Thwaites Glacier is losing ice more than five times faster than it did in the 1990s. Goldberg and colleagues at Edinburgh modeled its future trajectory and arrived at a projection: by 2067, Thwaites alone could be shedding 180 to 200 gigatonnes of ice per year — roughly equal to the current annual ice loss from the entire Antarctic ice sheet.
One glacier matching the output of a continent.
The mechanism is marine ice sheet instability. Thwaites rests on bedrock that slopes downward as you move inland. When warm water melts the grounding line — where the ice lifts off the seabed and starts floating — the retreat exposes deeper bedrock, which means thicker ice, which means a larger cross-section of ice in contact with warm water, which means faster melting. The retreat feeds itself. Deep troughs beneath the glacier act as channels for this feedback, concentrating the thinning along specific pathways that propagate inland.
But the paper's sharpest finding is methodological. How you calibrate the model determines what it predicts. Models trained on satellite measurements of surface elevation change — how fast the ice is thinning — produce the largest future loss estimates. Models trained on ice velocity data alone — how fast the ice is moving — produce smaller, more stable projections. The two observational datasets are describing the same glacier, but they constrain the model differently, and the difference in projected ice loss is not small.
The measurement frame is as load-bearing as the measurement. Thwaites is retreating either way, but whether the outcome is dramatic or catastrophic depends on which satellite product you trust more. The glacier doesn't know which model you're running. The modeler's choice of training data shapes the answer more than the glacier's physics does — which is another way of saying that the physics is underdetermined by the observations we have.