The standard answer is that leaves are full of chlorophyll, chlorophyll is green, and when it drains away in autumn the yellows and oranges underneath are revealed. That answer is half right, and the half that is wrong is the interesting half.
I wanted numbers rather than a story, so I ran a leaf through a physical model of its own optics and measured what each pigment actually does to its color. Three things came out that the standard answer does not prepare you for. A green leaf absorbs 78.9 percent of the green light that lands on it. A tree can lose 71.2 percent of its chlorophyll before it has made it halfway to gold. And red is not an unmasking at all: across 2,594,025 combinations of pigment and structure, no anthocyanin-free leaf gets within 22 CIEDE2000 units of a red maple.
How I measured it
This runs on the same spectral toolkit as what color the sun is and why the sky is blue, with the atmosphere swapped out for a leaf.
- The leaf. PROSPECT-D, a radiative transfer model that takes a leaf apart into chlorophyll a+b, carotenoids, anthocyanins, brown pigment, water, dry matter, and a structure parameter for how many air and cell interfaces the light has to cross. It returns reflectance and transmittance from 400 to 2500 nm. I use the published specific absorption coefficients from Feret and colleagues without touching them.
- The light. D65 for the color numbers, and the ASTM G173-03 global tilt spectrum when I count photons, because a photon budget should be against real sunlight rather than a studio illuminant.
- The color. CIE 1931 two degree observer, sRGB, and CIEDE2000 for every difference quoted below. The just noticeable difference threshold of 2.3 units comes from the JND measurements on this site.
- The summer reference leaf. Chlorophyll 40 micrograms per square centimetre, carotenoids 10, no anthocyanin, structure 1.5. Those are ordinary values for a healthy broadleaf in July, not a tuned fit.
Two checks say the model is behaving. The summer leaf renders as #4A6133, a dark olive, which is what a leaf in hand actually looks like rather than the bright green of a crayon. And with chlorophyll set to zero it renders as #C79121, an autumn gold, without anyone telling it what autumn is.
One honest limit. A single leaf lit by an even sky is not a canopy in low October sun, where shadow, gloss and sky color all get a vote. Every number here is the leaf itself, which is the part the pigments control.
A green leaf is not green by reflecting green
This is the claim I most wanted to check, because the way it is usually told is that chlorophyll absorbs red and blue and reflects green, which makes it sound as though green light bounces off a leaf the way it bounces off a mirror.
Here is what the summer leaf does with the sunlight that reaches it.
- Blue, 400 to 500 nm: 95.5 percent absorbed, 4.2 percent reflected
- Green, 500 to 600 nm: 78.9 percent absorbed, 10.9 percent reflected, 10.2 percent transmitted
- Red, 600 to 700 nm: 90.6 percent absorbed, 5.5 percent reflected
Across the whole visible range the leaf swallows 87.8 percent of the photons that arrive. Even at 550 nm, the exact middle of the green window, reflectance is only 14.9 percent. A leaf is not reflecting green. It is absorbing green slightly less enthusiastically than it absorbs everything else, and that small margin is the entire color.
Sort the absorbed photons by band and it gets better. Of every photon the summer leaf takes in, 31.8 percent are green and only 29.3 percent are blue. The leaf absorbs more green light than blue light while looking green, because there is more green light in sunlight to begin with. The pigment is fussy, the sky is not.
The chlorophyll coefficients show why the margin exists. Absorption peaks at 436 nm in the blue and again at 676 nm in the red, and its lowest point across the visible range is at 525 nm, where it is only 4.8 percent of the blue peak. That is a deep dip in the pigment. It arrives at your eye as a 15 percent reflectance because the leaf is stacked with enough chlorophyll to catch most of the green anyway, and because light bouncing between cell walls gets many chances to be absorbed before it escapes.
There is a companion story that plants are green because they reject the sun’s peak output, which is supposedly green. That premise is false, as the solar spectrum measurement shows: per nanometre of wavelength the sun peaks at 451 nm, in the blue, and the peak moves to the infrared if you write the axis in frequency instead. Whatever plants are doing, they are not dodging a green spike that is not there.
The gold was there all summer
Now the part of the standard answer that survives. Take the summer leaf, hold the carotenoids where they are, and remove the chlorophyll.
