False-color astronomical images

Why is a picture of a nebula in colors no eye could ever see still honest?

The famous images of the Pillars of Creation are gold and teal in one version, orange and indigo in another. The nebula did not change. Neither picture shows what your eye would see if you were parked beside it, and one of them was taken at wavelengths your eye cannot detect at all.

The instinct is to feel slightly cheated — as though the colours were applied for the poster. But astronomers, who are careful people, publish these images as science and not as decoration. So the question worth asking is not "is it real". It is: what would it even mean for a picture in invented colours to be honest, and what would it take for one to be dishonest?

Start by trying to describe the alternative. Suppose we demanded "true colour" — the picture as the eye would see it. What is that, exactly?

Follow it through. Light arriving from the nebula is a continuous spectrum: some amount of energy at every wavelength. Your retina does not measure that spectrum. It has three cone types with broad, overlapping sensitivity curves, and it reports three numbers. An enormous number of physically different spectra produce identical triples and are literally indistinguishable to you — metamerism, the reason a screen can imitate a sunset with three phosphors. So the eye is already a mapping: from an infinite-dimensional signal to three numbers, chosen not for fidelity but by the accidents of primate evolution.

That reframes the whole problem. "True colour" is not the raw truth against which false colour is a departure. It is one particular projection, with one particular set of things it throws away. Once you see that, the question stops being whether to map and becomes which map, and by what rule.

So consider what a telescope does instead. It puts a filter in front of a detector that counts photons in some band. Choose a filter that isolates the light emitted by ionised hydrogen at 656 nanometres, another for doubly ionised oxygen at 501, another for singly ionised sulphur at 672. Three numbers again — but now each one means something specific, because each traces a particular element in a particular ionisation state. To display them you must assign each to a screen channel, and the traditional narrowband assignment gives sulphur to red, hydrogen to green and oxygen to blue.

Now ask the key question. Has anything been fabricated? The brightness values are measurements. What has been invented is only the colour, and colour is not in the sky in the first place — it is a label your visual system attaches. The image is a deliberate re-coding: it takes three real measurements and puts them into the three channels your perceptual system happens to read fastest. Nothing is added; a different projection is chosen than your retina's.

Which suggests the actual criterion for honesty, and it is not "matches the eye". A representation is honest when the mapping is stated, and when the structure a viewer is invited to read off the picture is structure that exists in the data. If two regions look different in the image, they must differ in the measurement. If a boundary appears, there must be a boundary.

Test that criterion against the practices. The infrared images from Webb are usually assigned by chromatic ordering — shortest wavelength to blue, longest to red — so that the ordering relation in the data survives into the picture, and "redder means longer wavelength" remains a true statement about the image. That is honest by the criterion. Chandra's X-ray images do the same with photon energy bands. And now test the failure case. The rainbow colour scale, once ubiquitous, fails it: because human sensitivity to hue is not uniform, a smooth gradient rendered in rainbow shows sharp-looking bands at the yellow and cyan transitions, so viewers read boundaries in data that has none. That is a genuine defect — documented well enough that perceptually uniform scales were built to replace it — and notice that it fails for a reason having nothing to do with the colours being invented. It fails because the mapping introduced structure.

So the pair of images at the start are not two competing truths. They are two different measurement sets, mapped by two different stated rules. Compare them without knowing the rules and you learn nothing; know the rules and each one tells you something the other cannot.

Three refinements.

First, "false colour" is an unfortunate name that has caused decades of confusion. Nothing is false. The convention distinguishes representative colour, which tries to approximate what an eye would see, from assigned colour, where channels stand for chosen bands. Both are mappings; only the choice of map differs.

Second, honesty is a property of the mapping, not of the palette. A pretty image can be perfectly rigorous and an ugly one badly misleading. The questions to ask are: which bands, assigned how, and stretched how — because the brightness stretch, usually logarithmic to make faint structure visible next to a bright core, is a transformation as consequential as the colour and far less often mentioned.

Third, the honest position on the poster images: they are real measurements, mapped by public rules, and also chosen among many defensible mappings partly because they look striking. Both things are true, and saying so is more useful than defending them as pure data or dismissing them as art.

• How the brightness stretch is chosen, and why it deserves the same scrutiny as the colour assignment.
• Why perceptually uniform colour scales were designed, and what "uniform" is measured against.
• What a species with four cone types would consider the true colour of the same nebula.

Every term the collection defines is gathered in the glossary.