Ocean current models vs satellite imagery for predicting debris convergence zones

A drift model will hand you a probability field. A patch of ocean shaded in gradients, high confidence here, low confidence there, built from surface currents, Stokes drift, windage coefficients, and however many particles you seeded the run with. It's a forecast of where debris should be converging, three days out, five days out, based on physics that mostly hold up. What it doesn't tell you is what's floating in that patch of ocean right now. That gap is where a lot of cleanup planning goes sideways.

What a drift model gives you

Lagrangian particle tracking, OSCAR, HYCOM, whatever stack your team runs, the output is a likelihood surface, not a sighting. You seed thousands of virtual particles at known or estimated source points, apply surface current fields and a leeway term for windage, and let the model advect them forward. Convergence zones fall out of that as regions where particle density pools: the seams between current systems, the edges of eddies, the downwind side of a front. For debris with real buoyancy and real windage, those particles are a reasonable proxy for where plastic accumulates over time.

The catch is that the model is only as good as its forcing data and its assumptions about what's drifting. A ghost net behaves nothing like a fishing crate, which behaves nothing like a sheet of microplastic film. Most operational drift products use one or two generic windage classes, so the convergence zone they predict is really the convergence zone for a hypothetical average object.

Storms shuffle things faster than the model updates. A gyre's core can shift position by tens of kilometers between the run you pulled last week and the one you'd pull today. None of that makes the model wrong. It makes it a planning layer, not a sighting report.

Where satellite imagery earns its keep

This is the part a drift model can't do on its own: confirm that something is actually sitting on the surface in the zone it flagged. Floating plastic and organic windrows have a distinct spectral signature against open water, visible in multispectral imagery even where the debris is too thin or too scattered to show up in a standard RGB scene. A wide-swath pass over a predicted convergence zone either lights up with that signature or it doesn't, and that single bit of information changes what a vessel does next.

Satellite coverage is event driven and the resolution runs wide, in the 10 to 30 meter range, so it's not reading individual bottles off the water. What it's good for is telling you whether the windrow the model predicted has formed, and roughly where along that seam it's sitting today, before a boat burns fuel finding out the hard way. Treat the drift model as the shortlist of places worth watching and the imagery as the filter that tells you which of those places has something worth a transect.

Using the two together

Neither one replaces the other, and running only one leaves a real blind spot. A model with no imagery confirmation sends vessels chasing a probability that may have already dispersed. Imagery with no model context means scanning open ocean with no prior on where to even start looking, which is an expensive way to spend swath coverage.

A workable sequence looks like this: pull the current drift forecast for your basin, rank the predicted convergence zones by particle density and persistence, then check the top candidates against recent multispectral passes before committing a route. If a zone shows a signature consistent with aggregated floating material, it moves up the list. If it doesn't, you've saved a day of transit and a tank of fuel without ever leaving port.

Plastic Detection builds that second layer: an event-driven read on where floating debris is aggregating, pulled from the spectral signature multispectral imagery picks up, meant to sit alongside your drift model output rather than replace it. If that's the gap in your current workflow, the early-access list is open.