One of the key findings of the new study, published in Geophysical Research Letters, is that the way models are “trained” against observations strongly influences their long-term predictions. Models that were constrained using satellite measurements of surface elevation change (how the height of the glacier is decreasing over time) projected the largest future mass losses. Those projections suggest that by 2067, the rate of ice loss from Thwaites Glacier could equal what the entire Antarctic ice sheet currently contributes to sea level rise each year.
By contrast, calibrating models using only ice velocity data (how fast ice is moving toward the ocean) produced lower and more stable future loss rates. This discrepancy shows that what data scientists choose to emphasize when calibrating models can dramatically change the future picture scientists paint for Thwaites Glacier. While both approaches capture important behaviors, the models incorporating surface elevation changes appear to match the recent observed patterns of thinning most consistently.