The third row is the control, and it settles the question. A leaf with no chlorophyll and no carotenoid is a pale greenish grey with a chroma of 10.5. Put the carotenoid back and it jumps to a gold with a chroma of 62.9, 24.82 CIEDE2000 units away. The autumn yellow really is a pigment that was sitting in the leaf all along, and it really is revealed rather than manufactured.
What is not true is that it was invisible. Comparing the summer leaf with and without its carotenoid, the pigment is worth 10.0 CIEDE2000 units and pulls the hue angle 15 degrees toward yellow. A leaf with no carotenoid renders #2E7635, a cleaner and lighter green than any real leaf. The carotenoid is not hidden in summer. It is doing what a dark filter does in a photograph, muting and warming, and you only read it as yellow once the green stops shouting.
A fair objection: carotenoids do not really sit still. Keskitalo and colleagues followed a single aspen through senescence and found lutein and beta carotene degrading in parallel with chlorophyll. So I ran the parallel case too, with carotenoid falling from 10 to 5 while chlorophyll goes to zero. The leaf still lands on gold, #C39C3F instead of #C79121, chroma 52.7 instead of 62.9. The gold does not depend on the carotenoid staying put. It depends on chlorophyll leaving faster.
Nothing happens for weeks, then everything happens at once
This is the result I did not expect. Colour change is not proportional to chlorophyll loss, and it is not even close.
The whole journey is 35.66 units. Halfway along it, at 17.83 units, the leaf still holds 11.5 micrograms of chlorophyll per square centimetre. Which is to say: half the color change waits until 71.2 percent of the chlorophyll has already gone. The last eighth of the pigment does more visible work than the first half of it.
The reason is saturation. A summer leaf carries far more chlorophyll than it needs to black out the red and blue, so the first losses come out of a surplus and change nothing you can see. The first just noticeable difference, 2.3 units, does not arrive until 14.4 percent of the chlorophyll is gone. Along the full path from green to gold there are only 16 distinguishable steps, and most of them are crowded into the final stretch.
Anyone who has watched a hillside turn knows the feeling this produces. The woods look stubbornly green for weeks, then seem to go over in a few days. Nothing sudden has to happen in the tree for that to be true. A steady, boring, linear drain of pigment produces exactly this experience, because your eye is reading the tail of the curve.
The change is mostly happening in a color nobody names
We describe autumn as leaves turning yellow, so you would expect the big spectral move to be in the green to yellow region. It is not.
- At 550 nm, green: reflectance goes from 14.9 to 38.6 percent, a factor of 2.6
- At 660 nm, red: reflectance goes from 4.1 to 44.7 percent, a factor of 10.9
- At 450 nm, blue: 4.1 to 4.9 percent, essentially no change
The dominant physical event of autumn is a leaf becoming a good reflector of red light, because chlorophyll’s 676 nm absorption band is the thing that disappears. The carotenoid that stays behind keeps absorbing blue, so blue never opens up. Red opening while blue stays shut is what yellow is. The name we give the season describes the sum rather than the move.
It also explains why the leaf gets so much lighter. Overall absorption of visible photons falls from 87.8 to 40.6 percent. A golden leaf is not just a different hue, it is throwing back more than twice as much light, which is why a stand of birches in October reads as luminous rather than merely yellow.
Red is a new build, not an unmasking
Plenty of explanations treat red the same way as yellow, as another pigment revealed by the retreat of the green. It is worth stating plainly that this is wrong, and it is testable.
Anthocyanins are made in autumn, from sugars trapped in the leaf, at a point when the tree is dismantling everything else. So I asked the model: can any leaf without anthocyanin be red? I swept 2,594,025 combinations covering chlorophyll from 0 to 80, carotenoid from 0 to 30, structure from 1.0 to 3.0, and the full plausible range of water and dry matter, with anthocyanin fixed at zero and no brown pigment.
- Lowest hue angle reachable: 66.8 degrees. That is still an orange yellow, #A46829.
- Highest redness a* reachable: 23.14, and it arrives at hue 71.1 as the orange #D9880C, not as a red.
- Closest approach to a red maple leaf: 22.12 CIEDE2000 units, which is roughly ten times the threshold at which two colors stop looking the same.
