What color is lightning: #A2BBFF at the channel, and the air decides the rest

Search for the colour of lightning and you will be handed a ranking. Blue lightning is the hottest. White is next. Yellow means dust in the air, orange means rain, red means the storm is weakening. Purple is the rare one and green is rarer still. The ranking shows up on weather sites, in safety guides and in the summary Google writes above them all, and the whole thing is presented as a way of reading a storm off its flashes.

It is the blacksmith’s scale. Heat iron and it goes red, then orange, then yellow, then white, and at the top of a furnace it starts to look blue. That scale is real and it is useful up to a few thousand kelvin. A lightning return stroke runs at about 30,000 K, five times the surface of the sun, and the question nobody seems to ask is whether the scale still has any resolution left up there.

It does not. I computed the thermal colour across the entire range of channel temperatures that spectroscopy has ever measured in real lightning, and from one end to the other it moves by 0.64 CIEDE2000. That is below the threshold at which anyone can see a difference. Then I computed what ten kilometres of ordinary clear air does to the same bolt, and the answer is 23.19, which is 36 times further. The colour of lightning is a measurement of the air in front of it.

The honest hex of the channel

Start with the bolt itself, with no air in the way. A thermal emitter at 30,000 K has one chromaticity, and converting it through the CIE 1931 observer into sRGB gives a single colour.

Look at those three hex codes next to each other. The temperature at the bottom of that list is 8,000 K higher than the temperature at the top, a jump larger than the entire surface temperature of the sun, and the colour has moved by two digits on one channel.

Put a number on it. Between 28,000 K and 36,000 K the chromaticity moves 0.00344 in the CIE 1976 u′v′ plane and 0.636 CIEDE2000. A just noticeable difference is around 2.3 for two colours sharing an edge, and these would not be sharing an edge. They would be separated by one storm and several minutes. Nobody has ever seen that difference and nobody ever will.

For scale, here is what 0.636 is worth in the temperature range people actually have intuitions about.

The third row is the one that gives the game away. Fifteen thousand kelvin to fifty thousand kelvin spans 47 mired, almost exactly the same reciprocal distance as 5,000 K to 6,500 K, and it produces a third of the colour change. Even on the scale built to make colour temperature behave linearly, the top of the range is running out of colour.

Why the scale dies at the top

Colour temperature lives in the reciprocal. That is why photographers use mired, which is a million divided by the kelvin figure. Every equal step in mired is supposed to be a roughly equal step in appearance, and it works well enough from candlelight to daylight.

• 1,900 K, a candle flame, sits at 526 mired
• 3,000 K, a warm bulb, sits at 333 mired
• 6,500 K, average daylight, sits at 154 mired
• 28,000 K sits at 35.7 mired
• 36,000 K sits at 27.8 mired

The entire measured range of lightning channel temperatures is 7.9 mired wide. A single click on the white balance dial of most cameras is five or ten. The reason is simple once you see it: the visible band sits far down the long tail of a 30,000 K Planck curve, where the spectrum is close to the Rayleigh-Jeans limit and the shape across 400 to 700 nm barely changes with temperature at all. The peak left the visible long ago and went into the ultraviolet. Doubling the temperature mostly buys you more ultraviolet that never reaches your eye.

So the folk ranking fails on its own terms. It does not merely get the order wrong. There is no order available. Anything hot enough to be lightning is sitting in the part of the Planckian locus where colour has stopped being a function of temperature. The same compression is visible at the other end of the colour temperature chart, where kelvin turns out to be 52 times non-uniform across its usual range.

The thermal premise was wrong anyway

Everything above grants the folk claim its own assumption, which is that lightning glows the way hot iron glows. It does not. A lightning channel is an air plasma, and air plasma emits in lines.

At 30,000 K most of the nitrogen in the channel is singly ionised, which spectroscopists write as N II. I pulled every N II transition between 360 and 800 nm out of the NIST Atomic Spectra Database, computed the local thermodynamic equilibrium emission for each one, and converted the result through the same observer. One detail makes this cleaner than it sounds: within a single species the partition function is a constant factor, and a constant factor cancels out of chromaticity entirely. The colour of a pure N II spectrum has no free parameters except temperature.

Cyan, not blue. And it is cyan for a specific reason that is worth looking at, because the visible output of ionised nitrogen is not spread across the band. It is concentrated in a handful of multiplets.

Those are wavelengths in nanometres and the percentages are shares of the luminous output. The single line at 568.11 nm carries 16.8 percent of everything your eye receives from ionised nitrogen. Its multiplet carries a quarter, the 500 nm group carries another fifth, and between them those two clusters, both sitting in the green and cyan, account for 48 percent of the visible emission. That is why the computed colour lands where it does.

Here is the part I did not expect. I did not choose those eight wavelength windows. They are the N II features that Kieu and colleagues labelled on their measured spectra of metre-long lightning-like discharges in 2021, widened at each end by the 0.75 nm spectral resolution they report. My line-by-line calculation, which knew nothing about their data, puts 63.5 percent of its luminous output inside those windows, and independently agrees that 567.9 to 571.0 nm is the single biggest feature in the visible band. Two unrelated routes landing on the same ranking is about as much validation as this kind of calculation gets.

