Telling plastic debris from kelp rafts in satellite imagery

Anyone who has stared at a Sentinel-2 scene after a storm knows the problem. A long, thin streak sits on the water exactly where you'd expect a debris line to collect, and you can't tell from the thumbnail whether it's a plastic mat worth a vessel diversion or a raft of torn-loose kelp doing the same thing in the same current.

Both show up for the same physical reason. Wind-driven Langmuir circulation pushes buoyant material into long parallel bands, the windrows you see converging at odd angles across a bay or along a frontal boundary. Plastic bottles, fishing gear, kelp, pumice, foam: anything that floats gets swept into the same lines. Geometry alone won't separate them. You need the pixels themselves to tell a different story.

What the spectral signature shows

Live kelp is vegetation. It carries chlorophyll, and chlorophyll has a signature that's hard to fake: strong absorption in the red band, a sharp jump in reflectance right at the red edge, and a secondary bump in the near-infrared from the leaf (or in this case, blade) structure. A Macrocystis or Sargassum raft that's still mostly alive will show that red-edge rise pretty clearly, even smeared across a 10 to 30 meter pixel that's mixing in open water on either side.

Plastic doesn't do that. Most common polymers reflect more flatly across visible and NIR, with brightness driven by the plastic's own color and how clean or biofouled the surface is rather than by a pigment absorption feature. A dense mat of consumer plastic and net debris usually reads as broadly bright without the chlorophyll dip-and-rise, which is the main thing that lets a multispectral pass flag it apart from a living kelp raft in the same windrow.

The catch is degradation. Kelp that's been dead and bleaching on the surface for a week loses chlorophyll and starts reading flatter, closer to the plastic side of the spectrum. Biofouled plastic covered in algae can pick up a faint red-edge signal of its own. At the edges of the brightness range, the two start to converge, and no amount of band math fully closes that gap from orbit.

Where false positives still creep in

A few things reliably fool a floating-debris flag, and it's worth knowing them before you commit a vessel:

Sun glint off a calm patch of water can mimic the brightness of a dense debris mat, especially near the specular point in a wide-swath pass. Foam lines from breaking waves or ship wakes aggregate the same way windrows do and can sit right on top of a real debris line, making it hard to say which part of the signature is foam and which is plastic. Sediment plumes near a river mouth or after heavy rain carry their own flat, bright signature that can read similarly to a debris mat at 10 to 30 meter resolution, particularly where pixel mixing blends plume edge with open water. And large, dense Sargassum blooms, now a recurring feature across parts of the Atlantic and Caribbean, are themselves worth flagging for different reasons but get lumped in with plastic aggregation alerts if the spectral read isn't checked against what's known about bloom timing in that area.

None of this means the spectral approach doesn't work. It means a flagged event is a strong reason to look closer, not a finished identification. The pixel tells you something is aggregating and gives you a read on whether it leans toward vegetation or something flatter and more plastic-like. It doesn't replace someone on deck with binoculars or a drone pass confirming what's actually in the water.

That's the gap Plastic Detection is built to sit in: an event-driven alert, built off that same spectral contrast, pointing a research or cleanup team at a coordinate worth checking before they burn fuel steaming out to look at every windrow a storm leaves behind.

If you're scoping the next cleanup run and want a shortlist of aggregation events instead of a blank chart to search, that's the gap worth closing first.