The sea turtle “lost years”

Can you take a lost-years turtle and back-track it to the beach it came from? We measured the answer three independent ways. A perfect-model twin experiment: origin uncertainty exceeds the spacing between nesting beaches within weeks and reaches gyre scale within a year. A model-free measurement from 6,670 real drifter pairs: trajectories that start within 15 km separate beyond rookery spacing in about 13 days. And real tracked turtles, which disperse even faster because they actively swim. That two-week horizon holds on the full 1989–2024 drifter record (171,823 pairs, an order of magnitude more data than the pilot), so the impossibility is not a small-sample artifact.

Turbulent ocean mixing is an information-destroying operation. Forward, it produces a predictable corridor. Inverted, it maps many origins onto overlapping endpoints almost immediately. That is precisely why provenance is assigned with genetics rather than drift, and now we have measured the horizon that forces the choice.

The forward-works, inverse-fails asymmetry is not turtle-specific. It collapses to a single inequality: origin reconstruction is possible only while the age at observation stays below S² / 2K, where S is the spacing between candidate origins and K the dispersal diffusivity of the medium.

This places eel larvae (caught days old, far below the frontier, invertible, which is exactly why our eel inversion worked) and lost-years turtles (observed months old, far above it, erased) on one phase diagram, and predicts the frontier for untested species. It also has a useful corollary: the frontier is scale-dependent, so while the exact birthplace is unrecoverable within weeks, the coarse management-unit origin survives much longer, and combining a genetic signal with a drift signal can assign most turtles to a region.

Virtual hatchlings seeded at the real nesting beaches and drifted through actual ocean currents reproduce the textbook lost-years loop: Florida, into the Sargasso Sea, out toward the mid-Atlantic and the Azores, where oceanic juveniles are really found. Compared like-for-like against real satellite-tracked turtles, the model predicts where they are at AUC 0.91. (An honest caveat: a fully independent test against basin-wide occurrence density is only near chance, so the model reproduces tracked trajectories well and standing occurrence less so.)

As an independent check we used satellite-observed Sargassum, the floating seaweed the hatchlings literally ride. Within the Sargasso nursery the corridor positively tracks the real Sargassum. (A useful aside: the much-publicized post-2011 Great Atlantic Sargassum Belt is a separate tropical feature, not the turtle nursery.)

We drove the dispersal model across 24 cohorts of real 1993–2016 ocean currents. The physical transport corridor that carries the lost years shows no detectable poleward drift (centroid trend +0.07° per decade, not significant), while the thermal envelope migrates poleward under greenhouse forcing. It was tempting to call this “warming in place” — the conditions moving while the conduit stays put.

We could not make it stand up, and we say so plainly. The corridor’s drift flips sign depending on which ocean-current dataset you use: one model flat, an observation-based current drifting equatorward, a second model a noisy non-significant northward. When we extended the model past 2016 — where the Gulf Stream gate that feeds the corridor has begun migrating north — one model hinted the corridor follows it, but the observation-based current did not reproduce that turn.And there is a hard floor: a turtle drifts about five years before you can see where it landed, so the most recent cohorts have not finished drifting, in the model or in the ocean. The post-2016 window is only a handful of points in any dataset, and won’t be resolvable until those cohorts mature around 2028.

So the published paper makes no climate claim — whether warming moves the corridor is left as an explicitly open question. The result we stand behind is the asymmetry above: you can predict where the lost years go, and you can prove you can never trace an individual home. That one does not depend on the climate question at all.

More computing sharpened those bounds further, all against an over-confident reading. Extending the model past 2016, the corridor centroid does begin to shift poleward with the gate, but on only four cohorts, too few to confirm. On an independent observation-based current product the corridor trends the other way (equatorward), so its apparent stationarity is specific to one current dataset. Beneath the surface there is no thermal refuge: at diving depth the corridor warms as fast or faster than the surface, so the relief is a surface-cue story, not a three-dimensional one. And the corridor is quietly losing productivity, with satellite chlorophyll declining about 4 percent per decade, a non-thermal pressure independent of temperature. The honest read: the corridor is less of a fixed conduit than it first appeared, and the relief is narrower than it looked. The durable result remains the asymmetry, not the climate story.

We then pushed to the hardest version of the question: not whether the corridor moved in the record we can see, but whether it holds under the forcing still to come. We drove the corridor model with the forced current change taken directly from climate models, at both coarse (1 degree) and eddy-permitting (0.25 degree) resolution, where the Gulf Stream actually lives. The rest-of-century answer is genuinely uncertain, and we can say exactly why. The corridor tracks its current, so the verdict turns entirely on how far the forced Gulf Stream shifts, and that is the one thing the models will not agree on. Coarse models overstate the shift about two-fold, with a stiff and mis-placed Gulf Stream. Resolving it halves the projected shift and tips the eels and the lobster toward warming-in-place holding, while the turtle, whose corridor is the most sensitive, still leans the other way. Even the highest-resolution models that exist disagree by a factor of several, and the truly eddy-resolving forced ocean currents that would settle it have never been produced. So the future of this pattern rests on a property of the forced circulation that current data cannot measure, and we say that plainly rather than pick a side.

Putting real seafloor depth under every tracked turtle, the species split is stark: loggerheads spend the lost years over the deep open ocean (median 1,457 m), while Kemp’s ridley stay over the continental shelf (median 92 m). The modeled corridor sits over a median depth of nearly 5,000 m.

That one fact ties together two puzzles. Kemp’s ridley are the turtles that get cold-stunned in New England bays, because they are a shelf species, trapped when coastal water cools rapidly. And the Gulf of Mexico is the exception to the warming story, because its signature lost-years turtle is not an open-ocean drifter at all. Warming’s effect on cold mortality even flips sign between fixed structures (relief) and expanding range edges (a trap). The framework describes oceanic dispersers, and naming that boundary makes it more correct, not less.