A global machine learning system for glacier ice volumes

Knowledge of glacier ice volumes is crucial for constraining future sea level potential, evaluating freshwater resources, and assessing impacts on societies, from regional to global. Motivated by the disparity in existing ice volume estimates, we developed a global machine learning framework to model the ice thickness of individual glaciers. IceBoost is a gradient-boosted tree regression model, trained with 3.7 million global ice thickness measurements and an array of 34 numerical features. The model's error aligns within 10% of existing models outside polar regions, and is 30% to 40% lower at high latitudes. We find that providing supervision through available thickness measurements can further reduce the error of individual glaciers by up to a factor 2 to 3. A feature ranking analysis reveals that geodetic information is the most informative variable, while incorporating ice velocity improves model performance by 6% at high latitudes. A major feature of IceBoost is its ability to generalize globally, including in ice sheet peripheries. We present the model, discuss the advantages and shortcomings of a machine learning approach, estimate errors, and provide updated regional glacier volumes.