How much the species mix matters

One honest caveat. Real lightning is not pure N II. There is neutral nitrogen, singly ionised oxygen, and a hydrogen alpha line at 656.3 nm from water vapour in the air. The cyan figure above is the colour of the dominant visible emitter on its own, and adding the others moves it.

Hydrogen barely registers, which is a small surprise given how prominent that line looks on a spectrograph. Ionised oxygen matters a lot, because its strongest visible lines sit near 465 nm and pull the whole thing blue. So the thin-plasma colour is pinned down to within about 20 CIEDE2000 and no better, and the honest statement is that the sheath around a lightning channel is somewhere between cyan and blue depending on how ionised the oxygen in it happens to be. Compare that with the thermal calculation, which is pinned to 0.64 across all of real lightning. The uncertainty has moved entirely out of the temperature and into the chemistry.

What actually changes the colour

Now put air in front of it. Lightning is normally seen across kilometres, and the atmosphere is a selective filter: it removes short wavelengths faster than long ones, which is the same mechanism behind the blue of the sky and the red of a sunset.

I built the extinction the way the sunset piece did. Rayleigh scattering comes from the Bodhaine formula, which gives a sea-level optical depth of 0.09707 at 550 nm and a horizontal extinction coefficient of 0.01213 per kilometre. Aerosol comes from the Koschmieder relation, which turns a stated meteorological visibility into an extinction coefficient, with an Angstrom exponent of 1.3 for continental haze. Then I ran the 30,000 K channel through increasing path lengths.

This is clear air, a 23 km visibility, the sort of day a forecast calls good.

A bolt two kilometres away is already a just noticeable difference away from its own colour. At five kilometres it has lost most of its blue. At just under ten kilometres it passes through neutral white, and past that it keeps going into cream, amber and orange. The same bolt, the same temperature, the same spectrum, read as six different colours depending on nothing but how far away you were standing.

The crossing point moves with the air. Here is where the bolt turns white under different conditions.

In haze, a bolt two and a half kilometres away is already white, and one five kilometres away is #FFF8F7 going on cream. In very clear air you would have to be nearly eighteen kilometres out, at which point you cannot hear it.

Which gives the comparison that I think is the real answer to the question. The full measured range of lightning temperatures is worth 0.636 CIEDE2000. Ten kilometres of clear air is worth 23.19.

Every row is how many times the whole temperature range that one amount of air is worth. The smallest entry on that list, five kilometres on an exceptionally clear day, already beats the temperature effect six times over. There is no realistic viewing situation in which the bolt’s own temperature is the thing you are seeing.

Heat lightning is just this, named wrongly

The National Weather Service defines heat lightning as lightning far enough away that the thunder is no longer audible, and puts that limit at around ten miles. It is not a kind of lightning. It is ordinary lightning with the sound subtracted by distance.

Sixteen kilometres is ten miles, so the table above has already answered what heat lightning should look like. On a clear night it is #FFE3BD, a warm cream. In light haze it is #FFA949, properly orange. The folk explanation for that orange is usually dust or a dying storm, and the real explanation is that the colour and the silence have the same single cause. If you can see the flash and not hear it, the light has crossed enough air to be reddened, by arithmetic. The two observations are the same observation.

Which lightning colours are impossible

Once the colour is a function of temperature and path length, the set of colours lightning can have becomes something you can map rather than argue about. So I mapped it: every channel temperature from 14,000 to 60,000 K, every visibility from 1 km to effectively infinite, every path length up to the point where the bolt is too dim to see. That is 1.46 million chromaticities.

Taking only the ones with enough chroma to look like a colour rather than a shade of white, the reachable set covers 80.1 degrees of the CIELAB hue circle, which is 22.3 percent of it, in two disconnected arcs:

• a warm arc from 47.7 to 109.0 degrees, running orange through yellow to the edge of yellow-green
• a cool arc from 263.4 to 282.3 degrees, which is the blue the bolt starts at

Between them sit two holes. One is 154.4 degrees wide and contains all of green and all of cyan. The other is 125.4 degrees wide and contains magenta, pink and violet. Lightning cannot thermally reach any of it.

That argument can be made stronger. Adding light together moves you to a point inside the convex hull of what you started with, always, so any superposition of bolts, paths, haloes and reflections is still inside the hull of the reachable set. I built that hull from all 1.46 million samples and tested where various named colours fall. The number is the margin by which each one sits outside, in chromaticity units.

Not one of them is reachable, and the greens are not close. A pure 520 nm green, which no screen can show anyway, misses by 0.528, and even cadet blue, which is barely a colour, misses by 0.035. Green lightning, by the thermal route, is not rare. It is impossible, in the same sense that impossible colours are impossible, which is that no amount of the available ingredients combines into it.

So where does purple come from

Purple lightning gets reported constantly, and it sits squarely in the 125 degree hole. Both of those things are true, so one of my assumptions needs loosening, and the loose one is obvious: I have been treating the bolt as a single colour arriving along a single path.

