GlobalWarming & ClimateChange News Desk – AI reveals how much ice is really left in the world’s glaciers
A towering glacier reveals only part of its ice, while new AI mapping shows how much remains hidden below.
AI mapped more than 215,000 glaciers worldwide, revealing their hidden thickness, total ice volume, and where the biggest uncertainties remain.
How much ice is left in the world’s glaciers, and where exactly is it sitting? A new machine learning model gives one of the most detailed answers yet.
It also does something unusual for a global dataset. It marks the spots where its numbers should not be trusted.
The model is called IceBoost v2.0. Niccolò Maffezzoli, a physicist at Ca’ Foscari University of Venice, built it in collaboration with the Institute of Polar Sciences at Italy’s National Research Council.
IceBoost learned from more than seven million ice thickness measurements. Most of those came from ground-penetrating radar, either dragged across the ice by snowmobile or flown overhead on survey aircraft.
NASA’s Operation IceBridge flights supplied a large share of the polar data.
The training set covers 1,661 glaciers. That sounds substantial until you check the denominator. Fewer than 1 percent of the world’s glaciers have ever been measured from the inside.
What the totals show
The model, applied to the 215,547 glacier outlines in version 6 of the Randolph Glacier Inventory, estimates the global glacier volume at roughly 36,000 cubic miles (150,000 cubic kilometers).
If that land-based ice were lost, it would contribute about 12.7 inches (323 millimeters) to global sea level.
That total barely moved. Two earlier global estimates landed at 141,000 and 158,000 cubic kilometers, so the new figure sits between them.
The advance is not the sum. It is the shape.
Where the ice actually sits
IceBoost predicts thickness point by point on a grid spaced every 330 feet (100 meters), using 26 inputs per pixel.
Surface slope, curvature, ice velocity, air temperature, distance to the nearest bare rock, and distance to the ocean all feed into the calculation.
Two gradient-boosted decision tree systems run in parallel, and their answers are averaged.
Against field measurements in the high Arctic, IceBoost’s error was 20 to 45 percent lower than that of competing models. Elsewhere it was comparable. The gains concentrate where the training data is thickest.
One region stands out. On the Geikie Plateau in coastal East Greenland, IceBoost finds nearly twice as much ice as earlier maps reported, with thicknesses reaching roughly 1.2 miles (2 kilometers).
Bed topography under that plateau has been poorly constrained for years, and the model says it is deeper than assumed.
“The distribution of glacier ice thickness is a fundamental variable,” said Maffezzoli, the study’s lead author.
A map of its own doubt
Every glacier file in the release ships with two extra layers that most datasets skip. One is a per-pixel uncertainty estimate, built by perturbing all the inputs 50 times and watching how much the answer wobbles.
The second is stranger – and potentially even more useful.
The team calls it the Jensen Gap. It measures whether the model’s response to input error is lopsided, and over thick ice on gentle terrain, it very much is.
Nudge the slope upward slightly, and the predicted thickness drops hard. Nudge it downward by the same amount, and the thickness gains far less.
That asymmetry has a direction. Where the ice is thickest and flattest, IceBoost is likely underestimating thickness. Publishing that admission alongside the numbers is rare, and it tells modelers exactly which pixels to treat carefully.
Confidence runs highest across the Arctic, where decades of airborne radar have sampled the ice thoroughly. It falls over steep mountain terrain, over small glaciers, and across lower latitudes where almost nobody has flown a radar line.
Where to fly next
The uncertainty maps double as a shopping list for field campaigns.
Three areas come up repeatedly as data-starved: the Himalaya, the Karakoram, and the two Patagonian icefields.
Those are also, awkwardly, among the regions where glacier loss carries the heaviest downstream consequences.
Survey flights cost money and helicopter time. Knowing which valleys would most improve the global picture is a practical result, not an abstract one.
Why this matters downstream
Mountain water systems fed by snow and glaciers help support agriculture, hydropower, and drinking water for nearly 2 billion people.
Water managers need to know not just how fast ice is vanishing but how much remains upstream, because that reserve sets the timing of the eventual shortfall.
Earth.com has covered how shrinking ice reshapes freshwater supplies and coastlines at the same time.
The ice is going quickly. A recent international assessment found glaciers lost about 5 percent of their mass in two decades and up to 39 percent in some regions. Every one of the 19 major glacier regions registered a net loss in 2025.
Projection models have to start somewhere, and that starting point is a thickness map.
Researchers with the Glacier Model Intercomparison Project, whose simulations feed IPCC assessments of glacier change through 2100, plan to use IceBoost v2.0 as their single description of present-day ice.
Open data, open code
All 274,531 glacier files for the newest inventory version, plus regional mosaics, sit on Zenodo under an open license. The trained model and training data are also posted. An interactive web app lets anyone click a glacier and pull its thickness map.
Maffezzoli argues that the next step is a hybrid: physical models and learned patterns working together rather than competing. He also thinks the clock is short.
“Glaciers at mid-latitudes, including those in the Alps, are expected to disappear,” he said, referring to the coming decades. That warning arrives during a period when glacier preservation has moved up the international agenda.
For now, the world has a thickness map that is good enough to build projections on, along with an honest accounting of where that map goes soft.
At Sterling Publishing & Media Service Agency, we prioritise transparency and accountability in all our operations. We wish to clarify that we are not responsible for any external content, hyperlinks, or costs associated with our services. Nevertheless, we remain committed to delivering outstanding services and greatly value your continued support. Thank you for your trust in us.