IceBoost v2.0 maps global glacier ice volume with machine learning, Scientific Data study says

A study led by Ca' Foscari University of Venice with Italy's Institute of Polar Sciences (CNR-ISP) released an updated global map of glacier ice volume built with a machine-learning model called IceBoost v2.0, phys.org reported, citing work published in Scientific Data. Physicist Niccolò Maffezzoli developed IceBoost v2.0, trained on more than 7 million ice-thickness measurements and 26 physical and geometrical variables including slope, curvature, ice velocity, and temperature. The model reconstructs thickness and volume for every glacier in the Randolph Glacier Inventory, excluding the polar ice sheets. An interactive web app lets users explore the results glacier by glacier. The study estimates the world's glaciers hold about 150,000 cubic kilometers of ice, equal to 32.3 centimeters of global mean sea-level rise if fully melted, excluding Antarctica and Greenland. That global total aligns with earlier estimates, but IceBoost v2.0 matches field observations up to 40% more accurately on where ice sits. On eastern Greenland's Geikie Plateau, where ice is up to 2 kilometers thick, the model estimates nearly twice as much ice as previously reported. Maffezzoli said glacier thickness distribution is fundamental for glaciological and climate models and that researchers in GlacierMIP4 will use IceBoost v2.0 as their sole representation of the present-day situation for IPCC-facing simulations through 2100. The maps also flag where more field data are needed, including the Himalaya, Karakoram, and major Patagonian ice fields. Glaciers support freshwater for about 1.9 billion people, the writeup said. DOI: 10.1038/s41597-026-07744-9.