Thwaites Glacier in Antarctica is one of the most concerning glaciers when it comes to sea level rise. It’s losing ice rapidly—over five times faster than it was in the 1990s. To estimate how much ice it might lose in the future, scientists rely on computer models that must be guided by real satellite data.

In this study, we compared different ways of training these models using two types of satellite information: ice movement and changes in surface height over time. Surprisingly, the models trained only with surface height data gave much more accurate predictions than those using ice speed. These better-performing models suggest that by 2067, Thwaites could be shedding 180–200 billion tons of ice each year—similar to the current total loss from all of Antarctica.

The models also show that ice thinning could extend deep inland through a valley beneath the glacier, possibly triggering unstoppable ice loss. These findings highlight how the way we train models can significantly change the results. Accurate predictions are essential, as a full collapse of Thwaites could lead to more than two feet of global sea level rise.