arithmetic, published years ahead. The only decision it supports is which month you come, which is the month-by-month page's subject rather than any forecast's. It is worth saying once what it is competing with: the emission forms between about 80 and 500 km up, on NOAA's figures, against cloud tops that stop around ten kilometres even for high cirrus. Aurora is not dimmed by cloud. It is hidden by it, the way a ceiling hides a light fitting.

Geomagnetic activity is the condition everybody reads about and the one nobody can act on. The term that decides whether a display happens is not known until the wind carrying it is minutes away — the solar-wind section below is what that leaves you with — and everything published further out is a probability of disturbance rather than a schedule. Because Iceland lies under the oval, the gate is low enough that on most clear nights the answer is yes, at some point, probably late.

Cloud is the odd one out on all three counts. It is the only one an ordinary weather model forecasts with real skill two to three days ahead, the only one that is meaningfully different sixty kilometres away, and the only one you can change by moving the van.

The definition published by NOAA's Space Weather Prediction Center — spaceweather.gov, where every NOAA product named on this page lives — repays reading slowly. The K-index quantifies disturbances in the horizontal component of the Earth's magnetic field on a 0 to 9 scale, derived from the maximum fluctuations recorded on a magnetometer during a three-hour interval; the planetary index Kp is the mean standardised K-index from 13 observatories between 44° and 60° geomagnetic latitude. Not one of the 13 is in Iceland or anywhere near the auroral oval — the closest is Lerwick in Shetland. The instrument was built to detect the oval expanding down towards mid-latitudes, which is a different question from whether the oval is lit up overhead.

It is also a measurement rather than a prediction, computed after the event from ground magnetometer traces. What a phone shows tonight is a near-real-time estimate; the definitive series, in 28 steps of a third, follows later, can differ, and arrives on GFZ Potsdam's monthly cycle. A bare "Kp 3" could mean anything from 2⅔ to 3⅓.

The three-hour bin is the failure that bites on a specific evening, and one pair of numbers shows it. On 11 August 2026 the GFZ series gave Kp 2.667 for the 12:00–15:00 UTC window and 3.0 for 15:00–18:00. The half-hourly Hp30 index, computed by the same institute over the same hours, contained a 4.333 at 14:00 and another 4.333 at 16:00. Most of each window was quiet and a half-hour of each was not, and Kp cannot tell them apart. Hp30 nowcasts are themselves revised after the fact — the 16:00 value we first read was higher than the one that survived — so re-read them rather than quoting ours.

The geometry is what makes Iceland unusual, and it is where two conventions collide. NOAA puts the oval's equatorward edge near 66° geomagnetic latitude at Kp 0 and moves it roughly two degrees per Kp step. Computed with the AACGM-v2 reference implementation, Reykjavík sits at 64.15° geomagnetic and Akureyri at 65.42° — essentially Fairbanks's latitude, two degrees south of Tromsø. Pages describing Iceland as sitting at geomagnetic latitude 70 are using the older centred-dipole convention, which puts Reykjavík at 69.45°. Both are correct; they are different coordinate systems.

The page is en.vedur.is/weather/forecasts/aurora, and the layout is a cloud-cover map of Iceland with a time slider plus a box in the upper right whose visible heading is only the weekday evening — the phrase "Aurora activity, sun and moon" sits in the markup as an accessibility label that sighted readers never see. The activity forecast renders as a row of ten numbered cells with one highlighted and a word underneath, described by the page as the forecast of auroral activity at midnight on a scale of 0 to 9. There is one value per night, for the entire country, and the sunset, darkness and moon times printed beside it are valid for Reykjavík only — the page says so in one line. The moon line matters less than people expect, and what it costs is a photography trade-off rather than a reason to stay in — the photography article in this hub works through it. In midsummer, when the sun never falls 6° below the horizon at all, the darkness line is replaced outright by the words "Not dark".

Where the activity value comes from is not stated anywhere on the page, and the answer looks to be Maryland rather than Reykjavík. On 11 August 2026 the page's data block carried ten nights of activity values, and nine of the ten matched — digit for digit — the "largest Kp index" column of NOAA's 27-day outlook issued at 01:53 UTC the day before. The exception was the current day, which the Met Office appears to overwrite from a shorter-range product. That is a single reading of one ten-night table rather than a documented data flow, so we would not state it as fact — and it sits awkwardly against the page's own words, because a largest-daily-Kp figure can be set by a disturbance at nine in the morning while the page calls its number the activity at midnight. Read it as a national figure of uncertain provenance and do not read a time of night into it.

