Why underwater photos look blue

Because water absorbs the long wavelengths first, and red is the longest one your camera records. What is less widely known is that red does not gradually fade with distance. It reaches zero and stays there, often within the first few meters, and the distance at which that happens has almost nothing to do with how deep you are.

Measured on four sites in a public dataset · Last updated 2 August 2026

Water absorbs the long wavelengths first

Light traveling through water loses its long wavelengths fastest. Red goes first, then orange, then yellow. This is why a photograph taken a few meters down arrives blue or green even though your eyes, which adapt, saw something closer to normal at the time.

The loss happens in the water between your subject and your lens, so what governs it is the distance light travels through water, not your depth gauge. A fish a meter in front of you and the reef eight meters behind it are in the same photograph and have lost entirely different amounts of light. That is the part most explanations skip, and it is the part that decides whether a photograph can be corrected.

A diver photographing a crocodilefish on sand: the uncorrected original on the left is almost entirely blue, the restored version on the right shows sand and skin tones. — Left, as the camera recorded it. The average red value across that entire frame was 1.1 out of 255. Right, the same file after the water is reversed.

Red does not fade with distance. It stops.

At every site we measured, the red channel did not trail off toward darkness. It reached exactly zero, and it did so well inside recreational diving range. On a standard color chart photographed underwater, the red patch, the yellow patch and the skin-tone patch all read red = 0. Not dim, not buried in noise. Zero.

This matters more than it sounds. A dim red channel can be amplified back. A red channel that is exactly zero contains no information to amplify, and no amount of processing recovers what was never recorded.

Line-of-sight distance from subject to lens at which the red channel measured zero.
Site	Water	Red measured zero at	Blue vs green
MichmoretMediterranean	Shallow coastal green	3.4 mstill alive at 1.12 m	1.25
NachsholimMediterranean	Green, five meters	4.1 mstill alive at 3.00 m	—not measurable
KatzaaRed Sea	Tropical, 10 to 15 m	≤ 6.1 m	0.73
SatilRed Sea	Tropical, on a wreck	≤ 10.2 m	1.10
		06 m12 mDistance through water

Sites ordered by how soon red died. A bright segment marks distances at which red was measured still present and the red band marks distances at which it measured zero; plain rule between them is distance at which nothing was measured. Where two charts bracket the transition the distance given is the nearer one at which red was already zero. Where red was already gone at the closest chart in the scene, the figure is an upper bound and red may have died sooner.

Light surviving 10 meters of water, by channel

red: 0%
green: 10.5–13.1%
blue: 6.2–10.9%

Measured off the two color charts in the Red Sea wreck scene, which sat at 10.2 m and 10.9 m from the camera; the range is the difference between them rather than a margin of error. Roughly a tenth of the green and a twelfth of the blue made it that far. None of the red did.

Shallow water does not protect red

The shallowest site we measured lost red the fastest. In coastal Mediterranean water, red was still present at 1.12 m from the lens and had reached zero by 3.4 m, in water where a diver is barely below the surface. In clear tropical water on a Red Sea wreck, red survived past 10 m.

So the widespread assumption that shallow diving keeps your color is false. What governs red is the clarity and type of the water and the distance through it, and a murky three-meter dive can strip red more completely than a clear twenty-meter one. In one shallow frame, 94% of the picture had no red left in it at all.

Blue does not always travel furthest

The familiar version of this story ends “and blue travels furthest”. The first half, that red goes first, held everywhere we looked. The second half did not. Which of green and blue survives better changed from site to site, and at one site the order reversed: green was attenuated faster than blue, the opposite of the other two.

Across the three sites where the measurement was sound the ratio of blue to green ran 0.73, 1.10, 1.25. Above one means blue went first; below one means green did. That spread is why a single set of constants cannot describe underwater color, and why the app measures each photograph instead of applying a preset for “tropical” or “temperate” water.

A gorgonian sea fan at depth: the uncorrected left half is very nearly monochrome blue, the restored right half shows the fan's red. — Blue tropical water. Mean red across the original frame: 10 of 255.

A gorgonian reef in green water: the uncorrected left half is a flat green murk, the restored right half shows golden sea rods and sea plumes. — Green water, the same kind of reef structure. Uncorrected it is a flat green murk: 20.3 of 255.

A single photograph is not a single problem

Within one frame, with scene distances running from 6.5 m to 29.7 m, the correction needed to undo the water varied by 113× in green and 254× in blue between the nearest pixel and the furthest. One photograph, one moment, two orders of magnitude of difference across it.

A correction applied evenly to the whole frame can therefore be right in at most one plane of the scene. Whatever value fixes the fish in front of you is wrong for the reef behind it, and the wider the scene, the more wrong it gets.

A reef photographed from above with a diver far out in the water beyond it, uncorrected on the left and restored on the right. — Coral a meter or so away, and a diver far out in the water beyond it, in one exposure. The near coral comes back. The far water stays blue, because by the time its light reached the lens there was no red left in it to bring back.

How this was measured

The measurements were taken on SQUID, the Stereo Quantitative Underwater Image Dataset, published by Berman, Levy, Avidan and Treibitz, which is the reason they can be checked. It contains stereo photographs from four sites in Israel, two in the Red Sea and two in the Mediterranean, each with standard color charts placed in the scene at known distances from the camera, plus distance maps for the scenes themselves.

Color charts at known distances are what make any of this quantitative rather than impressionistic. A grey patch has a known reflectance, so what the camera records from it at a known distance says exactly what the intervening water did. We measured the channel values of those patches, and the attenuation implied between charts at different distances in the same scene.

One limit, stated because it belongs in the result: at the fourth site the blue-to-green ratio could not be measured honestly. Sunlight caustics in shallow water put moving bright bands across the scene, which breaks the assumption of comparing like with like across distance, and the estimate came back unstable in sign across a robustness battery. Rather than publish the single run that looked plausible, that cell is empty.

Dataset: Berman, Levy, Avidan and Treibitz, “Underwater Single Image Color Restoration Using Haze-Lines and a New Quantitative Dataset”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020. 10.5281/zenodo.5744037. The measurements above are ours; the photographs on this page are our own and are not from the dataset.

What this means if you shoot underwater

Get closer than feels necessary. Distance through water is the variable that decides everything else, and halving the water between you and your subject does more for color than any setting on the camera.

Judge a photograph by whether red is dim or absent. If there is any red left, it can be brought back. If the channel is empty, no editing recovers it, and what you are being offered instead is invention.

And do not trust a single preset across a dive, let alone across a trip. The water changed the rules between two sites in the same sea.