In haze or rain you do not see only the bolt. You see the bolt plus the glow it scatters into the air around it, and that glow has travelled a longer and more scattered route, so it arrives reddened. Those two arrive together and your eye adds them. A near-blue direct channel plus its own orange halo is exactly the recipe for something the bolt cannot produce alone.

I computed it. The direct component is the channel at 1 km through clear air. The halo is the same channel after 25 km of light haze, which is a reasonable stand-in for a heavily scattered path. The percentage is the halo’s share of the total luminance.

There it is. Between roughly 15 and 30 percent halo fraction the combination sits in the violet and purple band, and the window is narrow, which is the satisfying part. Purple lightning is real, it is a colour neither the bolt nor the air can make by itself, and it requires the two to be mixed in roughly a one to four ratio. Get the ratio wrong in one direction and it is blue, wrong in the other and it is pink. That is a decent explanation for why purple is reported often enough to be a known thing and rarely enough to be remarked on.

Note also that the purple it produces is a pastel. Chroma stays around 30 across the whole window. Saturated violet of the sort a screen can show stays outside the hull and stays unreachable. If a photograph shows you a vivid purple bolt, the vividness came from the camera.

Green is the genuinely rare one

Green survives as the real outlier, and the line spectrum is the only door into it. Recall where the N II emission sits: two clusters at 500 and 568 nm, both in the green and cyan. Run that thin-plasma spectrum through the same atmosphere rather than the thermal one, and it enters the hole the thermal bolt cannot.

That is the whole answer to why green lightning reports are rare, inconsistent and disputed. Producing green needs the line emission to dominate over the thermal continuum, which means a channel segment carrying relatively little current, and it needs a particular amount of air in front of it, and it needs the ionised oxygen fraction to be low enough not to drag the result back towards blue. Three conditions at once, each with a narrow window. Meanwhile the thermal route, which is what you see in almost every flash, cannot get there at all.

Worth adding that most photographs of green lightning are not lightning. A distribution transformer failing throws copper and the copper burns green, brightly and at ground level, and it happens during storms. That is a chemistry colour, not a temperature colour, and it is the same reason copper turns a flame green.

What the calculation assumes

Four things are worth stating plainly, because they are where the numbers could be wrong.

The thermal calculation uses the Planck function as a reference rather than a claim. I am not asserting that the channel is a blackbody. I am asserting that the folk ranking is a thermal claim, so testing it against a thermal spectrum is testing it on its own ground. On its own ground the scale has 21 times less resolution at lightning temperatures than it has between a warm bulb and daylight. The real spectrum is a mix of continuum and lines whose proportions change microsecond by microsecond during a stroke, and the line calculation above is the other bound.

The LTE line model assumes optically thin emission, which the strong lines in a real channel core are not. Self-absorption flattens the strongest features and would pull the cyan towards white. That pushes the true channel colour towards the thermal answer, not away from it, so it does not threaten the conclusion.

The atmosphere is a clean horizontal sea-level path with Rayleigh scattering and continental aerosol. Real lightning is usually seen through or near cloud, and cloud droplets scatter almost neutrally across the visible, so their effect is to dim and diffuse rather than to tint. That is why the halo in the purple calculation uses a long clear-air path as its stand-in. A full multiple-scattering treatment would change the halo fraction needed, not the fact that the window exists.

Finally, the eye. A bolt at close range is many orders of magnitude brighter than anything else in the scene, and vision at that intensity desaturates towards white. So the most common honest report, that lightning is simply white, is correct as a report and tells you about the observer rather than the bolt. The chromaticity is #A2BBFF whether or not your cones are in a position to say so.

The thing worth keeping

What I like about this one is that the folk belief is not a wrong answer. It is a correct answer imported from a place where it worked. The blacksmith’s colour scale is genuinely informative between about 800 and 3,000 K, where 200 kelvin shows up plainly. Carry it up to 30,000 K and the same scale has a resolution of nothing, because colour lives in reciprocal temperature and the reciprocal has collapsed. The belief did not get falsified so much as it ran out of range and nobody checked the gauge.

The second thing is how often the dramatic variable turns out to be geometry. Snow and glacier ice are the same substance and differ only in path length. The ocean’s blue is mostly reflected sky. Lightning runs the same way: the bolt contributes one hex code and the distance contributes everything else. If you want to know something about a storm from a flash, the colour is telling you how far away it is, which is genuinely useful, just not the thing the ranking promised.

If you want to find out how well you can actually read a colour without a reference beside it, that is what this site is for. The colour memory game takes the comparison away and makes you hold a target in your head, which is the hard version and the honest one. Spot the difference keeps the comparison and shrinks the gap until you lose it, which is the closest thing to the 0.64 measurement at the top of this piece, and hue sort will tell you whether small hue steps or small purity steps are what defeat you. There is more in the full set.

For the neighbouring questions, what colour fire is has the same argument for flames and finds that blue flame is too saturated to be a temperature reading either, why sunsets are red has the atmosphere model this piece borrowed, and the visible spectrum covers why none of these pure-wavelength colours fit inside a screen. The method behind every number here is in what CIEDE2000 is.