How much does wind push floating debris? We scored ourselves.

Anything floating at sea gets carried two ways: by the water it sits in, and by the wind pushing on whatever sticks up above the surface. Windage is the share of the wind's speed a floating object picks up. A piece riding low in the water feels almost none of it. A light piece riding high can move at 4 to 6 percent of the wind speed, and over a five-day drift that is the difference between staying far out at sea and landing on a beach.

When a wind turbine blade breaks over the sea, the debris is a mix: heavy sections that float almost entirely underwater, shell fragments that sit at the surface, and light core foam that rides on top of it. Each behaves differently in the wind. Treat them as one object and you get one confident, wrong answer.

What happened. A blade broke at the He Dreiht wind farm in the German North Sea. A 20-meter section floated off, almost entirely underwater, and drifted for about five days before a German Federal Police patrol vessel found it between two wind farms.

What we asked. We ran our drift model four times, each with a different windage setting, using the real wind and current data for those days. Which setting put the modeled section closest to where it was actually found, at the time it was found?

What it means. The wind barely mattered, and that makes physical sense: this piece floated almost entirely underwater, so the wind had almost nothing to grab. For a big submerged section, follow the current and nearly ignore the wind.

Read the numbers with their uncertainty. Neither the exact failure position nor the exact interception position is public; the scoring point is a between-farms estimate with roughly 15 km of stated uncertainty, and the spread between the class fits above is of the same order. So the ranking is real but not sharp: the low-windage end fits this object class best, and no single kilometer figure on this table deserves more precision than its inputs carry.

What happened. A blade failed at Vineyard Wind, off Massachusetts. Over the following week, foam and shell fragments washed onto Nantucket-area beaches, and the town published dated notices of where debris arrived: Tuckernuck and Muskeget on July 18, Nobadeer on July 20.

What we asked. Same question, run backwards in time with the real 2024 weather: which windage setting lands modeled debris nearest the right beaches on the right days?

What it means. The opposite answer from Case 1. Here the strong-wind settings won, and pretending the wind away was four times worse. The beach debris was dominated by light foam riding high on the surface, exactly the material the wind pushes hardest.

The table above already includes a correction, and the correction has a story. Our first run of this case used reanalysis wind as published. Its best track came no closer than about 12 kilometres to Nantucket while the real debris was on the beaches. We published that miss, then checked the forcing against reality.

Two NOAA buoys sat inside the event window. During the critical ten hours, the reanalysis wind showed a northwest peak the buoys never recorded, and it brought the wind back to south southwest three to four hours late. We replaced those ten hours of reanalysis with the observed buoy wind and changed nothing else. The closest approach dropped from 12 kilometres to under 4.

Still a miss, and we say so. The corrected tracks pass just offshore, and the remaining gap now carries the signature of the current field rather than the wind. And the family table held: rerun on the corrected wind, the best windage setting keeps the same 10.2 km score it had before, while the standard planning class tightens toward it. A headline number that survives its own correction is worth more than one that was never questioned.

The core of a large blade contains closed-cell foam. Closed-cell means the bubbles inside are sealed: it cannot soak up seawater the way a sponge does, so it keeps floating, day after day, riding high where the wind can push it. That is why foam showed up on beaches first and kept arriving in the days and weeks after the failure, while heavier shell pieces trailed behind and the big sections barely left the current's path.

So a real debris forecast is several forecasts in one. Foam runs ahead, downwind, at 4 to 6 percent of wind speed. Surface shell fragments follow at a slower rate. Submerged sections follow the current and nearly ignore the wind. Search planes, boom crews, and town notices each end up chasing a different part of the same debris field, and the forecast should tell each of them where their part is going.

One input we refuse to guess at: how fast other core materials, like end-grain balsa, take on water once the shell is breached. There is no good published data for debris-sized pieces. We are measuring it ourselves in bench tests, and the numbers will be published on this page as inputs anyone can check.

The July 2026 event has also been reconstructed independently by others, using a different modeling tradition, without either side seeing the other's work first. The overall picture agrees with ours. For the wind-caught light material, the independent reconstruction runs somewhat faster, reaching the coast on the order of 24 to 48 hours earlier, a spread that is small next to the total drift time.

One comparison caveat we state against ourselves: "arrival" is not one number. In our own runs, the leading edge of the light-material cloud and the closest approach of its bulk differ by several days and by coastal sector. A between-model spread of one to two days therefore sits inside the spread that a single model already carries between its first fragments and its main mass. Since no exact arrival ground truth exists for this material, we read the spreads themselves as the honest answer: plan beach response against the leading edge, plan recovery effort against the bulk, and treat the earlier edge of every window as the right side to be wrong on.

We also replayed both cases through a second, independent model tradition: the Leeway drift model used in search and rescue planning, with its published object library, exactly as shipped. Same currents, same wind, nothing adjusted to make the comparison friendly.

Where the two traditions overlap, they agree. For the North Sea blade section, the search and rescue model's zero windage class lands 8 km from the vessel recovery position, inside the stated observation uncertainty, and its ordering matches ours: submerged object, follow the current. For the Nantucket foam it flips the same way reality did: zero windage misses worst, wind driven classes do best. For the Nantucket case both sides of this comparison ran on the original pre-correction forcing, so the wind correction above affects neither side; the finding is about the models, and it holds either way.

Where they differ is reach. The search and rescue library's classes top out below the wind response the surfaced foam actually showed. Its best class stops tens of kilometres short of the dated beach notices that our five to six percent setting matched to within a few kilometres. That difference is the reason this page shows a family of windage settings instead of a fixed object list: real debris does not always fit a catalogue.

One honest technical footnote. That model's coefficients were calibrated from field drifters whose motion already included wave drift, so it deliberately ignores explicit wave forcing. We verified this the hard way: adding wave transport on top made its North Sea answer worse, not better. The disagreement taught us more than the agreement did.

One interception point and three dated beach notices are a small test: enough to score settings against each other, not enough to calibrate a physics model. Position uncertainties are stated per case and carried through the scoring. The failure times are not known to the minute, so we model a spread of release times, not a single instant. And these are after-the-fact scorings against known outcomes: nothing on this page claims we predicted anything in advance. When the next incident grades us, the score goes up here whether it flatters us or not.