The map is the useful half. Its green is graded rather than binary — one downloaded frame carries at least three distinct greens as well as white — but no legend for the gradient is published, so darker green reads as more cloud without the page ever saying so. Five layer tabs sit above the map: low and mid-level cloud together as the default, then lower, middle and high cloud separately, and a composite. All come from HARMONIE-AROME, which Veðurstofa Íslands runs at 2.5 km resolution, four times a day. The Met Office's own model page still documents a 66-hour range, but the map loaded 72 hourly frames on the afternoon we read it, so treat three days as the horizon.

That split is not decoration. High cirrus is translucent and aurora is frequently visible straight through it, while low stratus is opaque. The default tab bundles low and mid cloud, which is the pessimistic view, so a green patch there may be a blocker or may be a veil, and the only way to tell is to click the high-cloud tab. It is the most useful thirty seconds available on the page and almost nobody spends it. The two halves also reach different distances: the activity box lists ten nights, the map three days. Where our own autumn guide still describes an activity forecast running six days, ten nights and three days are the figures that survived this reading.

The real-time solar wind is measured at L1, the balance point roughly 1.5 to 1.6 million km upstream of Earth, and that distance is the entire lead time. NOAA calls its 30-minute aurora forecast a 30 to 90 minute product and says outright that the lead time is the travel time from L1 to Earth. Do the arithmetic yourself and that is about an hour at 400 km/s and half of it at 800. Its coronal-mass-ejection page puts shock warning at a shorter 15 to 60 minutes, because the front of a coronal mass ejection is travelling far faster than the wind ahead of it — that band is a shock-arrival figure, not the ordinary transit time. You can watch the figure move: on 11 August 2026 the gap between observation and forecast time in NOAA's hemispheric-power file narrowed from 72 minutes at 00:55 UTC to 59 by 16:35 as the wind picked up.

Bz is the term that matters and the one nobody can predict. It is the north–south component of the field carried in the wind, and when it points south it couples with Earth's field and opens the door. The published thresholds disagree, and the disagreement is about latitude: SpaceWeatherLive, writing for mid-latitude Europe, wants sustained values of −10 nT or lower; NOAA's own real-time plot shades southward Bz from zero by default and offers −5 nT as an alternative highlight, which is the level its own display treats as worth separating out; Aurora Reykjavík, writing for Iceland, puts the gate at −2 nT. All three are compatible once you see they answer different questions — how far south the oval must come, against whether the oval you are already under will light up.

The supporting numbers are easier. Total field strength above about 6 nT alongside a southward Bz is the Icelandic operator's rule of thumb; SpaceWeatherLive calls 10 nT moderately strong and 30 very strong. Speed sits somewhere between 300 and 400 km/s in the ambient slow wind — SpaceWeatherLive quotes the lower figure, NOAA's own solar-wind page the higher — faster in a coronal-hole stream, and it can exceed 1000 behind a coronal mass ejection. Density above 20 particles per cubic centimetre is called a good start for a strong storm and explicitly not a guarantee.

One correction, because it tells you how fresh a page you are reading. The spacecraft feeding this data changed in January 2026. On 11 August 2026 NOAA's feed carried three sources — SOLAR1, ACE and IMAP — and the "active" flag marking what its forecasters use was set on SOLAR-1 for every one of the 1,436 magnetometer records in the file. DSCOVR did not appear at all. SOLAR-1, formerly SWFO-L1, launched on 24 September 2025 and reached L1 on 23 January 2026, and NOAA's product page states that it and IMAP will replace the legacy measurements from ACE and DSCOVR. NOAA's own CME page and SpaceWeatherLive's help pages both still name DSCOVR. A site that has not noticed a spacecraft handover is a site to check the dates on.

Two quite different solar structures produce Icelandic aurora and they give completely different notice. The workhorse is the coronal hole: a region of open magnetic field, dark in extreme-ultraviolet images, which lets the wind escape at 500 to 800 km/s. A compression called a co-rotating interaction region runs ahead of the fast stream, and NOAA notes that these typically drive G1 to G2 storming rather than headline events, can buffet Earth for many days, and often deposit more energy in the magnetosphere over a longer interval than a coronal mass ejection does. That is the better shape of event for somebody with a week here.

Coronal holes also come back. The Sun rotates in about 27 days as seen from Earth and persistent holes survive several rotations, so NOAA builds the recurrence into two products: the 27-day outlook table, and an overlay on the real-time solar wind plot drawing the data from 27 days ago on top of today's. A stream that produced good aurora last month is a candidate to do it again on roughly the same date — up to four weeks of notice, at low confidence, saying nothing about Bz.