Allowing brown pigment as well, which is a decay product rather than anything the summer leaf was hiding, the sweep can crawl to within 8.23 units, and it does so by producing a dark brown, #604137, at the extreme edge of the parameter space. Even cheating toward death does not make a scarlet.
The de-greened summer leaf is 43.29 units from a red maple. Whatever the red is, it is not what was underneath.
What the new pigment actually does
Anthocyanin absorbs at 535 nm, which is green light, almost exactly where chlorophyll is weakest. Autumn red is made by plugging the gap chlorophyll leaves behind.
The measurement that makes the mechanism obvious: a red maple leaf reflects only 4.5 percent at 550 nm. A green summer leaf reflects 14.9 percent there. The red leaf is darker in the green than the green leaf is, by a factor of three. This is the cleanest illustration I have found of a rule that runs through every article on this site, from what color a mirror is to why the ocean is blue: the color of a thing is set by what it removes, not by what it contains.
It also explains why autumn reds are dark. Anthocyanin adds absorption to a leaf that was busy getting brighter, so lightness drops from L* 58.6 to 30.6 as the pigment builds. Gold is a leaf giving light back. Red is a leaf taking more of it, and it costs two thirds of the lightness to do it.
Why a dying leaf would bother making a pigment
A leaf about to be discarded spending energy on new chemistry is odd enough that it has its own literature. The leading explanation is photoprotection. Autumn is when the tree pulls nitrogen and other nutrients back out of the leaf, and that recovery runs on cellular machinery that is easily damaged by bright light once the photosynthetic apparatus is being dismantled. Cold, sunny days are exactly the worst combination. Hoch, Zeldin and McCown argued that anthocyanin acts as a sunscreen over the top of the leaf during that window, and the timing fits: the pigment shows up in the upper cell layers, in exposed leaves, on the sunny side of the tree.
The measurement here is consistent with that reading. Anthocyanin sits at 535 nm, which is a wavelength the leaf can no longer use once chlorophyll is going and which would otherwise pass straight through to the tissue underneath. It is well placed as a screen and badly placed as a signal to anything with human style color vision, since it makes the leaf darker rather than more conspicuous.
There is a competing idea, from Hamilton and Brown, that bright autumn color is a signal to insects about the tree’s defensive strength. It is not settled, and I am not going to pretend a color model can settle it. What the model can say is that the yellow and the red are two completely different kinds of event, and any explanation that treats them as one process is starting from the wrong place.
The autumn palette, measured
Pigment mixes for the usual suspects, each one rendered from the model rather than eyedropped from a photograph.
Two things stand out in the pairwise distances. Gold and scarlet are 38.81 units apart, which is about as far as two natural surface colors get from each other. But sugar maple orange and oak russet are only 2.93 units apart, barely over the threshold where two colors stop being the same one. Autumn has a vocabulary of half a dozen words for a region of color space that contains, in the strict sense, about two colors and a lot of lightness.
Notice also how dark all of this is. Not one of these leaves has a lightness above L* 59. The autumn palette as we picture it, and as it gets rendered on packaging and greeting cards, is considerably brighter and more saturated than the leaves themselves. That is the same overshoot measured in memory colors, where remembered versions of familiar objects come back more saturated than the originals every single time.
What to take from it
If you want the short version: chlorophyll is not a green filter over a yellow leaf, it is an almost total absorber with one weak spot at 525 nm. Autumn removes the absorber, which mostly opens up red light, which combines with the carotenoid still blocking blue to make yellow. And the reds are a separate, deliberate, freshly built thing that works by absorbing green harder than the chlorophyll ever did.
The claim I would most like to retire is that leaves reflect green. They absorb four fifths of it. Almost every everyday color works this way, which is why thinking in terms of subtraction rather than emission gets you much further with real surfaces.
The other thing worth keeping is the shape of the curve. A change that looks sudden can be produced by a process that is perfectly steady, as long as the thing you are watching saturates. That is worth remembering the next time something appears to happen overnight.
If you want to feel how small some of these distances are, the 2.93 units between orange and russet is the kind of judgement the color memory game is built around, and it is much harder than it sounds when the two swatches are not side by side. The gradient mode walks a ramp of the sort a leaf takes through October. Shades of orange and shades of brown cover the naming end of the same region, and training your eye for color covers why the muted, low chroma part of the space is where people improve fastest.