Experts fed 18 years of satellite images of 3 Red Sea reef sites through a machine learning model, and two of the three had lost more than a tenth of their coral before any dredger reached that coast

Along one stretch of Saudi Arabia’s northern Red Sea coast, a cutter suction dredger works a channel.

Behind it the water carries a pale plume that spreads with the current.

Somewhere under that water sits reef, in the shallowest and most light dependent band there is.

The two things are a few miles apart.

What nobody has published is a measurement connecting them.

What has been published is something else, and it is worth reading carefully.

What sediment does to an animal that farms light

A reef building coral is a partnership. The animal builds the structure and algae living inside its tissue make most of its food from sunlight.

That arrangement makes light the currency, and anything between the sun and the tissue is a tax on it.

Suspended sediment scatters and absorbs light before it arrives, so a colony under a plume is running at a fraction of its normal income.

Settled sediment is worse. Silt lands in the hollows and on the flat upper surfaces, and clearing it costs mucus and energy the colony has to take from growth.

Below a certain light level and above a certain silt load the partnership stops paying, the algae are expelled, and what is left is a white skeleton.

None of that requires a spill or a chemical. It only requires cloudy water for long enough.

What a satellite can and cannot see down there

A Landsat pixel covers about 100 feet on a side, which is a large area of seabed to reduce to one classification.

Light also has to make the round trip. It passes through the atmosphere, through the water column, off the bottom and back, and the water takes a bite out of it in both directions.

So a satellite is not seeing coral. It is seeing a color signature and being asked whether that signature belongs to coral, sand, rubble or seagrass.

Doing that consistently across two decades means correcting for tide, sun angle, water clarity and two different sensors.

Which is why this kind of work is done by a trained classifier rather than by eye.

It also means the answer comes with an accuracy figure attached, and that figure is part of the result.

What the imagery actually showed

The study compared imagery from 2000 against imagery from 2018, an eighteen year window across three reef sites.

At Al Wajh on the Saudi coast, coral cover fell 13.6 percent. In the Gulf of Aqaba it fell 11.4 percent.

At Umluj it fell 3.4 percent, which is a different order of change entirely and does not belong in the same sentence as the other two.

The method was a support vector machine trained on classified imagery and then applied to a site it had not seen.

Its accuracy was 78 percent at Aqaba and about 73 percent at Umluj.

That matters when reading the smallest number. A 3.4 percent shift measured by a classifier that is right roughly three times in four is not a finding you can lean on.

The eighteen years and the two years that do not overlap

The imagery window closed in 2018.

The marine works at the port were awarded in 2023 and the berths were built across the two years to late 2025.

So whatever the satellites recorded, the dredgers were not there for any of it.

The losses are real and the port is real, and the study does not connect them.

What the region did experience in that window is repeated thermal stress, the same mechanism now bleaching reefs on the other side of the world.

Reading the decline as construction damage is a story the published work does not support, however tidy it looks on a map.

What is actually being built on that shoreline

The port is an expansion of the existing harbor at Duba.

The completed marine package runs to more than 2.8 miles of quay wall across seven berths, at depths from 34 to 61 feet.

The container terminal is 2,953 feet long and rated at 1.5 million standard containers a year, with operations starting in 2026.

The octagonal floating industrial platform that dominated the early renderings has not been built.

None of this is small, and none of it has been measured against the reef in public. The contractor reports zero material discharge and reuse of dredged material, which is a stated practice rather than an audited outcome.

The honest position is that the satellites have given a good baseline for these reef sites and nothing more, and closing the gap would take turbidity logging and repeat surveys through the construction window, the way better imaging keeps improving what a camera can resolve underwater.

Until somebody publishes that, the port and the coral are two separate stories that happen to share a coastline.