Machine-learned global glacier ice volumes
We present a global dataset of glacier ice thickness modeled with IceBoost v2.0, a machine learning model trained on 7 million ice thickness measurements and informed by physical and geometrical predictors. We model the distributed ice thickness for every glacier in the two latest Randolph Glacier Inventory releases (v6.0 and v7.0), totaling 215,547 and 274,531 glacier outlines, respectively, plus 955 ice masses contiguous with the Greenland Ice Sheet. We find a global glacier volume of $(149 \pm 38)\times 10^3$ km$^3$, consistent with the previous ensemble estimate of $(147 \pm 28)\times 10^3$ km$^3$. The corresponding sea-level equivalent, $323 \pm 91$ mm, is likewise consistent with the earlier value of $315 \pm 63$ mm. Compared to measurements, IceBoost error is 20-45% lower than the other solutions in the high Arctic, highlighting the value of machine-learning approaches. Confidence in our solution is highest at higher latitudes. Over mountainous terrain, small glaciers, and under-represented lower-latitude regions, confidence is lower. IceBoost v2.0 demonstrates strong generalization at ice sheet margins. On the Geikie Plateau (East Greenland), we find nearly twice as much ice as previously reported, highlighting the method's potential to infer bed topography in parts of the ice sheets. The quality of the solutions depends on the accuracy of the training data, the Digital Elevation Model, ice velocity fields, and glacier geometries, including nunataks. Using the Jensen Gap, we probe the model's curvature with respect to input errors and find it is strongly concave over low-slope, thick-ice regions, implying a potential downward bias in predicted thickness under input uncertainty. The released dataset can be used to model future glacier evolution and sea-level rise, inform the design of glaciological surveys and field campaigns, as well as guide policies on freshwater management.