Coronal mass ejections are the spectacular, rare half. They are spotted in coronagraph imagery, travel at anything from 250 to nearly 3000 km/s, and NOAA's heliosphere model gives one to four days of warning of an Earth-directed one. The fastest arrive in 15 to 18 hours; NOAA says the rest can take several days, the Icelandic Met Office says one to three, and the spread between two official sources is the honest answer. Then the sentence SpaceWeatherLive states more plainly than either: nobody can predict the strength, orientation and duration of the field inside the cloud until it reaches L1. A CME can be tracked for three days and do nothing, because its field happened to point north.

Put those together and the ordering is upside down. The panel worth planning on reaches about three days, and it is an ordinary weather forecast with geography in it. The panel that reaches ten nights has nothing local in it at all. There is no such thing as picking the aurora night of a trip in advance; there is such a thing as deciding, thirty-six hours out, which region to be in.

NOAA's description of a normal night is the best short account there is. During the evening the rays form arcs stretching from horizon to horizon. Late in the evening, near midnight, the arcs begin to twist and sway as if a wind were blowing on the curtains. At some point they may expand to fill the whole sky, moving rapidly and becoming very bright — the peak of an auroral substorm. Then in the early morning the forms become diffuse patches, blinking on and off for hours. The phases have been named since the 1960s as growth, onset, expansion and recovery, and the important word is cycle.

The timing is one of the better-measured things in the field. Frey and colleagues published 2,437 substorm onsets imaged from orbit by IMAGE-FUV and found a median onset at 23:00 magnetic local time and 66.4° magnetic latitude. Their own table collects four earlier surveys using different satellites, and every one lands between 22:30 and 23:24 magnetic local time and between 65.0° and 67.0° magnetic latitude. More than 91% of onsets in their sample fell between 60° and 70°, and Iceland sits in the middle of that band — which is why the aurora here is so often overhead rather than distant.

Converting that to an Icelandic clock is unusually easy, because Iceland keeps UTC all year and UTC is the time on your phone. Magnetic midnight — the moment a place is on the far side of the magnetic pole from the Sun — computed for mid-November 2026 falls at 00:14 UTC at Höfn, 00:20 at Akureyri, 00:30 at Vík, 00:36 at Reykjavík and 00:40 out on Snæfellsnes. Median onset an hour before that puts the statistical peak at roughly 23:15 in the east and 23:40 in the west. Icelandic aurora sites give a different figure and it is worth recording rather than smoothing over: they place magnetic midnight itself at about 23:00, and 22:30 in the east. The likeliest reconciliation is that the popular figure describes peak onset rather than magnetic midnight — in which case both accounts agree that the busy part of the night starts around eleven and runs past midnight.

The trap is the growth phase, and it is a trap because no part of the forecasting stack separates a tail quietly loading from a dead night. The Kp bin does not change across that hour, because it is one bin. The 30-minute oval reports the state of the oval now rather than what is coming. Hemispheric power moves at the moment it happens. No instrument tells you at 22:30 what will happen at 23:40.

Follow the routine above and you have committed to something specific: parked at a dark spot before it is properly dark, outside repeatedly for several hours, and not leaving before one in the morning on any genuinely clear night. In late September that is a short late window. In January it is most of an evening. Either way it is time spent in a stationary vehicle in an Icelandic winter, and the vehicle does all the work.

Running the engine is not the answer. It is unlawful — Icelandic regulation forbids letting a stationary vehicle idle beyond a very brief moment, and the heating article in this hub sets out the rule and the carbon-monoxide risk behind it — and in this context it also fouls the view you drove out for, for the reason the photography article works through.

That leaves the heater, and the aurora-specific consequence is one the heating article does not draw. A camper whose heat works only when it is plugged into a bollard can only be warm at a powered pitch. Powered pitches are at campsites, campsites have amenity blocks, and amenity blocks have lights on all night. Hookup-only heat and a genuinely dark stand are mutually exclusive, and the choice was made when you picked the camper.

The second-order cost is the leisure battery, with an aurora-specific twist. Overnight heating claims assume a day of driving recharged the bank — the heating article does that arithmetic — and an aurora day is a parked day, so the deficit lands on the night you least want it. If tomorrow is another long wait, drive somewhere in the afternoon.

Two failures after dark belong to the tools rather than to the sky. The first is the alert on your phone, which fires on a planetary number and therefore cannot know whether it is dark or clear where you are standing; the FAQ below says which alerts are built differently and what to carry instead.

The second is quieter and worse: the service can stop without telling you. On the day this was compiled, the SpaceWeatherLive forum carried an open thread titled "SWL site and app frozen since july 21th 18z" with 21 replies — three weeks of stale data on a site thousands of aurora chasers rely on. A notification that never arrives is indistinguishable from a quiet night, which is the argument for a routine that runs on a clock rather than on a push.