Your phone sees the aurora before your eyes do

There is a pale cloud above the treeline behaving like no cloud you have watched before. Grey, faintly green if you look slightly to one side of it. You rest the phone on a wall, give it three seconds, and the screen comes back with a green curtain, sharp along the bottom edge, folded in vertical pleats. No creative filter, no software enthusiasm. The sensor recorded what was actually there and you did not, and the reason sits inside your own eye.

The eye has no night mode

Below a hundredth of a candela per square metre the retina is running on rods alone. One type of photoreceptor means one sensitivity curve, and with nothing to compare that curve against, the brain has no way to construct a colour: it receives a greyscale and makes do. Rod sensitivity peaks at 507 nanometres, in the blue green, while the daylight system peaks around 555. The gap between those two peaks is the Purkinje shift, which is why poppies go black at dusk while the leaves stay bright.

The joke, at this point, writes itself. The green auroral line sits at 557.7 nanometres, almost exactly where rods do their best work. You can see that light perfectly well. What you cannot do is tell what colour it is, which is a different problem entirely.

Then there is integration time. Your eye gathers for something like a tenth of a second and then delivers the picture, whether or not it has enough to work with; the sensor gathers until you tell it to stop. Three seconds is thirty of your turns, and thirty times the photons is roughly what it takes to push the scene up into the range where colour exists at all. Between 3 and 0.01 candelas per square metre lies mesopic vision, rods and cones together. Auroras that get there are the ones that make people on the hillside shout. The rest you photograph, and that is that.

Green at a hundred kilometres, red at two hundred

The 557.7 nanometre emission comes from atomic oxygen dropping from the O(¹S) state to the O(¹D) state, and it lights up between 100 and 150 kilometres. The red line at 630.0 nanometres comes out of the same atom one energy step lower down, but from between 200 and 300 kilometres.

The altitudes differ because the lifetimes do. O(¹S) holds its excitation for about seven tenths of a second, O(¹D) for around 110 seconds. Nearly two minutes waiting to emit: down where the air is still thick, that atom collides with something long before its time is up and gives away the energy silently. Only where collisions thin out does the red get to escape.

Which explains the view from well outside the auroral oval. What clears your northern horizon is the top of the curtain, so the high part, so the red part, often with no hint of green underneath it at all. And it carries: SWPC notes that a bright aurora can be seen from as much as a thousand kilometres away, which for a structure hanging at two hundred kilometres is simple geometry.

Knowing whether tonight is the night

The number every app quotes is Kp, it runs from 0 to 9, and what it reports has already happened. It is the standardised mean of the K index gathered by thirteen geomagnetic observatories sitting between 44 and 60 degrees geomagnetic latitude, computed over three hour intervals. From 5 upwards you are in storm territory, and NOAA’s G scale maps onto those values from G1 to G5.

In its quiet state the auroral oval is a thin band around 67 to 68 degrees geomagnetic latitude, widening equatorward as activity rises. The word to read twice is «geomagnetic». The magnetic pole does not sit on the geographic one, it loiters over the Canadian Arctic, and that skews the sums hard depending on your meridian: at the same latitude on the map, a reader in northern Europe is further from the oval than one in Canada. Scotland is not Saskatchewan, whatever the satnav implies.

On actual forecasting, make your peace with the clock. The SWPC’s OVATION model produces a thirty minute map, and those thirty minutes are precisely how long the solar wind takes to cover the last stretch: the data comes from DSCOVR, parked at L1 a million and a half kilometres out, and the warning runs anywhere from fifteen to sixty minutes depending on flow speed, with half an hour corresponding to roughly 800 kilometres per second. Anything circulating days in advance is a probability, not an appointment. The one solid alert arrives when you are already dressed and outdoors.

The rest is statistics worth obeying. The best hours fall within an hour or two of local midnight, spreading towards evening and dawn as activity climbs. The best weeks cluster around the equinoxes, thanks to the way the solar wind couples to the magnetosphere then. A single substorm, the episode that actually sets the sky going, runs thirty to ninety minutes end to end, and the initial brightening of the arc arrives within a few minutes. Anyone who has gone back to the car to warm up misses it.

And a reality check for readers at mid latitudes: the night of 10 to 11 May 2024 was rated G5, the first since October 2003, and it put aurora over the Canary Islands, Ladakh and south Florida. Twenty years of waiting for one night. This far south, the aurora is something you chase when it happens, with the bag already packed.

The one subject that moves faster than the sky

Here the usual reasoning collapses. Normally your exposure ceiling is set by the Earth’s rotation, and the arithmetic behind how many seconds you get before the stars stretch hands you a generous number on a phone’s wide lens. With aurora that number is irrelevant, because the subject sets the limit: during the expansion phase the pleats of the curtain rearrange themselves in seconds.

In practice: two seconds while a substorm is running and the structure is sprinting, four to six when what you have is a quiet arc lying along the horizon. Overdo it and you get a flat green fog where the columns were, a kind of blur that looks nothing like the one you are used to. Check the screen after the first frame, because the situation changes every quarter of an hour.

The second instinct to suppress is stacking. AstroStackerPro aligns frames after the shot by comparing the stars, which is exactly why it works so well on deep sky: stars are a rigid lattice, they superimpose, and the noise retreats. An active aurora has no such lattice. Align on the stars and you get perfect stars with green mush where the folds were, the same mistake already covered here for meteor showers. Stacking earns its keep on a static arc, and then only across a handful of closely spaced frames.

Green clips first

A strong aurora is far brighter than memory suggests, and it saturates the green channel well before the overall histogram gets near the top. What you end up with is a curtain that is white in the middle and coloured only at the edges, all internal structure gone. Watch the separate channels: the combined curve lies to you, and you would only find out at home.

White balance is the usual argument, made worse. Auto sees a scene dominated by green and corrects it, because it has been trained to remove colour casts: you are asking it to delete your subject. Lock it before you start, as we have explained before, and leave it alone until the night is over.

Where to put the horizon

What is left is the part no measurement decides for you. A green curtain on black, with nothing around it, is a photograph people look at for four seconds. You need something on the ground for scale: a ridge line, a lake doubling the display, a house with one window lit. The phone’s wide lens, a handicap on every other target in the sky, is finally the right optic here: the curtain spans tens of degrees and no telephoto will hold it.

The framing, though, gets decided in advance. At mid latitudes you point north and shoot the red top of the curtain; under the oval it goes straight over your head, and the good frame turns nearly vertical, corona at the zenith. Two different photographs, two different compositions.

The phone, in this story, is not a compromise. It is the instrument that shows you a colour your eyes, built the way they are, will not see that night.